<%BANNER%>

Effect of HMO coverage on the choice of outpatient or inpatient surgery

University of Florida Institutional Repository

PAGE 1

EFFECT OF HMO COVERAGE ON THE CHOICE OF OUTPATIENT OR INPATIENT SURGERY By HSOU MEI HU A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORID A IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2003

PAGE 2

Copyright 2003 by HSOU MEI HU

PAGE 3

This work is dedicated to my family.

PAGE 4

iv ACKNOWLEDGMENTS I thank Dr. Niccie L. McKay, chair of my graduate advisory committee, for her support, guidance, and encouragement thr ough all the stages. Her straightforward teaching style has been extremely helpful in clarifying my research goals. I thank Dr. R. Paul Duncan for his support, counsel, and valuable discussion. I appreciate all the opportunities to work on his research projects that have enhanced my analytical skills. I also thank Dr. Tiffany A. Radcliff for shar ing her knowledge in econometrics, especially her frequent advice on my analyses. I would also like to thank Dr. Cynthia Garvan for her help on improving my skills in statistical analysis. I appreciate Dr. Alan Agresti in the Department of Statistics for his help during the early stage of my dissertation. I extend my thanks to all professors and sta ff from the Department of Health Services Administration for all the disc ussions, advice, and friendship. I am grateful for my parents’ supports that allowed their grown-up daughter to pursue her dream. I would also like to th ank my brothers and sisters in Taiwan, especially my sister, Shoushing, and my brothe r, Fonlin, who have assumed most of the responsibility of taking care of our parent s and our families while I am in the United States. Finally, I must thank my husband, Shiuhyang. During my doctoral study, he has shared so much responsibility with me in ta king care of our daughter, Rachel. His crucial support has made my study possible.

PAGE 5

v TABLE OF CONTENTS Page ACKNOWLEDGMENTS.................................................................................................iv TABLE OF CONTENTS.....................................................................................................v LIST OF TABLES...........................................................................................................viii LIST OF FIGURES.............................................................................................................x ABSTRACT....................................................................................................................... xi CHAPTER 1INTRODUCTION........................................................................................................1 2BACKGROUND..........................................................................................................5 The U.S. Health Insurance System...............................................................................5 Types of Private Health Plans.......................................................................................6 FFS Plans...............................................................................................................7 HMO Plans............................................................................................................8 PPOs and POS Plans...........................................................................................12 3LITERATURE REVIEW...........................................................................................14 Utilization...................................................................................................................1 4 Inpatient Care......................................................................................................14 Primary Care vs. Specialty Care..........................................................................15 Non-Physician Practitioner Care.........................................................................16 Preventive Care...................................................................................................16 Expensive Procedures..........................................................................................17 Summary: Utilization..........................................................................................18 Patient Outcomes and Quality of Care.......................................................................19 Mortality..............................................................................................................19 Survival Rate.......................................................................................................19 Physical and Mental Health.................................................................................20 Patient Satisfaction..............................................................................................21 The Shift from Inpatient to Outpatient Surgery..........................................................21 Utilization............................................................................................................22

PAGE 6

vi Outcomes.............................................................................................................22 Costs....................................................................................................................23 Research Questions.....................................................................................................24 4CONCEPTUAL FRAMEWORK...............................................................................25 A Two-Stage Model of P hysician Decision-Making..................................................25 Stage One: Determine a Set of Feasible Treatments...........................................25 Stage Two: Choose a Specific Treatment Method..............................................28 Choice of Treatment Methods.............................................................................29 Professional Uncertainty.....................................................................................29 The Conceptual Model I.............................................................................................30 Physician Characteristics.....................................................................................31 The Relationship between Physicia n and Payer Characteristics.........................34 The Final Model.........................................................................................................36 5EMPIRICAL SPECIFICATION................................................................................39 Dependent Variable....................................................................................................39 Independent Variables................................................................................................41 Primary Independent Variable.............................................................................41 Control Variables: Patient Characteristics...........................................................41 Control Variables: Physician Characteristics......................................................47 Control Variables: Other Payer Characteristics..................................................49 The Analytic Model....................................................................................................50 6DATA, VARIABLES AND STATISTICAL ANALYSIS........................................51 Data........................................................................................................................... ..51 Sample Design and Sample Weights...................................................................52 MEPS Data Collection........................................................................................53 Construction of Dataset for Analysis..................................................................55 A Subset of the Constructed Dataset...................................................................59 Variables.....................................................................................................................7 1 Dependent Variable.............................................................................................71 Primary Independent Variables...........................................................................71 Control Variables: Patient Characteristics...........................................................77 Control Variables: Physician Characteristics......................................................80 Control Variables: Other Payer Characteristics..................................................80 Other Control Variable........................................................................................80 Statistical Analysis......................................................................................................80 Sample Size.........................................................................................................82 Analytical Issues..................................................................................................83 7FINDINGS..................................................................................................................87 Contingency Tables....................................................................................................87

PAGE 7

vii Two-way Tables: Surgery Setting by Health Plan Type.....................................88 Three-way Tables for Control Variables: Surgery Setting by Health Plan Type90 Mean Charge and Payment by Surgery Setting and by Health Plan Type..........97 Logistic Regression Analysis...................................................................................100 Univariate Regression Analysis........................................................................100 Multivariate Regression Analysis......................................................................103 Summary............................................................................................................117 8DISCUSSION AND CONCLUSION......................................................................118 APPENDIX ATWO-DIGIT ICD-9 PROCEDURE CO DES REPORTING BOTH INPATIENT AND OUTPATIENT PROCEDURES IN 1996, AGE 0 TO 64..............................123 BSAS PROGRAM FOR CONSTRUCTING DATASETS........................................126 CREGRESSION MODELS FOR ALL SURG ICAL CASES: MAIN EFFECT AND INTERACTION.......................................................................................................138 DREGRESSION MODELS FOR THE SUBSET OF CASES: MAIN EFFECT AND INTERACTION.......................................................................................................145 BIBLIOGRAPHY............................................................................................................156 BIOGRAPHICAL SKETCH...........................................................................................168

PAGE 8

viii LIST OF TABLES Table page 2-1 Health Insurance Coverage for Americans under Age 65, 2000................................5 2-2 Employer-Sponsored Health Insuran ce Coverage by Plan Type, 1988-2000............8 5-1Independent Variables Affecting the Proba bility of Choosing a Surgery Setting...50 6-1Number of Cases by Type of Procedure..................................................................60 6-2Number of the Subset of Cases by Type of Procedure............................................62 6-3Independent Variables by Type of Charac teristic, Conceptual Basis, and Questions from the MEPS.........................................................................................................63 6-4Description of Variables...........................................................................................70 6-5Descriptive Statistics for Discrete Variables (Unweight ed n=814; Weighted n=9,595,657)............................................................................................................73 6-6Descriptive Statistics for Continuous Variables (Unweight ed n=814; Weighted n=9,595,657)............................................................................................................74 6-7Comparing Cases with a Non-HMO NonGa tekeeper Plan and Cases with No SelfReported Plan Information (Nom inal and Ordinal Variables).................................75 6-8Comparing Cases with a Non-HMO Non-Gate keeper Plan and Cases with No SelfReported Plan Information (Continuous Variables).................................................77 6-9Cases for the Subset of the Dataset by Health Plan Coverage.................................77 7-1Weighted Number (Percent) of Outpatie nt and Inpatient Surgical Cases by Health Plan Coverage (Based on Unweighted n=814)........................................................88 7-2Unweighted Number (Percent) of Outp atient and Inpatient Surgical Cases by Health Plan Coverage (Based on Unweighted n=814).............................................88 7-3Weighted Outpatient and Inpatient Surgi cal Cases of the Subset of Data by Health Plan Coverage (Based on Unweighted n=391)........................................................89

PAGE 9

ix 7-4Unweighted Outpatient and Inpatient Su rgical Cases of the Subset of Data by Health Plan Coverage...............................................................................................90 7-5Distribution of Selected Control Vari ables by Surgical Setting and Health Plan Type (n=814)............................................................................................................93 7-6Distribution of Selected Control Vari ables by Surgical Setting and Health Plan Type, the Subset of Cases (n=391)...........................................................................95 7-7Mean Charges and Payment to Surgeries by Health Plan Type, All Cases (n=814)99 7-8Mean Charges and Payment by HMO, the Subset of Cases (n=391)....................100 7-9Univariate Regression Anal ysis: Nominal Variables.............................................102 7-10Univariate Regression Analysis: Ordinal and Continuous Variables....................103 7-11Logistic Regression for A ll Cases: Main Effect Only............................................107 7-12Stage One: Logistic Regression Results that Predict HMO Status........................109 7-13Mean Predicted Probability of HMO Status by Observed HMO Membership......110 7-14Stage Two: Logistic Regression Results Using Predicted HMO Membership as the Primary Independent Variable...............................................................................110 7-15Logistic Regression for All Case s: Main Effect and Interaction...........................111 7-16Estimated Probability of Having an Outpatient Surgery by HMO Status..............113 7-17Logistic Regression for the Subset of Cases: Main Effect and Interaction............114 7-18Goodness of Fit of the Final Models......................................................................116

PAGE 10

x LIST OF FIGURES Figure page 2-1Employer-Sponsored Health Insuran ce Coverage by Plan Type, 1988-2000............6 2-2Percent of Americans Enrolled in HMOs, 1976-2000...............................................8 2-3Privately Insured HMO Enrollment by Model Type from 1976 to 2000.................10 4-1The Two-Stage Model of Physician Decision-Making............................................26 4-2Conceptual Model I of Choice between Inpatient and Outpatient Surgery.............31 4-3The Conceptual Model II of Choice betw een Inpatient and Outpatient Surgery.....37 6-1The MEPS Data Collection......................................................................................54 6-2Dataset Construction................................................................................................58 6-3Types of Health Plan Coverage for Cases with Two-plan Coverage.......................72 6-4The Possible Self-Selection Effect...........................................................................83 6-5The Effect of Unobserved Variable: Surgeon’s Specialty.......................................84 6-6The Effect of Unobserved Variab les: Payment Method, and Utilization Management.............................................................................................................84 7-1Correlation Matrix for Number of C ondition (COND), Health Status (HLTH), And Total Charge (TCH)...............................................................................................105 7-2Correlation Matrix of Charge and Payment Variables...........................................106

PAGE 11

xi Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy EFFECT OF HMO COVERAGE ON THE CHOICE OF OUTPATIENT OR INPATIENT SURGERY By Hsou Mei Hu August 2003 Chair: Niccie L. McKay Major Department: Health Services Administration This dissertation studies th e effect of health main tenance organization (HMO) coverage and gatekeeping on the choice of surgery setting. The study population is people under age 65 who require a surgery that is feasible in either the outpatient or inpatient setting. This dissertation cons tructs a dataset using the pooled 1997, 1998, and 1999 Medical Expenditure Panel Survey (ME PS). The constructe d dataset includes 814 cases; a subset of the data (391 cases) ex cludes surgeries that were primarily done in either the inpatient or outpatient set ting, with few done in the other setting. Because the dependent variable is dichot omous (outpatient or inpatient), logistic regression is specified to an alyze the relationship between the likelihood of choosing an outpatient surgery (vs. inpatient surgery) and the primary independent variables (HMO coverage and gatekeeper plan coverage), co ntrolling for severity, patient characteristics, and payer characteristics.

PAGE 12

xii This dissertation found that having HMO coverage did not increase the odds of having an outpatient surgery. Rather, the in teraction between HMO status and facility payment had a significant effect on the lik elihood of choosing an outpatient surgery. When facility payment increased, the likelihood of having an outpatient surgery for HMO patients dropped more than that for non-HMO patients. For example, when facility payment was increased by $400, reduced the pr obability of having an outpatient surgery for HMO patients decreased by 2%, bu t for non-HMO patients by only 0.6%. Gatekeeping did not significantly affect th e likelihood of having an outpatient surgery. For the subset of cases, HMO status did not show a stronger effect on the use of outpatient surgery than for all cases with surgeries in general. These conclusions appear to be inconsistent with the general belief that HMOs control costs by directly contro lling the use of care. Rath er, this dissertation found that HMOs paid less for a surgery than non-HMO s. However, when payment for outpatient surgery increased, HMOs were more aggressive in controlling the use of this type of care. These findings on HMO utilization patterns may help to identify strategies that promote the appropriate use of care and reduce healthcare costs.

PAGE 13

1 CHAPTER 1 INTRODUCTION This dissertation studies th e effect of HMO coverage on the choice of inpatient or outpatient surgery. During the past two decades, outpatient surgery has become a generally acceptable practice for many medical conditions, and is perceived to cost less than inpatient surgery. Because HMOs ar e thought to control costs by promoting enrollees’ wellness and preventing health pr oblems while reducing the use of expensive care, HMOs are expected to prefer outpatien t surgeries to inpatient surgeries given that outpatient surgeries cost less. This disserta tion compares utilization patterns of inpatient and outpatient surgeries according to type of health plan, thus helping to identify strategies that promote appropriate use of care and reduce healthcare costs. As national healthcare expe nditures continue to grow, controlling costs remains an important issue for the U.S. health car e system. In 1999, national health care expenditures totaled $1.2 trillion, a 5.6% in crease from 1998 (Kramarow et al. 2001). While the percent of gross domestic produc t (GDP) spent on heal thcare, 13%, remained the same as in 1998, the 1999 growth rate ex ceeded the 1998 rate of 4.8%. This growth has resulted in payers’ continued e fforts to control healthcare costs. Managed care has been one of the primar y methods used by payers to contain healthcare costs. Both priv ate payers and public payers offer managed care plans. In 1999, 6.1 million Medicare beneficiaries enrolled in the capitated program called Medicare+Choice, which was triple the enro llment in 1994 (Centers for Medicare and Medicaid Services 2001). As of June 1997, 19.5 million Medicaid recipients were

PAGE 14

2 enrolled in managed care plans, including 8.4 million in HMOs and the rest in primary care case management (PCCM) programs (Fox 2001). For people with employmentbased health insurance, enro llment in managed care plans increased from 27% in 1988 to 92% in 2000 (Kaiser Family Foundation and H ealth Research Educational Trust 2001). HMO enrollment rose from 16% of all Americans with employment-based health insurance in 1988 to 31% in 1996, and has rema ined stable since then (Kaiser Family Foundation and Health Research Educationa l Trust 2001). Nationally, HMO enrollment was 30% of the population in 2000 (Kramarow et al. 2001). The efforts to control healthcare expendi tures have focused on specific types of services. During the early 1980s 40% of national health expenditures went to hospital care (Kramarow et al. 2001). In 1983, Medi care introduced the Prospective Payment System (PPS) that paid hospitals a fixed fee based on the specified Diagnostic Related Group (DRG) regardless of the le vel of services provided. Inpatient surgery was one of the services immediately affected by PPS. B ecause the payment for an inpatient surgery was often too low to cover the costs, there were incentives to shift to an outpatient setting. The 1989 physician payment reforms introduced a schedule of prices for physician services based on Resource-Based Relative Va lue Scales (RBRVS) to facilitate control of physician expenditure growth. The fees for so me technically oriented services, such as certain surgeries, were considered excessive, while services like primary care were considered to be under-compensated. The reform was intended to ameliorate fee inequities. Consequently, the introduction of PPS and RBRVS had a major impact on the use of inpatient vs. outpatient surgery.

PAGE 15

3 Technological advances and patient preferen ces have also promoted the growth of outpatient surgery. The development of less invasive surgical techniques enabled some surgeries to be performed in ambulatory setti ngs. Minimally invasive procedures usually produce less postoperative pain, smaller scars, and faster recovery. The new anesthetic agents reduce postoperative nausea, headache s, and drowsiness, and orally administered analgesics represent another breakthrough that provides convenience and saves costs while retaining potency. Some surgical pro cedures, such as tubal ligation, laparoscopic cholecystectomy, and endoscopy, have become predominantly outpatient procedures. Patients favored outpatient procedures because of the convenience of recovering at their own home. Thus, while technology made the sh ift from inpatient to outpatient settings possible, patient preferences also have co ntributed to the incr easing use of outpatient surgeries. Previous research has focused largel y on how HMO and non-HMO health plans affect utilization, expenditures, and patien t outcomes. But few studies have examined how the type of health plan affects the choi ce of treatment method, out patient or inpatient surgery in particular. The purpose of this dissertation is to as sess the effect of the type of health plan on the choice of an outpatient or inpatient surgery, given a person who is diagnosed to undergo a surgery that is feasib le in either setting. The studied population is people under age 65 with private health insurance coverage. This dissertation uses the 1997, 1998, and 1999 MEPS data to test the hypoth esis that an HMO patient is more likely to receive an outpatient surgery th an a non-HMO patient, holding other factors constant. Conclusions from the dissertation will be useful to payers, consumers, and

PAGE 16

4 policy makers seeking to identify strategies to both improve appropriate use of care and control healthcare costs.

PAGE 17

5 CHAPTER 2 BACKGROUND Health insurance is the main source of payment for health care in the U.S. In 1999, 39.1% of personal health care expenditures, totaling $1.06 trillion, was from private health insurance or other private funding (Kramarow et al. 2001). Therefore, how insurers pay for care has a major effect on shaping the U.S. healthcare system. This chapter summarizes the development of the U.S. health insurance system. Table 2-1. Health Insurance Cove rage for Americans under Age 65, 2000 PayerTypes of CoveragePercenta Private 73.9% Employer-sponsored 68.2% Individual-purchased 5.7% Public Medicaid 10.4% Medicare 2.2% Militaryb 2.8% Uninsured 15.8% Source: Mills RJ, Health Insurance Coverage:2000 (2001) Current Population Report. Washington, D.C., U.S. Census Bureau.a This column represents the percentage of total popula tion under age 65 in 2000. The estimates by type of coverage are not mutually exclusive; people can be covered by more than one type of health insurance during the year.b Includes CHAMPUS (Comprehensive Health and Me dical Plan for Uniformed Services), Tricare, Veterans, and military health care.The U.S. Health Insurance System Most Americans have at least one type of health insurance (Table 2-1). In 2000, 73.9% of Americans under age 65 had private health insurance, with 68.2% obtaining their coverage from the workplace. About 15% of Americans under age 65 had coverage from public programs, including Medicaid, Medicare, and military health insurance, but

PAGE 18

6 73% 46% 27% 14% 9% 8% 16% 21% 31% 27% 28% 29% 11% 26% 28% 35% 38% 41% 14% 24% 25% 22% 7% 0% 20% 40% 60% 80% 100% 198819931996199819992000 Year POS PPO HMO FFS 15.8% were uninsured. Because most of Americans under age 65 are covered by private health insurance, this disse rtation will focus on this subpopulation, and will investigate how the type of plan for these individuals affects the choice be tween inpatient and outpatient surgery. Types of Private Health Plans Until the 1970s, traditional fee-for-service (FFS) plans were the usual type of health plan in the U.S. During the pa st two decades, however, FFS plans have lost market share, as managed care enrollm ent accelerated. Between 1988 and 2000, for example, the percent of employees in FFS plans declined from 73% to 8% (Figure 2-1).Source: The Kaiser Family Foundation and Health Research and Educational Trust. (2001). Employer Health Benefits, 2000 Annual Survey, Kaiser Family Foundation.Figure 2-1. Employer-Sponsored Health Insu rance Coverage by Plan Type, 1988-2000 Managed care represents a relatively new concept that integrates the insurance function with the delivery of care as a mean of controlling costs. While HMOs remain the prototype of managed care organizations (MCOs), managed care approaches are not limited to HMOs or even to MCOs. Rather managed care techniques have been adopted

PAGE 19

7 industrywide, with traditiona l FFS plans also using managed care techniques to control utilization. Preauthorizati on for hospitalization and a s econd opinion for surgery are examples of managed care utilization manageme nt used in FFS plans. Consequently, the definition of type of health plan has become blurred. Yet some fundamental characteristics can be used to diffe rentiate among types of health plans. FFS Plans Under a FFS plan, patients have free choi ce of physicians and access to whatever care they request that is cove red by the plan (Enthoven 1978). Under the original form of FFS, physicians have complete autonomy in clin ical decisions, and ther e is little internal or external assessment, such as ut ilization review or quality assurance. The FFS system produces incentives for over-utilization. Physicians are reimbursed per service provided; consequen tly, physicians have in centives to provide more services in order to increase their in come. Patients pay only a fraction of the costs of care while the third party payers (i.e., in surance companies) are responsible for the major portion of the costs of care. Therefore, the full price has little effect on patients’ demand for care. Because the FFS system does not manage utilization, patients tend to request more care than they w ould if they paid full costs, and physicians are likely to meet patients’ demands. To control health care costs, employers started to offer their employees managed care plans. As shown in Table 2-2, enrollme nt in various types of managed care plans has increased steadily during the past decad e. In 1988, 27% of insured workers were enrolled in managed care plans, but by 2000, enrollment in managed care plans had grown to 92%. Given that employer-sponsored health insurance covers the majority of

PAGE 20

8 2.8% 4.0% 8.9% 13.4% 19.4% 22.3% 25.2% 28.6% 30.1% 30.0% 0% 5% 10% 15% 20% 25% 30% 35% 1976198019851990199519961997199819992000 Year the privately insured population, this dramatic increase illustrates the impact of managed care on the U.S. healthcare system. Table 2-2. Employer-Sponsored Health Insu rance Coverage by Plan Type, 1988-2000 198819931996199819992000 FFS73%46%27%14%9%8% Managed Care Plans27%54%73%86%91%92% HMO16%21%31%27%28%29% PPO11%26%28%35%38%41% POS 7%14%24%25%22% Source: The Kaiser Family Foundation and Health Research and Educational Trust. Employer Health Benefits, 2000 Annual Survey.HMO Plans The original form of MCOs was the HMO. Figure 2-2 shows that HMO enrollment rose from 4% of the U.S. population in 1980 to 30% in 2000, totaling 81 million Americans (Kramarow et al. 2001). The num ber of HMO plans has also increased. In 1970, there were fewer than 30 HMO plans; by 1980, the number had grown to 235 plans, and by 1990, to 572 plans (Robinson a nd Steiner 1998; Kramarow et al. 2001).Source: Kramarow, E., Lentzner, H., Rooks, R. (2 001). Health, United States, 2001. Hyattsville, MD, National Center for Health Statistics.Figure 2-2. Percent of Ameri cans Enrolled in HMOs, 1976-2000 HMOs promote prevention and encourage ear ly identification of disease, while adopting approaches to control the use of care, such as limited choice of provider,

PAGE 21

9 financial incentives for providers, gatekeep ers, and utilization management. HMO enrollees see only HMO physicians or physicians contracted with th eir health plan, but bear less out-of-pocket costs. Enrollees in gatekeeping plans are required to go to primary care physicians for the initial contact of care, a nd must obtain referrals for specialty care. Providers are paid on a capit ation basis or discoun ted FFS, and are subject to utilization management. However, not a ll HMOs adopt the same set of approaches, and thus variations among HMOs are observed. Variations also come from the developmen t of different models of HMOs. These models differ in terms of the relations hip between HMOs and physicians. Some physicians are employed by HMOs or exclusiv ely take care of HMO enrollees, while others have contractual but not exclusive relationship with HMOs. However, because data on HMO models were not available, this dissertation focuses on differences between HMO and non-HMO plans. The following discu ssion of five HMO models is included to illustrate the variation among HMOs. A staff model HMO employs physicians w ho are recruited based on a selection process that meets organizational needs. HMO physicians are salaried, and thus do not bear financial risk. The HMO assumes full fi nancial risks, and usually uses extensive utilization management to ensure enrollees ’ access to necessary ca re without high costs. A group model HMO has an exclusive arrange ment with one or more large medical practice groups that treat enrollees of a sing le HMO. The HMO performs the insurance function, while the medical group provides clin ical care. Although th e two functions are separated, the HMO and medical groups work cl osely together, and t hus have a similar degree of control as in a staff model. Kais er Permanente HMO in California, consisting

PAGE 22

10 0% 20% 40% 60% 80% 100% YearPercent of Total Population IPA 6.6%18.7%30.4%41.6%39.4%44.1%39.9%42.6%40.3%41.3% Group 93.4%81.3%69.6%58.4%26.0%23.7%16.5%18.0%19.6%18.9% Mixed 34.5%32.2%43.4%39.2%40.1%39.9% 1976198019851990199519961997199819992000 of Kaiser Foundation Health Plan Inc. a nd Permanente Medical Group, is a typical example of a group model HMO. A network model HMO has an “arm’s length” relationship with medical groups. The HMO does not own, control or contract on an exclusive basis with medical groups, and medical groups treat patients from mu ltiple HMOs. Medical groups treat only HMO patients, not those covered by non-HMO plans. Compared with staff and group models, a network HMO has a relatively diffuse management structure.Source: Kramarow, E., Lentzner, H., Rooks, R. (2 001). Health, United States, 2001. Hyattsville, MD, National Center for Health Statistics. Note: Group model includes staff, group, and network model types. Figure 2-3. Privately Insured HMO Enro llment by Model Type from 1976 to 2000 In the early years of managed care, sta ff, group, and network models predominated (Figure 2-3). Over the year s, individual practice associat ions (IPAs) and mixed model HMOs have become the more common models. During the 1980s, the proportion of privately insured HMO enrollees in staff, group, or network models declined. By 1995,

PAGE 23

11 IPAs and mixed model HMOs had approxima tely 74% of all private HMO enrollees (Kramarow et al. 2001). An IPA is a hybrid between group model HMO and FFS plans. There are two types of IPA HMOs. One type is formed between an HMO and physician practices. The HMO contracts directly with solo or small group practices on a non-exclusive basis. The second type of IPA is formed by solo or small-group practices, and the IPA contracts with HMOs. Under the second type of IPA, HMOs pay IPAs on a capitation basis, but IPAs may pay physicians on a modified FFS ba sis. Therefore, physicians bear little financial risk but are subject to extensiv e utilization review. Unlike network model HMOs that provide care exclusively to HM O patients, physicians of IPA HMOs see both HMO and non-HMO patients. Due to the cont ractual relationship between the HMO and physicians, IPAs can usually be expanded wit hout large capital investment, unlike staff or group HMOs (Gabel 1997). Perhaps for this reason, enrollment in IPAs increased considerably between 1976 and 2000 (Figure 2-3). A mixed model HMO can be the combination of one of the four models and a contractual arrangement with different provi der organizations or networks. In some cases, a staff model HMO may c ontract exclusively with larg e group practices to form a mixed model HMO. In other cases, a network model HMO may contract, but nonexclusively, with solo or small group practi ces. Enrollment in mixed model HMOs has grown over the past few years. In 2000, a pproximately 40% of HMO enrollees were in mixed model HMOs (Figure 2-3). Regardless of the variation among HMOs, HMOs are considered to have tighter control over the use of care th an non-HMOs. Meanwhile, as newer forms of HMO plans

PAGE 24

12 evolve, physicians face different utilization ma nagement and financial incentives. During the 1970s, staff and group models were the dominant types of HMO. Physicians’ practices were subject to mainly internal ut ilization review, and th e costs of providing care were under a budget constraint. Newe r models of HMOs often contract with physicians in different locations and enroll more members. Physicians may be paid on a capitation basis, but many of them receive discounted fee-for-service payments that may reintroduce the incentive s of over-utilization. As physicians take on more financial responsibility, HMO enrollees may face restrictions in accessing medical care. Using a gatekeeper is one of the approaches that HMOs adopt to control utilization whil e non-HMO managed care plans also may use gatekeeper to manage patient care. Newe r forms of preferred provider organizations (PPOs) and point-of-service (POS) plans ha ve modified some of their managed care approaches and incorporated certain trad itional FFS plan charac teristics to allow enrollees access to provider s outside of the network. PPOs and POS Plans PPOs are the fastest growing type of ma naged care (Table 2-2). In 1988, 11% of Americans covered by employer-sponsored insu rance were enrolled in PPOs. One decade later, enrollment in PPOs rose to 35%. PPOs contract direc tly with a network of providers for services on a discounted fee. Enrollees are not restricted to in-network providers, but are required to pay a larger out-of-pocket payment when using out-ofnetwork providers. Because they are paid by discounted FFS, PPO providers do not usually bear financial risks. In 2000, 22% of Americans with empl oyer-sponsored health insurance were enrolled in POS plans (Table 2-2). POS pl ans are a hybrid of HMOs and indemnity plans

PAGE 25

13 that pay HMO providers on a capitation basis, but pay providers in the indemnity plans on a FFS basis. Enrollees decide whether to use HMO providers or out-of-network providers at the time of a medical event. If enrollees self-refer to a specialist or an outof-network provider, they are responsible for a larger share of the costs. However, a recent study found that most POS enrollees did no t exercise the POS option (Forrest et al. 2001). Because studies of newer forms of MCOs are limited, discussion of MCO performance in the following section will focus on HMOs. Findings related to specific managed care approaches, such as using a ga tekeeper, pre-authoriz ation, and financial incentives, are also summarized. Studies us ing data after 1980 are more relevant to the current situation, so the literatu re review in the next chapter on the effect of managed care includes only studies using data collected after 1980.

PAGE 26

14 CHAPTER 3 LITERATURE REVIEW This chapter reviews the literature on the performance of HMOs vs. non-HMOs during the past two decades. The focus is on the comparison between HMOs and nonHMOs because the newer forms of managed care (i.e., PPOs and POS plans) have much looser utilization management than HMOs, and thus the effect of managed care may be diluted in PPOs and POS. The following lit erature review will summarize the findings on utilization, patient outcome and quality, and on how type of health plan affects choice of treatment method, with particular atte ntion to the choice between inpatient and outpatient surgery for a given medical condition. Utilization Because HMOs are thought to control co sts through controlling utilization, many studies have examined utilization patterns between HMOs and non-HMO plans. Studies has assessed different types of utilization, including inpati ent care, specialty care, nonphysician practitioner care, prev entive care, and expensive care. This section summarizes the findings. Inpatient Care HMO enrollees use less inpatient care than non-HMO enrollees. The Medical Outcome Study examined utilization and outcom es of patients with chronic conditions, including hypertension, diabetes mellitus, re cent myocardial infarction, and congestive heart failure, between 1986 and 1990. The findings indicated that chr onically ill patients enrolled in prepaid plans, which consisted of group model HMOs and IPA HMOs, used

PAGE 27

15 41% less inpatient care than patients in FFS plans (Greenfi eld et al. 1992). Another study of Medicaid elderly beneficiar ies found that patients in cap itated plans were 11.2% less likely to have an inpatient vi sit than patients in traditiona l Medicaid (Lurie et al. 1994). Miller and Luft (1997) reviewed ar ticles published between 1993 and 1997, and concluded that HMO enrollees use less inpatient care. However, HMOs do not decrease the use of inpatient care across the board. The recent findings of the Medical Outcome Study showed that, between1986 and 1990, patients of IPA and group HMOs, who had better physical functioning, experienced a slightly higher rate of hospi talization than FFS patients. On the other hand, sicker HMO patients had lower hospitalization rates th an FFS patients with comparable clinical characteristics (Nelson et al. 1998). Based on data collected between 1987 and 1989, Pearson et al. (1994) found that lowto mid-risk patients of a staff-model HMO were more likely to be hospitalized than FFS patient s with a comparable level of risk. These findings suggest that HMOs trea t patients aggressively during the early stages of illness to prevent more expensive care later on. Primary Care vs. Specialty Care HMOs tend to use more primary care physic ian services while reducing specialist care. Specialty care is more expensive than primary care, usually between 1.5 to 2.0 times as costly as primary care (Kongstved t 2001). HMOs often use a gatekeeper to constrain enrollees’ access to specialists. Although having a gatekeeper may reduce the use of specialty care, it may not have a substantial effect on overall costs. Based on the 1996 MEPS, one study found that total per capita annual health expe nditures for children in gatekeeping plans were approximately 8 dol lars less than for those in indemnity plans (Pati et al. 2003).

PAGE 28

16 In California, to reduce the use of sp ecialty care, some medical practices encouraged primary care physicians to retain patients who otherwise required referral to specialists for simple procedures or ex aminations. These medical groups rewarded primary care physicians by offering a higher fee for services that are in the border between primary and specialty care, such as well-women examinations, suturing and wound treatment, and drainage of abscesses and cysts (Robinson 1999). Nine large national surveys, including th e RAND Health Insurance Experiment, the Medical Outcome Study, and the 1996-97 Community Tracking Survey reported that HMO enrollees used less specialty care (Hellinger 1998; Tu et al.1999). Not surprisingly, HMO enrollees are usually less sa tisfied with their acc ess to specialty care than non-HMO enrollees (Hellinger 1998; R obinson and Steiner 1998; Tu et al. 1999). Non-Physician Practitioner Care Studies also have found that HMOs enro llees used more non-physician practitioner care. Findings from the 1996-97 Commun ity Tracking Survey showed that HMO enrollees in general used more ambulator y visits, including nonphysician practitioner visits, than patients in non-HMO plans (Tu et al. 1999). Claims data from 1995 and 1996 showed that, in mental health care settings, managed care patients were more likely to be treated by non-physician providers, such as psychologists, psychiatric nurses, and psychiatric social workers (Sturm 1997). Preventive Care Previous studies found that HMO enro llees used significantly more cancer screening. A study based on the 1987 National Health Interview Survey showed that HMO enrollees received more cancer screening tests, including Pap smears, mammography, breast physical examinations, di gital rectal examinations, and blood stool

PAGE 29

17 tests (Bernstein et al. 1991) Another study of Medicare recipients between 1983 and 1986 found that HMO enrollees of staff/group HM Os and IPAs were more likely to have tonometry, mammography, pelvic examinations rectal examinati ons, and fecal occult blood tests than FFS enrollees (Retchin a nd Brown 1990). Similar findings were also reported among enrollees of a network HMO (Udvarhelyi et al. 1991). The 1996-97 Community Tracking Survey found that HM O enrollees were more likely to have mammography and flu shots than non-HMO enrolle es (Tu et al. 1999). Therefore, HMO enrollees tend to use more preventive care, cancer screenings in particular. However, a recent study using the 1996-97 Community Tracking Survey found no significant difference in the use of preventive care between HMO and non-HMO enrollees (Reschovsky et al. 2000). Us ing the 1987 National Medical Expenditure Survey (NMES) and the 1992 National Health Interview Survey (NHIS), another study compared the difference in the use of preventive care between HMO and non-HMO enrollees in 1987 and 1992, including blood pr essure checks, pap smears, breast examinations, and mammograms, among female non-elderly HMO enrollees (Weinick et al. 1998). Weinick found that, in 1987, HMO enrollees used more preventive care than non-HMO enrollees, but, by 1992, there was no significant difference in the use of preventive care. These findings suggest that HMOs may be losing their edge of providing more preventive care. Expensive Procedures HMOs enrollees also tend to use fewer costly tests and procedures than FFS enrollees, particularly when there are altern ative approaches available. HMO enrollees had a significantly lower rate of Caesaria n section than FFS enrollees, with the ratio ranging from 0.68 to 0.97 (Stafford 1990; McCloskey et al. 1992; Tussing and

PAGE 30

18 Wojtowycz 1994). Based on 45,425 births in California, one study further examined the likelihood of vaginal birth for women who ha d a previous Caesarian section, and found that in 1986 enrollees of group model HMOs we re 23% more likely to have vaginal birth than FFS enrollees, but there was no significant difference between IPA and FFS enrollees (Stafford 1991). Cardiovascular procedures are another example of different utilization patterns between HMO and non-HMO plans. Based on data from the 1994-95 National Registry of Myocardial Infarction, FFS patients und er age 65 were more likely to undergo angiography than HMO patients of the same age (Sada et al. 1998). HMO patients who enrolled in the National Registry of Myocar dial Infarction 2 (NRMI 2) also used less coronary arteriography, catheter-based reva scularization and coronary artery bypass surgery than FFS patients (Canto et al. 2000). The finding regarding utilization of angiography was also reported for Medi care patients. Data from the 1994-95 Cooperative Cardiovascular Project of th e Health Care Financing Administration (HCFA) showed that Medicare patients enro lled in HMO plans (Medicare+Choice) were less likely to undergo coronary angiography after acute myocardial infarction. The difference between patients in Medicare+ Choice and traditiona l FFS Medicare plans persisted even when patients were initially admitted to hospitals without an angiography facility (Guadagnoli et al. 2000). Summary: Utilization HMO enrollees use less specialty care, a nd less expensive tests and procedures. HMO enrollees use less inpatient care, but there are exceptions that appear to reflect HMOs’ strategies in treating mild condi tions more aggressively than FFS plans. Likewise, HMOs use more early detection and early treatment, although some studies

PAGE 31

19 showed that non-HMO plans might be catchi ng up with HMOs in the use of preventive care. Patient Outcomes and Quality of Care Given that HMO enrollees use less expe nsive care, a subs equent question is whether HMO enrollees also receive lower quality of care and have worse outcomes. Patient outcomes have been measured in term s of mortality rate, survival rate, health status, and physical health status. Quality of care is measured in various ways, including process of care, continuity of care, and patient satisfaction. Mortality Several studies have found that the outco me of HMO patients with cardiovascular conditions did not differ from that of FFS pa tients although the processes of care varied. HMO patients were more likely to be trea ted by cardiologists than FFS patients, but FFS enrollees were more likely to receive vasc ular catheterization than HMO enrollees. However, there was no difference in mortal ity rates due to cardi ovascular conditions. Based on the data from the National Regist ry of Myocardial Infraction between June 1994 and October 1995, the mortality of myocar dial infraction patie nts enrolling into HMOs was not significantly different from th e mortality of those in FFS plans (Sada et al. 1998). Similar findings were reported for patients admitted to hospitals participating in the Global Unstable Angina Registry and Treatment Evaluation Registry during 1995 and 1996 (Every et al. 1998). Survival Rate Survival rates for breast cancer and prostate cancer differ between HMO and FFS patients. Based on the population-based br east cancer registry of Orange County, California, between 1984 and 1990, the outcome of patients in HMO hospitals was worse

PAGE 32

20 than patients in community hospitals or te aching hospitals (Lee-Feldstein and Feldstein 1994). Another study of prostate cancer pati ents who were diagnosed between 1985 and 1992 showed that the ten-year survival rate was worse for patients in group staff HMO plans than patients in FFS plans (Potosky et al. 1999). Physical and Mental Health Longitudinal studies of the outcomes of chronically ill patients show no significant differences between FFS and HMO patients. The Medical Outcomes Study (MOS) evaluated the outcomes of patients with non-insulin-dependent diabetes mellitus, hypertension, recent acute myocardial infarcti on, congestive heart failure, or depression disorder. The study sampled patients who made office visits to physicians of family medicine, internal medicine, cardiology, or endocrinology in three major U.S. cities during the period of time that interviews were conducted. In a four-year follow-up between 1986 and 1990, and in a ten-year fo llow-up evaluation, there was no difference in physical and mental health between HM O and FFS patients (Greenfield et al. 1995). Another study of rheumatoid arthritis patients also reported no difference in physical health status between HMO and FFS patients. The Rheumatoid Arthritis Study was conducted between 1982 and 1994 in norther n California, and en rolled participants in 1982-83 and in 1989 (Yelin et al. 1996) The study assessed the outcome of rheumatoid arthritis patients for up to 11 years, and found no si gnificant difference in physical health and several physical function measures between patients of prepaid group practices and FFS plans. However, based on the Medical Outcome Study, the elderly in Medicare HMOs and the poor chronically il l HMO enrollees experienced worse physical health than those in FFS plans (Ware et al. 1996).

PAGE 33

21 Patient Satisfaction Patient satisfaction of care also diffe rs between HMO and FFS enrollees. Some studies have found that HMO enrollees ar e less satisfied with their care than FFS enrollees (Brown et al. 1993; Adler 1995; Da vis et al. 1995). However, another study, a 1993 survey of three large employers, f ound that HMO enrollees of prepaid group practices and IPAs were more satisfied with their plans than FFS en rollees (Allen et al. 1994). Safran examined HMO enrollee’s satisfaction between 1986 and 1990. He found that, compared to FFS enrollees, HMO enrollees of IPA and group-model HMOs were satisfied with financial access and coordina tion of care, but were not satisfied with number of physician visits, and physicians’ in terpersonal and techni cal skills (Safran et al. 1994). The same findings were also repor ted for Medicare patients (Brown et al. 1993; Adler 1995). In general, it appears that HMO enrollees are less satisfied with their access to care and more satisfied with the financial aspects of the care. The Shift from Inpatient to Outpatient Surgery As outpatient surgery has become genera lly accepted, more surgical procedures have been performed in outpatient settings. In 1983, 24% of surgeries were performed in the outpatient department of community hos pitals (American Hospital Association 1987). By 1996, more than half of the surgical pro cedures in the U.S. were on an outpatient basis (Detmer and Gelijns 1994; Owings and Ko zak 1998). Medicare has attributed this trend to the introduction of the Prospective Payment System (PPS) and Resource-based Relative Values (RBRVs). Under the new pa yment systems, some inpatient surgeries were paid at a lower rate that provided an incentive to shift inpatient procedures to outpatient settings. Private payers took up this trend and soon after several studies reported that surgical outcomes were similar in both settings while the charges incurred

PAGE 34

22 were lower in outpatient settings (Davis and Detmer 1972; Stephenson 1985). During the 1970s, only 35% of payers covered ambulat ory surgeries; by the 1980s, 96% included ambulatory surgeries in their coverage (Detmer and Buchanan-Davidson 1989). Therefore, the payment system played an important role in promoting ambulatory surgery. Utilization Two studies reported that HMO enrollees we re more likely to receive outpatient, rather than inpatient surgery. Betw een 1983 and 1993, the growth of hospital expenditures in California was less rapid in areas with high HMO penetration than in areas with low HMO penetration; one of the factors accounting for these differences was that HMOs substituted outpatient for in patient surgery, including hysterectomies, coronary artery bypass grafting (CABG), chol ecystectomies, and inguinal hernia repair (Robinson 1996; Trauner and Chesnutt 1996). A recent study based on seven years ( 1990-1996) of data from the Healthcare Cost and Utilization Pr oject found that the likelihood of having an outpatient mast ectomy was higher for HMO breast cancer patients than for non-HMO plan patients (Case et al. 2001). However, more studies are still needed to provide evidence regardi ng the effect of HMO coverage on the choice between inpatient and outpatient surgery. Outcomes Many studies have compared the outcomes of specific surgical procedures that were done in either inpatient or outpatient settings, but none has further compared the outcomes between HMO and non-HMO patients. Among children with comparable health status and family support who underw ent a tonsillectomy, those with outpatient procedures had no higher risk of postopera tive bleeding than children with inpatient

PAGE 35

23 procedures (Guida and Mattuc ci 1990; Reiner et al. 1990; Truy et al. 1994; Rakover et al. 1997). A randomized clinical trial was undertaken between 1993 and 1996 to study the outcome of cataract surgery (Castells et al. 2001). All participating patients with health status suitable for outpatie nt cataract surgery were ra ndomly assigned to either an outpatient or inpatient setting. Although pa tients with ambulatory surgery experienced higher complication rates within 24 hours af ter the surgery, the study found no difference in visual acuity and postoperative complicati ons four months after the surgery. Other studies compared the complication rates betw een outpatient and inpa tient catheterization, but, probably because the number of cases was too small, did not find a statistically significant difference (Block et al. 1988; Skinner and Adams 1996). Patient satisfaction has been found to be higher for patients and their families with ambulatory procedures than those with inpa tient procedures. For those who underwent ambulatory cardiac catheterization, both patients and their families were more satisfied with the process and recovery (Kern et al 1990; Lee et al. 1990; Seckler and Held 1990). Patients having ambulatory cath eterization were more satisfied with the convenience, continuity, and technical aspect of the ca re, as well as interpersonal communication. Costs Studies have reported that outpatient surg ery costs less than inpatient surgery. In a randomized clinical trial, an outpatient cataract surgery cost $1, 001 while an inpatient surgery costs $1,218 (Castells et al. 2001). On the other hand, if patients stayed in the hospital overnight only, outpatient cardiac catheteri zation costs were similar to the costs of an inpatient procedure. Once catheterized patients stayed more than one night in the hospital, the costs of an inpatient procedur e were higher than that of an outpatient

PAGE 36

24 procedure (Lee et al. 1990). Based on these fi ndings, inpatient surg eries generally cost more than outpatient surgeries. Research Questions Given that HMOs are thought to control ut ilization in order to reduce costs, HMOs would be expected to prefer out patient to inpatient surgeries. This dissertation will assess the effect of the type of private health insurance plan, HMO or non-HMO, on the choice between an outpatient and an inpatient surg ery. Controlling for patients’ medical condition, are HMO patients more likely to receive outpatient surgery than non-HMO patients? Specifically, for a patient under age 65 who is diagnosed to undergo a surgical procedure that is feasible in both outpatient and inpatien t settings, does an HMO patient have a higher likelihood of receiving an ou tpatient surgery than a non-HMO patient? When excluding surgeries that are done primar ily in one setting with few in the other setting, does HMO status have a greater imp act on these surgeries than on all surgical procedures generally feasible in either inpatient or outpatient setting? Does a specific managed care approach, having a gatekeeper, affect the choice of a surgery setting?

PAGE 37

25 CHAPTER 4 CONCEPTUAL FRAMEWORK This chapter develops a model to pr ovide the foundation for the empirical specification and subsequent data analysis. While this dissertation assesses the effect of HMO coverage on the choice of outpatient or inpatient surgery, physicians make the final decision on the choice of a surgery setting; consequently, this ch apter begins with a discussion of the process of physician decisi on-making. This chapter first presents a general model of the process by which physicia ns make clinical decisions, then focuses on factors that influence the choice among treatments for a given diagnosis (i.e., outpatient or inpatient surgery), which fa ll into three major categories: physician characteristics, patient characteristics, and payer characteristics. The chapter then concludes with a summary of the model th at will be used as the basis for the identification of variables and empirical analysis. A Two-Stage Model of Physician Decision-Making A general model of the process of physic ian decision-making must distinguish between two stages of treatment choice : diagnosis based on presented symptoms, followed by choice of treatment method. Wolff (1989) developed a model that disentangles the relationship be tween the first stage of dia gnosis based on the presented symptoms and the subsequent stage of c hoosing suitable treatment (Figure 4-1). Stage One: Determine a Set of Feasible Treatments In the first stage, patients initiate the contact with physicians for their medical

PAGE 38

26Source: Wolff, N., Professional uncertainty and physician medi cal decision-making in a multiple treatment framework, 1989, Soc. Sci. Med., 28 (2):99-107Figure 4-1: The Two-Stage Mode l of Physician Decision-Making Medically available care Regional availability and accessibility factors (C, L, A) Feasible treatment set (R1, R2, ---Rn) Health outcome assessment Patient medical needs (SYM) Physician knowledge (K) Illness diagnosis (DIA) Health outcome estimates (HSi) Treatment selection (Ri) Objective function Preferences Physician Patient Constraints Financial incentive/ disincentive Delivery organization factor Stage 1: Medically technical decision-making Stage 2: Treatment selection decision-making Determine a set of feasible treatments Choose a specific treatment method

PAGE 39

27 needs. Based on the presented symptoms (SYM), physicians use their medical knowledge and judgment (K) to diagnose (DIA) the condition: DIA=K(SYM). Given a diagnosis, physicians evaluate which treatment methods are medically appropriate, and which are available and a ccessible in the community, based on factors such as capital (C), labor (L), and access constraint (A). This evaluation yields a menu of feasible treatments (R1, R2, ---Rn): (R1, R2, ---Rn)= f(C,L,A|DIA). Physicians then assess the efficacy of each available treatment (Ri), by means of a health status (HSi) production function: HSi=K(Ri|DIA), where i=1,2,---n. Under the assumption that physicians are homogeneous producers of medical decisions, the resulting menu of treatment optio ns and the assessment of each treatment's efficacy only differ according to the availabili ty and accessibility of resources in the community. However, the first stage of decision-maki ng is subject to prof essional uncertainty and error that may violate this assumption. Decision-making at this stage relies heavily on technical factors. Many medical condi tions are not well defined, and cannot be diagnosed or treated unambiguously. Thus, physicians’ decisions can be heterogeneous, based on their educational background, clinical training, expertise, and local acceptable standard of practices. The menu of feasible treatment options (R1, R2, ---Rn) and the assessment of each option’s efficacy (HSi) are potentially different among physicians.

PAGE 40

28 Stage Two: Choose a Specific Treatment Method In Stage Two of the decision-making pr ocess, one treatment method is selected from the menu of options (R1, R2,---,Rn). Now the agency re lationship between patient and physician becomes key. Because patient s typically delegate a major portion of medical decision-making to their physicians, pa tient and physician have a principal-agent relationship. If acting as a perfect agent, the physician makes decisions based solely on what is in the best interest of the patient. The physician knows the patient’s biomedical and psychosocial characteris tics. The physician is aw are of the patient’s budget constraint and preferences regarding health and other goods and services. And the physician has better information about treatm ent costs, resource availability and accessibility. Under these conditions, a physic ian acting as a perfect agent seeks to maximize the patient’s utility function (U), which is based on the patient’s preferences regarding health outcomes (HS) a nd all other goods and services (Z): Choose Rk from (R1, R2,---,Rn) to maximize the utility function Max U=U(HS,Z). In the case of perfect agency, the physic ian determines the treatment choice that would have been the patient’s choice if th e patient had the physician’s knowledge and training. However, during the decision-making proce ss, physicians face financial incentives and institutional factors associ ated with the type of deliv ery organization. Physicians may want to maximize their income, or ma y be subject to certain organizational constraints governing choice of treatment. On the other hand, patients who have acquired medical knowledge formally or informally may have a major influence on the physician’s decision-making. When patient and physician preferences diverge, the final outcome

PAGE 41

29 depends on the extent of asymmetric info rmation and the bargaining power between patient and physician. While the physician is the executor of the fi nal decision, patients, physicians, and payers all play role s in the final choice of treatment. Choice of Treatment Methods For purposes of this dissertation, the fo cus is on the second stage of treatment choice. That is, the diagnosis is taken as given, and the question then becomes what factors influence the choice of treatment method for a partic ular diagnosis. For example, given the diagnosis of breast cancer, the choi ces of treatment are mastectomy and breastconserving surgery. There is a trade-off between benefits and risks in the choice of treatment. Some women prefer breastconserving surgery followed by a course of radiation therapy because of co smetic factors, and are willi ng to bear the increased risks of recurrence. Others would rather undergo a mastectomy to reduce the risk of recurrence, but need to overcome the ps ychological effect due to the change in appearance (Potosky et al. 1997; Hadley a nd Mitchell 1997/98). Thus, a patient’s preference may contribute to the physician ’s decision-making, as well as physician factors, such as the fina ncial return associated with a particular treatment. Professional Uncertainty A key factor influencing choice of trea tment method is professional uncertainty. For some conditions, sufficient scientific evidence exists to identify one appropriate and preferred treatment. For example, there is li ttle uncertainty in ap pendicitis diagnosis and treatment. Once appendicitis is diagnosed, an appendectomy is the generally accepted treatment. Similarly, for i nguinal hernia, repair procedur es are the generally accepted treatment. In contrast, for diagnoses such as low back pain, breast cancer, and hip replacement, there is little consensus among phys icians as to the best choice of treatment,

PAGE 42

30 and physicians tend to have different opini ons on whether a certain procedure is more valuable than others (Wennberg et al. 1982 ; Cherkin et al. 1994; Birkmeyer et al. 1998). Consequently, we observe variati ons in the choice of treatment. Professional uncertainty t hus affects the choice of treatment method in the following ways. If there is a generally accep ted treatment for a particular diagnosis, the treatment method is given and there is no choi ce per se (i.e., at the end of Stage One in the Wolff model, there is only one feasible treatment method). This dissertation focuses on cases in which professional uncertainty le ads to a set of possible treatment methods for a particular diagnosis. More specificall y, the focus is on diagnoses for which either inpatient or outpatient surgery may be c hosen. For example, Case (2001), using 1996 data from the Healthcare Cost and Utiliz ation Project (HCUP), found that 6.8% of all breast cancer patients who underwent a ma stectomy had an ambulatory surgical procedure while 93.2% had inpatient surgery. When professional uncertainty exists, factor s such as patient, physician, and payer characteristics can play an important ro le on the choice of treatment. Payer characteristics, such as reimbursement method, can affect physician decision-making. The study by Case (2001) also found that HMO breast cancer patients were more likely to receive an outpatient mastectomy than patie nts with other types of health plan after controlling for clinical characteristics. Therefor e, the next step is to develop a model that incorporates factors affecti ng the choice between inpatien t and outpatient surgery for a given diagnosis. The Conceptual Model I Figure 4-2 presents a genera l model of the choice between inpatient and outpatient surgery (Model I). For individuals facing su ch a choice, this dissertation assesses the

PAGE 43

31 effect of the type of private health plan on the likelihood of receiving outpatient surgery when controlling for patient characteristic s, physician characteristics, and other payer characteristics. As discussed earlier, patient preferences can affect the choice between inpatient and outpatient surgery (Elit et al. 1996). The e ffect of patient characteristics on morbidity and mortality has been extensively studied. These characteristics also may have an impact on the communication process between physician and patient, on how patients present their symptoms, and on treatment pref erences. Key patient characteristics include health status, race/ethnicity, age, gende r, income, education, and family support. Figure 4-2. Conceptual Model I of Choice between Inpatien t and Outpatient Surgery Physician Characteristics When there is professional uncertainty regarding the prefer red treatment method, physician characteristics can play a major role. Key physician ch aracteristics include practice style, specialty, years of practice, and practice environment. The Choice of Outpatient or Inpatient Surgery Payer Characteristics Patient’s Characteristics Treatment Option: Inpatient or Outpatient Surgery Physician Characteristics Dia g nosis Re q uirin g Sur g er y

PAGE 44

32 Principal-agent relationship A principal-agent relationship occurs when ever an individual delegates decisionmaking authority to another individual (Lev inthal 1988; Arrow 1991). In the healthcare sector, patients often rely on physicians’ superior medical knowledge in the choice of medical treatment. Thus, physicians (the ag ents) are making decisions for patients (the principals) (Dranove and White 1987). Information asymmetries exist in any pr incipal-agent relationship. For example, physicians have superior knowledge about the treatment choices and the expected efficacy in improving health status, while onl y patients can judge how changes in their health status will affect thei r well being. Uncertainty ar ises because patients have no perfect and costless way to monitor physicia ns’ information and actions. Therefore, patients rely on other approaches, such as a long-standing principa l-agent relationship, to motivate physicians to act in thei r best interests. If physicians act as perfect agents, they will make the treatment choices that the patients would have chosen if they had the full knowledge and information that their physicians have. Imperfect agents The principal-agent relationship raises the po ssibility that agents will put their selfinterest above that of the principals. In particular, th e concern is that physicians could use their position of superior medical knowledge to recommend decisions that benefit the physician financially with little or no accompanying health benefit to the patient (Wolff 1989). Because patients generally have little ability to assess the quality of physicians’ decisions, approaches such as a second opini on can induce physicians to become a better agent (Rochaix 1989).

PAGE 45

33 Certain factors encourage physic ians to make treatment choi ces in the best interests of their patients. For example, the re lationship between patie nts and primary care physicians plays an important role (Dranove and White 1987). When a long-term relationship exists, patients expect their primar y care physicians, as sophisticated medical consumers themselves, to refer them to a qua lified specialist. By doing so, primary care physicians may expect a reward in the form of a better reputation, and attract more new patients to increase their income. Besides monetary rewards, non-monetary f actors such as professional, legal, or ethical standards can also compel physicians to make decisions that benefit patients rather than themselves. During their medical educ ation, physicians are so cialized to become professionals who make decisions based on obj ective science that focuses on the relevant aspects of the patient’s circumstances (Arr ow 1991; Clark et al. 1991). The threats of malpractice litigation and license withdrawal may also prevent doctors from treating patients inappropria tely (Pontes 1995). Physician practice patterns Due to professional uncerta inty and the principal-ag ent relationship, physician practice patterns may play a ro le in the choice of treatment method. For some conditions, such as breast cancer, there is little c onsensus in the medical community regarding a preferred choice of treatment method, so phys icians typically make the final decision. According to Wennberg (1982), it is the exerci se of clinical judgment under conditions of uncertainty that produces different practice patterns among physicians. Practice patterns are shaped by prior medi cal training, clinical experience and expertise, as well as personality characte ristics, value judgments, and professional or patient convenience (Burns et al. 1995; Rutchi k et al. 2001). Physician specialty, years of

PAGE 46

34 practice, and practice environment (e.g., rura l or urban area, geographic region) also influence practice patterns. Practice styles can change if physicians take actions to modify their clinical decisions. Some managed care approaches, such as physician profiling followed by feedback on utilization ra tes, are especially designed to modify physician practice patterns. The resulting chan ges can lead to the development of local standards of acceptable practice (Wennberg 1984). The Relationship between Physician and Payer Characteristics The principal-agent relationship may also l ead to medical decisions that adversely affect payers (i.e., health plans). Even if phys icians act as perfect agents for patients, the decisions made by physicians may not be what payers prefer. Th erefore, the challenge for payers is to draw up a contract ensu ring that physicians act in the payer’s best interests. Like patients, payers also encounter th e problem of monitoring physician actions. Information such as blood pressure, wei ght, and immunization compliance is readily available from medical records, but the ap propriateness of diagnosis and treatment is difficult to obtain, especially for condi tions that lack consensus among physicians (Randall 1993). Payers would like to contro l costs by providing only the appropriate use of care, but usually do not have an indicator to judge that appropri ateness. Thus, payers often use other approaches, such as a gatekeeper, preauthori zation, and utilization review, to ensure that their best interests are served. Financial incentives Payers often use financial incentives to induce desired beha viors. Financial incentives consist of methods of payment, physician payment amounts, withholds, and bonuses. Different payment methods encour age different practice styles. Under

PAGE 47

35 capitation, physicians are paid prospectively for services provided to a defined population, regardless of the level of care us ed. Thus, capitation encourages a costconscious practice style, but may result in unde rtreatment, especially for sicker patients. On the other hand, FFS rewards physicians w ho are willing to treat sicker patients because they are paid for services they pr ovide, but may induce physicians to overtreat patients to increase their own income. Thus, neither FFS nor capitation payment is without shortcomings. Innovative payment methods that incor porate both capitation and FFS attempt to correct financial incentives, but have been found to be costly to develop and administer. For example, medical groups in California have blended capitation and FFS to induce desired physician behavior, but the ble nded payment methods require extensive administration in development, execution, and negotiation (Robinson 1999). Thus, FFS and capitation remain the predominant fo rms of payment for physician services. While there is limited evidence of the e ffectiveness of capita tion, the percent of physicians paid by capitation has either remain ed about the same or increased slightly. One study indicated that the pe rcent of physicians in office practice receiving capitation rose from 40% in 1996 to 44% in 1998 (Terry 1999). However, an AMA study found that the percent of physician revenues derive d from capitation remained constant at 24% between 1996 and 1998 (Kane 1999). Utilization management Managed care plans usually have a system of utilization management in place. Utilization management comprises a range of techniques, such as utilization review, gatekeeping, and preauthorization. Depe nding on the local practice environment, managed care plans in different geographic re gions adopt different approaches to manage

PAGE 48

36 utilization (Kongstvedt 2001). While there is no standard set of techniques, managed care plans that emphasize utilization manage ment generally have lower use and lower cost of care than others. A 1999 survey of eight major managed care organizations showed that those investing more in utili zation management had lower utilization and medical costs (Kongstvedt 2000). Summary: payer characteristics The payer characteristic of primary intere st is enrollment in an HMO or non-HMO plans. Other payer characteristics that ma y influence the choice between inpatient and outpatient surgery are financial incentiv es, including method of physician payment (capitation vs. FFS) and payment rates (total payment, physician payment, and patient’s share of payment). Finally, the use of util ization management is a payer characteristic expected to affect the choice betw een inpatient and outpatient surgery. The Final Model The research question focuses on how a sp ecific payer characteristic, namely HMO coverage, affects the choice of inpatient or outpatient surgery while controlling for patient, physician, and other payer characteristi cs. This dissertation also studies the effect of a specific managed care approach, ga tekeeping, on the choice of surgery setting. Based on the physician, patient, and payer char acteristics identified in the above section, Figure 4-3 presents Model II of the choi ce between inpatient and outpatient surgery. Key patient characteristics include heal th status, race/ethnicity, age, gender, income, education, and family supports. Several indicators measure patient’s health status, including self-reported health status, number of c onditions associated with a surgical event, and total ch arge. Important physician char acteristics incl ude specialty, years of practice, and practice environmen t (e.g., urban or rural area, and geographic

PAGE 49

37 region). Finally, other payer characteristic s, such as physician payment method, payment rates, and how utilization is managed, also influence the choice between outpatient and inpatient surgery. Figure 4-3. The Conceptual Model II of Choice between Inpatient and Outpatient Surgery Model II represents a simplified relationship among the three groups of characteristics, while inter action between groups of charact eristics may exist. For instance, physician’s practice environment (urb an or rural area, and geographic region) is included in the final model under physician characte ristics, but this char acteristic also can affect payment and charge levels, and some patient characteristics (such as income, race Choice of Inpatient or Outpatient Surgery A Diagnosis Requiring a Surgery Patient Characteristics -Health Status -Number of Condition -Total Charge -Race/ethnicity -Age -Gender -Income -Education -Family support Treatment Options: In p atient or Out p atient Sur g er y Physician Characteristics -Specialty -Years of Practice -Practice Environment (Area Characteristics) Urban or Rural Area Geographic Region Payer Characteristics -HMO vs. Non-HMO Plans -Having a Gatekeeper -Financial incentive Physician Payment Method Rate Setting Total Payment Physician Payment Patient’s Share of Payment

PAGE 50

38 and ethnicity). Therefore, area character istics may a better way to describe these environment characteristics.

PAGE 51

39 CHAPTER 5 EMPIRICAL SPECIFICATION This chapter presents the empirical specif ication for the subse quent data analysis. The specific research questions are to assess the effect of HMO enrollment status, and the effect of gatekeeper plan enrollment on th e choice of outpatient or inpatient surgery, given that a patient is diagnosed to have a surgery that is feasib le in either setting. Chapter 4 presented a conceptual model of treatment choice in which patient, physician, and payer characteristics jointly determin e the choice between inpatient and outpatient surgery for a given diagnosis (see Figure 4-3). Based on the conceptual model, the most general form of the empirical specification is: (Choice of Outpatient or Inpatient Surgery | Diagnosis) = f (HMO/non-HMO; patient characterist ics; physician characteristics; other payer characteristics). Dependent Variable The dependent variable is the choice of an outpatient (vs. inpatient) surgery when the choice set for a given diagnosis contains both inpatient and outpatient surgery. An outpatient surgery is a medical ev ent that occurs in an outpat ient setting, and patients are discharged in the same day. On the othe r hand, an inpatient surgery is done in an inpatient setting. However, some patients iden tified as having an i npatient procedure are discharged from a hospital without an overn ight stay. For example, the 1996 National Hospital Discharge Survey estimated that 383,000 hospital discharges, approximately 2% of all inpatient discharges w ith procedures, were reported w ith zero night’s hospital stay

PAGE 52

40 (Owings 1998). In this disserta tion, an inpatient surgical case is defined as an inpatient event with at least one overnight stay. Otherwise, the surgical case is considered outpatient. Surgical procedures for which there is littl e variation in the choice of inpatient vs. outpatient setting will be excluded from the analysis. Some surgeries are done exclusively in inpatient sett ings due to the intensity a nd required equipment of the surgery or post-surgical care. Surgeries requiring intense monitoring and post-surgical care, such as a coronary artery bypass graft, a partial excision of the large in testine, or a colostomy, are usually done in inpatient setti ngs. Other surgical procedures, such as a vasectomy, ligation or stripping of varico se veins, are typically performed on an outpatient basis (Owings and Kozak 1998). These types of surgeries, in which there is no variation in setting, are ex cluded from the analysis. This dissertation thus focu ses on surgical procedures that are performed in both inpatient and outpatient settings. In 1996, 83 out of 99 two-digit ICD-9 procedure codes were reported as both inpatient and outpatie nt procedures (see Appendix A). The 1996 National Health Care Survey (NHCS) estimated that 71.9 million procedures were performed in the United States, with 44% be ing done in outpatient settings (Owings and Kozak 1998). Based on the 1996 NHCS, surgic al cases with one of these 83 ICD-9 procedure codes are included in the analysis. This dissertation further studies the effect of HMO coverage on a subset of the 83 two-digit ICD-9 procedure codes, after excl uding surgical procedures done mostly in one of the settings (outpatient or inpatient) w ith only a few done in the other setting. This subset of surgery cases includes only surger ies that had a reported ratio of inpatient to

PAGE 53

41 outpatient cases (I/O ratio) between 0.2 and 5 in the 1996 National Health Care Surveys (see Appendix A). Independent Variables Primary Independent Variable The primary independent variable is he alth plan type (PLAN), which has four possible values: one-HMO coverage, one nonHMO gatekeeper plan coverage, one nonHMO non-gatekeeper plan coverage, and twoplan coverage. Given a diagnosis for which both inpatient and outpatient surgery are feasible, it is hypothesized that HMO patients are more likely to have outpatient surgery, all else equal. Because HMOs are designed to control costs, HMOs are more likel y to manage the use of costly care, such as inpatient care. Therefore, enrollment in an HMO plan is expected to increase the likelihood of a patient receiv ing an outpatient surgery. This dissertation does not attempt to expl ore the effect of managed care in general on the choice of treatment because managed car e plans also include alternatives such as preferred provider organization (PPO) plans. Although PPOs typically manage utilization, they provid e enrollees freedom in choice of pr ovider and are considered to be a less restrictive form of managed care. PPO plans are considered non-HMO plans in this analysis. However, this dissertation doe s examine the effect of one managed care approach, having a gatekeeper, on the choice of outpatient surgery. Control Variables: Patient Characteristics Three patient characteristics (health stat us, number of conditions, and total charge) are used to control the severi ty of a surgery case. Other patient characteristics include race/ethnicity, age, gender, income, education, and family support. Patient characteristics control for differences in mortality and morbidity, as well as differences in the

PAGE 54

42 communication process between physicians and patients, presentation of symptoms, and treatment preferences (Mort et al. 1994). Health status A key patient determinant of the choice of treatment is health status. Substantial evidence indicates that self-rated health is a significant predictor of mortality and morbidity. A 1971 study of 3,128 randomly se lected non-institutional elderly in Manitoba, Canada, found that se lf-rated health was significant in pr edicting mortality within three years, independent of clinically assessed health status (Mossey and Shapiro 1982). For persons with “poor” self-rated health status, the risk of mortality within three years (1971-1973) was 2.92 times the risk of mo rtality for persons with “excellent” selfrated health status. Similarly, a study of 2,682 males in Finland, aged 42 to 60, reported that the level of self-rated health was signif icantly associated with all-cause death, death from cardiovascular disease, the incidence of myocardial infarction, and the extent of carotid atherosclerosis (Kaplan et al. 1996). While most previous studies have used self-rated health status as a baseline predictor to examine the a ssociation with mortality, a recent study found that repeated observations of self-rated health status over time detect dec lines of physical health status. Based on 20 years of data from the National Health and Nutrition Examination Survey-I Epidemiological Follow-up Study, the study argue d that self-rated health status reflects not only existing illness but also undiagnose d but preclinical condi tions (Ferraro and Kelley-Moore 2001). Therefore, self-rated heal th status with five levels (poor, fair, good, very good, and excellent) will be used to cont rol for the severity of a surgical case.

PAGE 55

43 Number of conditions Number of conditions is associated with th e severity of a pati ent’s condition at the time of surgery. Patients who have complex c linical conditions usually require a surgery involving a hospital stay (Pepine 1991; Skinner and Adams 1996). Similarly, a preexisting condition is important in dete rmining the feasibili ty of an outpatient tonsillectomy (Truy et al. 1994). In gene ral, the number of conditions reported along with a surgical event reflects the patient’s health status (Hadley and Mitchell 1997/98). Total charge Total charge reflects the intensity of services provided. Unlike payment that depends on the payer’s negotiation power, to tal charge represents the provider’s assessment of the intensity of care needed for a certain condition. Because severe conditions usually require extensive care and ha ve a high charge, total charge was used to control for the severity of illness (Manning et al. 1984). Race and ethnicity Race and ethnicity are key patie nt characteristics affecti ng the use of health care. Other studies have found that black patients us ed less care than white patients. Kressin (2001) reviewed articles published be tween 1966 and 2000 and found that African Americans were less likely to have invasive cardiovascular pro cedures, including cardiac catheterization, percutaneous transluminal coronary angiopla sty (PTCA), and CABG (Kressin and Petersen 2001). A study of th e Northern California Kaiser Permanente Medical Care Program showed that, among women younger than 60, African Americans were more likely to be hospitalized for cong estive heart failure than white Americans (Alexander et al. 1995). Based on the 1988 Mass achusetts hospital discharge data, blacks had lower rates of eight procedures, incl uding abdominal aortic aneurysm repair,

PAGE 56

44 appendectomy, and cardiac valve replacemen t, but higher rates for two procedures, hysterectomy and prostatectomy (Mort et al 1994). A study of end-stage renal disease patients between 1996 and 1997 showed that bl ack patients were less likely to receive kidney transplantation than whites (Ayanian et al. 1999). One study on claims data from the 1992 Colorado’s fee-for-service Medi caid program found that total annual expenditures per child with otitis media were higher for white children than for Hispanic or black children (Bondy et al. 2000). Other studies have also reported vari ation in the use of care for non-black minorities. Hispanics and Asian Americans were more likely to report barriers to health care than non-Hispanic whites (Phillips et al. 2000). In fact, between 1977 and 1996, Hispanics had a decreased likelihood of having at least one ambulatory visit (Weinick et al. 2000). Mexican Americans were less aware of the available treat ment of stroke and less trusting of their health care providers th an non-Hispanic whites (Morgenstern et al. 2001). Therefore, race and ethnicity is incl uded in the study to control for possible effects on the use of health care. Age Age is an important determinant of use of health care. Older patients have a higher likelihood of experiencing preventable hospita lizations (Blustein et al. 1998). Based on the National Health Care Survey, the number of surgical procedures done in inpatient settings increases with patient age (Owi ngs and Kozak 1998). Older respondents were more likely to be screened for cholesterol (D avis et al. 1998). A ccording to the American College of Cardiology/American Hosp ital Association (ACC/AHA) guidelines, outpatient cardiac catheteriz ation is recommended for patients younger than 75 years old (Pepine et al. 1991).

PAGE 57

45 Between 1980 and 1995, a study reported that the pattern of using inpatient vs. outpatient surgical procedures varied by ag e. For children under age 15, substitutions between inpatient and outpatient surgeries occurred in tonsill ectomies, repairs of inguinal hernia, reductions of fracture and dislocat ion and operations on muscle, tendon, fascia, and bursa. For patients aged 15 to 44, substitutions were seen in tubal sterilizations and reductions of fracture and dislocation, whil e substitutions for the operations on the urinary system, cystoscopies, and cholecyst ectomies were observed among patients aged 45 to 64 and those 65 or older (Kozak et al. 1999). Other studies found that age was not a significant factor in tonsillectomies. T onsillectomies for younger children were no longer considered risky and could be done either in ou tpatient or inpatient se ttings (Reiner et al. 1990; Truy et al. 1994; Rakover et al. 1997). Therefore, age may affect the choice of inpatient vs. outpatient surgery. Gender Gender also influences the choice of treatment. Males and females are different physiologically, and thus vary in the morbid ity of different condi tions. In one study, women tended to report more symptoms, us ed more ambulatory care, and were more likely to present chronic conditions than me n (Clark et al. 1991). In 1996, the ten most common surgical procedures for males were different from those for females (Owings and Kozak 1998). While most of these pr ocedures had similar proportions done in ambulatory settings, males tended to rece ive diagnostic ultrasounds in ambulatory settings while females had a smaller perc entage of procedures done in ambulatory settings. Since men use care differently th an women, gender can affect the choice of treatment.

PAGE 58

46 Family income The effect of low-income on the use of health care remains significant even after adjusting for other measures of socioeconom ic status, such as no insurance or low education level. Based on 1995 New York St ate hospital discharge data, patients who lived in lower income neighborhoods were less likely to be treated with cardiac catheterization, PTCA, CABG, and any revascul arization procedure (P hilbin et al. 2000). Income was a significant determinant of recei ving these invasive procedures even for patients who were admitted into a hospital with on-site CABG and PTCA. Lower income Medicare patients are more likely to experience a preventable hospitalization. However, after controlling for patients’ de mographic factors, socioeconomic status, history of chronic illness, and health status low income became insignificant in affecting the likelihood of preventable hospi talization (Blustein et al. 1998). Education Lack of knowledge about health care may discourage the use of care. Studies show that less-educated Americans tend not to us e preventive care or r ecognize symptoms that need medical attention (Davis 1998; Gornic k 2000). Medicare beneficiaries with lower education levels tend to experience preven table hospitalization th at could have been managed in ambulatory settings in 1992 (Blu stein et al. 1998). In addition, physicians tend to follow-up intensively on patients with professional careers. Therefore, patients’ level of education may affect the choice of treatment method. Family support Social characteristics, such as living w ith a spouse or being from a two-parent family, are considered favorable criteria for outpatient surgery w ithout a hospital stay. For example, children are recommended for out patient tonsillectomies if they have strong

PAGE 59

47 family support (Rakover et al. 1997). The cl inical guideline for ambulatory anorectal surgery developed by the American Society of Colon and Rectal Surgeons recommends family support as a criterion for outpatient su rgery (The Standards Task Force 1998-99). Because patients need immediate rest and as sistance after an out patient surgery, those having family support are more likely to get sufficient post-operation care. Control Variables: Physician Characteristics Physician characteristics include physician specialty, years of pr actice, and practice environment (Selby et al. 1999). In this st udy, urban or rural area and geographic region control for the practice environment although these practice environment variables can also affect some payer and patient character istics. Urban or rural area and geographic region are included to control for area char acteristics. Two other characteristics, physician specialty and years of practice, are unobserved and thus inevitably omitted from this study. Urban or rural area The area where physicians practice can aff ect their decisions about the choice of treatment. Physicians practicing in urban ar eas have different patterns of care than those in rural areas. For instance, breast cancer pa tients who were treated in urban areas were more likely to receive breast-conserving trea tment than those treated in rural areas (Nattinger and Goodwin 1994). Furthermore, hos pitals and surgeons’ practices tend to be located in Metropolitan Statistical Areas (MSAs) where sufficient labor and capital resources are available and where advanced te chnology is more likely to be adopted for outpatient surgery. Freestanding surgery cente rs are usually located in urban areas. Pauly and Erder (1993) found that patients were more likely to undergo outpatient surgery when freestanding surgery centers were available. Patients living in an MSA are

PAGE 60

48 likely to have better access to outpatient su rgical centers and thus are more likely to undergo outpatient surgery. Geographic region Variations in using different types of treatment are also found among geographic regions. For example, the use of breast-cons erving surgery was higher in Middle Atlantic states and New England, and lower in East South Central states and the West South Central states (Nattinger et al. 1992). Birkmeyer et al. ( 1998) found that some surgeries, such as lower extremity revascularization, carotid endarterectomy, and back surgery, were more frequently performed in certain re gions of the country than others. Based on national fee-for-service Medicare claims from 1996 through 1997, the rate of major amputations per year for individuals without diabetes were higher in the Southern and Atlantic states, but the patter n was not consistent among diabetes patients (Wrobel et al. 2001). A recent study found that the use of out patient mastectomy, rather than inpatient mastectomy, was more prevalent in some states, such as Maryland, Colorado, and Connecticut (Case et al. 2001). Similarly, the use of outpatient hernia repair also varied by state. For Medicare patients who had hern ia repair in 1987-88, over one third of the cases were done in an outpatie nt setting, but the rates of out patient surgery by state varied tremendously, ranging from 89.9% in Washi ngton to 6.3% in Georgia (Mitchell and Harrow 1994). Consequently, geographic regi on is included to account for regional variation in the use of outpatient surgery. Unobserved physician characteristics Physician specialty and years of practice may affect physicians’ decisions regarding the choice of treatment. Studies have f ound that patients seeing physicians of different specialties used different levels of care and incurred different costs (Bartman et al. 1996;

PAGE 61

49 Levetan et al. 1999; Selby et al. 1999). The number of years in practice also influenced physician referral and prescr ibing treatment (O'Leary et al. 2000; Rutchik et al. 2001). Although physician specialty and years of practice may affect the choice between inpatient and outpatient su rgery, they are no t included in this study due to data limitations. Control Variables: Other Payer Characteristics In addition to HMO vs. non-HMO covera ge, other payer characteristics may influence the choice of treatment method. In particular, payment rate, payment method, and utilization management may play a role. Payment rate Payment rate is likely to have an impor tant influence on the choice of treatment method. Total payment for a surgical procedure includes the insurer’s payments to physician and facility, as well as the patient’ s out-of-pocket payment. Payers set the rates, and the level of rate s can encourage or discourage the use of certain types of services. For example, a surgical procedur e that requires a hi gh out-of-pocket payment could lead to reduced use of the procedur e (Pauly and Erder 1993). Payment rates included are total payment, physician payment (the sum of insurer and patient payment), and the patient’s share of paym ent for a surgical procedure. Unobserved payer characteristics Studies show that payment method and utilization management affect the use of care, but unfortunately these variables are not available for this analysis. One study found that physician group practices that pay physicians based on capitation tend to use fewer resources (Kralewski et al. 2000). Ut ilization review led to reducing the use of hospital care and surgical pr ocedures (Feldstein et al 1988; Rosenberg et al. 1995).

PAGE 62

50 Utilization review, physician prof iling and clinical guidelines had a significant effect on reducing the variation in phys ician practice (Rizzo 1993; Wolff and Schlesinger 1998; Kralewski et al. 2000). The Analytic Model Based on the empirical specific ation, an analytic model is formulated to incorporate relevant factors affecting th e choice of outpatient surgery: Choice of (Outpatient Surgery) = f (PLAN, health status, number of cond itions, total charge, race/ethnicity, age, gender, income, education, family support, MSA, geographic area, total payment, physician payment, patient out-of-pocket payment) Table 5-1 lists the independ ent variables expected to influence the probability of choosing outpatient rather than inpatient surgery. The next chapter describes the implementation of the analytic model using a specific dataset. Table 5-1. Independent Variables Affecti ng the Probability of Choosing a Surgery Setting VariablesCharacteristics T yp e of health p lan: HMO vs. non-HMOPa y er Havin g a g atekee p erPa y er Self-re p orted health statusPatient Number of conditionsPatient Total char g ePatient Race/ethnicit y Patient A g ePatient GenderPatient Famil y incomePatient EducationPatient Famil y su pp ortPatient MSAPh y sician Re g ionPh y sician Total p a y mentPa y er Ph y sician p a y mentPa y er Out-ofp ocket p a y mentPa y er

PAGE 63

51 CHAPTER 6 DATA, VARIABLES AND STATISTICAL ANALYSIS The data that test the rese arch questions of this diss ertation come from a nationally representative household survey, the MEPS. A description of the MEPS dataset is followed by a detailed description of variable s obtained from the survey. The chapter concludes with a discussion of statistical methods used for the analysis. Data The MEPS is a national probability survey cosponsored by the Agency for Healthcare Research and Quality (AHRQ) a nd the National Center for Health Statistics (NCHS) to address policy issues in financ ing for health care. Begun in 1996, the MEPS collects information about hea lth care utilizati on and expenditures to provide nationally representative estimates for the U.S. ci vilian non-institutionalized population. This dissertation relies on the combined 1997, 1998, and 1999 MEPS data to assess the effect of HMO coverage on the choice of outpatient or inpatient surgery. The MEPS consists of four components: the Household Component (HC), the Medical Provider Component (MPC), the In surance Component (IC), and the Nursing Home Component (NHC). The HC is the core survey that determines the samples of the MPC and part of the IC. As part of the HC and the IC, insurance and employment data are collected from respondents’ employers, unions, and other sources of private health insurance. The information includes the num ber and types of private insurance plans offered, benefits associated with these pl ans, premium contributions by employers and employees, eligibility requirements, and em ployer characteristics. The NHC surveys

PAGE 64

52 nursing homes and persons residing in or admitted to nursing homes at any time during the survey year. Because the NHC has a di fferent sample design from the HC, which is the primary data source for this disse rtation, the NHC is not addressed here. Every year, the HC samples a new panel of households from the respondents to the previous year’s National Hea lth Interview Survey (NHIS). Each panel of households is surveyed through face-to-face interviews five times during a two-year period by using an overlapping panel design. For example, th e first panel was in terviewed between 1996 and 1997, while the interviews of the sec ond panel of households began in 1997 and continued in 1998. Due to the overlapping pane l design, after the firs t year of the MEPS (1996), two panels of households are interviewed each year. This dissertation pools the 1997, 1998, and 1999 MEPS data to study the effect of health plan type on the choice of outpatient surgery. The response rate for the 1997 survey was 74.0%, 67.9% for the 1998 MEPS, and 64.3% for the 1999 survey (MEPS 2001a; MEPS 2001b; MEPS 2002a). The 1997 MEPS dataset has 36,340 respondents, with 22,953 respondents in the 1998 MEPS and 23,565 individuals in the 1999 survey. Sample Design and Sample Weights Because the MEPS interviews a portion of the previous year’s NHIS sample, the MEPS uses the same complex sampling design as the NHIS. In the sampling design for the 1995 to 2004 NHIS, the nation is partitioned into 1,995 primary sampling units (PSUs) that are counties or groups of adjacent counties, and 52 of the largest metropolitan areas are assigned to 52 PSUs (Botman et al. 2000). PSUs are further partitioned into 237 design strata. PSUs are selected from each stratum, and households are sampled from each PSU based on known probabilities. Because of the MEPS’

PAGE 65

53 overlapping sampling design, when pooling three years’ MEPS data, as this dissertation does, some adjustment to the assigned PSUs and strata is required to obtain national estimates (MEPS 2003a). As a nationally representative survey, each case in the MEPS data represents a group of Americans that share similar charact eristics used to sample from the population. A sample weight for each case is developed to incorporate in the estimation processes. These sample weights are constructed to account for sample design, including unequal probability sampling of the population (i.e., oversampling minority groups), as well as nonresponse rates and partial res ponses from some survey participants. This dissertation uses sample weights, adjusted by pooled st rata and PSUs, to estimate variances and test hypotheses. MEPS Data Collection The MEPS collects data beginning with household interviews (Figure 6-1). Households are asked about demographics, employment, income, health status, health insurance, utilization of care, and payment. Healthcare utilizati on data include hospital stays, other hospital care, office-based physic ian care, other medical provider care, dental services, home health, prescribed medications, medical equipment and supplies, and alternative care. Detailed data for each he althcare encounter are collected, including type of practitioner, time spent with provider, ty pe of care, medical conditions, main surgical procedures, charges and/or payments. When a household reports any use of care, the MEPS subsequently surveys their medical providers, which becomes the MPC of the MEPS. Medical providers include office-based physicians, doctors of osteopa thy, other practitioners practicing under physician supervision, hospitals that provide inpatient, outpatient, and emergency room

PAGE 66

54 care, home health care agencies, long term care institutions, and pharmacies. Information from the MPC is used to verify respondent-re ported medical events and collect additional information. In particular, the MPC validates respondent-reported payments or charges and types of services received. Figure 6-1. The MEPS Data Collection In the MPC, providers are interviewe d based on a predetermined coverage: 100% of the providers of Medicaid recipients; 75% of the providers of managed care enrollees; and 25% of the providers of the remaining HC respondents who ar e either enrolled in non-managed care plans or other public plans. The data of the MPC are collected thr ough telephone or mailed questionnaires with a telephone follow-up for nonresponses. Combining information from household respondents and medical providers, the data on medical events are categorized into Hospital Inpatient Stay, Outpatient Depa rtment Visits, Office-Based Provider, Emergency Room, Dental, Other Medical, Home Health, and Prescription Medication. The Household Component (HC) Hospital inpatient stay Outpatient visits Home Health Visits Prescription drug Dental visits ERvisits The Medical Provider Component (MPC) Medical officevisits

PAGE 67

55 Medical conditions and type of medical proce dures come from respondents’ reports, and are verified through the MPC survey. Profes sional coders then code medical conditions and procedures into a fully specified ICD9-CM condition or procedure code. Due to confidentiality restrictions, the ICD-9-CM codes are furt her collapsed into 3-digit condition codes or 2-digit procedure codes. Construction of Dataset for Analysis In order to assess the effect of HMO coverage on the choice of outpatient or inpatient surgery, the dataset includes only surgeries that were performed in both the inpatient and outpatient settings. The dataset constructed for this dissertation comes from 16 MEPS public use data files and three Nati onal Health Interview Survey public used data files, including: The 1997, 1998, and 1999 Hospital Inpatient Stays Files (HC-016DF, HC-026DF, and HC-033DF) (MEPS 2001c; MEPS 2001d; MEPS 2002b): containing information about types of surgical pr ocedure and medical condition, expenditure, days of hospital stay, and type of payer. The 1997, 1998, and 1999 Outpatient Depart ment Visits Files (HC-016F, HC026F, and HC-033F) (MEPS 2001e; MEPS 2001f; MEPS 2002c): containing information about types of surgical pr ocedure and medical condition, expenditure, and type of payer. The 1997, 1998, and 1999 Consolidated Data Files (HC-020, HC-028, and HC038) (MEPS 2001a; MEPS 2001b; MEPS 2002a): providing demographic information. The 1997, 1998, 1999 Person Round Plan Files (HC-047) (MEPS 2003b): providing detailed information on health pl an coverage, including number of plans covering a specific period of time, t ype of health plan, and managed care information. The 1998 and 1999 Medical Conditions Files (HC-027 and HC-037) (MEPS 2001g; MEPS 2002d): linking to surgical events in the 1998 and in the 1999 Hospital Inpatient Stays Files and Outpatient Department Visits Files. Number of medical conditions associated with a surgi cal event is used as a control independent variable, COND. The 1997 data files do not need this linkage because COND is provided in the public use files, named NUMCOND.

PAGE 68

56 Three linking files (96HIS_97MEPX.DAT, NHMEP98X.DAT, and NHMEP99X.DAT) (MEPS 2001h; MEPS 2002e; MEPS 2002f): linking survey information reported by the same respondent s in the MEPS and the previous year’s NHIS; these files are necessary because e ach year a new MEPS-HC panel is drawn from the previous year’s NHIS sample. 96HIS_97MEPX.DAT links the 1997 MEPS with the 1996 NHIS, and NHMEP98X.DAT links between the 1998 MEPS dataset and the 1997 NHIS public use da taset. NHMEP99X.DAT links the 1999 MEPS with the 1998 NHIS. The 1996 NHIS, the 1997 NHIS, and the 1998 NHIS (DA2658.DAT, DA2954.PERSON, and DA3107p.DAT) (U.S Dept. of Health and Human Services 1998a; U.S. Dept. of Health and Human Services 1998b; U.S. Dept. of Health and Human Services 1999): providi ng health status information reported by the MEPS respondents who also participated in the previous year’s NHIS and had a surgical event during the period of the first MEPS interview. The 1996 to 1999 Pool Estimates Files (H36.ssp) (MEPS 2003a): containing adjusted strata and PSU for pooling multiple years' MEPS data. Figure 6-2 shows a step-by-step data c onstruction, and Appendix B contains the corresponding SAS programming: 1. Step 1: Take the 1997, 1998, and 1999 MEPS Ho spital Inpatient Stays files (HC016D, HC-026D, and HC-033D), keep only sc heduled surgical cas es, i.e., exclude cases that “operation or surgical procedure” is not the main reason for hospitalization (in the MEPS dataset, RSNINHOS not equal to “1”); take the 1997, 1998, and 1999 MEPS Outpatient Department Visits files (HC-016F, HC-026F, and HC-033F), exclude cases that indicate no surgical procedure performed on the visit (in the MEPS dataset, SURGPROC=“0”); 2. Step 2: Merge the above two files with the 1997, 1998, and 1999 MEPS Full Year Consolidated Data files (HC-020, HC-028, and HC-038), keep only cases for age 0 to 64, and add types and number of health plan coverage in the time when a surgery occurred by linking the constructed data files with person round plan files (HC047). 3. Step 3: Refine the dataset by selecting surg ical procedures that are reported as both inpatient and outpatient surgeries. This step is first based on a variable in the MEPS datasets: SURGPROC in Hospital Stay File, and SURGNAME in Outpatient Visit File. If SURGPROC or SURGNAME do not provide specific type of surgery, then the selecti on process is based on the va riables of two-digit ICD-9 procedure codes (IPPRO1X and IPPRO2X in Hospital Stay File, and OPPRO1X in Outpatient Visit File). a. First, choose cases having one of six main surgical procedures (arthroscopic surgery, cardiac cathet erization, tonsillectomy, dilation and

PAGE 69

57 curettage (D&C), cataract surgery, and pacemaker) in the MEPS variable (SURGPROC or SURGNAME). b. Second, match the remaining cases that have valid ICD-9 procedure codes (from 01 to 99) with the list of IC D-9 procedure codes reported as both inpatient and outpatient cases in the 1996 National Health Care Survey (Table 6-1). 4. Step 4: Exclude cases with non-private insurer payers. 5. Link cases from Step 4 with the 1998 a nd 1999 medical condition files (HC-027 and HC-037) to get the number of conditions asso ciated with the surgical events in the 1998 and 1999 MEPS. 6. Then, merge the data file with the 199699 pooled estimate file (HC-036) to obtain the reassigned strata and PSU after pooling the 1997, 1998, and 1999 MEPS data. This dissertation uses the 1996 National Hea lth Care Survey as the reference for constructing the datase t because the 1996 National Health Care Survey contains the most recent and complete national information on inpatient and outpatient surgeries. However, since new technology could be developed over the time, it is possible that certain inpatient surgeries in 1996 might have become mostly outpatient surgeries in 1997, 1998, or 1999. One component of the National Hea lth Care Surveys, the National Hospital Discharge Survey, collects inpatient data every year. Thus the 1997, 1998, and 1999 inpatient surgical cases in the National Hospit al Discharge Surveys are used to verify that surgical procedures that had inpatient and ou tpatient cases in 1996 were still reported as inpatient surgery in 1997, 1998, and 1999 ( Owings and Lawrence 1999; Popovic and Kozak 2000; Popovic 2001). Table 6-1 shows the number of cases of su rgeries that were reported as inpatient and outpatient surgery. Ratios of inpatient to outpatient cases (I/O ratio) are included for both the MEPS data and the data from the 1996 National Health Care Surveys. While

PAGE 70

58 Figure 6-2. Dataset Construction The 1997 MEPS Dataset Outpatient (16,035 cases) Inpatient (3,710 cases) Surgery is performed during this visit. 1 555 cases 1 176 cases Step 2. Merging with the Full Year Consolidated Data files which provide demographic and socioeconomic status information, keeping only cases of age 0 to 64 768 cases Step 3. Select procedures that were reported as inpatient and outpatient procedures Step 4. Keeping cases whose surgical expenses were paid by private insurance companies only 260 Inpatient Surgery Cases 554 Outpatient Surgery Cases 1 110 cases 230 cases 344 cases 125 cases 233 cases The 1998 MEPS Dataset Outpatient (10,470cases) Inpatient (2,588 cases) Surgery is performed during this visit. 1 032 cases 821 cases 536 cases 763 cases 151 cases 244 cases 67 cases 181 cases The 1999 MEPS Dataset Outpatient (9,551 cases) Inpatient (2,420 cases) Surgery is the main reason of hospitalization. Surgery is performed during this visit. 1 012 cases 768 cases 490 cases 687 cases 119 cases 181 cases 68 cases 140 cases Surgery is the main reason of hospitalization. Surgery is the main reason of hospitalization. Step 1.

PAGE 71

59 the National Health Care Surveys collect data from providers, the MEPS interviews households from the population. Because of di fferent sources for the information, the I/O ratios reported in the MEPS data differ from those in the 1996 National Health Care Survey. In 1996, only 11.4% of the population under age 65 had an outpatient visit or inpatient discharge with procedures (Owi ng and Kozak 1998). Given that the MEPS sampled a household based on characteristics othe r than type of surgery, it is likely that people having certain types of surgeries are not sampled into MEPS. Therefore, the 1996 National Health Care Surveys serve as an exte rnal reference to select procedures for this dissertation. A subset of data containing only cases of surgic al procedures with an I/O ratio between 0.2 and 5 is also constructed to test one of the research questions. The resulting dataset contains 814 cas es representing 9,595,656 surgical cases in the nation between 1997 and 1999, includi ng 3,333,464 surgical cases in 1997; 3,466,624 cases in 1998; and 2,795,568 in 1999. The dataset requires an additional modification to the original inpatient vs. outpatient coding. Because this dissertation defines inpatient surgery as a surgery with at least one nigh t inpatient stay, cases from Hospital Inpatient Stay files with zero night stay s are recoded as outpatient surg ical cases. Table 6-1 shows the unweighted number of cases by type of su rgical procedures in the final dataset. A Subset of the Constructed Dataset A subset of the final constructed datase t is prepared to include only cases of surgical procedures that had ratios of inpati ent to outpatient cases (I/O ratios) between 0.2 and 5 in the 1996 National Health Care Survey (Table 6-2). Thus, cases excluded from the subset are surgeries that were mostly done in one of the settings (i.e., inpatient or outpatient setting). This subset contains 391 cases, including 203 inpatient cases and 188 outpatient cases.

PAGE 72

60 Table 6-1. Number of Cases by Type of Procedure The 1997, 1998, 1999 MEPS (Private Insured) SURGERY (based on a variable in the MEPS dataa) TOTALInpatientOutpatientI/Ob1996 NHCS I/Ob Arthrosco p ic sur g er y 157181390.130.04 Cardiac Catheterization7043271.592.50 Dilation and Curetta g e796730.080.16 Tonsillectom y 6614520.270.09 Cataract Sur g er y 3503500.04 Pacemaker6515.006.5 Two-Digit ICD-9 Procedure Codec and the Description 03S p inal cord and canal o p erations330#1.20 06Th y roid and p arath y roid o p erations7431.331.85 08E y elids o p erations20200.17 16Orbit/e y eball o p erations5140.250.87 18External ear o p erations20200.27 19Middle ear reconstructure110#0.07 20Other mid and inner ear o p erations1901900.05 21O p erations on nose70700.11 22Nasal sinus o p erations40400.07 23Tooth removal & restoration10100.17 27Other mouth and face o p erations10100.49 28Tonsil and adenoid o p erations5140.250.09 29O p erations on p har y nx10100.40 31Lar y nx trachea o p erations30300.81 33Other o p erations on lun g bronchus10101.63 36O p erations on heart vessels440#21.00 37Other heart and p ericardium o p erations220#2.70 38Vessel incision, excision, and20205.11 39Other o p erations on vessels4313.006.94 40L y m p hatic s y stem o p erations10101.05 44Other gastric operations4130.331.61 45Intestine incision, excision, anastomosis 16214 0.14 0.31 48Rectal & perirectal operations110#0.51 49Operations on anus40400.21 51Biliary tract operations6132291.101.22 53Repair of hernia195140.360.21 54Other abdomen region operations3120.500.97 56Operations on ureter20200.93 57Urinary bladder operations7616.000.47 59Other urinary tract operations220#1.45 60Prostate & seminal vesicle operations330#1.69 63Operations on Sperm cord, epididymis, vas deferens 60600.05 64Operations on penis20200.25

PAGE 73

61 Table 6-1. Continued The 1997, 1998, 1999 MEPS (Private Insured) Two-Digit ICD-9 Procedure Codesc and the Description TOTALInpatientOutpatientI/Ob1996 NHCS I/Ob 65O p erations on ovar y 3212.003.84 66Fallopian tube operations122100.201.07 67Operations on cervix40400.11 68Other uterine incision and excision54522261.73 69Other uterus and support operations60600.20 70Vagina and cul-de-sac operations220#1.90 76Facial bone and joint operations21112.04 77Incision excision division bone161150.070.48 78Other bone o p erations exce p t face84410.73 80Incision excision j oint7616.000.24 81Joint re p air3719181.061.04 83Other muscle, tendon, fascia, bursa operations 32120.57 84Other musculoskeletal p rocedures9450.803.13 85O p erations on the breast154110.400.18 86Skin & subcutaneous o p erations1301300.65 97Re p lace & remove devices220#0.49 Total cases based on ICD-94011742270.77 GRAND TOTAL8142605540.47 Note: # These procedures have zero outpatient cases, so the ratio of inpatient to outpatient cases is invalid.a These six surgeries are based on a MEPS’ variable, SURGNAME in Outpatient Department Visit data files and SURGPROC in Hospital Inpatient Stay files.b I/O represents the ratios of inpatient to outpatient cases.c These cases did not have information in "SURGNAME" or "SURGPROC", but had reported information in “OPPRO1X” (the MEPS Outpatient Department Visit data files) or “IPPRO1X” and/or “IPPRO2X” (the MEPS Hospital Inpatient Stay data files).

PAGE 74

62 Table 6-2. Number of the Subset of Cases by Type of Procedure The 1997 1998 1999 MEPS SURGERY (based on a variable in the MEPSa) TOTALInpatientOutpatientI/Ob1996 NHCS I/Ob Cardiac Catheterization 70 4 3 2 7 1 .59 2 .50 Two-Di g it ICD-9 Procedure Code and the Descri p tion 03 S p inal cord and canal o p erations 3 3 0 # 1.20 06 Th y r o i d, p arath y r o i d ope rati o n s 7 4 3 1 .33 1 .85 1 6 Orbit e y eball o p erations 5 14 0. 2 5 0.87 1 8 External ear o p erations2 0 2 0 0. 2 7 2 9 O p erations on p har y nx1 0 1 0 0. 4 0 3 1Lar y nx trachea neck o p erations 3 0 3 0 0.8 1 33Other o p erations on lun g, bronchus10101.63 37Other heart p ericardium o p erations220#2.70 40L y m p hatic s y stem o p erations10101.05 44Other g astric o p erations4130.331.61 45Intestine incision, excision, anastomosis162140.140.31 48Rectal & p erirectal o p erations110#0.51 49O p erations on anus40400.21 51Biliar y tract o p erations6132291.101.22 53Re p air of hernia195140.360.21 54Other abdomen re g ion o p erations3120.500.97 57Urinar y bladder o p erations7616.000.47 59Other urinar y tract o p erations220#1.45 60Prostate seminal vesicle o p erations330#1.69 64O p erations on p enis20200.25 65O p erations on ovar y 3212.003.84 66Fallo p ian tube o p erations122100.201.07 68Other uterine incision excision54522261.73 69Other uterus su pp ort o p erations60600.20 70Va g ina & cul-de-sac o p erations220#1.90 76Facial bone & j oint o p erations21112.04 77Incision excision division bone161150.070.48 78 O th e r bo n e ope rati o n s e x cep t fa ce 8 441 0.73 80Incision excision j oint7616.000.24 81Joint re p air3719181.061.04 83Other muscle, tendon, fascia, bursa o p erations32120.57 84Other musculoskeletal p rocedures9450.803.13 86Skin & subcutaneous o p erations1301300.65 97Re p lace & remove devices220#0.49 Total cases based on ICD-93211601610.77 GRAND TOTAL3912031880.47 Note: # These procedures have zero outpatient cases, so the ratio of inpatient to outpatient cases is invalid. a This surgery was from a MEPS’s variable, SURGNAME in Outpatient Department Visit data files and SURGPROC in Hospital Inpatient Stay files.b I/O represents the ratios of inpatient to outpatient cases All cases in this Table had surgeries with an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Surveys.

PAGE 75

63Table 6-3. Independent Variables by Ty pe of Characteristic, Conceptual Basis, and Questions from the MEPSVariable Type of Characteristic Conceptual BasisQuestions in the MEPS Questionnaires and Possible Answers PLAN (Type of Health Plan) PayerHMO status increases the likelihood of having an outpatient surgery. Since HMOs tend to manage care by reducing expensive care, such as inpatient surgeries, when there are less costly alternatives, i.e., outpatient surgeries, HMO patients may be more likely to have a surgery on an outpatient setting than those in nonHMO plans. If the person is covered by private insura nce and answered one of the following questions using the term “HMO,” HMO is set to “yes.” 1. From which of the sources on this card did anyone in the family purchase health insurance? (HX23) Possible Answers: 1: From a group or association; 2: From a health insurance purchasing alliance; 3: Directly through a school; 4: Directly fro m an insurance agent; 5: Directly from insurance company; 6: Directly from an HMO; 7: From a union; 8: From anyone’s previous employer (cobra); 9: From any one’s previous employer (not cobra); 10: From spouse’s/deceased spouse’s previous employer; 11: From some other employer; 12: Under plan of someone not living here; 91: Other source; -7: Refuse; -8: Don’t know 2. {You mentioned that (PERSON) {(ar e/is)/(were/was)} self-employed and had health insurance through that business.} Which category on this card comes closest to {the main/another} way (PERSON) (purchase/purchases) this insurance? (HX03) Possible Answers: 1: From a professional association; 2: From a small business group; 3: From a union; 4: From a health insurance purchasing alliance; 5: Directly from an insurance agent; 6: Directly from insuran ce company; 7: Directly from an HMO; 8: From a previous employer; 9: From a previous employer (cobra); 91: Other; -7: Refuse; -8: Don’t know. 3. What is the name of the in surance company or HMO from which (POLICYHOLDER) receives hospital and physician benefits? (HX51) Possible Answer: 1: Insurance company; 2: HMO; 3: Self-insured company.

PAGE 76

64Table 6-3. ContinuedVariable Type of Characteristic Conceptual BasisQuestions in the MEPS Questionnaires and Possible Answers PLAN (Type of Health PlanContinued) Payer(Continued)4. From which of the sources on th is card did anyone in the family purchase health insurance? Possible Answers: 1: From a group or association; 2: From a health insurance purchasing alliance; 3: Directly through a school; 4: Directly fro m an insurance agent; 5: Directly from insurance company; 6: Directly from an HMO; 7: From a union; 8: From anyone’s previous employer (cobra); 9: From any one’s previous employer (not cobra); 10: From spouse’s/deceased spouse’s previous employer; 11: From some other employer; 12: Under plan of someone not living here; 91: Other source; -7: Refuse; -8. Don’t know. 5. What is the name of the {oth er} insurance company or HMO for (POLICYHOLDER)’s (ESTABLISHMENT) insurance? (HX54) Possible Answers: 1: Insurance company; 2: HMO; 3: Self-insured company. 6. If the person answered yes to the following question (MC01): Now I will ask you a few questions about how (POLICYHOLDER)’s health insurance through (ESTABLISHMENT) works for non-emergency care. We are interested in knowing if (POLICYHOLDER)’s (ESTABLISHMENT) plan is an HMO, that is, a Health Maintenance Organization. With an HMO, you must generally receive care from HMO physicians. For other doctors, the expense is not covered unless you were referred by the HMO or there was a medical emergency. Is (POLICYHOLDER)’s (INSURER NAME) an HMO? Possible Answers: Yes, No

PAGE 77

65Table 6-3. ContinuedVariable Type of Characteristic Conceptual BasisQuestions in the MEPS Questionnaires and Possible Answers PLAN (Type of Health PlanContinued) PayerGatekeeper, as a managed care approach, could reduce the use of inpatient surgery {(Do/Does)/As of (END DATE), did} (POLICYHOLDER)’s insurance plan require (POLICYHOLDER) to sign up with a certain primary care doctor, group of doctors, or a certain clinic which (POLICYHOLDER) must go to for all of (POLICYHOLDER)’s routine care? (MC02) Possible Answers: 1: Yes; 2: No; -7: Refuse; -8: Don’t know HLTH (Selfreported Health Status) PatientPatients with worse presurgery health tend to have an inpatient surgery. In general, compared to other people of (PERSON)'s age, would you say that (PERSON)'s health is excellent, very good, good, fair, or poor (E01) Possible Answers: 1: Excellent; 2: Very good; 3: Good; 4: Fair; 5: Poor; -9: Not ascertained; -8: Don’t know; -7: Refused; -3: No Data in round; -1: Inapplicable. COND (Number of conditions) PatientThe more conditions associated with a surgical event, the higher the level of severity tends to be. Thus, an inpatient surgery tends to be chosen. A constructed variable based on what respondents reported on their conditions, which are then coded by professional coders into ICD-9 Codes. COND is the counts of the ICD-9 codes. Was this hospital stay related to any specific health condition or were any conditions discovered during this hospital stay? (HS03) Possible Answers: Yes, No What conditions were discovered or led (PERSON) to enter the hospital? (HS04) TCH (Total Charge) PatientThe level of charge represents the intensity of the procedure. High charges are set for procedures requiring extensive care. A constructed variable from a series of charge/payment questions. Please refer to payment information at the end of this Table since this variable is imputed from a series of charge/payment questions.

PAGE 78

66Table 6-3. Continued VariableType of CharacteristicConceptual BasisQuestions in the ME PS Questionnaires and Possible Answers AGEPatientYoung patients are more likely to have an outpatient surgery. What is (READ NAME BELOW)'s date of birth? Enter MM/DD/YYYY. GENDERPatientMale patients tend to have outpatient surgeries. I have (PERSON) recorded as (READ GENDER BELOW). Is that correct? Possible Answers: 1: Male; 2: Female RACE (Race/ Ethnicity) PatientPeople of different races/ethnicities may use health care differently due to the differences in morbidity and mortality, presentation of symptoms, and communication with physicians. Please look at this card and tell me the group which best describes (PERSON)'s racial background. (RE101) Possible Answers: 1: American Indian; 2: Aleut, Eskimo; 3: Asian Or Pacific Islander; 4: Black; 5: White; 6: Other; -7: Refused; -8: Don’t knowWhich group represents (PERSON)'s main national origin or ancestry (RE99)Possible Answers: 1: Puerto Rican; 2: Cuban; 3: Mexican, Mexican-American, Mexicano, Chicano; 4: Other Latin American; 5: Other Spanish; 91: Other; -7: Refused; -8: Don’t Know INCOME (Annual family income as a percent of Federal Poverty Level) PatientAffordability of a surgery can affect the decision of surgery settings A constructed variable based on all income sources reported by respondents. The following are several key questions asked: How much money did (READ NAME(S) ABOVE) receive from wages or salary, tips, commissions, or bonuses? (IN18)How much did (READ NAME(S) ABOVE) receive in interest from savings accounts, bonds, NOW accounts, money market accounts, or similar types of investments? (IN19)How much money did (READ NAME(S) ABOVE) receive from alimony? (IN22) How much money was (READ NAME(S) ABOVE)’s net gain or net loss from the sale of property or other assets, including the sale of (his/her/their) home, if it was taxable? (IN24) Looking at this card, which range best estimates how much money was received (from Social Security and equivalent tie r 1 Railroad Retirement benefits)? (IN32)

PAGE 79

67Table 6-3. ContinuedVariable Type of Characteristic Conceptual BasisQuestions in the MEPS Questionnaires and Possible Answers EDU (Education) PatientLack of knowledge about health care may discourage the use of certain types of care What is the highest grade or year of regular school (PERSON) ever completed? (RE103) Possible Answers: 0-17 Years. SUPPORT (Family support) PatientFamily support increases the likelihood of outpatient surgery. For adults, this is a constructed variable based on the interview roster of household members and the following question: {(Are/Is) (PERSON) now/As of December 31, 1996, (were/was) (PERSON)} married, widowed, divorced, separated, or never married? (RE97) Possible Answers: 1: Married; 2: Widowed; 3: Divorced; 4: Separated; 5: Never Married. SUPPORT is set to “1”—married and living with spouse; otherwise, “0”. For children, if both father and mother ID is indicated, i.e. mother and father live in the household, SUPPORT is set to “1”—From 2-Parent Family; otherwise, “0.” MSA (Living inside MSA/ non-MSA) PhysicianPhysicians practicing in MSAs have better access to ambulatory surgical center, than those in non-MSAs, thus are more likely to prescribe an outpatient surgery. Assigned when sampled REGION (Geographic Region) PhysicianPhysicians practicing in some regions of the country tend to use more outpatient surgeries than those in other regions. Assigned When sampled

PAGE 80

68Table 6-3. ContinuedVariable Type of Characteristic Conceptual BasisQuestions in the MEPS Questionnaires and Possible Answers SPECIALTY (Physician’s Specialty) PhysicianPhysician training affects clinical decision-making. Not available due to data limitation YEARS OF PRACTICE (Number of years practicing) PhysicianYear of practicing reflects physician’s training, and thus affects clinical decision Not available due to data limitation Physician Payment Method (Capitation, discount FFS, or FFS) PayerHow a physician is paid produces different incentives. Capitation encourages a costconscious practice style, while FFS rewards productivity. Not available due to data limitation PTPAY (Out-ofpocket payment) PayerHigher patient out-ofpocket payment for a surgery (inpatient or outpatient) encourages patients to express the concern of affordability. A constructed variable from a series of charge/payment questions. Please refer to payment information at the end of this Table since this variable is imputed from a series of charge/payment questions.

PAGE 81

69Table 6-3. ContinuedVariable Type of Characteristic Conceptual BasisQuestions in the MEPS Questionnaires and Possible Answers TOTAL (Total Payment) PayerHigher payment for inpatient surgery encourages doing a surgery in inpatient settings. PAYMD (Physician Payment) PayerHigher physician payment for a surgery (inpatient or outpatient)discourages doing a surgery in a particular setting.A constructed variable from a series of ch arge/payment questions, and supplemented with survey results from the Medical Provi der Survey. The followings are some of the key questions: To whom was the bill sent? How much was the total charge for (PERSON)’s stay at (HOSPITAL) that began on (ADMIT DATE)/(PERSON)’s visit to (PROVIDER) on (VISIT DATE)/the last purchase of {NAME OF PRESCRIBED MEDICINE...} for (PERSON)/the services for (FLAT FEE GROUP) for (PERSON)/the {OME ITEM GROUP NAME} used by (PERSON) since (START DATE)/ services received at home from (PROVIDER) during (MONTH) for (PERSON)/(PROVIDER)’s services as part of the visit made on (VISIT DATE)}? (CP09) Is this a situation in which (PERSON) (are/is) required to pay a certain set amount each time {(PERSON) (visit/visits) (PROVIDE R) regardless of what happens during the visit/(PERSON) (receive/receives) services of this type}? (CP10) How much of the {{AMT TOT CH}/total charge} did anyone in the family pay for (PERSON)’s stay at (HOSPITAL) that began on (ADMIT DATE)/ (PERSON)’s visit to (PROVIDER) on (VISIT DATE)/the last purchase of {NAME OF PRESCRIBED MEDICINE...} for (PERSON)/the services for (FLAT FEE GROUP) for (PERSON)/the {OME ITEM GROUP NAME} used by (PERSON) since (START DATE)/services received at home from (PROVIDER) during (MONTH) for (PERSON) /(PROVIDER)’s services as part of the visit made on (VISIT DATE)}? (CP11) How much did (SOURCE) pay? (CP13) Has any source reimbursed or paid back anything to (PERSON) (or anyone in the family) for the amount paid ‘out-of-pocket’? That is, has any source reimbursed any of the {$/% FAMILY PAID} paid? (CP14)

PAGE 82

70Table 6-4. Description of VariablesVariableLabelTypeDescription Dependent Variable INOUTOutpatient SurgeryNominalTw o classes: outpatient or inpatient Independent Variables PLANType of health planNominalFour classes: One-HMO, One non-HMO gatekeeper, one non-HMO nongatekeeper, Two-plan HLTHSelf-reported Health StatusOrdinalFive levels: poor, fair, good, very good, excellent CONDNumber of conditionsOrdinalThree levels: 0-1, 2, 3-4 conditions TCHTotal ChargeContinuousTotal charge RACERace/EthnicityNominalTwo classes: Non-Hispanic White, Non-white or Hispanics AGEAgeContinuous0 to 64 years old GENDERGenderNominalTwo classes: Male, Female INCOMEAnnual family income as % FPLaOrdinalFive levels: Less than 100% FPL, 100-124%FPL, 125-199% FPL, 200-399% FPL, 400% FPL or more EDUEducationOrdinalThree levels of education: less than high school, high school, some college or higher SUPPORTFamily supportNominalTwo classes: “1”-“Living with spouse” (adults), or “from a two-parent family" children)”; “0”-“not living with spouse” (adults), or “from a one-parent family” (children). MSAMSANominalTwo classes: MSA, Non-MSA REGIONRegionNominalFour classes: Northeast, Midwest, South, West TOTALTotal PaymentContinuousSum of payer share of facility payment and physic ian payment, and patient’s share of payment PAYMDPhysician PaymentContinuousPhysician payment from payer and from patient PTPAYOut-of-pocket paymentContinuousPatient share of facility and physician payment YEARYearNominalThree classes: 1997, 1998, and 1999 a FPL: Federal Poverty Level.

PAGE 83

71 Variables This section describes variables used in th e analysis. Some variables come directly from the MEPS dataset, such as age, inco me, and health status, while others are constructed from two or more MEPS variables, such as family support, race/ethnicity, and payment. Table 6-3 shows the origin al questions in the MEPS used to derive independent variables. Based on the conceptu al model of treatment choice developed in Chapter 4 (see Figure 4-3), each variable in Table 6-3 is specified as one of the patient, physician, and payer characteristics that join tly determine the choice between inpatient and outpatient surgery. Table 6-3 presents the characteristics, the conceptual basis, and the origins of each variable. Table 6-4 furthe r describes the final formats of the variables in the constructed dataset. Table 6-5 contains descriptive statistics for discrete variables, and Table 6-6 provides descriptive statistics for continuous variables. Dependent Variable The dependent variable, INOUT, is dic hotomous (INOUT=1 for outpatient surgery, INOUT=0 for inpatient surgery). An inpa tient surgery was defined as a surgical procedure performed during a patient hospita l stay of one or more nights, while an outpatient surgery was performed in an outpati ent setting or in an inpatient setting with zero nights stay. In the da taset, 68.1% of the cases were outpatient surgeries while 31.9% were inpatient surgical procedures (Table 6-5). Primary Independent Variables To assess the effect of HMO enrollment status on the choice of surgery setting, PLAN was the primary independent variab le. PLAN had four possible values: 1. One HMO: people with one health insurance plan, which was an HMO;

PAGE 84

72 2. One non-HMO gatekeeper plan: people with one health insurance plan, which was a non-HMO gatekeeper health plan; 3. One non-HMO non-gatekeeper plan: people w ith one health insurance plan, which was a non-HMO non-gatekeeper health plan; and 4. Two-plan coverage: those covered by two health plans. The MEPS identified an individual’s HMO enrollment status through the questions listed in Table 6-3. When a respondent was not enrolled in an HMO, or was not certain about their health plan type, he/she was clas sified as a non-HMO enrollee. Among all the surgical cases in the constructed dataset, 46.3% were covered by one HMO health plan, 6.0% were enrollees of a non-HMO gatekeeper plan, and 38.9% were enrolled in a nonHMO, non-gatekeeper plan. Figure 6-3 show s the breakdown of the 8.7% of respondents having two-plan coverage (71 cases). The ma jority of respondents with two health plans (54 cases) were covered by at leas t one non-HMO non-gatekeeper plan. Figure 6-3. Types of Health Plan Covera ge for Cases with Two-plan Coverage Two HMOs n=13 One HMO & One Non-HMO Gatekeeper plan n=4 One HMO & One Non-HMO Non-Gatekeeper plan n=30 One Non-HMO Gatekeeper & One Non-HMO NonGatekeeper plan or Two Non-HMO Non-Gatekeeper plans n=24 Covered by Two Health Plans n=71

PAGE 85

73 Table 6-5. Descriptive Statistics for Disc rete Variables (Unwei ghted n=814; Weighted n=9,595,657) WeightedaUnweighted Level/ClassFrequency%Frequency% Out p atient 6,771,182 70.6 554 68.1 INOUT In p atient 2,824,475 29.4 260 31.9 One-HMO 4,406,708 45.9 377 46.3 One non-HMO Gatekee p er 577,144 6.0 49 6.0 One non-HMO nonGatekeeper Plan 3,671,750 38.3 317 38.9 PLAN Two p lans 940,056 9.8 71 8.7 Poor 324,875 3.4 32 3.9 Fair 1,045,211 10.9 105 12.9 Good 2,606,597 27.2 222 27.3 Ver y g ood 3,056,292 31.9 258 31.8 HLTH Excellent 2,552,196 26.6 195 24.0 0 45,606 0.5 2 0.3 1 8,535,207 89.9 728 89.4 2 853,355 8.9 68 8.4 COND 3-4 161,490 1.7 16 2.0 Non-His p anic White 8,258,534 86.1 644 79.1 RACE Non-White or His p anic 1,337,123 13.9 170 20.9 Male 4,187,401 43.6 313 38.5 GENDERFemale 5,408,257 56.4 501 61.5 Less than 100% FPL 339,691 3.5 36 4.2 100-124% FPL 294,664 3.1 31 3.8 125-199% FPL 797,722 8.3 83 10.2 200-399% FPL 3,278,588 34.2 298 36.6 INCOME400% FPL or More 4,884,992 50.9 366 45.0 Less than Hi g h School 682,436 7.2 79 9.8 Hi g h school 4,776,133 50.1 397 49.1 EDU Some colle g e or hi g her 4,069,169 42.7 333 41.2 Livin g with s p ouse/ two 6,970,556 72.6 588 72.2 SUPPORTNo s p ouse/sin g le p arent 2,625,101 27.4 226 27.8 MSA 7,865,969 82.0 633 77.8 MSA Non-MSA 1,729,688 18.0 181 22.2 Northeast 1,942,467 20.2 166 20.4 Midwest 2,614,114 27.2 205 25.2 South 3,613,755 37.7 314 38.6 REGIONWest 1,425,322 14.9 129 15.9 1997 3,333,464 34.7 358 44.0 1998 3,466,624 36.1 248 30.5 YEAR 1999 2,795,568 29.1 208 25.6 a Number of weighted cases is the product of number unweighted cases and the corresponding sample weight of each unweighted case.

PAGE 86

74 Table 6-6. Descriptive Statistics for Con tinuous Variables (Unwei ghted n=814; Weighted n=9,595,657) WeightedaUnweighted Mean Standard Error Mean Standard Error AGE 37.84 0.76 38.82 0.58 TCH (Total Charge)6,456.13249.476,808.49245.13 TOTAL (Total Payment)3,800.08159.764,094.96167.18 PAYMD (Physician Payment) 994.95 51.341,013.58 47.57 PTPAY (Patient Out-of-pocket Payment) 151.91 16.57 154.57 12.70 a weighted data are based on information of each unw eighted case and its corresponding sample weight. The means and standard errors of this column are the results of 9,595,657 weighted cases.Cases that did not report health plan info rmation (23 cases) were included in the non-HMO non-gatekeeper plan category. Table 6-7 and Table 6-8 compare cases reporting a non-HMO non-gatekeeper plan with those that did not provide health plan information. Most of the characteristics did not differ signifi cantly between the two groups, although three variables were signif icantly different, in cluding percentage of high-income cases (income 400% FPL or higher), total payment, and patient payment. Because the two groups of surgery cases had similar characteristics, cases with no plan information were included in the non-HMO non-gatekeeper category. When excluding surgeries performed primarily in either an inpatient or outpatient setting, as defined in Table 6-2, the subset of cases had a similar di stribution across types of health plan coverage (Table 6-9). Over 48% were HMO enrollees, while 36.3% had non-HMO non-gatekeeper health plan coverage Only 3.8% indicated that they were enrolled in a non-HMO gatekeeping pl an, and 11.3% were covered by two plans.

PAGE 87

75 Table 6-7. Comparing Cases with a Non-HMO NonGatekeeper Plan and Cases with No Self-Reported Plan Information (Nominal and Ordinal Variables) Self-Reported Non-HMO Non-Gatekeeper Plan (n=294) No Plan Information (n=23) VariablesLevel/Class Percenta Out p atient70.773.9 In p atient29.326.1 INOUT p value 0.7480 0-191.591.3 2 7.5 4.4 3-4 1.0 4.4 COND p value 0.5255 Poor3.1 8.7 Fair14.013.0 Good27.330.4 Ver y g ood33.517.4 Excellent22.230.4 HLTH p value 0.3607 Yes55.447.8 No44.652.2 Excellent/ Very Good Health p value 0.4808 Yes17.021.7 No83.078.3 Fair/Poor Health p value 0.5648 Non-His p anic White82.778.3 Non-White or His p anic17.321.7 RACE p value 0.5957 Male39.139.1 Female60.960.9 GENDER p value 0.9989 Living with spouse/ two parents 74.860.9 No s p ouse/sin g le p arent25.239.1 SUPPORT p value 0.1441 MSA66.078.3 Non-MSA34.021.7 MSA p value 0.2300 Northeast16.713.0 Midwest34.421.7 South34.447.8 West14.617.4 REGION P value 0.4969

PAGE 88

76 Table 6-7. Continued Self-Reported Non-HMO Non-Gatekeeper Plan (n=294) No Plan Information (n=23) VariablesLevel/Class Percent Less than 100% FPL 3.7 4.4 100-124% FPL 6.5 8.7 125-199% FPL10.2 8.7 200-399% FPL36.156.5 400% FPL or More43.521.7 INCOME p value 0.2869 Yes43.521.7 No56.578.3 High Income (400% FPL or more ) p value 0.0425 Some Hi g h School or11.017.4 Hi g h School Graduate46.739.1 Some Colle g e or42.343.5 EDU p value 0.5974 Yes11.017.4 No (High School Graduate89.082.6 Some High School or Less p value 0.3450 199744.626.1 199832.034.8 199923.539.1 YEAR p value 0.1467 a Percent is based on unweighted number of cases.

PAGE 89

77 Table 6-8. Comparing Cases with a Non-HM O Non-Gatekeeper Plan and Cases with No Self-Reported Plan Information (Continuous Variables) Self-Reported Non-HMO Non-Gatekeeper Plan (n=294) No Plan Information (n=23) Mean Standard Error Mean Standard Error p value AGE 39.58 1.01 36.30 3.330.3471 TCH (Total Charge)6,982.87 420.225,814.33 906.020.2429 TOTAL (Total Payment)4,812.24 338.023,185.44 692.880.0356 FACILITY (Facility Payment)3,381.75 299.732,255.85 613.560.1002 PAYMD (Physician Payment)1,161.20 86.47 875.78 212.790.2149 PTPAY (Patient Out-ofpocket Payment) 269.29 28.85 53.81 34.410.0000 Table 6-9. Cases for the Subset of the Dataset by Health Plan Coverage WeightedUnweighted CasesPercentCasesPercent One HMO plan2,166,381 49.5 190 48.6 One non-HMO gatekeeper plan 154,471 3.5 15 3.8 One non-HMO non-gatekeeper plan1,510,203 34.5 142 36.3 Two plans 545,084 12.5 44 11.3 TOTAL4,376,139100.0 391100.0 Note: This subset of data includes only cases of surgical procedures having an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Surveys. Number of weighted cases is the product of the number of unweighted cases and the corresponding sample weight of each unweighted case.Control Variables: Patient Characteristics For the purpose of this anal ysis, self-perceived health status (HLTH) represents pre-surgery health in ge neral. In the MEPS, self-perceiv ed health status was reported in each interview. Thus, the health status reported in the interview prior to the surgical procedure was used. For example, if a re spondent had a surgery during the second round of the interview (Round Two), the health st atus reported in the previous round (Round One) was used. Almost 56.0% of the surgical ca ses indicated that thei r health status was “excellent” or “very good,” while 27.3% reporte d “good” health status. Only a small

PAGE 90

78 percent (3.9%) reported “poor” health status, with 12.9% re porting “fair” health status (Table 6-5). For surgical cases occurring during the firs t round of the MEPS interview, the presurgery health status came from the previ ous year’s NHIS by linking the MEPS file with the 1996, 1997, and 1998 NHIS data. There were 63 cases in the 1997 MEPS, 51 cases in the 1998 MEPS, and 49 cases in the 1999 MEPS having health status (HLTH) information from the previous year’s NHIS data files. Number of conditions (COND) was the c ount of ICD-9 condition codes reported in a surgical event. COND was taken directly from a variable (NUMCOND) in the 1997 MEPS dataset, but for 1998 and 1999 MEPS data, COND was the count of medical conditions that could be linked to the su rgical event. Most cases (89.4%) had one condition, while 10.4% had two to four conditi ons associated with the surgical event (Table 6-5). Total charge (TCH) was the sum of facility, physician, and patient charge, and was taken directly from the MEPS dataset. Tota l charge was used as an indication of the intensity of the service rendered, and was in tended to control for the severity of the surgery. The mean charge per case was $6,808.49 (Table 6-6). Patients’ race (RACE ) was a constructed variable based on two MEPS variables, RACEX and RACETHNX. Over 79.0% of the su rgical cases in the constructed dataset were non-Hispanic Whites. Due to the small number of cases, Black, Hispanic, American Indian, Aleut, Eskimo, Asian, a nd Pacific Islander were grouped into one category, non-White or Hispanic (RACE=0), wh ich comprised almost 21.0% of all cases (Table 6-5).

PAGE 91

79 Gender (GENDER) is either male or female A majority of the cases were female (61.5%), while 38.5% were male (Table 6-5). Two socioeconomic status variables, family annual income and education were based on two MEPS variables, POVCAT and EDUCYR. Family annual income (INCOME) was reported as the percent of fe deral poverty level (FPL), which took into account family size (Federal Register 2001). INCOME had five levels: less than 100% FPL, 100% to 124%, 125% to199%, 200% to 399% and 400% FPL or more. A small proportion of cases (8.0%) were respondents from families with less than 125% FPL, while 45.0% were from families with incomes of at least 400% FPL. Over 10.0% of the cases were from families with incomes between 125 and 199% FPL, and 36.6% were in the range of 200% to 399% FPL income (Table 6-5). Education (EDU) was based on the MEPS variable EDUCYR. For people younger than 18 and not married, EDU was taken from moth er’s education level, or father’s in the case of single-parent (father) family. Almost half of the cases (49.1%) were high school graduates, and 41.2% had at least some co llege education, while 9.8% did not graduate from high school (Table 6-5). Family support (SUPPORT) was derived from SPOUSEIN in the MEPS dataset. For adults, a person had a value of “1” if ma rried and living with spouse. Children who came from a two-parent family, i.e., mother’s and father’s ID were valid (MOMPID and DADPID) in the MEPS data, also have a va lue of “1.” Otherwise, SUPPORT was assigned “0.” Over 72.0% of the cases had a value of “1” (Table 6-5).

PAGE 92

80 The variable AGE is based on the re ported age (AGE97X, AGE98X, and AGE99X in the MEPS dataset). Age is a continuous variable. The mean AGE is 38.8 years (Table 6-6). Control Variables: Physician Characteristics Two geographic variables, MSA and REGION, represent physician practice environment. Both variables were taken di rectly from the MEPS dataset. Almost 78.0% of the surgical cases were people residing in an MSA (Table 6-5). A large proportion of respondents (38.6%) lived in the South, with 20.4% living in the Northeast and 25.2% in the Midwest. Only 15.9% lived in the West (Table 6-5). Control Variables: Other Payer Characteristics Other than enrollment in an HMO, paye r characteristics in cluded three payment rates: total payment (TOTAL), physician payment (PAYMD), and patient out-of-pocket (PTPAY). TOTAL was the sum of the payment by a private insurer for a surgical event and the patient’s out-of-pocket payment for both facility and physician. As shown in Table 6-6, based on unweighted sample s, mean TOTAL was $4,094.96, mean PAYMD was $1,013.58, and mean PTPAY was $154.57. Other Control Variable Because the datasets used in this analys is were combined from three years data, YEAR, i.e. 1997, 1998, and 1999, was included to account for other changes over time (Table 6-5). Over 30.0% of cases were from the 1998 MEPS, 44.0% were from the 1997 data, and 25.6% were from the 1999 MEPS data. Statistical Analysis This section describes the statistical an alysis, based on the conceptual framework (Chapter 4) and the empirical specification (Chapter 5). The unit of observation is a

PAGE 93

81 privately insured non-elderly person who had a surgical procedure that could be done in either an outpatient or inpatient setting. Fo r patients having such a surgical procedure, the models test the following hypotheses: HMO patients were more likely to have ou tpatient surgery than non-HMO patients, when controlling for patient, phys ician, and payer characteristics. Patients enrolled in a gatekeeper plan were more likely to have outpatient surgery than patients with a non-gatekeeper plan. When excluding surgeries that are done prim arily in one setting and only occasionally done in the other setting, HMO status has a stronger effect on the likelihood of having an outpatient surgery than for surgeries in general. The analysis uses logistic regression because the dependent variable, the choice of outpatient surgery or inpatient surgery, has tw o outcomes. In this analysis, outpatient surgery is set to “1” while inpatient surger y is “0”. For binary dependent variables, logistic regression predicts the probability of one outcome [()ix] assuming that the cumulative density function of th e error term is logistic. Th e specification of the logit regression is: () log[()]log() 1()i i ix itx x =17 1 ii ix (6-1)where the probability of INOUT=1 is () ()() 1ii iix i xe x e (6-2) is the constant is the error term, and i is the parameter of the ith independent variable (ix). To test the effect of HMO enrollment status (HMO) and the effect of gatekeeper enrollment status (GATE), the adjusted Wald statistic is used to test the null hypothesis that equals to zero. It is e xpected that the null hypothesis will be rejected. In other

PAGE 94

82 words, the expectation is that the effect of HMO enrollment status will be significant and positive. Sample Size Sample size is an important issue in statistical analysis. Based on maximum likelihood estimation, i is consistent, efficient and asym ptotically normal if sample size is large enough. Small sample size can lead to two problems (Peduzzi et al. 1996). First, estimated i’s can be biased. Second, sample variance of the estimated i can be too large, resulting in false insignificance. To avoid these problems, the general guidelines for sample size are: At least 10 events per variab le (EPV) (Peduzzi et al. 1996; Long 1997). For example, the EPV of the dataset used in this dissertation is computed as: E: the smaller number of events betw een outpatient surgical events (554 cases) a nd inpatient surgical events (260 cases); V: number of independent variables, (1221,,, xxx ); EPV= E V = 260 21 =12.4 Total sample size should be at least 100 (Long 1997; Cohen et al. 1999). However, if there is little variation in the dependent variable, e.g., almost all outcomes are “1,” a larger total sample size is needed. If the independent variables are highly coll inear, a large sample size is required for better estimation. Since the EPV is greater than 10, total samp le size is 814, and the number of events for each value of the dependent variable, i.e., “1” and “0,” have some variation, (554 and 260 respectively), the sample size is sufficient.

PAGE 95

83 Analytical Issues Other analytical issues in clude possible selection bias unobserved factors, and collinearity between independent variables. Self-selection Self-selection may confound the effect of type of health plan on the likelihood of receiving an outpatient surgery (Figure 6-4). People with certain characteristics, such as being younger or healthier, may prefer HMO plans to non-HMO plan, and younger or healthier patients may be more likely to receive an outpatient surgery. In such a situation, the estimated coefficient for HMO will be biased upward. A two-stage estimation method will be used to account fo r self-selecting into an HMO (or non-HMO). The estimator from the first stage predicts th e probability of choosing a health plan given a person’s characteristics. The first stage estimator will then be incorporated into the error term of the second stag e equation, which is the equa tion estimating the probability of having an outpatient surgery. Figure 6-4. The Possible Self-Selection Effect. Unobserved variables Three relevant variables: phys ician specialty, payment method (capitation vs. FFS), and the extent of utilization management are not available for this analysis. As shown in Figure 6-5, HMOs may use different specialtie s than non-HMO plans, which would have an unobservable effect “b.” In turn, phys icians of different specialties may make HMO Having an outpatient surgery Patients self-select into HMO + + Biased toward

PAGE 96

84 different choices between outpatient and inpati ent surgery, and this “c” effect is also not observed. Having an HMO as the payer ha s effect “a” on the likelihood of receiving outpatient surgery. Given that both “b” and “c” are not observed, “a” can still be unbiased in terms of measuring the total effect, but partial effects, “b ” and “c,” will not be identified. This omission may result in fa lse insignificance of the primary independent variable (i.e., having HMO as the payer) (Greene 2000). Figure 6-5. The Effect of Unobser ved Variable: Surgeon’s Specialty Figure 6-6. The Effect of Unobserved Vari ables: Payment Method and Utilization Management A similar situation holds for the unobser ved payment method and the extent of utilization management (Figure 6-6). As be fore, the effect of having HMO coverage on Surgeon’s specialty HMOOutpatient surgery a b c Payment Method HMOOutpatient surgery a b c Utilization managementHMOOutpatient surgery a b c

PAGE 97

85 the likelihood of receiving outpat ient surgery is the total eff ect, “a,” but the partial effects “b” and “c” will be missed. Unfortunately, th ese variables are not available for the analysis, and therefore are omitted from the analysis. Collinearity between independent variables In this study, multicollinearity is another important issue. Independent variables may be highly correlated, i.e., HLTH and CON, INCOME and EDU. For example, previous studies have found that self-perceiv ed health status a nd clinically measured health status have a para llel predictive power (Mossey and Shapiro 1982; Idler and Benyamini 1997). Correlation between each pa ir of independent variables will be examined. If collinearity is found to be a significant problem, the final model may omit certain variables. Statistical software STATA version 7.0 is used for the statistic al analysis. STATA’s survey estimates can appropriately adjust for complex sample design and yield robust variance estimates (Cohen 1997; Cohen et al. 1999). Plan of analysis All analyses are performed on both the c onstructed dataset (all 814 surgical cases) and a subset of cases (391 cases) as defined in Table 6-2. First, inpatient/outpatient surgery cases are tabulated for each of the four possible values of the primary independent variable, PLAN. STATA uses the usual Pearson statistic for two-way tables to assess whether the distribution of inpatient/outpatient su rgery cases depends on health plan type. To account for the survey design, the Pearson statistic is transformed into an F statistic with non-integer degr ees of freedom using a second-order Rao and Scott correction (StataCorp 2001).

PAGE 98

86 Means and standard errors are calculat ed for charge and payment variables, including total charge total payment, physician payment, and patient out-of-pocket payment. Statistical significance of the diffe rence between health plan types in the same surgery setting (i.e., within outpatient settings or within inpatient setting) is tested. Univariate analyses are performed to estimate the magnitude and significance of the one-to-one correlation betw een one independent variable and the dependent variable, having an outpatient surgery. One logistic regression model is first fitted for each independent variable (uni variate analysis), and estimates a coefficient ( ) for the independent variable as shown in formula 6-1. Based on the preliminary findings, multivariate regression fits multiple variables into the model to estimate th e effects of health plan ty pe on the likelihood of having an outpatient surgery, holding other factors c onstant. Statistical significance of the coefficients is based on an adjusted Wald te st, and p-values are in dicated. Odds ratios (ORs) are reported.

PAGE 99

87 CHAPTER 7 FINDINGS This dissertation examines the effect of HMO and gatekeeper health plan coverage on the choice between outpatient and inpatient surgery. Using a cons tructed dataset from the pooled 1997, 1998, and 1999 MEPS public use data files, analyses are performed to answer the research questions: 1. For a patient under age 65 who is diagnosed to undergo a surgical procedure that is feasible in either outpatient or inpatient settings, after control ling for severity, does an HMO patient have a higher likelihood of receiving an outpa tient surgery (vs. inpatient) than a non-HMO patient? 2. As a widely used managed care approach, do es using a gatekeeper affect the choice of outpatient or inpatient surgery? Are pa tients with gatekeeper plans more likely to have an outpatient surgery? 3. When excluding surgeries that are done prim arily in one setting and only occasionally done in the other setting, does HMO status have a stronger effect on the likelihood of having an outpatient surgery th an for surgeries in general? To study the third research question, a s ubset of the data was constructed by excluding surgeries done primarily in one setti ng (inpatient or outpatient). The subset of data contains cases of surgeries having re ported inpatient cases between 0.2 and 5.0 times of that of outpatient cases (I/O ratios between 0.2 and 5.0) in the 1996 National Health Care Surveys (see Table 6-2). Contingency Tables The results come from several types of an alyses. First, the distribution of nominal variables is tabulated in two-way tables by type of hea lth plan (Tables 7-1 through 7-4). Three-way tables then present the distribu tion of selected cont rol variables for each

PAGE 100

88 surgery setting by health plan type (Table 7-5 and Table 7-6). Next, mean charges and payments for outpatient and inpatient surgery by health plan type are analyzed (Table 77 and Table 7-8). Univariate regression an alyses assess the co rrelation between each dependent variable and each independent vari able (Table 7-9 and Table 7-10). Finally, results are presented from multivariable re gression analyses (Tables 7-11 through 7-17). Two-way Tables: Surgery Setting by Health Plan Type Table 7-1. Weighted Number (Percent) of Ou tpatient and Inpatient Surgical Cases by Health Plan Coverage (Based on Unweighted n=814) Number of Cases (Percent) Outpatient SurgeryInpatient SurgeryTotal One HMO Plan3,094,861 (70.2%)1,311,847 (29.8%)4,406,708(100.0%) One Non-HMO Gatekeeper Plan 445,918 (77.3%) 131,226 (22.7%) 577,144(100.0%) One Non-HMO Non-Gatekeeper Plan 2,627,267 (71.5%)1,044,483 (28.5%)3,671,750(100.0%) Two Plans 603,135 (64.2%) 336,921 (35.8%) 940,056(100.0%) p value 0.5532 One Plan6,168,046 (71.3%)2,487,556 (28.7%)8,655,602(100.0%) Two Plans 603,135 (64.2%) 336,921 (35.8%) 940,056(100.0%) p value 0.2669 Note: Weighted number of cases were the product of unweighted number of cases and sample weights. Table 7-2. Unweighted Number (Percent) of Outpatient and Inpatient Surgical Cases by Health Plan Coverage (Based on Unweighted n=814) Number of Cases (Percent) Outpatient SurgeryInpatient SurgeryTotal One-HMO Plan249 (66.0%)128 (34.0%)377 (100.0%) One Non-HMO Gatekeeper Plan 36 (73.5%) 13 (26.5%) 49 (100.0%) One Non-HMO Non-Gatekeeper Plan 225 (71.0%) 92 (29.0%)317 (100.0%) Two Plans 44 (62.0%) 27 (38.0%) 71 (100.0%) p value 0.2831 One Plan510 (68.6%)233 (31.4%)743 (100.0%) Two Plans 44 (62.0%) 27 (38.0%) 71 (100.0%) p value 0.2502

PAGE 101

89 Table 7-1 and Table 7-2 show the distribut ion of weighted and unweighted cases in each setting by health plan type. A Pearson chisquare test indicates that the setting of a surgery was not significantly associated with health plan type. As shown in Table 7-1, 70.2% of weighted HMO cases were outpatie nt, and 71.6% of cases covered by a nonHMO non-gatekeeper plan were outpatient su rgeries. For non-HMO gatekeeper plan enrollees, outpatient surgeries accounted for 77.3% of all cases. Surgical cases covered by two health insurance plans had a lower pe rcent of outpatient cases (64.2%). However, the difference was not statistically significan t, probably due to the small number of cases, (as indicated in Table 7-2, only 71 cases had two health plans and 49 cases were covered by a non-HMO gatekeeper plan). Table 7-3. Weighted Outpatient and Inpatient Surgical Cases of the Subset of Data by Health Plan Coverage (Based on Unweighted n=391) Number of Cases (Percent) Outpatient SurgeryInpatient SurgeryTotal One HMO Plan1,206,498 (55.7%) 959,882 (44.3%)2,166,381(100.0%) One Non-HMO Gatekeeper Plan 79,551 (51.5%) 74,921 (48.5%) 154,472(100.0%) One Non-HMO Non-Gatekeeper Plan 763,159 (50.5%) 747,044 (49.5%)1,510,202(100.0%) Two Plans 267,569 (49.1%) 277,515 (50.9%) 545,084(100.0%) p value 0.8444 One Plan 2,049,208 (53.5%)1,781,847 (46.5%)2,059,362(100.0%) Two Plans 267,569 (49.1%) 277,515 (50.9%) 545,084(100.0%) p value 0.6443 Note: The subset of data contains cases of surgical procedures with an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Survey (see Table 6-2) Weighted number of cases is the product of unweighted number of cases and sample weights.Among the subset of cases that had a reported ratio of inpatient to outpatient cases between 0.2 and 5.0, the percent of HMO cases th at were outpatient was larger than nonHMO non-gatekeeper cases, although the differe nces were not statistically significant (Table 7-3). A majority of surgical cases covered by an HMO plan (55.7%) were

PAGE 102

90 outpatient while 50.5% of cases covered by a non-HMO non-gatekeeper plan were outpatient. As shown in Table 7-3, 49.1% of surgical cases that were covered by two health plans were outpatient. Among non-HM O gatekeeper cases, 51.5% were outpatient surgeries, but this percent was based on 15 unweighted cases (Table 7-4). Because too few cases were covered by one non-HMO gatekeep er plan, in this subset of data, health plan types are grouped into three cate gories (HMO, non-HMO, and two plans). Table 7-4. Unweighted Outpatient and Inpatient Surgical Case s of the Subset of Data by Health Plan Coverage Number of Cases (Percent) Outpatient SurgeryInpatient SurgeryTOTAL One HMO Plan 91 (47.9%) 99 (52.1%)190 (100.0%) One Non-HMO Gatekeeper Plan 7 (46.7%) 8 (53.3%) 15 (100.0%) One Non-HMO Non-Gatekeeper Plan 71 (50.0%) 71 (50.0%)142 (100.0%) Two Plans 19 (43.2%) 25 (56.8%) 44 (100.0%) p value 0.8859 One Plan169 (48.7%)178 (51.3%)347 (100.0%) Two Plans 19 (43.2%) 25 (56.8%) 44 (100.0%) p value 0.4908 Note: The subset of data contains cases of surgical procedures with an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Survey (see Table 6-2).Three-way Tables for Control Variables: Surgery Setting by Health Plan Type Within a surgery setting (outpatient or i npatient), two-way tabl es (health plan type by control variable) provide profiles of the ca ses (Table 7-5 and Table 7-6). Within each surgery setting, the Pearson statistic for two-way tables is used to test for a significant association between health plan type and th e control variables. Among all cases (n=814), at or four variables (MSA, REGION, RA CE, and AGE) had a statistically significant association with health plan type, especial ly among outpatient cases, while MSA, RACE, AGE, and INCOME were signi ficant variables for the subset of data

PAGE 103

91 (n=391). The percent distribution for all vari ables of all possible values represents the profiles of surgery cases of a health plan type (HMO, Non-HMO gatekeeper plan, nonHMO non-Gatekeeper plan, and two-plan coverage). Table 7-5 shows selected characteristics of all surger y cases (n=814), and the profil es of the subset of cases (n=391) are presented in Table 7-6. All cases For outpatient surgical cases, the di stribution of MSA and REGION varied significantly among the four health plan types (Table 7-5). In par ticular, approximately 89% of HMO cases were living in an MSA, while 71.6% of non-HMO non-Gatekeeper cases lived inside an MSA. Among those c overed by two health plans, 89.9% lived in an MSA. At 0.05, geographic region (REGION) for out patient cases also significantly differed by health plan type. More HMO outpatient surgery cases were from the Northeast (26.4%) than ot her health plan types, while 36.5% of non-HMO nongatekeeper cases were from the Midwest. Regardless of health plan type, many outpatient surgery cases were from the South; in particular, 59.1% of non-HMO gatekeeper cases were from this region. While most cases were white, the com position of race/ethnici ty (RACE) varied significantly by health plan type. Almost 82% of HMO patients were white, while over 90.0% of outpatient cases of the other three health plan types were white. Inpatient cases of the four health plan types also showed significant variation in their race/ethnicity composition. When grouped into three age categories, surgery cases covered by an HMO were the youngest, while cases covered by two plans we re the oldest. In fact, mean age for

PAGE 104

92 outpatient cases covered by two plans was 43.3, while mean age of HMO outpatient cases was 34.9. Mean ages for inpatient cases of HMO, non-HMO gatekeeper plan, non-HMO non-gatekeeper plan, and two-plan cove rage were 39.1, 42.4, 40.7, and 47.0, respectively. Number of medical conditions and health status were not sign ificantly associated with health plan type in either surgery se tting. Over 80% of the cases reported zero or one medical condition (Table 7-5). Because only a few cases had two or more medical conditions, and small sample size might lead to statistical insignificance, additional tests that compare two proportions (the cases with zero or one medical condition vs. those reporting two or more medical conditions) ar e reported in Table 7-5. A comparison of two proportions, excellent health status vs. less than excellent health status, is also performed. Neither showed a statistically significant associati on among different health plan types in eith er surgery setting. The subset of cases The subset of cases that had a reported ratio of inpatient to outp atient cases between 0.2 and 5.0 presented similar profiles as those for all surgery cases (Table 7-6). MSA remained significant in the outpatient sett ing among different health plans. HMO outpatient cases had more non-whites or Hi spanics than non-HMO cases did, while almost all outpatient cases covered by two plans were whites (98.1%). Non-HMO cases were older than cases with HMO coverage in both inpatient and outpatient settings. Outpatient surgical cases covered by one HM O had slightly higher incomes than those covered by one non-HMO plan. Over 87% of the outpatient cases covered by two health plans had 400% FPL or more annual income. Family income of outpatient surgery cases was significantly different among health plan types.

PAGE 105

93Table 7-5. Distribution of Selected Control Variables by Surgical Setti ng and Health Plan Type (n=814) Percent Distribution of Weighted Cases Outpatient SurgeryInpatient Surgery Control Variable One HMO Plan (na=249) One Non-HMOGatekeeperPlan (na=36) One Non-HMO NonGatekeeper Plan (na=225) Two Plans (na=44) One HMO Plan (na=128) One Non-HMOGatekeeperPlan (na=13) One Non-HMO NonGatekeeper Plan (na=92)Two Plans (na=27) 0-1 92.1% 98.8% 92.7% 88.2% 80.8% 82.3% 83.1% 85.9% 2 6.7% 1.2% 7.1% 8.5% 15.2% 17.7% 13.2% 12.2% 3-4 1.2% 0.0% 0.2% 3.3% 4.1% 0.0% 3.6% 1.9% TOTAL100.0% 100.0% 100.0%100.0%100.0%100.0%100.0%100.0% Number of Conditions (COND) p value 0.65160.8613 Yes 92.1% 98.8% 92.7% 88.2% 80.8% 82.3% 83.1% 85.9% No (2 or More) 7.9% 1.2% 7.3% 11.8% 19.2% 17.7% 16.9% 14.1% TOTAL100.0% 100.0% 100.0%100.0%100.0%100.0%100.0%100.0% 0-1 Condition p value 0.44090.9493 Excellent 28.8% 21.6% 29.5% 26.0% 22.6% 46.8% 19.3% 22.3% Very Good 30.1% 30.0% 33.4% 28.6% 34.9% 23.8% 34.7% 27.4% Good 28.0% 32.6% 24.7% 27.4% 27.9% 22.1% 28.6% 26.8% Fair 10.5% 13.1% 10.3% 10.9% 9.5% 7.3% 13.7% 15.3% Poor 2.6% 2.6% 2.1% 7.1% 5.1% 0.0% 3.8% 8.3% TOTAL100.0% 100.0% 100.0%100.0%100.0%100.0%100.0%100.0% Health Status (HLTH) p value 0.96810.9387 Yes 28.8% 21.6% 29.5% 26.0% 22.6% 46.8% 19.3% 22.3% No 71.2% 78.4% 70.5% 74.0% 77.4% 53.2% 80.7% 77.7% TOTAL100.0% 100.0% 100.0%100.0%100.0%100.0%100.0%100.0% Excellent Health Status p value 0.84890.4038

PAGE 106

94Table 7-5. Continued Percent Distribution of Wei g hted Cases Out p atient Sur g er y In p atient Sur g er y Control Variable One HMO Plan(na=249)One Non-HMO Gatekeeper Plan (na=36) One Non-HMO NonGatekeeper Plan (na=225) Two Plans (na=44) One HMO Plan (na=128) One Non-HMOGatekeeperPlan (na=13) One Non-HMO NonGatekeeper Plan (na=92)Two Plans (na=27) MSA 89.0% 83.3% 71.6% 89.9% 81.2% 97.4% 77.4% 93.6% Non-MSA 11.0% 16.7% 28.4% 10.1% 18.8% 2.6% 22.6% 6.4% TOTAL100.0%100.0%100.0%100.0%100.0%100.0%100.0%100.0% MSA p value 0.00010.1492 Northeast 26.4% 7.7% 15.3% 19.2% 25.5% 5.7% 17.7% 13.1% Midwest 19.3% 23.8% 36.5% 18.3% 19.3% 56.9% 40.3% 27.2% South 39.7% 59.1% 33.8% 38.2% 39.0% 16.8% 30.6% 45.2% West 14.6% 9.5% 14.4% 24.3% 16.2% 2.1% 11.4% 14.4% TOTAL100.0%100.0%100.0%100.0%100.0%100.0%100.0%100.0% Region p value 0.01610.2461 NonHispanic-White 81.9% 92.1% 90.3% 96.9% 79.2% 59.6% 89.5% 89.9% Non-White or Hispanic 18.1% 7.9% 9.7% 3.1% 20.8% 40.4% 10.5% 10.1% TOTAL100.0%100.0%100.0%100.0%100.0%100.0%100.0%100.0% Race p value 0.00280.0354 0-17 19.5% 10.3% 16.5% 6.9% 8.0% 0.0% 10.7% 5.2% 18-54 71.3% 78.0% 68.5% 61.7% 76.8% 80.2% 64.0% 52.6% 55-64 9.3% 11.7% 15.0% 31.4% 15.3% 19.8% 25.3% 42.2% TOTAL100.0%100.0%100.0%100.0%100.0%100.0%100.0%100.0% Mean (y ear ) 34.938.536.843.339.142.440.747.0 Age Group p value 0.02080.1814 a unweighted number of cases.

PAGE 107

95Table 7-6. Distribution of Select ed Control Variables by Surgical Setting and He alth Plan Type, the S ubset of Cases (n=391) Percent Distribution of Weighted Cases OutpatientInpatient Control Variables One-HMO (na=91) One Non-HMO (na=78) Two Plans (na=19) One-HMO (na=99) One Non-HMO (na=79) Two Plans (na=25) 0-1 91.0% 91.0% 76.9% 81.8% 83.4% 82.8% 2 8.6% 8.4% 15.6% 16.1% 14.4% 14.8% 3-4 0.4% 0.6% 7.5% 2.1% 2.2% 2.4% TOTAL100.0% 100.0%100.0%100.0% 100.0% 100.0% Number ofConditions p value 0.20030.9480 Yes 91.0% 91.0% 76.9% 81.8% 83.4% 82.8% No ( 2 or More ) 9.0% 9.0% 23.1% 18.2% 16.6% 17.2% TOTAL100.0% 100.0%100.0%100.0%100.0% 100.0% 0-1 Condition p value 0.32800.9755 Excellent 24.3% 28.0% 21.4% 20.5% 18.2% 20.7% Ver y Good 40.9% 32.5% 18.7% 36.9% 37.9% 33.2% Good 26.4% 26.5% 45.1% 28.5% 3.8% 17.5% Fair 6.9% 10.7% 8.3% 9.9% 15.4% 18.5% Poor 1.6% 2.4% 6.5% 4.2% 4.8% 10.1% TOTAL100.0% 100.0%100.0%100.0%100.0% 100.0% Health Status p value 0.70920.9223 Yes 24.3% 28.0% 21.4% 20.5% 18.2% 20.7% No, Less Than Excellent 75.7% 72.0% 78.6% 79.5% 81.8% 79.3% TOTAL100.0% 100.0%100.0%100.0%100.0% 100.0% Excellent Health Status p value 0.87740.9515 MSA 85.9% 67.5% 95.4% 79.5% 75.6% 92.2% Non-MSA 14.1% 32.5% 4.6% 20.5% 24.4% 7.8% TOTAL100.0% 100.0%100.0%100.0%100.0% 100.0% MSA p value 0.00310.2510

PAGE 108

96Table 7-6. Continued Percent Distribution of Weighted Cases Outpatient SurgeryInpatient Surgery Control Variables One-HMO (na=91) One Non-HMO (na=78) Two Plans (na=19) One-HMO (na=99) One Non-HMO (na=79) Two Plans (na=25) Northeast 16.5% 22.0% 24.2% 19.5% 18.8% 15.9% Midwest 24.0% 29.0% 14.9% 18.7% 36.6% 26.7% South 44.3% 34.4% 28.9% 43.1% 34.0% 54.9% West 15.2% 14.6% 31.9% 18.7% 10.5% 2.5% TOTAL100.0% 100.0%100.0% 100.0% 100.0%100.0% Region p value 0.59390.1361 Non-His p White 85.8% 93.5% 98.1% 79.3% 85.7% 87.7% Non-White/His p 14.2% 6.5% 1.9% 20.7% 14.3% 12.3% TOTAL100.0% 100.0%100.0% 100.0% 100.0% 100.0% Race p value 0.07460.4703 0-17 9.4% 7.8% 9.2% 2.3% 2.8% 0.0% 18-54 79.8% 80.8% 41.6% 83.7% 65.4% 63.9% 55-64 10.8% 11.4% 49.2% 14.0% 31.8% 36.1% TOTAL100.0% 100.0%100.0% 100.0% 100.0%100.0% Mean (y ears ) 38.5 40.648.1 41.9 45.4 47.8 Age Group p value 0.02820.0650 < 100% 0.5% 6.3% 0.0% 10.2% 5.1% 2.4% 100-124% 0.8% 2.4% 0.0% 4.8% 7.2% 0.0% 125-199% 6.4% 8.8% 6.3% 9.9% 16.2% 4.8% 200-399% 32.4% 33.4% 6.3% 28.7% 29.8% 42.9% 400% 59.9% 49.2% 87.4% 46.4% 41.8% 50.0% TOTAL100.0% 100.0%100.0% 100.0% 100.0%100.0% Income (%FPL) p value 0.07550.6078 Note: Percent Distribution was based on weighted number of cases. The subset of data contains cases of surgical procedures with an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Survey (see Table 6-2).a unweighted number of cases.

PAGE 109

97 Summary: contingency tables Among health plan types, outpatient case s had different proportions of patients living in an MSA and locating in each geogr aphic location (REGION). Age and race also varied significantly among surgery cases of di fferent health plan types. However, the severity of cases in each surgery setting did not appear to differ by he alth plan type. For the subset of cases that had a reported ratio of inpatient to outpatient cases between 0.2 and 5.0 (n=391), their profiles were si milar to that of all cases (n=814). Mean Charge and Payment by Surgery Setting and by Health Plan Type Tables 7-7 and 7-8 present the means of charge and payment variables in each setting (outpatient or inpatient) by health pl an type. Facility payment is also reported since total payment is the sum of physician payment, patient out-of-pocket payment, and facility payment. All cases HMO cases were charged and paid the least among all health plan types (Table 77). Using an adjusted Wald test, total payment was significantly different among health plan types in both settings. On the other hand, total charge, included as a control variable for severity, was not significantly different among health plan types in the inpatient setting. Like total payment, mean facility payment differed significantly in both settings. Interestingly, outpatient cases covered by two plans paid considerably higher facility costs than cases cove red by one plan only. However, inpatient surgical cases with two-plan coverage did not cost much highe r than cases covered by other health plan types. These findings on the payment for an inpatient surgery might be due to the PPS that pays hospitals a fixed fee based on the specified DRGs. Additionally, the cases covered by two plans were a mixture of HMOs, gatekeeper plans, and non-HMO non-

PAGE 110

98 gatekeeper plans (see Figure 63). Therefore, the mean payment for an inpatient surgery for cases covered by two plans was averag ed out and shows a mid-dollar amount among the four health plan types. Patient out-of-pocket payment tended to va ry by plan type and by number of plans, as did mean physician payment. However, if physician payment were based on the resource-based relative value scale (RBRVS) physicians would be paid similarly even when a case was covered by two health pl ans, although the patie nt’s out-of-pocket payment could vary if part of the patient's payment were covered by the second plan. The subset of cases Among the subset of cases that had a report ed ratio of inpatient for outpatient cases between 0.2 and 5, payment for a surgery in both settings also showed significant variation among the three health plan types (Table7-8). However, total charge to an inpatient surgery did not differ significantly among health plan types, and neither did patient out-of-pocket payment for an outpatient surgery. Summary: mean charge and payment In summary, mean total payment and facility payment was significantly different among health plan types and in bo th surgery settings. Patient payment also varied in that, when covered by two health plans, patients pa id less out of their pocket, but physician payment was similar among plan types. Fo r the subset of cases, mean charge and payment were different among health plan t ypes, but patient’s out-of-pocket payment for an outpatient surgery was relatively similar.

PAGE 111

99 Table 7-7. Mean Charges and Payment to Su rgeries by Health Plan Type, All Cases (n=814) Outpatient SurgeryInpatient Surgery Charge/ PaymentHealth Plan Type MeanSEMeanSE One HMO Plan$3,927.59187.29$10,156.82 902.20 One Non-HMO Gatekeeper Plan $4,666.04767.55$14,251.142,979.85 One Non-HMO Non-Gatekeeper Plan $4,926.84281.82$11,656.321,111.87 Two Plans$6,118.56595.19$11,015.181,452.05 Total Chargep value0.01110.3198 One HMO Plan$2,132.53113.46$5,227.02 554.83 One Non-HMO Gatekeeper Plan $2,625.30510.03$7,373.851,252.00 One Non-HMO Non-Gatekeeper Plan $2,974.89179.16$8,224.59 933.77 Two Plans$3,642.15489.25$6,725.77 942.42 Total Paymentp value0.00000.0198 One HMO Plan$1,383.70 82.29$3,960.11 466.03 One Non-HMO Gatekeeper Plan $1,756.19454.89$5,305.43 909.96 One Non-HMO Non-Gatekeeper Plan $1,773.12135.88$6,238.47 856.54 Two Plans$2,746.32432.67$4,961.91 833.02 Facility Paymentp value0.00130.0596 One HMO Plan $691.64 51.74$1,153.33155.33 One Non-HMO Gatekeeper Plan $760.18156.34$2,039.01811.78 One Non-HMO Non-Gatekeeper Plan $1,003.87 90.27$1,589.82212.59 Two Plans $752.14138.36$1,589.53249.50 Physician Paymentp value0.01430.1515 One HMO Plan $57.20 10.32 $113.57 29.51 One Non-HMO Gatekeeper Plan $108.92 36.96 $29.41 18.65 One Non-HMO Non-Gatekeeper Plan $197.90 26.74 $396.30 76.76 Two Plans $143.69 55.69 $174.34 82.07 Patient Out-ofPocket Paymentp value0.00000.0034 Note: Mean charge and payments were based on weighted number of cases.

PAGE 112

100 Table 7-8. Mean Charges and Payment by HMO, the Subset of Cases (n=391) OutpatientInpatient Charge/ Payment Health Plan Type MeanSEMeanSE One HMO Plan$3,816.01 291.13$10,065.491,044.46 One Non-HMO$4,561.96 302.90$13,342.461,294.11 Two Plans$5,527.081,003.43$12,079.681,586.22 Total Chargep value0.07080.1085 One HMO Plan$2,080.02 175.06 $5,364.94 659.54 One Non-HMO$3,036.70 303.91 $9,233.491,046.69 Two Plans$3,392.62 716.89 $7,392.161,273.78 Total Paymentp value0.00630.0067 One HMO Plan$1,392.29 158.25$4,004.49 550.88 One Non-HMO$1,987.03 249.83$7,017.25 996.87 Two Plans$2,762.81 591.59$5,388.641,122.90 Facility Paymentp value0.01360.0362 One HMO Plan $631.42 65.01$1,252.16 187.49 One Non-HMO $901.76 97.05$1,847.48 257.61 Two Plans $548.70 189.18$1,792.62 247.48 Physician Paymentp value0.07260.0862 One HMO Plan $56.31 17.35 $108.30 24.16 One Non-HMO $147.91 38.41 $368.77 94.08 Two Plans $81.11 40.64 $210.90 83.16 Patient’s Out-of-Pocket Paymentp value0.10820.0278 Note: Mean charge and payments were based on weighted number of cases. The subset of data contains cases of surgical procedures with an I/O ratio betw een 0.2 and 5.0 in the 1996 National Health Care Survey (see Table 6-2).Logistic Regression Analysis Table 7-9 and Table 7-10 present the estim ated coefficients based on univariate regression analysis. Multivariate regression analysis then fits a best model that estimates the effect of the primary independent va riable, health plan type, on the dependent variable (INOUT), when controlling for othe r independent variables (Tables 7-12 through 7-17). Univariate Regression Analysis Univariate regression analyses estimate co efficients that represent the one-to-one correlation between the dependent variable (I NOUT) and an independent variable. Table

PAGE 113

101 7-9 lists estimated coefficients for nominal variables, and Table 710 contains estimates for ordinal and continuous variab les. In the univariate analyses, the primary independent variable (PLAN) did not have a statistically significant effect on the choice of having surgery in an outpatient setting. However, nine control variab les were significantly associated with the dependent variab le (having an outpatient surgery). All cases As shown in Table 7-9, among all cases (n=814), family support (SUPPORT) and year (YEAR) were significantly associated with the dependent variable. Year, including two dummies (YEAR98, and YEAR99), was associated with having an outpatient surgery. Compared to 1997 cases, surgery cases in 1998 were more likely to be outpatient. Having family support (SU PPORT) also had a positive effect at 0.1. Married adults living with a s pouse or children in a two-parent family were more likely to have an outpatient surgery than an inpatient surgery. Two of the variables used to contro l for the severity, COND and TCH, were negatively associated with th e likelihood of having an outpa tient surgery (Table 7-10). The number of conditions (COND) was negativel y associated with having an outpatient surgery, indicating that patien ts having more conditions at the time of surgery were less likely to have an outpatient surgery. A highe r charge (TCH), assumed to indicate a more severe case, also was negatively associ ated with having an outpatient surgery. Five ordinal and continuous variables al so were significantly associated with outpatient surgery (Table 7-10). Patients having higher annual income (INCOME) were more likely to have an outpatient surgery, ra ther than an inpatient surgery. At the .05 confidence level, age negatively affected the odds of having an outpatient surgery,

PAGE 114

102 meaning that older patients were less likely to have an outpatient surgery. Three payment variables also had negative effects on th e likelihood of having an outpatient surgery, including total payment (TOTAL), physician payment (PAYMD), and patient's out-ofpocket payment (PTPAY). Table 7-9. Univariate Regression Analysis: Nominal Variables All CasesThe Subset of Casesa Dependent Variable: Having an Outpatient Surgery (y) Logit y= ix + Independent Variable (ix)Coef ( p value Coef ( ) p value One HMO Plan -0.0641 0.75800.20370.4560 One Non-HMO Gatekeeper Plan 0.3008 0.4480-----Two Plans -0.3401 0.2840-0.06140.8810 _cons 0.9224 0.0000 0.02490.8940 Non-Hispanic White 0.3030 0.1690 0.61250.0800 _cons 0.6157 0.0020-0.41370.1800 Male 0.2741 0.1180 0.55420.0200 _cons 0.7586 0.0000-0.09860.5010 Having Family Support 0.4239 0.0580 0.16330.5730 _cons 0.5735 0.0060 0.00030.9990 Living in MSA-0.0048 0.9840 0.04170.8890 _cons 0.8783 0.0000 0.08440.7340 Northeast-0.0792 0.7260 0.08870.7940 Midwest-0.2082 0.3390 -0.02910.9330 West-0.0394 0.8840 0.29550.5040 _cons 0.9544 0.0000 0.06400.7440 YEAR98 0.3711 0.0410 0.54610.0510 YEAR99 0.0994 0.5900-0.04750.8700 _cons 0.6227 0.0000-0.05270.7610 Note: Significant variables are in bold. Regression analysis was based on weighted number of cases.a The subset of data contains cases of surgical procedures with an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Survey (see Table 6-2).The subset of cases The analyses based on the subset of cas es (391 cases) also showed a negative association between charge and payment variables and the likelihood of having an outpatient surgery (Table 7-10). Family annual income and age were significantly

PAGE 115

103 associated with the likelihood of having an outpatient surgery at the .05 confidence level. Males were more likely to have an outpatient surgery compared to female patients. NonHispanic white patients were more likely to have an outpatient surgery than non-whites or Hispanic patients. Table 7-10. Univariate Regression Analys is: Ordinal and Continuous Variables All CasesHigh Variation Casesa Dependent Variable: Having an Outpatient Surgery (y) Logit y= ix + Independent Variable (ix )Coef ( p value Coef ( p value Health Status-0.12340.1580-0.18060.1430 _cons 1.16870.0000 0.54660.1070 Number of Conditions -0.65970.0240 -0.47410.1680 _cons 1.62330.0000 0.66590.1030 Total Charge -0.00020.0000-0.00030.0000 _cons 2.40100.0000 2.23840.0000 Income as % of FPLb 0.26990.0010 0.37620.0000 _cons-0.26230.4740-1.47660.0020 Education 0.06820.6180 0.04390.8170 _cons 0.71010.0250-0.00120.9980 Age -0.01610.0130-0.01980.0200 _cons 1.49950.0000 0.95550.0160 Total Payment -0.00030.0000-0.00040.0000 _cons 2.09380.0000 1.63180.0000 Physician Payment -0.00040.0000-0.00060.0000 _cons 1.26410.0000 0.75740.0000 Out-of-pocket payment -0.00070.0090-0.00120.0270 _cons 0.99050.0000 0.28910.0370 Note: Significant variables are in bold. Regression analysis was based on weighted number of cases.a The subset of data contains cases of surgical proce dures with an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Survey (see Table 6-2).b Federal Poverty LevelMultivariate Regression Analysis Because independent variables are usuall y associated with one another and may have different distributions between the two cl asses of dependent variable (outpatient and

PAGE 116

104 inpatient), multivariate analysis can better es timate the effect of each explanatory variable on the probability of having an outpatient surgery. Variable selection Based on the findings from contingency tabl es and univariate analysis, all control variables but one (EDU) were significantly associated with the dependent variable. Health plan type did not show a significant association with the su rgery setting (Table 7-1 and Table 7-3), but other health plan char acteristics (TOTAL, FACILITY, MDPAY, and PTPAY) were significantly different between HMO and other health plan types (Tables 7-7 and 7-8). Among outpatient surgery ca ses, Tables 7-5 and 7-6 show that the distribution of two area charac teristics (MSA, REGION) varied among health plan types. Patient characteristics (AGE, SUPPORT, RACE, GENDER, and INCOME) that could potentially affect the use of outpatient surg ery also varied by health plan type. The univariate analysis confirmed the correlation between many of the characteristics and the dependent variable, having an outpatient surgery (INOUT) (Table 7-9 and Table 7-10). Three variables (COND, HLTH, and TCH) were included in the multivariate analysis to control for patien t’s severity. Number of c onditions (COND) and total charge (TCH) demonstrated in the univariate analysis a significant association with the odds of having an outpatient surgery. Health status (HLTH) was not as strongly correlated as COND and TCH with the depe ndent variable. However, controlling for severity is critical, so unless the presen ce of HLTH affects model fit, HLTH should remain in the multivariate analysis. While the three control variables for severity were important in the analysis, assessing the correlation between these variables helps identify possible multicollinearity problems. Figure 7-1 shows that the corr elation between health status (HLTH) and

PAGE 117

105 number of conditions (COND) is not significan t. However, the correlation coefficient between HLTH and TCH is 0.1168 and is 0.1443 for COND and TCH, with both correlation coefficients being significant at =0.01. Figure 7-1 Correlation Matrix for Number of Condition (COND), Health Status (HLTH), and Total Charge (TCH). The correlation between total payment (TOTAL), physician payment (PAYMD), and patient out-of-pocket payment (PTPAY) was significant. Replacing TOTAL with facility payment (FACILITY) turns out to be important because the correlation coefficients between TOTAL and any of othe r payment variables and TCH were big and significant at =0.01. As shown in the correlation matrix below (Figure 7-2), the correlation coefficient of total charge (TCH ) and total payment (TOTAL) is 0.8271 for all cases and 0.8310 for high variation cases (n= 391). Moreover, total payment (TOTAL) has larger correlation coefficients ( ) with physician payment (PAYMD) ( =0.5404) and All cases (n=814) | hlth cond tch -------------+--------------------------hlth | 1.0000 | cond | 0.0389 1.0000 p value | 0.2678 | tch | 0.1168 0.1443 1.0000 p value | 0.0008 0.0000 The subset of Surgical Cases (n=391) | hlth cond tch -------------+--------------------------hlth | 1.0000 | cond | 0.0490 1.0000 p value | 0.3338 | tch | 0.1535 0.1621 1.0000 p value | 0.0023 0.0013

PAGE 118

106 with patient payment (PTPAY) ( =0.1590) than FACILITY does ( =0.2780 and 0.0388, respectively). While the correlation coe fficients between TOTAL and other payment variables and total charge (PAYMD, PTPAY, TCH) are all signifi cant (p value =0.000), FACILITY is significantly correlated with PAYMD and TCH (p value =0.000), but not with PTPAY (p value=0.2692). Consequently, TOTAL is not a suitable control variable, and FACILITY will be used instead. Figure 7-2. Correlation Matrix of Charge and Payment Variables All cases (n=814) | tch total facility mdpay ptpay -------------+--------------------------------------------tch | 1.0000 | total | 0.8271 1.0000 p value | 0.0000 | facility | 0.7538 0.9553 1.0000 p value | 0.0000 0.0000 | paymd | 0.5635 0.5404 0.2780 1.0000 p value | 0.0000 0.0000 0.0000 | ptpay | 0.1120 0.1590 0.0388 0.1729 1.0000 p value | 0.0014 0.0000 0.2692 0.0000 The subset of Surgical Cases (n=391) | tch total facility mdpay ptpay -------------+--------------------------------------------tch | 1.0000 | total | 0.8310 1.0000 p value | 0.0000 | facility | 0.7597 0.9621 1.0000 p value | 0.0000 0.0000 | paymd | 0.5676 0.5232 0.2793 1.0000 p value | 0.0000 0.0000 0.0000 | ptpay | 0.1008 0.1323 0.0447 0.1029 1.0000 p value | 0.0464 0.0088 0.3783 0.0419

PAGE 119

107 Table 7-11. Logistic Regression for All Cases: Main Effect Only Coef.ORP>|t|95% CI (OR) One HMO Plan-0.51030.60030.0690(0.3461 1.0413) One Non-HMO Gatekeeper Plan 0.41421.51310.3490(0.6337 3.6127) Two Plans-0.32450.72290.4710(0.2983 1.7519) Number of Conditions-0.54320.58090.1260(0.2893 1.1662) Health Status0.04541.04650.7010(0.8288 1.3213) Total Charge-0.00020.99980.0000(0.9997 0.9998) MSA0.03551.03610.9040(0.5805 1.8494) Northeast-0.38240.68220.2700(0.3451 1.3489) Midwest-0.34210.71030.2480(0.3970 1.2706) West-0.03940.96140.8970(0.5292 1.7466) Facility Payment-0.00020.99980.0010(0.9997 0.9999) Physician Payment0.00021.00020.0600(1.0000 1.0004) Patient Payment-0.00050.99950.0530(0.9989 1.0000) Age-0.00990.99020.1880(0.9757 1.0049) Race0.32511.38420.2610(0.7838 2.4447) Family Support0.39351.48210.0990(0.9280 2.3672) Gender0.56651.76210.0190(1.0974 2.8294) Income as % FPL0.24261.27450.0400(1.0117 1.6057) Year: 19980.60621.83340.0180(1.1122 3.0223) Year:19990.28031.32360.3380(0.7445 2.3530) Constant1.8405 n*813 Note: The regression results were based on weighted cases. *Unweighted sample size Model fitting for all cases: main effect only Table 7-11 shows the estimated coeffici ents and the corresponding ORs, along with 95% confidence intervals of the estimated ORs. HMO coverage was a significant predictor of the likeli hood of having an outpatient surgery when controlling for all other payer, physician, and patient characteristics. Other paye r characteristics were also significant, including facility payment (FACILITY), physician payment (PAYMD), and patient out-of-pocket payment (PTPAY). Signi ficant patient characteristics were family support (SUPPORT), gender (GENDER), annual income as a percent of federal poverty

PAGE 120

108 level (INCOME), and year (YEAR98). One of the control variables for severity, total charge (TCH), was significant at 0.01. However, the negative coefficient of HMO coverage (One-HMO) indicated that HMO patients were less likely to have an ou tpatient surgery than patients covered by a non-HMO non-gatekeeper plan (Table 7-11). The odds of having an outpatient surgery for an HMO patient were only 60% of that for a non-HMO non-gatekeeper patient. This estimated OR is contrary to expectations a bout the effect of HMO coverage on the choice of outpatient surgery. During model fitting, the payment effect overwhelmed the plan effect (Appendix C). When one of the payment variables was in the model, the estimated coefficient for one-HMO dropped to negative values (m eaning a smaller likelihood of having an outpatient surgery than the reference group). Additionally, the presence of at least one of charge/payment variables improved the fit of the model significantly (Appendix C). Because payment variables had such strong eff ect, and, as indicated in Tables 7-7 and 78, HMO patients were charged significantly less and paid less for a surgery than nonHMO patients, it is possible that multicolline arity might exist between the health plan variable and charge /payment variables. Model fitting for all cases: two-stage method A two-stage estimation method may improve the regression model in Table 7-11. In the two-stage method, HMO status is pred icted in the first st age of the regression model. The second-stage regression then es timates the effect of the predicted HMO status, rather than self-repo rted HMO status, on the like lihood of having an outpatient surgery.

PAGE 121

109 Using the pooled 1997, 1998, and 1999 MEPS data, Table 7-12 shows the regression results of the firs t-stage model. Characteristics found to influence HMO enrollment are included in the model: heal th status, employment-based insurance coverage status, residentia l area (MSA and geographic re gion), age, race/ethnicity, gender, marital status, income, and educa tion (Brown and Langwell 1988; Dowd et al. 1996; Garfinkel et al. 1986; The Kaiser Fa mily Foundation and Health Research and Educational Trust 2001; Lichte nstein et al. 1991). The pr edicted probability of HMO status for HMO enrollees was si gnificantly higher than that for non-HMO enrollees at the level of 0.05 (Table 7-13). Mean probability of HMO status for observed HMO enrollees was 0.5580 and 0.4946 for observed non-HMO enrollees. Table 7-12. Stage One: Logistic Regres sion Results that Predict HMO Status. Coefficient ( ORP value95% CI (OR) Health status 0.01421.01430.4270(0.9793 1.0505) Employ-based Insurance 0.85732.35680.0000(2.0001 2.7772) MSA 0.68411.98200.0000(1.5813 2.4842) Northeast 0.51081.66650.0000(1.3322 2.0847) Midwest-0.22120.80160.0640(0.6342 1.0132) West 0.55241.73740.0000(1.3429 2.2478) Age-0.00790.99210.0000(0.9891 0.9951) Black 0.29731.34620.0000(1.1594 1.5629) Hispanics 0.26221.29980.0060(1.0791 1.5656) Other Races 0.21431.23900.1160(0.9482 1.6190) Marry 0.06091.06280.1870(0.9709 1.1635) Male-0.06710.93510.0080(0.8900 0.9825) Income-0.03000.97040.2030(0.9265 1.0164) Education-0.01170.98830.6880(0.9333 1.0466) Year98 0.10811.11420.0030(1.0372 1.1969) Year99 0.06181.06370.2750(0.9519 1.1887) Constant 1.0826 n31,570

PAGE 122

110 Table 7-13. Mean Predicted Probability of HMO Status by Observed HMO Membership. Estimated Probability Self-reported HMO Membership MEANStandard Error95% CI HMO0.55800.0040(0.5502 0.5659) Non-HMO0.49460.0037(0.4874 0.5018) Based on the second-stage regression, the e ffect of predicted HMO status on the likelihood of having an outpatient surgery wa s even more negative than the estimate using self-reported HMO status as in Table 7-11. Table 714 shows that patients with a high probability of HMO membership were ex tremely less likely to have an outpatient surgery. The two-stage method did not change the unexpected result regarding the effect of HMO coverage on the choice of outpatient surgery. Table 7-14. Stage Two: Logistic Regressi on Results, Using Predicted HMO Membership as the Primary Independent Variable. Coef.ORP>|t|95% CI (OR) Predicted HMO Status-7.44080.00060.0090(0.0000 0.1479) Two Plans-0.82640.43760.0930(0.1668 1.1478) Number of Conditions-0.41000.66370.3330(0.2888 1.5250) Health Status 0.01821.01840.8920(0.7813 1.3274) Total Charge-0.00020.99980.0000(0.9997 0.9998) MSA 1.24223.46310.0140(1.2952 9.2597) Northeast 0.20141.22310.6400(0.5242 2.8540) Midwest-0.68060.50630.1080(0.2204 1.1633) West 0.51531.67420.2670(0.6728 4.1658) Facility Payment-0.00020.99980.0080(0.9996 0.9999) Physician Payment 0.00041.00040.0010(1.0002 1.0007) Patient Payment-0.00060.99940.0680(0.9987 1.0000) Age-0.01150.98860.2410(0.9697 1.0078) Race-0.12800.87990.7340(0.4194 1.8458) Family Support 0.60951.83960.0460(1.0101 3.3504) Gender 0.73572.08690.0090(1.2072 3.6079) Income as % FPL 0.09131.09560.4830(0.8479 1.4157) Year: 1998 0.78192.18550.0140(1.1745 4.0670) Year:1999 0.33461.39740.2930(0.7477 2.6116) Constant 5.2469 n*674 Note: The regression results were based on weighted cases. *Unweighted sample size

PAGE 123

111 Model fitting for all cases: main effect and interaction Another way to elucidate multicollinearity is to add interaction terms between correlated variables. For this analysis, th e interaction between HM O status and facility payment was the only significant interacti on (Appendix C). Table 7-15 shows a model with both a main effect and an interaction term between HMO and facility payment (One_HMO_Facility). Given th at mean facility payment of HMO cases was significantly associated with health plan t ype (see Table 7-7), the associati on is reflected in this final model. Table 7-15. Logistic Regression for All Cases: Main Effect and Interaction Coef.ORP>|t|95% CI (OR) One HMO-0.00820.99190.9830(0.4678 2.1030) One_HMO_Facility-0.00020.99980.0410(0.9996 1.0000) One Non-HMO Gatekeeper Plan 0.38371.46770.3740(0.6282 3.4289) Two Plans-0.34340.70940.4190(0.3078 1.6349) Number of Conditions-0.53370.58650.1470(0.2849 1.2072) Health Status0.05151.05290.6570(0.8377 1.3232) Total Charge-0.00020.99980.0000(0.9997 0.9998) MSA0.01461.01470.9610(0.5651 1.8221) Northeast-0.45250.63600.1940(0.3207 1.2614) Midwest-0.34870.70560.2410(0.3930 1.2666) West-0.08050.92260.7940(0.5029 1.6926) Facility Payment-0.00010.99990.0880(0.9998 1.0000) Physician Payment0.00021.00020.0550(1.0000 1.0004) Patient Payment-0.00050.99950.0690(0.9990 1.0000) Age-0.01010.98990.1790(0.9754 1.0047) Race0.39111.47860.1860(0.8272 2.6430) Family Support0.37601.45640.1140(0.9129 2.3235) Gender0.56431.75830.0200(1.0919 2.8315) Income as % FPL0.24081.27220.0430(1.0076 1.6064) Year: 19980.65661.92820.0100(1.1682 3.1827) Year:19990.29521.34340.3150(0.7540 2.3935) Constant1.6295 n*813 Note: The regression results were based on weighted cases. *Unweighted sample size.

PAGE 124

112 Other payer characteristics were also signi ficant in the model, including facility payment (FACILITY), physician payment (PAYMD), and patient out-of-pocket payment (PTPAY). While none of the physician character istics were significant, both variables (MSA, REGION) enhanced the model fit. One of the control variable s for severity, total charge (TCH), was significant. Two of th e patient characterist ics (gender and annual income) had a significant effect on the choice of outpatient surgery. Therefore, the final model can be written as Formula 7-1: ____ _(1) log()0.00820.00020.3837 (0) 0.34340.53370.05150.00020.0146 0.45250.34870.08050.0001oneHMOHMOFacilitynonHMOGate TwoplanCONDHLTHTCHMSA northeastmidwestwestFACILinout xxx inout xxxxx xxxx 98990.0002 0.00050.01010.39110.37600.5643 0.24080.65660.29521.6295ITYPAYMD PTPAYAGERACESUPPORTGENDER INCOMEYERAYEARx xxxxx xxx Multivariate results from the regression model for all cases Because the interaction between HMO status and facility payment is significant, the estimated coefficient for HMO status can no longer be interp reted directly from Formula 7-1. Rather, the effect of facility payment has to be taken into consideration. Therefore, controlling for all other characteristics, the proba bility of having an outpatient surgery can be obtained as Formula 7-2: __ __exp(0.00820.00020.00011.6295) 1exp(0.00820.00020.00011.6295)oneHMOHMOFACILITYFACILITY oneHMOHMOFACILITYFACILITYxxx xxx As indicated in Table 7-7, the mean facili ty payment for an HMO outpatient case was approximately $400 less than that of a non-HM O patient. Using Formula 7-2, Table 7-16 illustrates the comparisons of the probability of having an outpatient surgery by HMO status when facility payment increases by $400.

PAGE 125

113 Table 7-16. Estimated Probability of Ha ving an Outpatient Surgery by HMO Status One HMO (One HMO=1)Non-HMO (One HMO=0) Facility Payment, baseline0.8350.836 Facility Payment, + $4000.8180.831 % change-2.0%-0.6% Controlling for all other f actors, if facility payment remains the same, the probability of having an outpatient surger y (0.835) for HMO patients was almost the same as non-HMO patients’ (0.836). However, when facility payment was increased by $400, the probability of having an outpatient surgery for an HMO patient dropped 2% to 0.818. For a non-HMO patient, the probabil ity only decreased by 0.6% to 0.831. While facility payment had a negative effect on the probability of having an outpatient surgery, the probability for HMO patients was more se nsitive to the changes in facility payment than that of non-HMO patients. The effect of having a gatekeeper was no t a significant factor in affecting the choice of outpatient surgery. Although the estimat ed effect due to gatekeeper plan status could be positive, the estimated coeffi cient for having non-HMO gatekeeper plan coverage was not significant (T able 7-15). However, this result might be partly due to the small sample size of the non-HMO gatekeeper group (n=49). At the significance level of 0.1, two ot her payer characteristics also have a significant effect on the like lihood of having an outpatient surgery. Physician payment had a positive effect on the likelihood of havi ng an outpatient payment. The likelihood increased by 2% for every $100 increase in physician payment, while the likelihood decreased by 5% for every $100 increase in patient out-of-pocket payment. Three patient characteristics, including one controlling for severity (TCH), significantly affected the like lihood of having an outpatient surgery, when controlling for

PAGE 126

114 all other characteristics. Male patients we re 1.758 times as likely to receive an outpatient surgery as female patients. People having hi gher incomes were more likely to have an outpatient surgery. For every level increase in annual income (incl uding five levels: less than 100% FPL, 100 to 124% FPL, 125 to 199% FPL, 200 to 399% FPL, and 400% or more FPL), the likelihood of having an outpatient surgery increased by 27.2%. As expected, high total charge, which might i ndicate a more severe surgery case, was associated with a lower likelihood of having an outpatient surgery. Table 7-17. Logistic Regression for the Subset of Cases: Main Effect and Interaction Coef.ORP>|t|95% CI (OR) One HMO-0.00730.99280.9900 ( 0.3307 2.9800 ) One HMOFacilit y -0.00020.99980.2400 ( 0.9995 1.0001 ) Two Plans-0.23070.79400.6450 ( 0.2956 2.1327 ) Number of Conditions-0.29730.74280.4980 ( 0.3126 1.7653 ) Health Status-0.00160.99840.9910 ( 0.7523 1.3249 ) MSA 0.07611.07900.8070 ( 0.5837 1.9946 ) Northeast-0.04570.95530.9070 ( 0.4414 2.0677 ) Midwest 0.22411.25120.5630 ( 0.5827 2.6865 ) West 0.15741.17050.7110 ( 0.5055 2.7103 ) Facilit y Pa y ment-0.00040.99960.0000 ( 0.9995 0.9998 ) A g e-0.02010.98010.0750 ( 0.9586 1.0020 ) Gender 0.83202.29790.0070 ( 1.2564 4.2027 ) Income as % FPL 0.47491.60790.0030 ( 1.1836 2.1843 ) Year: 1998 0.45781.58060.2220 ( 0.7558 3.3055 ) Year:1999 0.03361.03420.9150 ( 0.5564 1.9223 ) Constant-0.0356 n*391 Note: The subset of data contains cases of surgical procedures with an I/O ratio between 0.2 and 5.0 in the 1996 National Health Care Survey (see last column of Table 6-1). The regression results were based on weighted cases. Unweighted sample size.Year is also significant. As advanced technology became available in the later years, the likelihood of having an outpatient surgery increased. Cases which occurred in 1998 had almost two times (1.928 times) the likelihood of that in 1997, and 1999 cases were 1.343 times as likely as that of 1997 cases.

PAGE 127

115 Model fitting for the subset of cases For the subset of cases, a model with ma in effect and interaction also helps to reduce possible multicollinearity. Like the model fitting for all cases, a model with a main effect only for the subset of cases is problematic (see Appendix D). Because the subset has a smaller sample size than the da taset of all cases, Table 7-17 presents a best model that contains only one of the payment va riables, facility payment. Total charge, physician payment, and patient payment were left out of the model due to multicollinearity. The evidence of multicollinearity between HMO status and payment variables appeared during model fitting when STATA showed a warning message, such as “Note: 2 failures and 0 successes completely determ ined” (see Appendix D). A model with the main effect only tended to have this warn ing message, particularly when one of the payment variables was in the model. Only after adding the inte raction between HMO status and facility payment into the model did the warning message disappear. Table 717 is the best model with as many vari ables as possible but with the least multicollinearity. For the subset of cases, findings from th is regression analysis showed that HMO status did not have a significant effect on the likelihood of having an outpatient surgery. The final regression model also indicated that the estimated ORs were not significant for any health plan type. Four control variables were significant, including facility payment, age, gender, and income. Because the model included the interaction between HMO status and facility payment, the effect of facility pa yment could not be interpreted alone. However, for patients with the same health plan covera ge, facility payment had a negative effect on

PAGE 128

116 the likelihood of having an outpatient surgery afte r controlling for all other factors. Older patients were less likely to receive an outpa tient surgery. For every one-year-older, the likelihood of having an outpatient su rgery dropped by 2%, only 0.9801 times the likelihood. Male patients were more likely to have an outpatient surgery, about 2.2979 times as likely as female patients were. This contrast between male and female rates of outpatient surgery was also observed in th e 1996 National Health Care Survey (Owings 1998). Patients with higher income tended to have an outpatient surgery, rather than inpatient one. For every one level incr ease on income, the likelihood of having an outpatient surgery rose by 60.8%. The goodness of fit STATA uses an adjusted Wald test to assess the model goodness of fit. The Wald statistic has an approximate F-distribution with k (number of independent variables) numerator degrees of freedom and 1 dk denominator degrees of freedom, where d = total number of sampled PSUs minus the total number of strata. (,1) Fkdk Table 7-18. Goodness of Fit of the Final Models Number of Independent Variables ( k ) Number of Strata Number of PSUs 1 dk aF(k,1 dk )p value Table 7-15 (n=814) 21612892088.790.0000 Table 7-17 (n=391) 15611991244.120.0000 a d= Number of PSUs Number of Strata.As indicated in Table 7-18, both fitted regression models (Tables 7-15 and 7-17) have high F statistics and low p values (0.0000) However, for survey data, there is no

PAGE 129

117 consensus on how to test model fit (Hosmer and Lemeshow 2000). Adjusted Wald tests may provide some evidence of model fit, but because multicollinearity exists in this analysis, a larger sample size is st ill needed to improve the estimation. Summary Controlling for all other ch aracteristics, the interaction between HMO status and facility payment significantly affects the likelihood of having an outpatient surgery, while HMO status alone did not have significant e ffect on the choice of outpatient surgery. While facility payment had a negative eff ect on the likelihood of having an outpatient surgery, the effect of facility payment wa s stronger for HMO patients than for non-HMO non-gatekeeper plan patients. For example, when facility payment increased by $400, the probability of having an outpatient surger y for an HMO patient dropped 2%, compared with 0.6% for a non-HMO non-gatekeeper plan pa tient. Based on the findings from this analysis, having a gatekeeper did not aff ect the choice of an outpatient surgery significantly. For the subset of cases, the effect of HMO status and the interaction between HMO status and facility payment were not significant.

PAGE 130

118 CHAPTER 8 DISCUSSION AND CONCLUSION The purpose of this dissertation is to assess the effect of type of health plan on the choice of an outpatient or inpatient surger y for people under age 65 who are diagnosed to undergo a surgery that is feas ible in either setting. Th e research hypotheses of this dissertation are: 1. HMO patients are more likely to have out patient surgery than non-HMO patients, when controlling for patient, phys ician, and payer characteristics. 2. A patient enrolled in a plan with a gateke eper is more likely to have an outpatient surgery, when compared to a patient with a non-gatekeeper plan. 3. When excluding surgeries that are primarily done in one setting and only occasionally done in the other setting, HMO coverage has a stronger effect on the likelihood of choosing an outpatient surgery than for surgeries in general. Based on the 1997, 1998, and 1999 MEPS data, this dissertation found that there was not sufficient evidence to support the research hypotheses. HMO enrollment status did not increase the likelihood of having an outpatient surgery when controlling for patient, physician, and payer characteristics. Patients covered by a gatekeeper health insurance plan were no more likely to have an outpatient surgery than non-gatekeeper patients were. After excluding surgeries that were mostly done in one of the two settings (inpatient or outpatient), HM O status did not show a stronger effect on the use of outpatient surgery. While HMO status alone is not a significant predictor, the interaction between facility payment and HMO status has a signif icant effect on the use of outpatient surgery. The higher the facility payment is, the less likely an outpatien t surgery is chosen. And, as

PAGE 131

119 facility payment increases, the probability of having an outpatient surgery for HMO patients decreases more than that for non-HMO patients. For gatekeeper patients, the likelihood of having an outpatient surgery was not significantly different from non-gatekeeper patients. Previous studies reported that having a gatekeeper alone did not control the use of care; rather, many managed care approaches together reduced HM O enrollees’ use of care (Ferris et al. 2001a; Ferris et al. 2001b; Pati et al. 2003). This dissertation doe s not find significant differences between non-HMO gatekeeper plan enrollees and non-ga tekeeper plan enrollees in the use of outpatient surgery. These findings are adjusted for patient, phys ician, and payer characteristics. Patient characteristics include age, income, and gender. While the location of patient’s residence (MSA and geographic region) is used as a proxy of physician practice environment, this area characteristic is also related to pa tient and payer characteristics. Payer characteristics include facil ity payment, physician payment, and patient’s out-of-pocket payment. Surgical cases were specificall y adjusted for severity, including number of medical conditions associated with a surgery, se lf-reported health stat us, and total charge. These conclusions appear to be inconsistent with pr evious findings on access to expensive care, such as inpatient care, in HMOs. For instance, a study based on seven years of data from the Healthcare Cost a nd Utilization Project (HCUP) found that HMO coverage was a significant factor in re ceiving an outpatient mastectomy. The study controlled for residing state, t ype of payer, clinical charact eristics, and ownership of a hospital. Importantly, this study did not include any payment information.

PAGE 132

120 If the goal of controlling th e use of inpatient surgery is to reduce the costs of care, the findings from this dissertation are consis tent with the general observation that HMOs tend to control the costs of care more than non-HMO plans do. While HMOs combine the functions of financing and delivering hea lth care, HMOs not only manage the use of care but also negotiate lower payment rates to providers. In fact, this dissertation has found that HMOs in general are paying less for a surgery. Because controlling the use of expensive care is not the only approach to reduce the costs of care, findings from this dissertati on provide insight into the effect of HMO status. As public criticisms of the denial of care by some HMOs remain a focus of the media, it is inevitable that HMOs strive to improve their image to attract more enrollment. As market share increases, HM Os have more negotiating power with respect to provider payment rates. One limitation of this analysis is using se lf-reported health plan coverage. The MEPS has indicated that its estimated percenta ge of Americans enrolled in managed care plans is higher than that reported by health pl ans. One of the reasons is that the MEPS data are based on self-reported informati on that may include reporting errors. In addition, the MEPS data on HMO/gatekeeping plans were about the current plan that respondents held. Given that the interviews are conducted three times a year, depending on the time when the information on health plan coverage is provided, the estimates of number of people covered by each health pl an type could be slightly different. Although the accuracy of self -reported HMO/gatekeeping plan information in the MEPS has not been studied, the MEPS asked similar questions about managed care as another study (Blendon et al. 1998). The MEPS respondents who have private insurance

PAGE 133

121 coverage are explicitly asked a set of questions using the term “HMO.” If respondents are not certain that their plan is an HMO, but indicate th eir insurance plan requires them to sign up with a certain primary care doctor, group of doctors, or a cer tain clinic for all of their routine care, they are categorized as gatekeeping plan enro llees. Given that the MEPS collected various information, such as self-reported health plan, payment source, and type of coverage from employers or paye r, using other data to verify self-reported health plan type may have helped to improve the accuracy of self-reported health insurance coverage. Another limitation is the small sample size. Because the MEPS is a populationbased survey, the number of surgical cases is not expected to be as large as the claims data. Since the MEPS is an ongoing survey, pooling multiple years’ data can provide a larger sample size. However, because this dissertation included several variables that led to multicollinearity, such as facility paymen t, physician payment, and patient payment, an even larger sample size than the generally acceptable rule of thumb (10 events per variable) is needed to achi eve sufficient power. As the 2000 or later MEPS data become available, more data can be pooled to improve the estimates. The advantages of using the MEPS data include no costs in acquiring the data, many individual characteristics, previous year ’s data being available from the National Health Interview Survey (NHIS), the capab ility of pooling multiple years of data, payment information, and detailed health insurance and managed care coverage information. Because the MEPS is a nationa l study, findings from this dissertation were nationally representative for the ye ars of 1997, 1998, and 1999. Although the MEPS’ managed care data were not released until Ap ril 2003, the data were accessible from the

PAGE 134

122 MEPS website without costs. Furthermore, th e recently released personal health plan files provided detailed information on type a nd number of health plans. This dissertation has the advantage of using this personal health plan file along with payment information to verify self-reported private health insurance coverage. The MEPS collects many individual characteristics for analyses that are potentially confounded by underlying characteristics. Because the MEPS sample s a portion of respondents from the previous year’s NHIS, additional information in the NHIS can also be used for analyses. As new regulations, competition, and deve lopments in medical technology continue to affect the way health care is delivered, st udies of the use and costs of care can help in monitoring the system. When newer types of managed care organizations are developed, it is expected that the patter n of using certain types of ca re will changed. Consumers’ preferences can also be a driving force in the use of care. Findings from this dissertation suggest that HMOs may have relaxed their co ntrol over the use of inpatient surgery but still manage the overall co sts of surgical care. Future studies can investigate the outcom e of surgeries for HMO patients. The MEPS interviews respondents five times in tw o years. Since the MEPS collects all types of health care use, the use of care after surgery is avai lable. Additionally, respondents report their physical functionali ty and mental health in each interview. While a major concern is the small sample size, as more year’s data become available, pooling multiple years’ data may provide sufficient data to compare surgery outcome by type of health plan.

PAGE 135

123 APPENDIX A TWO-DIGIT ICD-9 PROCEDURE CODES REPORTING BOTH INPATIENT AND OUTPATIENT PROCEDURES IN 1996, AGE 0 TO 64 InpatientOutpatientI/O Ratio ICD Cases (in 1,000) 03Spinal cord and canal operationsa5014181.20 04Operations on cranial and peripheral nerve 394770.08 05Sympathetic nerves, ganglia operationsa 11 440.25 06Thyroid and parathyroid operationsa 63 341.85 08Eyelids operations 271570.17 11Operations on cornea 5 620.08 12Operations on iris, ciliary body, sclera, anterior chamber 9 540.17 13Operations on lens 117100.02 14Operations on retina, chorioid, vitreous, posterior chamber 231480.16 16Orbit/ eyeball operationsa 13 150.87 18External ear operationsa 18 660.27 19Middle ear reconstructure 7 990.07 20Other mid and inner ear operations 346410.05 21Operations on nose 635920.11 22Nasal sinus operations 304200.07 23Tooth removal & restoration 372210.17 24Other operations on teeth & gumsa 8 280.29 25Operations on tonguea 12 230.52 26Operation on salivary gland or ducta 10 220.45 27Other mouth and face operationsa 41 830.49 28Tonsil and adenoid operations 495230.09 29Operations on pharynxa 19 470.40 30Excision of larynx 7 490.14 31Larynx trachea operationsa 861060.81 33Other operations on lung, bronchusa1761081.63 34Operations on chest wall, pleura, mediastinum, and diaphragm 228 15 15.20 36Operations on heart vessels609 29 21.00 37Other heart and pericardium operationsa7822902.70 (Continued on the next page)

PAGE 136

124 InpatientOutpatient ICDCases (in 1,000) I/O Ratio 38Vessel incision, excision, and occlusion5371055.11 39Other operations on vessels597 866.94 40Lymphatic system operationsa1081031.05 41Bone marrow and spleen operationsa 88 224.00 42Operations on esophagusa 711670.43 43Incision, excision of stomach 73 107.30 44Other gastric operationsa 87 541.61 45Intestine incision, excision, anastomosisa725 23130.31 47Operations on appendix301 20 15.05 48Rectal & perirectal operationsa 551080.51 49Operations on anusa 472210.21 50Operations on livera 66 511.29 51Biliary tract operationsa4083351.22 52Operations on Pancreasa 18 121.50 53Repair of herniaa1195700.21 54Other abdomen region operationsa5986170.97 55Operations on kidneya 95 204.75 56Operations on uretera 81 870.93 57Urinary bladder operationsa2184670.47 58Operations on urethra 261340.19 59Other urinary tract operationsa1571081.45 60Prostate & seminal vesicle operationsa 54 321.69 61Operations scrotum, tunica vaginalisa 8 160.50 62Operations on testes 11 640.17 63Operations on spermatic cord, epididymis, vas deferens 81630.05 64Operations on penisa 341350.25 65Operations on ovarya5151343.84 66Fallopian tube operationsa4263981.07 67Operations on cervix 191770.11 68Other uterine incision and excisiona5773341.73 69Operations on other uterus and supporting structurea1306480.20 70Operations on vagina and culdesaca173 911.90 71Operations on vulva & perineuma 24 680.35 75Other obstetric operations 2115 14 151.07 76Facial bone and joint operationsa 94 462.04 77Incision, excision, division of other bonesa2334870.48 (Continued on the next page)

PAGE 137

125 InpatientOutpatient ICDCases (in 1,000) I/O Ratio 78Other bone operations except facea1452000.73 79Reduction of fracture and dislocationa3952231.77 80Incision and excision jointa387 16420.24 81Joint repaira4574391.04 82Hand muscle, tendon, fascia operations 352660.13 83Other muscle, tendon, fascia, bursa operationsa1562750.57 84Other musculoskeleton proceduresa 72 233.13 85Operations on the breast1377610.18 86Skin & subcutaneous operationsa70210810.65 87Diagnostic radiologya6473411.90 88Other diagnostic radiologya 19395103.80 89Interview, evaluation, consultation, and examinationa5921763.36 90Microscopic exam—ia 15 170.88 91Microscopic exam—iia 5 130.38 92Nuclear medicine182 16 11.38 93Physical therapy, respiratory therapy, rehabilitation, and related procedures 702 997.09 94Psyche related procedures647 13 49.77 95Opthalmologic and otologic diagnosis and treatmenta 8 300.27 97Replacement and removal of therapeutic appliancesa 43 870.49 Source: Owings MF, Kozak LJ (1998). “Ambulatory and Inpatient procedures in the United States, 1996. National Center for Health Statis tics.” Vital Health Stat 13(139). Note: Information in inpatient surgeries is from the 1996 National Hospital Discharge Survey, and outpatient surgery is from the 1996 National Survey of Ambulatory Surgery.a Procedures that had an I/O ratio between 0.2 and 5.0 were included in the subset of data.

PAGE 138

126 APPENDIX B SAS PROGRAM FOR CONSTRUCTING DATASETS LIBNAME PUFLIB 'C:\MEPS'; FILENAME IN1 'D:\97MEPS\HC\H20.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; FILENAME IN1 'D:\97MEPS\HC\Hc16ff1.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; FILENAME IN1 'D:\97MEPS\HC\Hc16df1.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; FILENAME IN1 'C:\MEPS\Hc12.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; FILENAME IN1 'C:\MEPS\97_00 plan file\H47F1.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; FILENAME IN1 'C:\MEPS\97_00 plan file\H47F2.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; FILENAME IN1 'C:\MEPS\97_00 plan file\H47F3.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; FILENAME IN1 'C:\MEPS\97_00 plan file\H36.SSP'; PROC XCOPY IN=IN1 OUT=PUFLIB IMPORT; RUN; /*Figure 6-2 Step 1*/ DATA PUFLIB.hosp7; SET PUFLIB1.HC16DF1 (keep=DUPERSID EVNTIDX EVENTRN COND IPPRO1X IPPRO2X IPEXP97X RSNI NHOS IPFSF97X IPFMR97X IPFMD97X IPFPV97X IPFVA97X IPFCH97X IPFO F97X IPFSL97X IPFWC97X IPFOR97X IPFOU97X IPFOT97X IPTC97X SURG PROC NUMNIGHX IPDSF97X IPDMR97X IPDMD97X IPDPV97X IPDVA97X IPDC H97X IPDOF97X IPDSL97X IPDWC97X IPDOR97X IPDOU97X IPDOT97X IPDXP97X); IF RSNINHOS=1; LENGTH INSURANC 3; INSURANC=.; IF (IPFMR97X gt 0 OR IPDMR97X gt 0) AND (IPFPV97X=0 AND IPDPV97X=0) THEN INSURANC=1; /*MCARE*/ ELSE IF (IPFMR97X gt 0 OR IPDMR97X gt 0) AND (IPFPV97X GT 0 OR IPDPV97X GT 0) THEN INSURANC=2; /*MediGap*/

PAGE 139

127 ELSE IF (IPFPV97X gt 0 OR IPDPV97X gt 0) THEN INSURANC=3; /*Private Ins*/ ELSE IF (IPFMD97X gt 0 OR IPDMD97X GT 0) THEN INSURANC=4; /*MCAID*/ ELSE IF IPFVA97X gt 0 OR IPFCH97X gt 0 OR IPFOF97X gt 0 OR IPFSL97X gt 0 OR IPFOU97X gt 0 OR IPDVA97X gt 0 OR IPDCH97X gt 0 OR IPDOF97X gt 0 OR IPDSL97X gt 0 OR IPDOU97X gt 0 THEN INSURANC=5; /*OTHER Public*/ ELSE IF IPFOR97X gt 0 OR IPFWC97X gt 0 OR IPFOT97X gt 0 OR IPFSF97X ge 0 OR IPDOR97X gt 0 OR IPDWC97X gt 0 OR IPDOT97X gt 0 OR IPDSF97X ge 0 THEN INSURANC=6; /*Other Private*/ EVENTRN=PUT(EVENTRN,1.); TOTAL=IPEXP97X; MDPAY=IPDXP97X; Ptpay=IPFSF97X + IPDSF97X; RUN; DATA PUFLIB.OPD97; SET PUFLIB1.HC16ff1(KEEP=DUPERSID EVNTIDX EVENTRN SURGPROC SURGNAME OPPRO1X COND OPEXP97X OPTC97X OPFSF97X OPFMR97X OPFMD97X OPFPV97X OPFVA 97X OPFCH97X OPFOF 97X OPFSL97X OPFWC97X OPFOR97X OPFOU97X OPFOT 97X OPDSF97X OPDMR97X OPDMD97X OPDPV97X OPDVA97X OPDCH97X OPDOF97X OPDSL97X OPDWC97X OPDOR97X OPDOU97X OPDOT97X OPDXP97X OPDTC97X); IF SURGPROC=1; LENGTH INSURANC 3; IF (OPFMR97X gt 0 OR OPDMR97X gt 0) AND (OPFPV97X=0 AND OPDPV97X=0) THEN INSURANC=1; /*MCARE*/ ELSE IF (OPFMR97X gt 0 OR OPDMR97X gt 0) AND (OPFPV97X GT 0 OR OPDPV97X GT 0) THEN INSURANC=2; /*MediGap*/ ELSE IF (OPFPV97X gt 0 OR OPDPV97X gt 0) THEN INSURANC=3; /*Private Ins*/ ELSE IF (OPFMD97X gt 0 OR OPDMD97X GT 0) THEN INSURANC=4; /*MCAID*/ ELSE IF OPFVA97X gt 0 OR OPFCH97X gt 0 OR OPFOF97X gt 0 OR OPFSL97X gt 0 OR OPFOU97X gt 0 OR OPDVA97X gt 0 OR OPDCH97X gt 0 OR OPDOF97X gt 0 OR OPDSL97X gt 0 OR OPDOU97X gt 0 THEN INSURANC=5; /*OTHER Public*/ ELSE IF OPFOR97X gt 0 OR OPFWC97X gt 0 OR OPFOT97X gt 0 OR OPFSF97X ge 0 OR OPDOR97X gt 0 OR OPDWC97X gt 0 OR OPDOT97X gt 0 OR OPDSF97X ge 0 THEN INSURANC=6; /*Other Private*/ TOTAL=OPEXP97X; MDPAY=OPDXP97X; Ptpay=OPFSF97X + OPDSF97X; RUN; DATA PUFLIB.person7; SET PUFLIB1.h20 (KEEP=DUID PID DUPERSID AGE97X SEX RACEX RACETHNX MSA97 Panel97 REGI ON97 EDUCYR97 POVCAT97 MARRY97X SPOUIN97 MOPID31X MOPID42X MO PID53X DAPID31X DAPID42X DAPID53X RTHLTH31 RTHLTH42 RTHLTH53 WTDPER97); IF AGE97X GE 0 AND AGE97X LE 64; LENGTH AGEGR 3 RACETHN 3 MSA 3 REGION 3 EDU 3 INCOMFPL 3; AGEGR=.; IF AGE97X LT 18 THEN AGEGR=1; ELSE IF AGE97X GE 18 AND AGE97X LE 54 THEN AGEGR=2;

PAGE 140

128 ELSE IF AGE97X GE 55 AND AGE97X LE 64 THEN AGEGR=3; RACETHN=.; IF RACETHNX=1 THEN RACETHN=3; ELSE IF RACEX=4 THEN RACETHN=2; ELSE IF RACEX=5 THEN RACETHN=1; ELSE IF RACEX in (1,2,3,91) THEN RACETHN=4; MSA=.; IF MSA97=1 THEN MSA=1; ELSE IF MSA97=0 THEN MSA=0; REGION=.; IF REGION97=1 THEN REGION=1; ELSE IF REGION97=2 THEN REGION=2; ELSE IF REGION97=3 THEN REGION=3; ELSE IF REGION97=4 THEN REGION=4; EDU=.; IF EDUCYR97 GE 0 and EDUCYR97 LE 11 THEN EDU=1; ELSE IF EDUCYR97=12 THEN EDU=2; ELSE IF EDUCYR97 GE 13 THEN EDU=3; INCOMFPL=.; IF POVCAT97=1 THEN INCOMFPL=1; ELSE IF POVCAT97=2 THEN INCOMFPL=2; ELSE IF POVCAT97=3 THEN INCOMFPL=3; ELSE IF POVCAT97=4 THEN INCOMFPL=4; ELSE IF POVCAT97=5 THEN INCOMFPL=5; RUN; /*Figure 6-2 Step 2*/ PROC SORT DATA=PUFLIB.hosp7; BY DUPERSID; RUN; PROC SORT DATA=PUFLIB.person7; BY DUPERSID; RUN; DATA PUFLIB.GOODHOSP; MERGE PUFLIB.person7 (IN=A) PUFLIB.HOSP7 (IN=B); BY DUPERSID; IF B AND A; YEAR=97; INOUT=.; IF NUMNIGHX=0 THEN INOUT=1; ELSE IF NUMNIGHX GT 0 AND NUMNIGHX LT 188 THEN INOUT=0; RUN; /*Figure 6-2 Step 3*/ DATA PUFLIB.GDHOSP7A; SET PUFLIB.GOODHOSP (DROP= RACEX RACETHNX MSA97 REGION97 EDUCYR97 POVCAT97 IPEXP97X IPPRO1X IPPRO2X IPFSF97X IPFMR97X IPFMD97X IPFPV97X IPFVA97X IPFCH97X IPFOF97X IPFSL97X IPFWC97X IPFOR97X IPFOU97X IPFOT97X IPDSF 97X IPDMR97X IPDMD97X IPDPV97X IPDVA97X IPDCH97X IPDOF97X IPDSL97X IPDWC97X IPDOR97X IPDOU97X IPDOT97X

PAGE 141

129 IPDXP97X); IF SURGPROC IN (2,3,4,7,14,21); SURG=.; IF SURGPROC=2 THEN SURG =1; /*arthroscopic surgery*/ ELSE IF SURGPROC=3 THEN SURG=2; /*cardiac catheterization*/ ELSE IF SURGPROC=7 THEN SURG=3; /*D & C*/ ELSE IF SURGPROC=21 THEN SURG=4; /*tonsillectomy*/ ELSE IF SURGPROC=4 THEN SURG=5; /*cataract surgery*/ ELSE IF SURGPROC=14 THEN SURG=6; /*pacemaker*/ RUN; DATA PUFLIB.GDHOSP7B; SET PUFLIB.GOODHOSP (DROP= RACEX RACETHNX MSA97 REGION97 EDUCYR97 POVCAT97 IPEXP97X IPFSF97X IPFMR97X IPFMD97X IPFPV97X IPFVA97X IPFCH97X IPFOF97X IPFSL97X IPFWC97X IPFOR97X IPFOU97X IPFOT97X IPDSF97X IPDMR97X IPDMD 97X IPDPV97X IPDVA97X IPDCH97X IPDOF97X IPDSL97X IPDWC97X IPDOR97X IPDOU97X IPDOT97X IPDXP97X); IF SURGPROC NOT IN (2,3,4,7,14,21); RUN; DATA PUFLIB.INPT97A; SET PUFLIB.GDHOSP7B (DROP=IPPRO2X); IF IPPRO1X GE 1 AND IPPRO1X LE 99; RUN; DATA PUFLIB.INPT971; SET PUFLIB.INPT97A; LENGTH PRO $ 3; PRO=IPPRO1X; RUN; DATA PUFLIB.INPT97B; SET PUFLIB.GDHOSP7B (DROP=IPPRO1X); IF IPPRO2X GE 1 AND IPPRO2X LE 99; RUN; DATA PUFLIB.INPT972; SET PUFLIB.INPT97B; LENGTH PRO $ 3; PRO=IPPRO2X; RUN; DATA PUFLIB.INPT97; SET PUFLIB.INPT971 (DROP=SURGPR OC IPPRO1X) PUFLIB.INPT972 (DROP= SURGPROC IPPRO2X); RUN; DATA PUFLIB.INPT97c; SET PUFLIB.INPT97; IF PRO NOT IN ('02', '35', '46','47','55','74'); RUN;

PAGE 142

130 PROC SORT DATA=PUFLIB.person7; BY DUPERSID; RUN; PROC SORT DATA=PUFLIB.OPD97; BY DUPERSID; RUN; DATA PUFLIB.GOODOPD; MERGE PUFLIB.person7 (IN=A) PUFLIB.OPD97 (IN=B); BY DUPERSID; IF B AND A; YEAR=97; INOUT=1; RUN; DATA PUFLIB.OUT17; SET PUFLIB.GOODOPD (DROP= RACEX RACETHNX MSA97 REGION97 EDUCYR97 POVCAT97 OPEXP97X OPPRO1 X OPFSF97X OPFMR97X OPFMD97X OPFPV97X OPFVA97X OPFCH97X OPFOF97X OPFSL97X OPFWC97X OPFOR97X OPFOU97X OPFOT97X OPDSF97X OPDMR97X OPDMD97X OPDPV97X OPDVA97X OPDCH97X OPDOF97X OPDSL97X OPDWC97X OPDOR97X OPDOU97X OPDOT97X OPDXP97X); IF SURGNAME IN (1,2,4,7,9,11); SURG=.; IF SURGNAME=1 THEN SURG=1; /*arthroscopic surgery*/ ELSE IF SURGNAME=9 THEN SURG=2; /*Catheterization*/ ELSE IF SURGNAME=4 THEN SURG=3; /*D AND C*/ ELSE IF SURGNAME=7 THEN SURG=4; /*Tonsillectomy*/ ELSE IF SURGNAME=2 THEN SURG=5; /*cataract surgery*/ ELSE IF SURGNAME=11 THEN SURG=6; /*pacemaker*/ RUN; DATA PUFLIB.OUT2; SET PUFLIB.GOODOPD (DROP= RACEX RACETHNX MSA97 REGION97 EDUCYR97 POVCAT97 OPEXP97X OPFSF97X OPFMR97X OPFMD97X OPFPV97X OPFVA97X OPFCH97X OPFOF97X OPFSL97X OPFWC97X OPFOR97X OPFOU97X OPFOT97X OPDSF97X OPDMR97X OPDMD97X OPDPV97X OPDVA97X OPDCH97X OPDOF97X OPDSL97X OPDWC97X OPDOR97X OPDOU97X OPDOT97X OPDXP97X); IF SURGNAME NOT IN (1,2,4,7,9,11); RUN; DATA PUFLIB.OUT2a; SET PUFLIB.OUT2 (DROP=SURGNAME SURGPROC); IF OPPRO1X GE 1 AND OPPRO1X LE 99; RUN; DATA PUFLIB.OUT2b; SET PUFLIB.OUT2a ; LENGTH PRO $ 2; PRO=OPPRO1X; RUN;

PAGE 143

131 DATA PUFLIB.OUT97; SET PUFLIB.OUT1 (DROP=SURGNAME SURGPROC) PUFLIB.OUT2b (DROP=OPPRO1X); RUN; /*Figure 6-2 Step 4*/ DATA PUFLIB.HOSP97; SET PUFLIB.GDHOSP7A PUFLIB.INPT97c; IF INSURANC=3; RUN; DATA PUFLIB.OUT97c; SET PUFLIB.OUT97; IF INSURANC=3; RUN; DATA PUFLIB.HOSP97a; SET PUFLIB.HOSP97 ; LENGTH TCH 8 WTDPER 8 AGE 3; AGE=AGE97X; TCH=IPTC97X; WTDPER=WTDPER97; EVENT=PUT(EVENTRN, 2.); RUN; DATA PUFLIB.OUT97d SET PUFLIB.OUT97c ; LENGTH TCH 8 WTDPER 8 AGE 3; AGE=AGE97X; TCH=OPTC97X; WTDPER=WTDPER97; EVENT=PUT(EVENTRN, 2.); RUN; /*deleting cases with zero weight*/ DATA PUFLIB.INOUT97; SET PUFLIB.OUT97d (DROP=EVEN TRN OPTC97X WTDPER97 AGE97X) PUFLIB.HOSP97a (DROP=EVEN TRN SURGPROC RSNINHOS IPTC97X WTDPER97 AGE97X); IF WTDPER GT 0; RUN; /*deleting duplicated cases*/ PROC SORT DATA=PUFLIB.INOUT97 OUT=PUFLIB.I_O97 NODUPKEY; BY EVNTIDX; RUN; /*obtaining health status reported in the last run of the 1996 MEPS for panel 1 surgical cases occurred in the first run of 1997*/ DATA PUFLIB.PERSON6; SET PUFLIB.HC12 (KEEP=DUPERSID RTEHLTH2);

PAGE 144

132 RUN; PROC SORT DATA=PUFLIB.PERSON6; BY DUPERSID; RUN; PROC SORT DATA=PUFLIB.I_O97; BY DUPERSID; RUN; /*obtaining health status reported in the previous year's NHIS for panel 2 surgical cases occurred in the first run of 1997*/ PROC SORT DATA=PUFLIB.HIS96LNK; BY DUPERSID; RUN; DATA PUFLIB.I_O97a; MERGE PUFLIB.PERSON6 PUFLIB.HIS96l nk (DROP=HLTHa) PUFLIB.I_O97 (IN=A); BY DUPERSID; IF A; RUN; /* STAT TRANSFER I_O97a into EXCEL FILE to get Hlth */ DATA PUFLIB.I_O97b; SET PUFLIB.I_O97a (DROP=EDU); IF AGE LT 18; RUN; /* STAT TRANSFER I_O97b (containing children unde r age 18) into EXCEL FILE to get mom id (father' id, if no mom), save as MOM 97, and two-parent family status. */ DATA PUFLIB.MOMID97; SET PUFLIB.MOM97; MOMID=DUID*1000+MOM; MOMI=PUT(MOMID,8.); RUN; DATA PUFLIB.MOMID97a; SET PUFLIB.MOMID97 (DROP=DUID MOM MOMID); RUN; DATA PUFLIB.MOMID97b; SET PUFLIB.MOMID97a; LENGTH DUPERSID $ 8; DUPERSID=MOMI; RUN;

PAGE 145

133 DATA PUFLIB.MOMID97c; SET PUFLIB.MOMID97b (DRO P=MOMI PANEL97 EVENT); RUN; /* obtaining mother's (father', if no mom) education attainment for surgical cases under age 18*/ DATA PUFLIB.PERSON7a; SET PUFLIB.person7 (KEEP=DUPERSID EDU); RUN; PROC SORT DATA=PUFLIB.MOMID97i; BY DUPERSID; RUN; PROC SORT DATA=PUFLIB.PERSON7a; BY DUPERSID; RUN; DATA PUFLIB.MOMEDU; MERGE PUFLIB.PERSON7a PUFLIB.MOMID97i (IN=A); BY DUPERSID; IF A; RUN; DATA PUFLIB.MOMEDUi; SET PUFLIB.MOMEDU (DROP=DUPERSID PANEL97 EVENT); RUN; PROC SORT DATA=PUFLIB.MOMEDUi; BY EVNTIDX; RUN; PROC SORT DATA=PUFLIB.I_O97b; BY EVNTIDX; RUN; DATA PUFLIB.I_O97c; MERGE PUFLIB.I_O97b PUFLIB.MOMEDUi; BY EVNTIDX; LENGTH PANEL 3; PANEL=PANEL97; RUN; DATA PUFLIB.I_O97f; SET PUFLIB.I_O97c (DROP=PANEL97 MO PID31X MOPID42X MOPID53X DAPID31X DAPID42X DAPID53X MARRY97X SPOUIN97); RUN; DATA PUFLIB.I_O97d; SET PUFLIB.I_O97a; IF AGE GE 18; LENGTH SUPPORT 3 PANEL 3; PANEL=PANEL97; SUPPORT=.;

PAGE 146

134 IF SPOUIN97=1 THEN SUPPORT=1; ELSE IF SPOUIN97 in (2,3) THEN SUPPORT=0; RUN; DATA PUFLIB.I_O97e; SET PUFLIB.I_O97d (DROP=PANEL97 MO PID31X MOPID42X MOPID53X DAPID31X DAPID42X DAPID53X MARRY97X SPOUIN97); RUN; DATA PUFLIB.I_O7; SET PUFLIB.I_O97f PUFLIB.I_O97e; RUN; DATA PUFLIB.I_O7a; SET PUFLIB.I_O7 (KEEP=DUPERSID PANEL EVENT); RUN; /*obtaining health plan data*/ DATA PUFLIB.PLAN97; SET PUFLIB.H47F1 (KEEP=DUPERSID EPCPIDX RN PANEL PRIVCAT UPRHMO UPRMNC DRLIST VISITPYX); RUN; DATA PUFLIB.PLAN98; SET PUFLIB.H47F2 (KEEP=DUPERSID EPCPIDX RN PANEL PRIVCAT UPRHMO UPRMNC DRLIST VISITPYX); RUN; DATA PUFLIB.PLAN99; SET PUFLIB.H47F3 (KEEP=DUPERS ID EPCPIDX RN PANEL PRIVCAT UPRHMO UPRMNC DRLIST VISITPYX); RUN; PROC SORT DATA=PUFLIB.I_O7a; BY DUPERSID; RUN; PROC SORT DATA=PUFLIB.PLAN97; BY DUPERSID; RUN; DATA PUFLIB.I_O7b; MERGE PUFLIB.I_O7a (IN=A) PUFLIB.PLAN97; BY DUPERSID; IF A; RUN; DATA PUFLIB.I_O7PLAN; SET PUFLIB.I_O7b; IF EVENT=RN; RUN;

PAGE 147

135 DATA PUFLIB.NUM_PLAN1; SET PUFLIB.N_PLAN (DROP=PERCENT); IF COUNT=1; RUN; DATA PUFLIB.PLAN971; MERGE PUFLIB.NUM_PLAN1 (IN=A) PUFLIB.I_O7PLAN; BY DUPERSID; IF A; LENGTH HMO 3 GATE 3; HMO=.; IF UPRHMO=1 THEN HMO=1; ELSE IF UPRHMO in (-1, -7, -8, -9, 2) THEN HMO=0; GATE=.; IF UPRMNC=1 OR UPRHMO=1 THEN GATE=1; ELSE IF UPRMNC in (-1, -7, -8, -9, 2) AND UPRHMO~=1 THEN GATE=0; RUN; DATA PUFLIB.NUM_PLAN2; SET PUFLIB.N_PLAN (DROP=PERCENT); IF COUNT GT 1; RUN; PROC SORT DATA=PUFLIB.NUM_PLAN2; BY DUPERSID; RUN; PROC SORT DATA=PUFLIB.I_O7PLAN; BY DUPERSID; RUN; DATA PUFLIB.PLAN972; MERGE PUFLIB.NUM_PLAN2 (IN=A) PUFLIB.I_O7PLAN; BY DUPERSID; IF A; RUN; DATA PUFLIB.PLAN97f; SET PUFLIB.PLAN971 (DROP=UPRMNC UPRHMO RN) PUFLIB.PLAN972 (DROP=UPRMNC UPRHMO RN); RUN; PROC SORT DATA=PUFLIB.I_O7; BY EVENT DUPERSID; RUN; PROC SORT DATA=PUFLIB.PLAN97f; BY EVENT DUPERSID; RUN; DATA PUFLIB.I_O1997; MERGE PUFLIB.I_O7 PUFLIB.PLAN97f; BY EVENT DUPERSID; LENGTH RACE 3;

PAGE 148

136 RACE=.; IF RACETHN=1 THEN RACE=1; ELSE IF RACETHN in (2,3,4) THEN RACE=0; RUN; /*repeating the above procedures to get PUFLI B.I_O1998 for 1998 data, and PUFLIB.I_O1999 for 1999 data*/ DATA PUFLIB.FINAL; SET PUFLIB.I_O1997 PUFLI B.I_O1998 PUFLIB.I_O1999; RUN; DATA PUFLIB.FINAL1; SET PUFLIB.FINAL; LENGTH COUNT1 3 PLANTYP1 3 PLANTYP2 3; IF COUNT=. THEN COUNT1=1; ELSE IF COUNT in (1,2) THEN COUNT1=COUNT; PLANTYP1=.; IF COUNT=1 AND GATE=1 THEN PLANTYP1=1; ELSE IF (COUNT=1 AND GATE=0) OR COUNT=. THEN PLANTYP1=2; ELSE IF COUNT=2 THEN PLANTYP1=3; PLANTYP2=.; IF COUNT=1 AND HMO=1 THEN PLANTYP2=1; ELSE IF (COUNT=1 AND HMO=0) OR COUNT=. THEN PLANTYP2=2; ELSE IF COUNT=2 THEN PLANTYP2=3; RUN; PROC MEANS DATA=PUFLIB.FINAL1; VAR AGE; RUN; /*obtaining pooled strata and psu*/ PROC SORT DATA=PUFLIB.H36; BY DUPERSID; RUN; PROC SORT DATA=PUFLIB.FINAL1; BY DUPERSID; RUN; DATA PUFLIB.FINALa; MERGE PUFLIB.FINAL1 (IN=A) PUFLI B.H36 (KEEP=DUPERSID PSU9699 STRA9699); BY DUPERSID; IF A; LENGTH HMO_1 3 GATE_1 3; IF HMO=1 THEN HMO_1=1; ELSE IF HMO=0 OR HMO=. THEN HMO_1=0; IF GATE=1 THEN GATE_1=1; ELSE IF GATE=0 OR GATE=. THEN GATE_1=0; RUN; /*merging with data containing predicted HM O status and gatekeeper enrollment status*/

PAGE 149

137 PROC SORT DATA=PUFLIB.FINALa; BY DUPERSID YEAR; RUN; PROC SORT DATA=PUFLIB.PLANHAT; BY DUPERSID YEAR; RUN; DATA PUFLIB.FINALah; MERGE PUFLIB.FINALa (IN=A) PUFLIB.PLANHAT; BY DUPERSID YEAR; IF A; RUN; /* constructing a subset of data cont aining only high variation surgical cases*/ DATA PUFLIB.VARAh; SET PUFLIB.FINALah; IF SURG=2 OR PRO in ('03','06','16','18','29','31','33','37','40','44','45','48','49','51','53','54', '57','59','60','64','65','66','68','69','70','76','77','78','80','81','83','84','86','97'); RUN;

PAGE 150

138 APPENDIX C REGRESSION MODELS FOR ALL SURGICAL CASES: MAIN EFFECT AND INTERACTION This appendix contains additional regressi on models as a supplement to Table 7-11 and Table 7-15.Survey logistic regression pweight: wtdper Number of obs = 813 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9585171.3 F( 19, 210) = 8.16 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Coef. Std. Err. t P>|t| [95%conf.Interval] -------------+--------------------------------------------------------one_hmo | -.5271085 .2797117 -1.88 0.061 -1.078259 .024042 gateonly | .3895111 .4601603 0.85 0.398 -.5171995 1.296222 two_plan | -.3022361 .4480599 -0.67 0.501 -1.185104 .5806314 cond | -.5429074 .3492584 -1.55 0.121 -1.231094 .1452794 hlth | .0452521 .1183169 0.38 0.702 -.1878822 .2783864 tch | -.0002171 .0000333 -6.53 0.000 -.0002827 -.0001516 msa | .0095111 .2881522 0.03 0.974 -.5582706 .5772928 region1 | -.33057 .3473223 -0.95 0.342 -1.014942 .3538019 region2 | -.3076129 .2935383 -1.05 0.296 -.8860074 .2707817 region4 | -.028988 .3024245 -0.10 0.924 -.6248923 .5669163 facility | -.0001724 .0000532 -3.24 0.001 -.0002773 -.0000676 paymd | .0002127 .0001099 1.94 0.054 -3.85e-06 .0004293 ptpay | -.0005082 .0002587 -1.96 0.051 -.0010178 1.53e-06 age | -.0098173 .0074681 -1.31 0.190 -.0245327 .0048981 support | .4161508 .2344977 1.77 0.077 -.0459089 .8782105 male | .5761653 .2385277 2.42 0.017 .1061648 1.046166 income | .2517215 .1142763 2.20 0.029 .0265489 .4768941 year98 | .5930209 .2576307 2.30 0.022 .0853794 1.100663 year99 | .2735708 .2882769 0.95 0.344 -.2944566 .8415982 _cons | 2.065568 .7107741 2.91 0.004 .6650425 3.466094 ----------------------------------------------------------------------.

PAGE 151

139Survey logistic regression pweight: wtdper Number of obs = 813 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9585171.3 F( 20, 209) = 8.11 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .600333 .1677877 -1.83 0.069 .3461142 1.041274 gateonly | 1.513109 .668301 0.94 0.349 .6337391 3.612685 two_plan | .7228634 .3247526 -0.72 0.471 .2982675 1.751889 cond | .5808795 .205462 -1.54 0.126 .2893358 1.166192 hlth | 1.046482 .123858 0.38 0.701 .8287981 1.32134 tch | .9997834 .0000333 -6.50 0.000 .9997177 .999849 msa | 1.036116 .3046665 0.12 0.904 .5804701 1.849427 region1 | .6822266 .2360086 -1.11 0.270 .3450593 1.34885 region2 | .7102799 .2096498 -1.16 0.248 .3970488 1.270619 region4 | .9613743 .291309 -0.13 0.897 .5291647 1.746603 facility | .9998271 .0000529 -3.27 0.001 .9997228 .9999314 paymd | 1.000209 .0001108 1.89 0.060 .999991 1.000428 ptpay | .9994756 .0002697 -1.94 0.053 .9989443 1.000007 age | .9901748 .0073959 -1.32 0.188 .9757085 1.004856 support | 1.482135 .3521786 1.66 0.099 .9280033 2.367151 male | 1.762141 .4234951 2.36 0.019 1.097438 2.829447 income | 1.274533 .1493965 2.07 0.040 1.011681 1.605678 race | 1.384226 .3995617 1.13 0.261 .783779 2.444672 year98 | 1.833385 .4650921 2.39 0.018 1.112165 3.022305 year99 | 1.323585 .3864806 0.96 0.338 .7445238 2.353017 -----------------------------------------------------------------------

PAGE 152

140Survey logistic regression pweight: wtdper Number of obs = 808 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9517252 F( 21, 208) = 7.74 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .5846186 .1641107 -1.91 0.057 .3362431 1.016464 gateonly | 1.500892 .6543271 0.93 0.353 .6357409 3.543386 two_plan | .7040832 .3217225 -0.77 0.44 .2861554 1.732391 cond | .5949206 .2098702 -1.47 0.142 .2968781 1.192175 hlth | 1.0349 .1268533 0.28 0.780 .8128402 1.317624 tch | .9997805 .0000336 -6.52 0.000 .9997142 .9998468 msa | 1.064158 .3124387 0.21 0.832 .5967028 1.897816 region1 | .6648728 .2294947 -1.18 0.238 .3367912 1.312552 region2 | .6739095 .197915 -1.34 0.180 .3778201 1.202038 region4 | .9097265 .2788882 -0.31 0.758 .4972464 1.664371 facility | .9998255 .0000538 -3.24 0.001 .9997194 .9999316 paymd | 1.000206 .0001114 1.85 0.066 .9999861 1.000425 ptpay | .999482 .0002678 -1.93 0.054 .9989545 1.00001 age | .9911288 .0073273 -1.21 0.229 .9767955 1.005672 support | 1.471104 .3419002 1.66 0.098 .9305917 2.325558 male | 1.768907 .4242532 2.38 0.018 1.102717 2.837566 income | 1.291991 .1540176 2.15 0.033 1.02152 1.634076 race | 1.408934 .4128156 1.17 0.243 .7909684 2.509704 edu | .8647644 .1590042 -0.79 0.430 .6019385 1.242348 year98 | 1.771973 .4414585 2.30 0.023 1.084582 2.895022 year99 | 1.288096 .3903417 0.84 0.404 .7089666 2.340297 ----------------------------------------------------------------------. Survey logistic regression pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 3, 226) = 0.80 Prob > F = 0.4923 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9378958 .1950367 -0.31 0.758 .6225891 1.412888 gateonly | 1.350929 .5345997 0.76 0.448 .6194288 2.946279 two_plan | .7116787 .2251797 -1.07 0.284 .3815234 1.327538 ----------------------------------------------------------------------.

PAGE 153

141Survey logistic regression pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 4, 225) = 26.05 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .6832082 .1744542 -1.49 0.137 .4130881 1.129961 gateonly | 1.544174 .714583 0.94 0.349 .6204285 3.84327 two_plan | .8385198 .3291908 -0.45 0.654 .386866 1.817465 tch | .9997636 .0000237 -9.97 0.000 .9997169 .9998104 ---------------------------------------------------------------------. Survey logistic regression pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 5, 224) = 19.43 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | .6526963 .1684596 -1.65 0.100 .3925052 1.085368 gateonly | 1.642811 .7820323 1.04 0.298 .6430213 4.197106 two_plan | .8901615 .3684357 -0.28 0.779 .3938025 2.012144 tch | .9998225 .0000289 -6.13 0.000 .9997655 .9998795 facility | .9998138 .0000567 -3.28 0.001 .9997021 .9999256 ----------------------------------------------------------------------. Survey logistic regression pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 6, 223) = 16.61 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .6769144 .1726832 -1.53 0.127 .4094779 1.119018 gateonly | 1.709339 .8553341 1.07 0.285 .6377105 4.581766 two_plan | .9219563 .3871232 -0.19 0.847 .4030782 2.10878 tch | .9997941 .000031 -6.64 0.000 .999733 .9998552 facility | .9998288 .0000561 -3.05 0.003 .9997183 .9999393 paymd | 1.000158 .0001124 1.41 0.161 .9999366 1.00038 ----------------------------------------------------------------------.

PAGE 154

142Survey logistic regression pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 7, 222) = 15.41 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | .5943894 .1569426 -1.97 0.050 .3532811 1.00005 gateonly | 1.491076 .7433467 0.80 0.424 .5583227 3.982118 two_plan | .8484596 .3613921 -0.39 0.700 .3665527 1.96393 tch | .999803 .0000311 -6.33 0.000 .9997416 .9998643 facility | .9998177 .0000552 -3.30 0.001 .999709 .9999266 paymd | 1.000177 .0001108 1.59 0.112 .9999583 1.000395 ptpay | .9993787 .0002464 -2.52 0.012 .9988934 .9998643 ------------------------------------------------------------------. Survey logistic regression pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 8, 221) = 13.40 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| 95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .5735233 .1974903 -1.61 0.108 .2909908 1.130376 hmo_md | 1.000036 .000188 0.19 0.847 .9996659 1.000407 gateonly | 1.489463 .7442063 0.80 0.426 .556493 3.986575 two_plan | .8465946 .3621176 -0.39 0.697 .3644554 1.966557 tch | .9998027 .0000314 -6.28 0.000 .9997408 .9998646 facility | .9998178 .0000553 -3.29 0.001 .9997088 .9999268 paymd | 1.000166 .0001224 1.36 0.176 .9999248 1.000407 ptpay | .9993799 .0002454 -2.53 0.012 .9988965 .9998636 --------------------------------------------------------------------pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 8, 221) = 14.27 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .6705802 .1833532 -1.46 0.145 .3912633 1.149297 hmo_pt | .9989678 .0008335 -1.24 0.217 .9973268 1.000612 gateonly | 1.541043 .7704771 0.86 0.388 .575397 4.127262 two_plan | .8661186 .3673387 -0.34 0.735 .3755249 1.997634 tch | .9998007 .0000306 -6.51 0.000 .9997404 .9998611 facility | .9998192 .0000552 -3.28 0.001 .9997104 .9999279 paymd | 1.000186 .0001116 1.67 0.097 .9999663 1.000406 ptpay | .9995221 .0002362 -2.02 0.044 .9990568 .9999876 --------------------------------------------------------------------

PAGE 155

143pweight: wtdper Number of obs = 814 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9595657.1 F( 8, 221) = 17.03 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9212661 .3504844 -0.22 0.830 .4353392 1.949586 hmo_fac | .9998085 .0001056 -1.81 0.071 .9996004 1.000017 gateonly | 1.454939 .7028417 0.78 0.438 .56164 3.769047 two_plan | .8346639 .3385978 -0.45 0.656 .3752859 1.856355 tch | .9997998 .0000321 -6.24 0.000 .9997366 .999863 facility | .9998782 .0000675 -1.80 0.073 .9997453 1.000011 paymd | 1.000184 .0001082 1.70 0.091 .9999705 1.000397 ptpay | .9993966 .0002475 -2.44 0.016 .998909 .9998844 ---------------------------------------------------------------------. Survey logistic regression pweight: wtdper Number of obs = 813 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9585171.3 F( 10, 219) = 12.73 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9286212 .3566477 -0.19 0.847 .4356927 1.979233 hmo_fac | .9998134 .0001089 -1.71 0.088 .9995989 1.000028 gateonly | 1.418876 .6972957 0.71 0.477 .5387607 3.736743 two_plan | .8671073 .3433467 -0.36 0.719 .3973981 1.891995 cond | .6124635 .210666 -1.43 0.155 .3109813 1.206219 hlth | .9379961 .0946823 -0.63 0.527 .7688141 1.144408 tch | .9997975 .0000334 -6.06 0.000 .9997318 .9998633 facility | .9998787 .0000688 -1.76 0.079 .9997432 1.000014 paymd | 1.000191 .0001065 1.79 0.074 .9999812 1.000401 ptpay | .9994611 .0002443 -2.21 0.028 .9989798 .9999426 -----------------------------------------------------------------------

PAGE 156

144pweight: wtdper Number of obs = 813 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9585171.3 F( 14, 215) = 10.09 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9266086 .3397729 -0.21 0.836 .4498921 1.908465 hmo_fac | .9998057 .0001061 -1.83 0.068 .9995967 1.000015 gateonly | 1.381877 .6799816 0.66 0.512 .5240617 3.643814 two_plan | .838829 .3346346 -0.44 0.660 .3822002 1.841009 cond | .6181804 .2097373 -1.42 0.158 .316794 1.206295 hlth | .9421748 .0947302 -0.59 0.554 .7728431 1.148607 tch | .9997985 .0000341 -5.91 0.000 .9997314 .9998656 msa | 1.001967 .2710626 0.01 0.994 .5879609 1.70749 region1 | .7823518 .2659437 -0.72 0.471 .400415 1.5286 region2 | .8309176 .2446485 -0.63 0.530 .4651573 1.484281 region4 | .9362257 .2837542 -0.22 0.828 .5152513 1.701148 facility | .9998792 .0000684 -1.77 0.079 .9997444 1.000014 paymd | 1.000182 .0001099 1.66 0.099 .9999655 1.000399 ptpay | .9994365 .0002499 -2.25 0.025 .9989442 .999929 --------------------------------------------------------------------pweight: wtdper Number of obs = 813 Strata: newstr Number of strata= 61 PSU: newpsu Number of PSUs = 289 Population size = 9585171.3 F( 21, 208) = 8.79 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9918818 .3782959 -0.02 0.983 .4678278 2.102974 hmo_fac | .9997816 .0001064 -2.05 0.041 .999572 .9999912 gateonly | 1.467687 .6320419 0.89 0.374 .6282281 3.428857 two_plan | .7093553 .3005847 -0.81 0.419 .3077854 1.634857 cond | .5864538 .2148778 -1.46 0.147 .2848969 1.207202 hlth | 1.05287 .1221297 0.44 0.657 .837743 1.323239 tch | .9997803 .0000351 -6.25 0.000 .999711 .9998495 msa | 1.014721 .3014542 0.05 0.961 .5650953 1.822098 region1 | .6360131 .221021 -1.30 0.194 .3206907 1.261379 region2 | .7055754 .2095177 -1.17 0.241 .3930373 1.26664 region4 | .9226311 .284134 -0.26 0.794 .5029128 1.692636 facility | .9998915 .0000633 -1.72 0.088 .9997669 1.000016 paymd | 1.000211 .0001094 1.93 0.055 .9999954 1.000426 ptpay | .9994997 .0002739 -1.83 0.069 .9989602 1.00004 age | .9899342 .0074328 -1.35 0.179 .9753961 1.004689 race | 1.478569 .4358391 1.33 0.186 .8271667 2.642956 support | 1.456435 .3452471 1.59 0.114 .9129302 2.323511 male | 1.758297 .4251633 2.33 0.020 1.091867 2.831486 income | 1.272218 .1505867 2.03 0.043 1.00756 1.606394 year98 | 1.928231 .4903963 2.58 0.010 1.168214 3.1827 year99 | 1.343397 .393779 1.01 0.315 .7539931 2.393545 -------------------------------------------------------------------

PAGE 157

145 APPENDIX D REGRESSION MODELS FOR THE SUBSET OF CASES: MAIN EFFECT AND INTERACTION This appendix contains additional regressi on models as a supplement to Table 7-17.Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 18, 121) = 4.78 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .5064955 .1659148 -2.08 0.040 .2650191 .9679969 two_plan | .7155704 .411338 -0.58 0.561 .2296251 2.2299 cond | .8012655 .2671627 -0.66 0.507 .4144317 1.549173 hlth | .9948047 .1630974 -0.03 0.975 .7193652 1.375708 tch | .9996853 .0000806 -3.90 0.000 .9995259 .9998447 msa | 1.174319 .3529398 0.53 0.594 .6481795 2.127535 region1 | .8414798 .3615903 -0.40 0.689 .3597824 1.968101 region2 | 1.028175 .3571735 0.08 0.936 .5173161 2.043518 region4 | 1.271122 .6182464 0.49 0.623 .4858684 3.325493 facility | .999853 .0000621 -2.37 0.019 .9997301 .9999758 paymd | 1.000052 .0002297 0.23 0.821 .9995979 1.000506 ptpay | .9990819 .0007249 -1.27 0.208 .9976496 1.000516 age | .9864795 .0113337 -1.18 0.238 .964322 1.009146 support | .7682079 .2887134 -0.70 0.484 .3653793 1.615153 male | 2.665568 .809756 3.23 0.002 1.461912 4.860249 income | 1.613109 .2178217 3.54 0.001 1.235114 2.106785 year98 | 2.451122 .8974681 2.45 0.016 1.188348 5.055755 year99 | 1.883765 .6679969 1.79 0.076 .9343594 3.797864 ---------------------------------------------------------------------Note: 2 failures and 0 successes completely determined. .

PAGE 158

146Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 9, 130) = 3.86 Prob > F = 0.0002 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .7007766 .2436651 -1.02 0.308 .3523645 1.393693 two_plan | 1.033318 .5465831 0.06 0.951 .3630779 2.940819 cond | .6527698 .3525313 -0.79 0.431 .2243871 1.898988 hlth | .9171982 .1331256 -0.60 0.552 .6883727 1.222089 facility | .9996025 .0000723 -5.49 0.000 .9994595 .9997456 age | .9857231 .0119664 -1.18 0.238 .9623436 1.009671 male | 2.36661 .7475849 2.73 0.007 1.267253 4.419674 year98 | 1.59432 .5583386 1.33 0.185 .7977065 3.186454 year99 | 1.118608 .3447911 0.36 0.717 .6081198 2.057628 ----------------------------------------------------------------------. gen hmo_fac=one_hmo*facility Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 10, 129) = 3.89 Prob > F = 0.0001 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | 1.113062 .638503 0.19 0.852 .3580238 3.460406 hmo_fac | .9997933 .0001622 -1.27 0.205 .9994727 1.000114 two_plan | 1.015997 .5052721 0.03 0.975 .3800412 2.716152 cond | .657322 .3571693 -0.77 0.441 .2244753 1.924809 hlth | .9261546 .1318869 -0.54 0.591 .6988722 1.227352 facility | .9996648 .0000846 -3.96 0.000 .9994976 .9998321 age | .9857855 .0117988 -1.20 0.234 .9627297 1.009394 male | 2.331891 .7504385 2.63 0.009 1.234113 4.406172 year98 | 1.699404 .6056703 1.49 0.139 .8399298 3.43835 year99 | 1.104609 .3393605 0.32 0.747 .6017101 2.027824 ----------------------------------------------------------------------

PAGE 159

147Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 11, 128) = 5.71 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | .9806407 .545286 -0.04 0.972 .3265964 2.944479 hmo_fac | .9998 .0001699 -1.18 0.241 .9994641 1.000136 two_plan | .794439 .3814201 -0.48 0.632 .3074486 2.052809 cond | .7612465 .3382624 -0.61 0.540 .3161877 1.83276 hlth | .9986323 .1447027 -0.01 0.992 .7498503 1.329954 facility | .9996438 .0000758 -4.70 0.000 .999494 .9997936 age | .9799251 .0110938 -1.79 0.075 .958233 1.002108 male | 2.273567 .6859993 2.72 0.007 1.252 4.128682 income | 1.601971 .2456403 3.07 0.003 1.182987 2.169349 year98 | 1.595747 .5844184 1.28 0.204 .7735109 3.292015 year99 | 1.062955 .3343149 0.19 0.846 .5707303 1.979697 ----------------------------------------------------------------------. Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 12, 127) = 7.00 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | 1.01441 .5203599 0.03 0.978 .3678842 2.797149 hmo_fac | .9997696 .0001414 -1.63 0.106 .99949 1.000049 two_plan | .7612226 .3932869 -0.53 0.598 .2740604 2.114351 cond | .784524 .2563691 -0.74 0.459 .4111376 1.497012 hlth | 1.050741 .1685063 0.31 0.758 .7652114 1.442811 tch | .9996694 .000059 -5.60 0.000 .9995528 .9997861 facility | .9999668 .0000872 -0.38 0.704 .9997945 1.000139 age | .9841092 .0108048 -1.46 0.147 .9629751 1.005707 male | 2.567055 .7605229 3.18 0.002 1.428975 4.611535 income | 1.635831 .2041705 3.94 0.000 1.278083 2.093718 year98 | 2.674764 .9890229 2.66 0.009 1.287538 5.556623 year99 | 1.996643 .6764543 2.04 0.043 1.0218 3.901533 ----------------------------------------------------------------------Note: 1 failure and 0 successes completely determined.

PAGE 160

148Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 12, 127) = 6.66 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .7891584 .4101744 -0.46 0.649 .2823762 2.205466 hmo_fac | .9998178 .0001496 -1.22 0.225 .9995221 1.000114 two_plan | .7054114 .3347499 -0.74 0.463 .2760164 1.80281 cond | .8308559 .3399982 -0.45 0.651 .3699306 1.866084 hlth | 1.047174 .1579018 0.31 0.760 .7771987 1.410931 facility | .999686 .0000678 -4.63 0.000 .9995519 .9998201 paymd | .9993796 .0001681 -3.69 0.000 .9990472 .9997121 age | .9711105 .0113462 -2.51 0.013 .9489329 .9938065 male | 2.104691 .614026 2.55 0.012 1.182117 3.74728 income | 1.62728 .2276053 3.48 0.001 1.234107 2.145715 year98 | 2.263367 .850488 2.17 0.031 1.076653 4.758106 year99 | 1.285963 .410058 0.79 0.432 .6845459 2.415763 ----------------------------------------------------------------------Note: 1 failure and 0 successes completely determined. Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 11, 128) = 5.71 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9806407 .545286 -0.04 0.972 .3265964 2.944479 hmo_fac | .9998 .0001699 -1.18 0.241 .9994641 1.000136 two_plan | .794439 .3814201 -0.48 0.632 .3074486 2.052809 cond | .7612465 .3382624 -0.61 0.540 .3161877 1.83276 hlth | .9986323 .1447027 -0.01 0.992 .7498503 1.329954 facility | .9996438 .0000758 -4.70 0.000 .999494 .9997936 age | .9799251 .0110938 -1.79 0.075 .958233 1.002108 male | 2.273567 .6859993 2.72 0.007 1.252 4.128682 income | 1.601971 .2456403 3.07 0.003 1.182987 2.169349 year98 | 1.595747 .5844184 1.28 0.204 .7735109 3.292015 year99 | 1.062955 .3343149 0.19 0.846 .5707303 1.979697 ----------------------------------------------------------------------.

PAGE 161

149Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 12, 127) = 5.20 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9714537 .5487133 -0.05 0.959 .3179642 2.968014 hmo_fac | .9998015 .0001711 -1.16 0.248 .9994632 1.00014 two_plan | .784068 .3847846 -0.50 0.621 .297117 2.069093 cond | .7599254 .3368157 -0.62 0.537 .3163457 1.825492 hlth | .9994671 .145302 -0.00 0.997 .7497673 1.332326 msa | 1.056013 .3038577 0.19 0.850 .5978299 1.865354 facility | .9996433 .0000762 -4.68 0.000 .9994927 .999794 age | .980004 .0110099 -1.80 0.074 .9584741 1.002018 male | 2.282538 .687246 2.74 0.007 1.25853 4.139733 income | 1.602842 .2468019 3.06 0.003 1.18213 2.173283 year98 | 1.592616 .5812989 1.28 0.204 .7738859 3.277517 year99 | 1.061375 .3344383 0.19 0.850 .5692236 1.97904 ----------------------------------------------------------------------. Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 16, 123) = 5.03 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | .8013306 .4198197 -0.42 0.673 .2843932 2.257897 hmo_fac | .9998127 .0001523 -1.23 0.221 .9995115 1.000114 two_plan | .7142376 .3556033 -0.68 0.500 .2668695 1.911554 cond | .8350038 .3336268 -0.45 0.652 .3789493 1.839907 hlth | 1.05719 .1565216 0.38 0.708 .7888857 1.416746 msa | 1.021697 .3330773 0.07 0.948 .5362562 1.94658 region1 | .7697334 .3029106 -0.67 0.507 .3535138 1.676001 region2 | 1.071037 .3776562 0.19 0.846 .5333452 2.150801 region4 | .9240187 .4169502 -0.18 0.861 .3786084 2.255128 facility | .9996832 .0000711 -4.46 0.000 .9995427 .9998238 paymd | .9993612 .0001634 -3.91 0.000 .9990382 .9996842 age | .9709659 .0112964 -2.53 0.012 .9488845 .9935611 male | 2.100462 .6126144 2.54 0.012 1.179939 3.739125 income | 1.639512 .2329114 3.48 0.001 1.238004 2.171237 year98 | 2.308628 .8591601 2.25 0.026 1.106051 4.81873 year99 | 1.25526 .4038171 0.71 0.481 .6644763 2.371307 ----------------------------------------------------------------------Note: 1 failure and 0 successes completely determined.

PAGE 162

150Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 16, 123) = 4.65 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | .8031505 .4225186 -0.42 0.678 .2838169 2.272771 hmo_fac | .9997838 .0001615 -1.34 0.183 .9994644 1.000103 two_plan | .7126214 .3538517 -0.68 0.496 .2669661 1.902224 cond | .8612739 .3335961 -0.39 0.700 .4004328 1.852477 hlth | .92399 .1388556 -0.53 0.600 .6864652 1.243701 msa | 1.044234 .3446472 0.13 0.896 .5437197 2.00549 region1 | .9025447 .3664417 -0.25 0.801 .404404 2.01429 region2 | 1.15314 .4395927 0.37 0.709 .5426522 2.45043 region4 | 1.17073 .4938464 0.37 0.709 .5084164 2.695837 facility | .999623 .0000743 -5.07 0.000 .9994761 .99977 ptpay | .9983253 .0008064 -2.07 0.040 .996732 .9999211 age | .9815697 .0111312 -1.64 0.103 .9598048 1.003828 male | 2.388619 .7007413 2.97 0.004 1.337283 4.266485 income | 1.479376 .2072874 2.79 0.006 1.121385 1.951652 year98 | 1.7101 .6091849 1.51 0.134 .8455072 3.458801 year99 | 1.067614 .3470711 0.20 0.841 .5613698 2.030392 ----------------------------------------------------------------------------Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 17, 122) = 4.04 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9095046 .4874638 -0.18 0.860 .3151763 2.624558 hmo_fac | .9997377 .0001674 -1.57 0.119 .9994067 1.000069 two_plan | .6820931 .3325736 -0.78 0.434 .2601027 1.78872 cond | .8636156 .3349657 -0.38 0.706 .4010966 1.859482 hlth | .9403804 .1405025 -0.41 0.681 .6998422 1.263592 msa | 1.126653 .4005036 0.34 0.738 .557863 2.275375 region1 | .8028496 .3213691 -0.55 0.584 .363828 1.771627 region2 | 1.072771 .4025279 0.19 0.852 .510848 2.2528 region4 | 1.178798 .5044714 0.38 0.701 .5057564 2.747497 facility | .9996332 .0000727 -5.05 0.000 .9994895 .9997768 ptpay | .9982185 .0008629 -2.06 0.041 .9965137 .9999261 age | .9792819 .0110512 -1.86 0.066 .9576723 1.001379 male | 2.248773 .6622654 2.75 0.007 1.25617 4.025714 race | 2.544654 1.07838 2.20 0.029 1.100811 5.882263 income | 1.460671 .2125435 2.60 0.010 1.095463 1.947634 year98 | 1.73435 .6144754 1.55 0.122 .8607767 3.494485 year99 | 1.053843 .3490139 0.16 0.874 .5474938 2.028488 ----------------------------------------------------------------------Note: 1 failure and 0 successes completely determined.

PAGE 163

151pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 16, 123) = 3.47 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | 1.011592 .5274938 0.02 0.982 .3607616 2.83655 hmo_fac | .9997201 .0001588 -1.76 0.080 .9994062 1.000034 two_plan | .8401443 .416465 -0.35 0.726 .3152643 2.238891 cond | .7758477 .3412644 -0.58 0.565 .3251282 1.851392 hlth | .8685803 .1274104 -0.96 0.338 .6498965 1.160849 msa | 1.073009 .3525228 0.21 0.830 .5603748 2.054605 region1 | .8409945 .337624 -0.43 0.667 .3802317 1.860107 region2 | 1.037758 .3758545 0.10 0.919 .5070856 2.123788 region4 | 1.129338 .492339 0.28 0.781 .4769325 2.674184 facility | .9996499 .0000752 -4.65 0.000 .9995011 .9997987 ptpay | .9978564 .000962 -2.23 0.028 .995956 .9997603 age | .9840899 .0109863 -1.44 0.153 .9626046 1.006055 male | 2.29143 .6824461 2.78 0.006 1.271612 4.129133 race | 2.638887 1.05757 2.42 0.017 1.194736 5.828671 year98 | 1.864843 .6313965 1.84 0.068 .9547587 3.642426 year99 | 1.119538 .3633337 0.35 0.728 .5893142 2.126821 ---------------------------------------------------------------------Note: 1 failure and 0 successes completely determined. Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 16, 123) = 4.65 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .8031505 .4225186 -0.42 0.678 .2838169 2.272771 hmo_fac | .9997838 .0001615 -1.34 0.183 .9994644 1.000103 two_plan | .7126214 .3538517 -0.68 0.496 .2669661 1.902224 cond | .8612739 .3335961 -0.39 0.700 .4004328 1.852477 hlth | .92399 .1388556 -0.53 0.600 .6864652 1.243701 msa | 1.044234 .3446472 0.13 0.896 .5437197 2.00549 region1 | .9025447 .3664417 -0.25 0.801 .404404 2.01429 region2 | 1.15314 .4395927 0.37 0.709 .5426522 2.45043 region4 | 1.17073 .4938464 0.37 0.709 .5084164 2.695837 facility | .999623 .0000743 -5.07 0.000 .9994761 .99977 ptpay | .9983253 .0008064 -2.07 0.040 .996732 .9999211 age | .9815697 .0111312 -1.64 0.103 .9598048 1.003828 male | 2.388619 .7007413 2.97 0.004 1.337283 4.266485 income | 1.479376 .2072874 2.79 0.006 1.121385 1.951652 year98 | 1.7101 .6091849 1.51 0.134 .8455072 3.458801 year99 | 1.067614 .3470711 0.20 0.841 .5613698 2.030392 -----------------------------------------------------------------------

PAGE 164

152Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 15, 124) = 3.89 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .8927474 .4569677 -0.22 0.825 .3244683 2.45632 hmo_fac | .9997667 .0001533 -1.52 0.130 .9994636 1.00007 two_plan | .8821879 .4399125 -0.25 0.802 .3291131 2.364705 cond | .7675056 .3320244 -0.61 0.542 .3262815 1.805389 hlth | .8520016 .1258355 -1.08 0.280 .6362252 1.140959 msa | .9966332 .3085043 -0.01 0.991 .5404032 1.838031 region1 | .9537882 .3845471 -0.12 0.907 .4297634 2.116774 region2 | 1.117344 .4100505 0.30 0.763 .5408088 2.308502 region4 | 1.128047 .4841359 0.28 0.779 .4828097 2.635593 facility | .9996397 .0000774 -4.65 0.000 .9994867 .9997928 ptpay | .9979508 .000922 -2.22 0.028 .9961294 .9997756 age | .9866954 .0110327 -1.20 0.233 .9651197 1.008753 male | 2.433336 .7235946 2.99 0.003 1.351585 4.380875 year98 | 1.837382 .6257954 1.79 0.076 .936965 3.603095 year99 | 1.129383 .3596291 0.38 0.703 .6017213 2.119762 ----------------------------------------------------------------------Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 15, 124) = 4.12 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9927538 .5518802 -0.01 0.990 .3307236 2.980011 hmo_fac | .9997993 .0001702 -1.18 0.240 .9994627 1.000136 two_plan | .793969 .3967562 -0.46 0.645 .2955864 2.132665 cond | .7428195 .3251976 -0.68 0.498 .3125652 1.76533 hlth | .9983802 .1428803 -0.01 0.991 .7523173 1.324924 msa | 1.07904 .3352726 0.24 0.807 .5837371 1.994607 region1 | .9553079 .3730517 -0.12 0.907 .4413722 2.067672 region2 | 1.251194 .4835218 0.58 0.563 .582732 2.686459 region4 | 1.170464 .4970459 0.37 0.711 .5054656 2.710346 facility | .9996419 .00008 -4.47 0.000 .9994837 .9998001 age | .9800596 .0109926 -1.80 0.075 .9585632 1.002038 male | 2.297937 .7016206 2.72 0.007 1.256448 4.202731 income | 1.607916 .2491239 3.07 0.003 1.183627 2.184298 year98 | 1.580579 .5897776 1.23 0.222 .7557682 3.305549 year99 | 1.034163 .3242239 0.11 0.915 .556372 1.922264 -----------------------------------------------------------------------

PAGE 165

153Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 14, 125) = 2.92 Prob > F = 0.0007 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | 1.130191 .6488915 0.21 0.832 .3631756 3.517119 hmo_fac | .9997924 .0001612 -1.29 0.200 .9994737 1.000111 two_plan | 1.028034 .5341757 0.05 0.958 .3679619 2.872185 cond | .6397089 .3421223 -0.84 0.405 .22219 1.841791 hlth | .9232396 .131108 -0.56 0.575 .6972155 1.222536 msa | 1.045839 .3173666 0.15 0.883 .5739537 1.905691 region1 | 1.035261 .3997693 0.09 0.929 .4824452 2.221529 region2 | 1.260633 .4659618 0.63 0.532 .6069877 2.618169 region4 | 1.136762 .4947939 0.29 0.769 .4807202 2.688106 facility | .9996625 .0000904 -3.73 0.000 .9994837 .9998413 age | .9858788 .0115731 -1.21 0.228 .9632588 1.00903 male | 2.351418 .7667314 2.62 0.010 1.234018 4.480621 year98 | 1.675607 .6069928 1.42 0.156 .8186427 3.429651 year99 | 1.082641 .3347083 0.26 0.798 .5874885 1.995121 ----------------------------------------------------------------------Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 14, 125) = 4.46 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+-------------------------------------------------------one_hmo | .6342747 .2026674 -1.42 0.156 .3372019 1.193067 two_plan | .8009117 .4218269 -0.42 0.674 .2826863 2.269157 cond | .735261 .3251321 -0.70 0.488 .3066977 1.762676 hlth | .9901582 .1450145 -0.07 0.946 .7412048 1.322729 msa | 1.106567 .3408889 0.33 0.743 .6017787 2.034785 region1 | .9697639 .3856695 -0.08 0.939 .4417263 2.129015 region2 | 1.249357 .4856255 0.57 0.568 .5792916 2.694486 region4 | 1.171753 .4815901 0.39 0.700 .519873 2.641038 facility | .9995847 .0000702 -5.92 0.000 .999446 .9997235 age | .9800503 .0112708 -1.75 0.082 .9580161 1.002591 male | 2.318338 .704719 2.77 0.006 1.270992 4.228736 income | 1.613653 .2554129 3.02 0.003 1.180016 2.206644 year98 | 1.475401 .5361284 1.07 0.286 .7192261 3.026596 year99 | 1.04365 .3295909 0.14 0.893 .5589366 1.948712 ----------------------------------------------------------------------Note: 1 failure and 0 successes completely determined.

PAGE 166

154Survey logistic regression pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 14, 125) = 4.46 Prob > F = 0.0000 ----------------------------------------------------------------------inout | Coef. Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | -.4552731 .3195262 -1.42 0.156 -1.087073 .1765273 two_plan | -.2220046 .5266834 -0.42 0.674 -1.263418 .8194083 cond | -.3075298 .4421996 -0.70 0.488 -1.181893 .566833 hlth | -.0098905 .1464559 -0.07 0.946 -.2994783 .2796972 msa | .1012623 .3080599 0.33 0.743 -.5078656 .7103902 region1 | -.0307027 .3976943 -0.08 0.939 -.8170649 .7556596 region2 | .2226291 .3887003 0.57 0.568 -.5459493 .9912075 region4 | .1585006 .4109998 0.39 0.700 -.6541707 .9711719 facility | -.0004154 .0000702 -5.92 0.000 -.0005542-.0002765 age | -.0201513 .0115002 -1.75 0.082 -.0428907 .002588 male | .8408505 .303976 2.77 0.006 .2397978 1.441903 income | .4785003 .1582825 3.02 0.003 .1655278 .7914728 year98 | .3889295 .3633782 1.07 0.286 .3295795 1.107438 year99 | .0427246 .3158058 0.14 0.893 -.5817193 .6671685 _cons | .1480349 1.016397 0.15 0.884 -1.861691 2.157761 ----------------------------------------------------------------------Note: 1 failure and 0 successes completely determined. pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 15, 124) = 4.12 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Odds Ratio Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | .9927538 .5518802 -0.01 0.990 .3307236 2.980011 hmo_fac | .9997993 .0001702 -1.18 0.240 .9994627 1.000136 two_plan | .793969 .3967562 -0.46 0.645 .2955864 2.132665 cond | .7428195 .3251976 -0.68 0.498 .3125652 1.76533 hlth | .9983802 .1428803 -0.01 0.991 .7523173 1.324924 msa | 1.07904 .3352726 0.24 0.807 .5837371 1.994607 region1 | .9553079 .3730517 -0.12 0.907 .4413722 2.067672 region2 | 1.251194 .4835218 0.58 0.563 .582732 2.686459 region4 | 1.170464 .4970459 0.37 0.711 .5054656 2.710346 facility | .9996419 .00008 -4.47 0.000 .9994837 .9998001 age | .9800596 .0109926 -1.80 0.075 .9585632 1.002038 male | 2.297937 .7016206 2.72 0.007 1.256448 4.202731 income | 1.607916 .2491239 3.07 0.003 1.183627 2.184298 year98 | 1.580579 .5897776 1.23 0.222 .7557682 3.305549 year99 | 1.034163 .3242239 0.11 0.915 .556372 1.922264 ----------------------------------------------------------------------.

PAGE 167

155pweight: wtdper Number of obs = 391 Strata: newstr Number of strata = 61 PSU: newpsu Number of PSUs = 199 Population size =4376139.5 F( 16, 123) = 3.74 Prob > F = 0.0000 ---------------------------------------------------------------------inout | Coef. Std. Err. t P>|t| [95%Conf.Interval] -------------+--------------------------------------------------------one_hmo | -.141811 .6211707 -0.23 0.820 -1.370054 1.086432 hmo_fac | -.0001563 .0001879 -0.83 0.407 -.0005279 .0002153 two_plan | -.7086121 .8317536 -0.85 0.396 -2.353242 .9360173 twop_fac | .0001462 .0001586 0.92 0.358 -.0001675 .0004599 cond | -.2999424 .4367384 -0.69 0.493 -1.163507 .5636219 hlth | -.0021305 .1412747 -0.02 0.988 -.2814735 .2772126 msa | .0867287 .3211957 0.27 0.788 -.5483726 .72183 region1 | -.0498685 .3855087 -0.13 0.897 -.8121361 .7123992 region2 | .2580433 .3999005 0.65 0.520 -.5326814 1.048768 region4 | .1720747 .4315868 0.40 0.691 -.6813034 1.025453 facility | -.0004035 .0001085 -3.72 0.000 -.000618 -.0001889 age | -.0200183 .0113102 -1.77 0.079 -.0423821 .0023454 male | .8195855 .3089827 2.65 0.009 .208633 1.430538 income | .4799508 .1588233 3.02 0.003 .1659089 .7939928 year98 | .4472838 .3756376 1.19 0.236 -.2954657 1.190033 year99 | .0150893 .3158141 0.05 0.962 -.609371 .6395496 _cons | .0737217 1.035767 0.07 0.943 -1.974304 2.121747 ----------------------------------------------------------------------Note: 1 failure and 0 successes completely determined

PAGE 168

156 BIBLIOGRAPHY Adler, G.S., 1995. “Medicare Beneficiaries Rate Their Medical Care: New Data from MCBS.” Health Care Financing Rev 16: 175-87. Alexander, M., K. Grumbach, J. Sel by, A. F. Brown, and E. Washington. 1995. “Hospitalization for Congestive Heart Fa ilure. Explaining Racial Differences.” JAMA 274 (13): 1037-42. Allen, H.M., H. Darling, B. McNeill, and F. Bastien. 1994. “The Employee Health Care Value Survey: Round One.” Health Affairs 13: 25-41. American Hospital Association (AHA). 1987. Hospital Statistics Chicago, IL, AHA. Arrow, K.J. 1991. "The Economics of Agency." in Principals and Agents: the Structure of Business edited by J.W. Pratt and R.J. Zeckhauser, Boston, MA, Harvard Business School Press: 37-54. Ayanian, J.Z., P. D. Cleary, J. S. Weissman, and A. M. Epstein. 1999. “The Effect of Patients' Preferences on Racial Differences in Access to Renal Transplantation.” N Engl J Med 341 (22): 1661-9. Bartman, B. A., C. M. Clancy, E. M oy, and P. Langenberg. 1996. “Cost Differences Among Women's Primary Care Physicians.” Health Aff (Millwood) 15 (4): 177182. Bernstein, A. B., G. B. Thompson, and L. C. Harlan. 1991. “Differences in Rates of Cancer Screening by Usual Source of Me dical Care. Data from the 1987 National Health Interview Survey.” Med Care 29 (3): 196-209. Birkmeyer, J. D., S. M. Sharp, G. Finlays on Sr., E. S. Fisher, and J. E. Wennberg. 1998. “Variation Profiles of Common Surgical Procedures.” Surgery 124(5): 917-923. Blendon, R.J., M. Brodie, J. M. Benson, D. E. Altman, L. Levitt, T. Hoff, and L. Hugick. 1998. “Understanding The Managed Care Backlash.” Health Aff (Millwood) 17(4): 80-94. Block, P.C., I. Ockene, R. J. Goldberg, J. Bu tterly, E. H. Block, C. Degon, A. Beiser, and T. Colton. 1988. “A Prospective Randomized Trial of Outpatient vs. Inpatient Cardiac Catheterization.” N Engl J Med 319(19): 1251-5.

PAGE 169

157 Blustein, J., K. Hanson, and S. Shea. 1998. “Preventable Hospitalizations and Socioeconomic Status; Failure to Consid er Patients' Characteristics May Lead to the False Conclusion that Care is of Poor Quality.” Health Affairs 17(2): 177-89. Bondy, J., S. Berman, J. Glazner, and D. Lezotte. 2000. “Direct Expenditures Related to Otitis Media Diagnoses: Extrapolations from a Pediatric Medicaid Cohort.” Pediatrics 105(6): E72. Botman, S.L., T. F. Moore, C. L. Moriarity, and V. L. Parsons. 2000. “Design and estimation for the National Health In terview Survey, 1995–2004." National Center for Health Statistics. Vital Health Stat 2(130). Brown, R.S., D. G. Clement, J. W. Hill, S. ,M. Retchin, and J. W. Bergeron. 1993. “Do Health Maintenance Organizations Work for Medicare?” Health Care Financing Rev 15(1): 7-23. Brown, R.S., and K. Langwell. 1988 "Enrollment Patterns in Medicare HMOs: Implications for Access to Care." Advances in Health Economics and Health Services Research. 9: 69-96. Burns, L. R., S. E. Geller, and D. R. Wholey. 1995. “The Effect of Physician Factors on the Cesarean Section Decision.” Med Care 33(4): 365-82. Canto, J.G., W. J. Rogers, W. J. French, J. M. Gore, N. C. Chandra, and H. V. Barron. 2000. “Payer Status and the Utilization of Hospital Resources in Acute Myocardial Infarction: A Report From the National Re gistry of Myocardial Infarction 2.” Arch Intern Med 160: 817-823. Case, C., M. Johantgen, and C. Steiner. 2001. “Outpatient Mastectomy: Clinical, Payer, and Geographic Influences.” Health Services Research 36(5): 869-84. Castells, X., J. Alonso, M. Castilla, C. Ribo, F. Cots, and J. M. Anto. 2001. “Outcomes and Costs of Outpatient and Inpatient Cataract Surgery: a Randomised Clinical Trial.” J of Clinical Epidemiology 54: 23-29. The Centers for Medicare & Medicaid Services (CMS). Managed Care Contract Reports. Washington, D.C.: CMS, 2001. Cherkin, D.C., R. A. Deyo, K. Wheeler, a nd M. A. Ciol. 1994. “Physician Variation in Diagnostic Testing for Low Back Pain : What You See Is What You Get.” Arthritis & Rheumatism 37(1): 15-22., Clark, J.A., D. A. Potter, and J. B. McKinl ay. 1991. “Bringing Social Structure Back into Clinical Decision Making.” Soc Sci Med 32(8): 853-66. Cohen, S. 1997. “An Evaluation of Alternative PC-based Software Packages Developed for the Analysis of Complex Survey Data.” The American Statistician 51(3): 285292.

PAGE 170

158 Cohen, S., A. DiGaetano, and H. Goksel. 1999. Estimation Procedures in the 1996 Medical Expenditure Panel Survey Household Component U.S. Department of Health and Human Services, Public Health Service, Agency for Health Research and Quality. Davis, J.E., and D. E. Detmer. 1972. “The Ambulatory Surgical Unit.” Ann Surg 175: 856-862. Davis, K., S. Collins, C. Schoen, and C. Morris. 1995. “Choice Matters: Enrollees' Views of Their Health Plans.” Health Affairs 14: 99-112. Davis, S.K., D. K. Ahn, and S. P. Fo rtmann. 1998. “Determinants of Cholesterol Screening and Treatment Patterns Insights for Decision-Makers.” Am J Prev Med 15(3): 178-86. Detmer, D.E., and D. J. Buchanan-Davidson. 1989. "Ambulatory Surgery." in Socioeconomics of Surgery edited by J. M. Rutkow. St. Louis, MO, Mosby: 31-50. Detmer, D.E., and A. C. Gelijns. 1994. “Ambulatory Surgery: A More Cost-effective Treatment Surgery?” Arch Surg 129: 123-127. Dowd, B.R., F. I. Moscovice, P. Wisner, P. Bland, and M. Finch. An Analysis of Selectivity Bias in the Medicare AAPCC. Health Care Financing Review. 1996; 17(3): 35-57. Dranove, D., and W. D. White. 1987. “Agenc y and the Organization of Health Care Delivery.” Inquiry 24: 405-415. Elit, L.M., M. N. Levine, A. Gafni, T. J. Wh elan, G. Doig, D. L. Streiner, and B. Rosen. 1996. “Patients' Preferences for Therapy in Advanced Epithelial Ovarian Cancer: Development, Testing, and Applicati on of a Bedside Decision Instrument.” Gynecol Oncol 62(3): 329-35. Enthoven, A.C. 1978. “Consumer-choice health plan (first of two parts). Inflation and Inequity in Health Care Today: Alterna tives for Cost Control and an Analysis of Proposals for National Health Insurance.” N Engl J Med 298(12): 650-8. Every, N.R., C. P. Cannon, and C. Granger. 1998. “Influence of In surance Type on the Use of Procedures, Medications, and Hosp ital Outcome in Patients with Unstable Angina: Results from th e GUARANTEE Registry.” Journal of American College of Cardiology 32(2): 387-92. Federal Register. 2001. 2001 HHS Poverty Guidelines 66: 10695-7. Feldstein, P.J., T. M. Wickizer, and C. Wh eeler Jr. 1988. “Private Cost Containment: the Effect of Utilization Review Programs on Health Care Use and Expenditures.” The New England Journal of Medicine 318: 1310-4.

PAGE 171

159 Ferraro, K.F., J. A. Kelley-Moore. 2001. “Sel f-Rated Health and Mortality among Black and White Adults: Examining the Dynamic Evaluation Thesis.” J Gerontol B Psychol Sci Soc Sci 56(4): S195-205. Ferris, T.G., Y. Chang, D. Blumenthal, and S. D. Pearson. 2001. “Leaving Gatekeeping Behind--Effects of Opening Access to Specialists for Adults in a Health Maintenance Organization.” N Engl J Med 345(18): 1312-7. Ferris, T.G., J. M. Perrin, J. A. Manganell o, Y. Chang, N. Causino, and D. Blumenthal. 2001. “Switching to Gatekeeping: Change s in Expenditures and Utilization for Children.” Pediatrics 108: 280-290. Forrest, C. B., J. P. Weiner, J. Fowles, C. Vogeli, K. D. Frick, K. W. Lemke, and B. Starfield. 2001. “Self-Referral in Po int-of-Service Health Plans.” JAMA 285(17): 2223-2231. Fox, P. 2001. "An Overview of Managed Care." in The Managed Health Care Handbook edited by P. R. Kongstvedt. Gaithersburg, MD, An Aspen Publication. Gabel, J. 1997. “Ten Ways HMOs Have Changed During the 1990s.” Health Affairs 16(3): 134-145. Garfinkel, S.A., W. E. Schlenger, K. R. McLeroy, F. A. Bryan Jr., B. J. York, G. H. Hunteman, and A. S. Friedlob. 1986. "Choice of Payment Plan in the Medicare Capitation Demonstration." Medical Care 24(7): 628-40. Gornick, M. E. 2000. “Disparities in Medicare Services: Potentia l Causes, Plausible Explanations, and Recommendations.” Health Care Financing Rev 21(4): 23-43. Greene, W. H. 2000. "Functional Form, Non-linearity, and Specification" in Econometrics. Upper Saddle River, NJ, Prentice-Hall, Inc. Greenfield, S., E. C. Nelson, M. Zubkoff, W. Manning, W. Rogers, R. L. Kravitz, A. Keller, A. R. Tarlov, and J. E. Ware Jr 1992. “Variations in Resource Utilization among Medical Specialties and Systems of Care. Results from the Medical Outcome Study.” JAMA 267(12): 1624-1630. Greenfield, S., W. Rogers, M. Mangotich, M. F. Carney, and A. R. Tarlov. 1995. “Outcomes of Patients with Hypertensi on and Non-Insulin-Dependent Diabetes Mellitus Treated by Different Systems and Specialties.” JAMA 274(18): 1436-44. Guadagnoli, E., M. B. Landrum, E. A. Peterson, M. T. Gahart, T. J. Ryan, and B. J. McNeil. 2000. “Appropriateness of Cor onary Angiography after Myocardial Infarction among Medicare Beneficiaries: Managed Care vs. Fee for Service.” N Engl J Med 343(20): 1460-66. Guida, R.A., and K. F. Mattucci. 1990. “Tonsil lectomy and Adenoidectomy: an Inpatient or Outpatient Procedure?” Laryngoscope 100(5): 491-3.

PAGE 172

160 Hadley, J., and J. M. Mitchell. 1997/98. “Breast Cancer Treatment Choice and Mastectomy Length of Stay: a Comparis on of HMO and Other Privately Insured Women.” Inquiry 34: 288-301. Hellinger, F.J. 1998. “The Effect of Mana ged Care on Quality: A Review of Recent Evidence.” Arch Intern Med 158: 833-41. Hosmer, D. W., and S. Lemeshow. 2000. Applied Logistic Regression New York, NY, John Wiley & Sons, Inc. Idler, E. L., Y. Benyamini. 1997. “Self-Rated He alth and Mortality: a Review of TwentySeven Community Studies.” J Health Soc Behav 38(1): 21-37. The Kaiser Family Foundation and Health Research and Educational Trust. 2001. Employer Health Benefits, 2000 Annual Survey Kaiser Family Foundation. Kane, C. K. 1999. Physician Marketplace Report: Ph ysician Involvement with Managed Care. Chicago, IL, American Medical Association, Agency for Health Research and Quality. Kaplan, G.A., D. E. Goldberg, S. A. Evers on, R. D. Cohen, R. Salonen R, J. Tuomilehto, and J. Salonen. 1996. “Perceived Health Status and Morbidity and Mortality: Evidence from the Kuopio Ischaemic Heart Disease Risk Factor Study.” Int J Epidemiol 25(2): 259-65. Kern, M. J., M. Cohen, J. D. Talley, F. Litvack, H. Serota, F. Aguirre, U. Deligonul, and T. M. Bashore. 1990. “Early Ambulation af ter 5 French Diagnostic Catheterization: Results of a Multicenter Trial.” J Am Coll Cardiol 15: 1475-1483. Kongstvedt, P. R. 2000. Managed Care Measures: Resu lts of the 1999 Benchmarking Study. Washington, DC and Walnut Creek, CA, Ernst & Young LLP. Kongstvedt, P. R. 2001. Managing Basic Medical-Surgical Utilization. in The Managed Health Care Handbook edited by P. R. Kongstvedt. Gaithersburg, MD, Aspen Publishers, Inc.: 315. Kozak, L. J., E. McCarthy, R. PoKras. 1999. “C hanging Patterns of Surg ical Care In the United States, 1980-1995.” Health Care Financing Rev 21(1): 31-49. Kralewski, J., E. C. Rich, R. Feldman, B. E. Dowd, T. Bernhardt, C. Johnson, and W. Gold. 2000. “The Effects of Medical Group Practice and Physician Payment Methods on Costs of Care.” Health Serv Res 35(3): 591-613. Kramarow, E., H. Lentzner, and R. Rooks. 2001. Health, United States, 2001. Hyattsville, MD, National Center for Health Statistics.

PAGE 173

161 Kressin, N. R., and L. A. Petersen. 2001. “R acial Differences in the Use of Invasive Cardiovascular Procedures: Review of th e Literature and Prescription for Future Research.” Ann Intern Med 135: 352-66. Lee, J. C., J. R. Bengtson, J. Lipscomb, T. M. Bashore, D. B. Mark, R. M. Califf, D. B. Pryor, and M.A. Hlatky. 1990. “Feasibility a nd Cost-Saving Potential of Outpatient Cardiac Catheterization.” J Am Coll Cardiol 15(2): 378-84. Lee-Feldstein, A. C., and P. J. Feldst ein. 1994. “Treatment differences and other prognostic factors related to breast cancer surv ival: delivery systems and medical outcomes.” JAMA 271: 1163-68. Levetan, C., M. D. Passaro, K. A. Jablonsk i, C. Levetan, M. D. Passaro, and K. A. Jablonski. 1999. “Effect of Physician Specialty on Outcomes in Diabetic Ketoacidosis.” Diabetes Care 22(11): 1790-5. Levinthal, D. 1988. “A Survey of Agency Models of Organizations.” Journal of Economic Behaviour and Organization 9: 153-185. Lichtenstein, R., J. W. Thomas, J. Adams-Watson, J. Lepkowski, and B. Simon. 1991 "Selection Bias in TEFRA At-Risk HMOs." Medical Care 29(4): 318-31. Long, J. S. 1997. Regression Models for Categorical and Limited Dependent Variables Thousand Oaks, CA, Sage Publications. Lurie, N., J. Christianson, M. Finch, and I. Moscovice. 1994. “The Effects of Capitation on Health and Functional Status of the Medicaid Elderly. A Randomized Trial.” Ann Intern Med 120(6): 506-11. Manning, W, G, A. Leibowitz, G. A. Goldber g, W. H. Rogers, and J. P. Newhouse. 1984. “A Controlled Trial of the Effect of a Prepaid Group Practice on Use of Services.” N Engl J Med 310(23): 1505-10. McCloskey, L., D. B. Petitti, and C. J. H obel. 1992. “Variations in the Use of Cesarean Delivery for Dystocia: Lessons about the Resource of Care.” Med Care 30: 126-35. MEPS. 2001a. HC-020: 1997 Full Year Consolidated Data File. Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=36. Accessed August 2002. MEPS 2001b. MEPS HC-028: 1998 Full Year Consolidated Data File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=69. Accessed April 2003. MEPS 2001c. HC-016D: 1997 Hospital Inpatient Stays File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail. asp?ID=29. Accessed at October 2002.

PAGE 174

162 MEPS 2001d. HC-026D: 1998 Hospital Inpatient Stays File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=44. Accessed April 2003. MEPS 2001e. HC-016F: 1997 Outpatient Visits File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=29. Accessed October 2002. MEPS 2001f. HC-026F: 1998 Outpatient Visits File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=46 Accessed April 2003. MEPS 2001h. HC-027: 1998 Medical Conditions File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=68. Accessed April 2003. MEPS 2001h. 97HC/NHISLinkFile: 1997 MEPS/1995 & 1996 NHIS Link File Agency for Healthcare Research and Quality. MEPS 2002a. MEPS HC-038: 1999 Full Year Consolidated Data File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=93. Accessed April 2003. MEPS 2002b. HC-033D: 1999 Hospital Inpatient Stays File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=92. Accessed April 2003. MEPS 2002c. HC-033F: 1999 Outpatient Visits File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=89. Accessed April 2003. MEPS 2002d. HC-037: 1999 Medical Conditions File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=94. Accessed April 2003. MEPS 2002e. 98HC/NHISLinkFile: 1998 MEPS/1996 & 1997 NHIS Link File Agency for Healthcare Research and Quality. MEPS 2002f. 99HC/NHISLinkFile: 1999 MEPS/1997 & 1998 NHIS Link File, Agency for Healthcare Research and Quality. MEPS 2003a. H36 1996-1999 Pooled Estimation File Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=86. Accessed April 2003. MEPS 2003b. HC-047: 1997-2000 Person Round Plan Files Agency for Healthcare Research and Quality Web Site. Available at: http://www.meps.ahrq.gov/Puf/PufDetail.asp?ID=122. Accessed April 2003.

PAGE 175

163 Miller, R. H., and H. S. Luft. 1997 "Does Managed Care Lead to Better or Worse Quality of Care?" Health Affairs 16(5):7-25. Mills, R.J. 2000. "Health Insurance Coverage:2000." Current Population Report Washington, D.C., U.S. Census Bureau. Mitchell, J. B., and B. Harrow. 1994. “Costs a nd outcomes of inpatient versus outpatient hernia repair.” Health Policy 28(2): 143-52. Morgenstern, L. B., L. Steffen-Batey, M. A. Smith, and L. A. Moey. 2001. “Barriers to Acute Stroke Therapy and Stroke Prevention in Mexican Americans.” Stroke 32: 1360. Mort, E., J. S. Weissmnn, and A. M. Ep stein. 1994. “Physician Discretion And Racial Variation In The Use Of Surgical-Procedures.” Archives Of Internal Medicine 154(7): 761-7. Mossey, J. M., and E. Shapiro. 1982. “Self-rate d health: a predictor of mortality among the elderly.” Am J Public Health 72(8): 800-8. Nattinger, A., and J. S. Goodwin. 1994. “Ge ographic and Hospital Variation in the Management of Older Women with Breast Cancer.” Cancer Control 1(4): 334-338. Nattinger, A., A. Gottlieb, J. Veum, D. Ya hnke, and J. S. Goodwin. 1992. “Geographic Variation in the Use of Breast-Conserving Treatment for Breast Cancer.” N Engl J Med 326: 1102-1107. Nelson, E. C., C. A. McHorney, W. G. Manning Jr., W. H. Rogers, M. Zubkoff, S. Greenfield, J. E. Ware Jr., and A. R. Tarlov. 1998. “A Longitudinal Study of Hospitalization Rates for Patients with Ch ronic Disease: Results from the Medical Outcomes Study.” Health Services Research 32(6): 759-774. O'Leary, M. P., W. D. Gee, H. L. Holtgrewe, M. L. Blute, T. P. Cooper, B. J. Miles, R. E. Nellans, R. Thomas, M. R. Painter, J. J. Meyer, M. J. Naslund, E. A. Gormley, R. Blizzard, and R. B. Fenninger. 2000. “ 1999 American Urological Association Gallup Survey: Changes in Physician Prac tice Patterns, Treatment of Incontinence and Bladder Cancer, and Impact of Managed care.” J Urol 164(4): 1311-6. Owings, M., and L. J. Kozak. 1998. “Ambulatory and Inpatient Procedures in the United States, 1996.” Vital Health Stat 13(139). Owings, M. F., and L. Lawrence 1999. “Deta iled Diagnoses and Procedures. National Hospital Discharge Survey, 1997." Nationa l Center for Health Statistics. Vital Health Stat 13(145). Pati, S., S. Shea, D. Rabinowitz, and O. Carrasquillo. 2003. “Does Gatekeeping Control Costs for Privately Insured Children? Findings from the 1996 Medical Expenditure Panel Survey.” Pediatrics 111(3): 456-60.

PAGE 176

164 Pauly, M. V., and M. H. Erder 1993. “Insurance Incentives for Ambulatory Surgery.” Health Serv Res 27(6): 813-839. Pearson, S. D., T. H. Lee, E. Lindsey, T. Hawkins, E. F. Cook, and L. Goldman. 1994. "The Impact of Membership in a Health Maintenance Organization on Hospital Admission Rates for Acute Chest pain." Health Serv Res 29(1): 59-74. Peduzzi, P, J. Concato, E. Kemper, T. R. Holford, and A. R. Feinstein. 1996. “A Simulation Study of the Number of Events per Variable in Logistic Regression Analysis.” J Clin Epidemiol 49(12): 1373-1379., Pepine, C. J., H. D. Allen, T. M. Bashore, J. A. Brinker, L. H. Cohn, J. C. Dillon, L. D. Hillis, F. J. Klocke, W. W. Parmley, and T. A. Ports. 1991. “ACC/AHA Guidelines for Cardiac Catheterization and Cardiac Catheterization Laboratories. American College of Cardiology/American Heart A ssociation Ad Hoc Task Force on Cardiac Catheterization." Circulation 84(5):2213-47. Philbin, E. F., P. A. McCullough, and T. G. DiSalvo. 2000. “Socioeconomic Status Is an Important Determinant of the Use of Inva sive Procedures Af ter Acute Myocardial Infarction in New York State.” Circulation 102(suppl III): III-107-III-115. Phillips, K.A., M. L. Mayer, and L. A. Aday. 2000. “Barriers to Care Among Racial/Ethnic Group Under Managed Ca re: Ethnic Minorities Continue to Encounter Barriers to Care in the Curre nt Managed Care Dominated U.S. Health Care System.” Health Affairs 2000(4): 65-75. Pontes, M.C. 1995. “Agency Theory: a Framew ork for Analyzing Physician Services.” Health Care Manage Rev 20(4): 57-67. Popovic, J. R. 2001. “1999 National Hospital Discharge Survey: Annual Summary with Detailed Diagnosis and Procedure Data." National Center for Health Statistics. Vital Health Stat 13(151). Popovic, J. R., and L. J. Kozak. 2000. “National Hospital Discharge Survey: Annual Summary, 1998.” Vital Health Stat 13(148). Potosky, A. L., R. M. Merrill, G. F. Riley, S. H. Taplin, W. Barlow. B. H. Fireman, and R. Ballard-Barbash. 1997. “Breast Cancer Survival and Treatment in Health Maintenance Organization and Fee-for-Service Settings.” J Natl Cancer Inst 89: 1683-91. Potosky, R. L., R. M. Merrill, G. F. Riley, S. H. Taplin, W. Barlow, B. H. Fireman, and J. D. Lubitz. 1999. “Prostate Cancer Tr eatment and Ten-Year Survival among Group Staff HMO and Fee-for-Ser vice Medicare Patients.” Health Services Research 34(2): 525-46.

PAGE 177

165 Rakover, Y, R. Almog, and G. Rosen. 1997. “The Risk of Postoperative Haemorrhage in Tonsillectomy as an Outpatient Procedure in Children.” Int J Pediatr Otorhinolaryngol 41(1): 29-36. Randall, T. 1993. “Varied Mammogram Readings Worry Researchers.” JAMA 269(20): 2616. Reiner, S. A., W. P. Sawyer, K. F. Clar k, and M. W. Wood. 1990. “Safety of Outpatient Tonsillectomy and Adenoidectomy.” Otolaryngol Head Neck Surg 102(2): 161-8. Reschovsky, J. D., P. Kemper, and H. Tu. 2000. “Does Type of Health Insurance Affect Health Care Use and Assessments of Care among the Privately Insured?” Health Serv Res 35(1): 219-237. Retchin, S. M., and B. Brown. 1990. “The Qu ality of Ambulatory Care in Medicare Health Maintenance Organizations.” Am J Public Health 80(4): 411-5. Rizzo, J. A. 1993. “Physician Uncertain ty and the Art of Persuasion.” Soc Sci Med 37(12): 1451-1459. Robinson, J. C. 1996. “Decline in Hospital Ut ilization and Cost Inflation under Managed Care in California.” JAMA 276(13): 1060-4. Robinson, J. C. 1999. “Blended Payment Methods in Physician Organizations under Managed Care.” JAMA 282(13): 1258-63. Robinson, R, and A. Steiner. 1998. Managed Health Care Buckingham, Open University Press. Rochaix, L. 1989. “Information Asymmetry and Search in the Market for Physicians' Services.” J Health Econ 8(1): 53-84. Rosenberg, S. N., D. R. Allen, J. S. Handte, T. C. Jackson, L. Leto, B. M. Rodstein, S. D. Stratton, G. Westfall, and R. Yasser. 1995. “Effect of Utilization Review in a Feefor-Service Health Insurance Plan.” The New England Journal of Medicine 333: 1326-30. Rutchik, S. D., M. Baudiere, M. Wade, G. Sullivan, W. Rayford, and J. Goodman. 2001. “Practice Patterns in the Diagnosis a nd Treatment of Erectile Dysfunction among Family Practice Physicians.” Urology 57(1): 146-50. Sada, M. J., W. J. French, and D. M. Carlisle. 1998. “Influence of Payor on Use of Invasive Cardiac Procedures and Patient Outcome After Myocardial Infarction in the United States.” Journal of American College of Cardiology 31(7): 1474-80. Safran, D. G., A. R. Tariov, and W. H. R ogers. 1994. “Primary Care Performance in Feefor-Service and Prepaid Health Care Syst ems: Results from the Medical Outcomes Study.” JAMA 271(20): 1579-86.

PAGE 178

166 Seckler, J., and D. A. Held 1990. “Patient Satisfaction with Diagnostic Cardiac Catheterization: Ambulatory vs. Inpatient Performance.” Mt Sinai J Med 57(6): 381-8. Selby, J., K. Grumbach, C. P. Quesenberry Jr J. A. Schmittdiel, and C. P. Quesenberry Jr, 1999. “Differences in Resource Use and Costs of Primary Care in a Large HMO According to Physician Ppecialty.” Health Serv Res 34(2): 503-18. Skinner, J. S., and P. C. Adams. 1996. “Outpatient Cardiac Catheterisation.” Int J Cardiol 53(3): 209-19. Stafford RS (1990). “Cesarean Section Us e and Source of Payment: An Analysis of California Hospital Discharge Abstracts.” Am J Public Health 80: 313-15. Stafford, R. S. 1991. “The Impact of Non-clin ical Factors on Repeat Caesarean Section.” JAMA 265(1): 59-63. The Standards Task Force, American So ciety of Colon and Rectal Surgeons. 1991. Practice Parameters for Ambulatory Anorectal Surgery American Society of Colon and Rectal Surgeons. Available at : http://www.fascrs.org/ascrspp-aas.html. Accessed October 2002. StataCorp 2001. Stata Reference Manual, Release 7 College Station, TX, Stata Press. Stephenson, S. V. 1985. “Ambulatory Surgical Centers.” JAMA 253(342-343). Sturm, R. 1997. “How Expensive Is Unlimite d Mental Health Care Coverage Under Managed Care?” JAMA 278(18): 1533-7. Terry, K. 1999. “Capitation on the Rise.” Med Econ 76(188-201). Trauner, J. B., and J. S. Chesnutt. 1996. “M edical Groups in Calif ornia: Managing Care under capitation.” Health Affairs 15(1): 159-170. Truy, E, F. Merad, P. Robin, B. Fantino, and A. Morgon. 1994. “Failures in Outpatient Tonsillectomy Policy in Children: a Retrospective Study in 311 Children.” Int J Pediatr Otorhinolaryngol 29(1): 33-42. Tu, H. T., P. Kemper, and H. J. Wong. 1999. “Do HMOs Make a Difference? Use of Health Services.” Inquiry 36(4): 400-410. Tussing, A. D., and J. A. Wojtowycz. 1994. “Health Maintenance Organizations, Independent Practice Associations and Cesarea Section Rates.” Health Serv Res 29: 75-93. U.S. Dept. of Health and Human Services, National Center for Health Statistics. 1998a. National Health Interview Survey, 1996. U.S. Dept. of Health and Human Services, National Center for Health Statistics.

PAGE 179

167 U.S. Dept. of Health and Human Services, National Center for Health Statistics. 1998b. National Health Interview Survey, 1997. U.S. Dept. of Health and Human Services, National Center for Health Statistics. U.S. Dept. of Health and Human Services, National Center for Health Statistics. 1999. National Health Interview Survey, 1998. U.S. Dept. of Health and Human Services, National Center for Health Statistics. Udvarhelyi, I. S., K. Jennison, R. S. Phil lips, and A. M. Epstein. 1991. “Comparison of the Quality of Ambulatory Care for Fe e-for-Service and Prepaid Patients.” Ann Intern Med 115(5): 394-400. Ware, J. E., M. S. Bayliss, W. H. Rogers, M. Kosinski, and A. R. Tarlov. 1996. “Differences in 4-year Health Outcomes for Elderly and Poor, Chronically Ill Patients treated in HMO and fee-for-service systems.” JAMA 276(13): 1039-47. Weinick, R. M., and K. M. Beauregard. 1998. “Women's Use of Preventive Screening Services: a Comparison of HMO vs. Fee-for-Service Enrollees.” Med Care Research and Review 54: 176-99. Weinick, R. M., S. H. Zuvekas, and J. W. Cohen. 2000. “Racial and Ethnic Differences in Access to and Use of Health Care Services, 1977 to 1996.” Medical Care Research and Review 57(Supp 1): 36-54. Wennberg, J. E. 1984. “Dealing with Medi cal Practice Variati ons: a Proposal for Action.” Health Affairs 3: 6-32. Wennberg, J. E., B. A. Barnes, and M. Zubkoff. 1982. “Professional Uncertainty and the Problem of Supplier-Induced Demand.” Soc. Sci. Med 16: 811-824. Wolff, N. 1989. “Professional uncertainty and physician medical decision-making in a multiple treatment framework.” Soc Sci Med 28(2): 99-107. Wolff, N., and M. Schlesinger. 1998. “Risk, Mo tives, and Styles of Utilization Review: A Cross-Condition Comparison.” Soc Sci Med 47(7): 911-26. Wrobel, J. S., J. A. Mayfield, and G. E. Reiber. 2001. “Geographic Variation of LowerExtremity Major Amputation in Individuals With and Without Diabetes in the Medicare Population.” Diabetes Care 24(5): 860-64. Yelin, E. H., L. A. Criswell, and P. G. Feigenbaum. 1996. “Health Care Utilization and Outcomes among Persons with Rheumatoid Arthritis in Fee-For-Service and Prepaid Group Practice Settings.” JAMA 276(13): 1048-53.

PAGE 180

168 BIOGRAPHICAL SKETCH Hsou Mei Hu was born on February 13, 1960, to Koa-Jin Hu and I-Tse Su Hu in Shanhua, Tainan County, Taiwan. She atte nded and graduated from Provincial Tainan Girls High School. She continued her e ducation at the National Yang-Ming Medical College, where she received a degree of Bach elor of Science in medical technology in 1983. During the following nine years, she wo rked as a medical technologist at the Taichung Veterans General Hospital, and as a research assistant in the Institute of Molecular Biology, Academia Sinica, Taiwan. She moved to the United States in 1992 with her husband, Shiuhyang Kuo, who began hi s graduate study at the University of Florida. She started her graduate study in health and hospital administration at the University of Florida in 1994, and graduate d in 1996 with dual degrees of Master of Business Administration and Master of Hea lth Science (MBA/MHS). She entered the doctoral program in health services res earch in 1999. Upon completion of the doctoral study, she will attend the Mental Health Se rvices and Systems Research Postdoctoral Training Program, funded by the National Inst itute of Mental Health, at Rutgers University, New Jersey. She is married to Shiuhyang Kuo, and has a daughter, Rachel Chiungshin Kuo.


Permanent Link: http://ufdc.ufl.edu/UFE0001083/00001

Material Information

Title: Effect of HMO coverage on the choice of outpatient or inpatient surgery
Physical Description: Mixed Material
Creator: Hu, Hsou Mei ( Author, Primary )
Publication Date: 2003
Copyright Date: 2003

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0001083:00001

Permanent Link: http://ufdc.ufl.edu/UFE0001083/00001

Material Information

Title: Effect of HMO coverage on the choice of outpatient or inpatient surgery
Physical Description: Mixed Material
Creator: Hu, Hsou Mei ( Author, Primary )
Publication Date: 2003
Copyright Date: 2003

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0001083:00001


This item has the following downloads:


Full Text












EFFECT OF HMO COVERAGE ON THE CHOICE OF OUTPATIENT OR
INPATIENT SURGERY














By

HSOU MEI HU


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2003




























Copyright 2003

by

HSOU MEI HU














This work is dedicated to my family.















ACKNOWLEDGMENTS

I thank Dr. Niccie L. McKay, chair of my graduate advisory committee, for her

support, guidance, and encouragement through all the stages. Her straightforward

teaching style has been extremely helpful in clarifying my research goals. I thank Dr. R.

Paul Duncan for his support, counsel, and valuable discussion. I appreciate all the

opportunities to work on his research projects that have enhanced my analytical skills. I

also thank Dr. Tiffany A. Radcliff for sharing her knowledge in econometrics, especially

her frequent advice on my analyses. I would also like to thank Dr. Cynthia Garvan for

her help on improving my skills in statistical analysis. I appreciate Dr. Alan Agresti in

the Department of Statistics for his help during the early stage of my dissertation. I

extend my thanks to all professors and staff from the Department of Health Services

Administration for all the discussions, advice, and friendship.

I am grateful for my parents' supports that allowed their grown-up daughter to

pursue her dream. I would also like to thank my brothers and sisters in Taiwan,

especially my sister, Shoushing, and my brother, Fonlin, who have assumed most of the

responsibility of taking care of our parents and our families while I am in the United

States.

Finally, I must thank my husband, Shiuhyang. During my doctoral study, he has

shared so much responsibility with me in taking care of our daughter, Rachel. His crucial

support has made my study possible.
















TABLE OF CONTENTS

Page

A C K N O W L E D G M E N T S ................................................................................................. iv

TA B LE O F C O N TEN T S................................................................... .......................... v

LIST O F TA BLE S ............................ .................... ..... ...... .. ... viii

LIST OF FIGURES ................................. .. ... ... ................. .x

ABSTRACT .............. ......................................... xi

CHAPTER

1 IN TRODU CTION ................................................. ...... .................

2 BACKGROUND ..................................... .......... ................. .5

The U .S. H health Insurance System ...................................................... .... ........... 5
T ypes of P private H health P lans.......................................................................... .... ... 6
F F S P lan s ................................................................................... . 7
H M O P lan s ....................................................... 8
PP O s and P O S P lans ............................................ .. .. .. ...... .......... 12

3 LITERATURE REVIEW ............................................................. ............... 14

U tiliz a tio n ....................................................................................................... 1 4
Inpatient Care ........................................... ......... 14
Prim ary C are vs. Specialty C are................................... ..................................... 15
N on-Physician Practitioner Care ....................................................................... 16
P rev en tiv e C are ............................................................................................. 16
E expensive Procedures................................................. ............................. 17
Su m m ary : U utilization ................................................................ .................... 18
Patient Outcomes and Quality of Care ......................................................... 19
M mortality ................................................................ ..... ........... 19
S u rv iv a l R a te ................................................................................................. 1 9
P hy sical and M ental H health ...................................................................... .. .... 20
Patient Satisfaction ............................................... ... ..... .............. ... 21
The Shift from Inpatient to Outpatient Surgery........................................................21
U utilization ............................................................... 22


v









O u tc o m e s ....................................................................................................... 2 2
C o sts ................................................................2 3
R research Q u estion s........... ................................................................ ........ .. ...... .. 24

4 CONCEPTUAL FRAM EW ORK ................................................................ ......... 25

A Two-Stage Model of Physician Decision-Making...............................................25
Stage One: Determine a Set of Feasible Treatments ........................... ........25
Stage Two: Choose a Specific Treatment Method.................... ..................28
Choice of Treatm ent M ethods ............................ ........................ ...... ......... 29
Professional Uncertainty .................................. .....................................29
T he C conceptual M odel I ............................................... ........................... .............30
Physician Characteristics.................. ..... ..... ...................... 31
The Relationship between Physician and Payer Characteristics.......................34
T h e F in al M o d el ..................................................... ................ 3 6

5 EM PIRICAL SPECIFICATION ........................................ .......................... 39

D dependent V variable ........................................... ... .... ........ ......... 39
Independent V ariables ...................................................... ........ .. ................4 1
Prim ary Independent V ariable....................................... .......................... 41
Control Variables: Patient Characteristics........................... ..........................41
Control Variables: Physician Characteristics ....................................... ......... 47
Control Variables: Other Payer Characteristics ...............................................49
The A nalytic M odel .................. ........................................... .............. 50

6 DATA, VARIABLES AND STATISTICAL ANALYSIS ........................................51

D a ta ...................................... ....... ........................ ................ 5 1
Sam ple Design and Sample W eights........................................ ............... 52
M E P S D ata C collection ............................................................. .....................53
Construction of Dataset for Analysis ......................................... ............. 55
A Subset of the Constructed Dataset...... .................. .............59
V ariab le s ............... ...... ....... ...................................................7 1
D dependent V variable .......... ............ .................... ..... .. .. ........ .... 71
Prim ary Independent V ariables ........................................ ........ ............... 71
Control Variables: Patient Characteristics........... ..... .................77
Control Variables: Physician Characteristics ..................... ................ ........... 80
Control Variables: Other Payer Characteristics ...............................................80
Other Control Variable ........... .. .................................. 80
Statistical A nalysis................................................... 80
S a m p le S iz e ................................................................................................... 8 2
A n aly tic al Issu e s ............................................................................................ 8 3

7 FINDINGS ........................................................ ..................87

C contingency T ables ..............................................................87









Two-way Tables: Surgery Setting by Health Plan Type.....................................88
Three-way Tables for Control Variables: Surgery Setting by Health Plan Type90
Mean Charge and Payment by Surgery Setting and by Health Plan Type..........97
Logistic Regression Analysis ............................................................................100
U nivariate Regression A analysis ............................................. ............... 100
M ultivariate Regression Analysis................................ ........................ 103
S u m m a ry ...................................................................................................... 1 1 7

8 DISCUSSION AND CONCLUSION ...........................................................118

APPENDIX

A TWO-DIGIT ICD-9 PROCEDURE CODES REPORTING BOTH INPATIENT
AND OUTPATIENT PROCEDURES IN 1996, AGE 0 TO 64 .............................123

B SAS PROGRAM FOR CONSTRUCTING DATASETS ........................................126

C REGRESSION MODELS FOR ALL SURGICAL CASES: MAIN EFFECT AND
INTERACTION ...................................... .. ........... ....... ........138

D REGRESSION MODELS FOR THE SUBSET OF CASES: MAIN EFFECT AND
INTERACTION ...................................... .. ........... ....... ........145

BIBLIO GRAPH Y .................. ............ .... .............. .......... .. ... ........ .. .. 156

BIOGRAPH ICAL SKETCH .................................... ............................................ 168
















LIST OF TABLES


Table page

2-1 Health Insurance Coverage for Americans under Age 65, 2000.............................5

2-2 Employer-Sponsored Health Insurance Coverage by Plan Type, 1988-2000............8

5-1 Independent Variables Affecting the Probability of Choosing a Surgery Setting ...50

6-1 Number of Cases by Type of Procedure ..... ......... ........................................ 60

6-2 Number of the Subset of Cases by Type of Procedure ............................................62

6-3 Independent Variables by Type of Characteristic, Conceptual Basis, and Questions
from the M E P S ......................................................................63

6-4 D description of V ariables................................................ .... .. ......................... 70

6-5 Descriptive Statistics for Discrete Variables (Unweighted n=814; Weighted
n= 9,59 5,6 57) .........................................................................73

6-6 Descriptive Statistics for Continuous Variables (Unweighted n=814; Weighted
n= 9,59 5,6 57) .........................................................................74

6-7 Comparing Cases with a Non-HMO Non- Gatekeeper Plan and Cases with No Self-
Reported Plan Information (Nominal and Ordinal Variables)...............................75

6-8 Comparing Cases with a Non-HMO Non-Gatekeeper Plan and Cases with No Self-
Reported Plan Information (Continuous Variables).....................................77

6-9 Cases for the Subset of the Dataset by Health Plan Coverage...............................77

7-1 Weighted Number (Percent) of Outpatient and Inpatient Surgical Cases by Health
Plan Coverage (Based on Unweighted n=814) .....................................................88

7-2 Unweighted Number (Percent) of Outpatient and Inpatient Surgical Cases by
Health Plan Coverage (Based on Unweighted n=814) ........................................88

7-3 Weighted Outpatient and Inpatient Surgical Cases of the Subset of Data by Health
Plan Coverage (Based on Unweighted n=391) ................................ ............... 89









7-4 Unweighted Outpatient and Inpatient Surgical Cases of the Subset of Data by
H health P lan C overage ............................................................... ........... ...... .90

7-5 Distribution of Selected Control Variables by Surgical Setting and Health Plan
Type (n=814) ..................................................................... .........93

7-6 Distribution of Selected Control Variables by Surgical Setting and Health Plan
Type, the Subset of Cases (n=391)...................................... ......................... 95

7-7 Mean Charges and Payment to Surgeries by Health Plan Type, All Cases (n=814)99

7-8 Mean Charges and Payment by HMO, the Subset of Cases (n=391) ..................100

7-9 Univariate Regression Analysis: Nominal Variables..............................102

7-10 Univariate Regression Analysis: Ordinal and Continuous Variables ..................103

7-11 Logistic Regression for All Cases: Main Effect Only ................. .. ...................107

7-12 Stage One: Logistic Regression Results that Predict HMO Status ......................109

7-13 Mean Predicted Probability of HMO Status by Observed HMO Membership...... 110

7-14 Stage Two: Logistic Regression Results, Using Predicted HMO Membership as the
Prim ary Independent Variable. ................................................................... ...... 110

7-15 Logistic Regression for All Cases: Main Effect and Interaction ...........................111

7-16 Estimated Probability of Having an Outpatient Surgery by HMO Status............. 113

7-17 Logistic Regression for the Subset of Cases: Main Effect and Interaction............14

7-18 Goodness of Fit of the Final Models ......................................................116
















LIST OF FIGURES


Figure p

2-1 Employer-Sponsored Health Insurance Coverage by Plan Type, 1988-2000...........6

2-2 Percent of Americans Enrolled in HMOs, 1976-2000 ............................................8

2-3 Privately Insured HMO Enrollment by Model Type from 1976 to 2000 ...............10

4-1 The Two-Stage Model of Physician Decision-Making................ .............. ....26

4-2 Conceptual Model I of Choice between Inpatient and Outpatient Surgery ............31

4-3 The Conceptual Model II of Choice between Inpatient and Outpatient Surgery.....37

6-1 The M EPS Data Collection ......... .................................... ....................... 54

6-2 D ataset C on struction ........................................................................ .................. 58

6-3 Types of Health Plan Coverage for Cases with Two-plan Coverage....................72

6-4 The Possible Self-Selection Effect. ............................................... ............... 83

6-5 The Effect of Unobserved Variable: Surgeon's Specialty ........................... 84

6-6 The Effect of Unobserved Variables: Payment Method, and Utilization
Management ............. ......... ..... ..................... 84

7-1 Correlation Matrix for Number of Condition (COND), Health Status (HLTH), And
Total Charge (TCH ). ........................ ....... ..... .. ...... .............. 105

7-2 Correlation Matrix of Charge and Payment Variables...................................106















Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy


EFFECT OF HMO COVERAGE ON THE CHOICE OF OUTPATIENT OR
INPATIENT SURGERY
By

Hsou Mei Hu

August 2003

Chair: Niccie L. McKay
Major Department: Health Services Administration

This dissertation studies the effect of health maintenance organization (HMO)

coverage and gatekeeping on the choice of surgery setting. The study population is

people under age 65 who require a surgery that is feasible in either the outpatient or

inpatient setting. This dissertation constructs a dataset using the pooled 1997, 1998, and

1999 Medical Expenditure Panel Survey (MEPS). The constructed dataset includes 814

cases; a subset of the data (391 cases) excludes surgeries that were primarily done in

either the inpatient or outpatient setting, with few done in the other setting.

Because the dependent variable is dichotomous (outpatient or inpatient), logistic

regression is specified to analyze the relationship between the likelihood of choosing an

outpatient surgery (vs. inpatient surgery) and the primary independent variables (HMO

coverage and gatekeeper plan coverage), controlling for severity, patient characteristics,

and payer characteristics.









This dissertation found that having HMO coverage did not increase the odds of

having an outpatient surgery. Rather, the interaction between HMO status and facility

payment had a significant effect on the likelihood of choosing an outpatient surgery.

When facility payment increased, the likelihood of having an outpatient surgery for HMO

patients dropped more than that for non-HMO patients. For example, when facility

payment was increased by $400, reduced the probability of having an outpatient surgery

for HMO patients decreased by 2%, but for non-HMO patients by only 0.6%.

Gatekeeping did not significantly affect the likelihood of having an outpatient surgery.

For the subset of cases, HMO status did not show a stronger effect on the use of

outpatient surgery than for all cases with surgeries in general.

These conclusions appear to be inconsistent with the general belief that HMOs

control costs by directly controlling the use of care. Rather, this dissertation found that

HMOs paid less for a surgery than non-HMOs. However, when payment for outpatient

surgery increased, HMOs were more aggressive in controlling the use of this type of care.

These findings on HMO utilization patterns may help to identify strategies that promote

the appropriate use of care and reduce healthcare costs.














CHAPTER 1
INTRODUCTION

This dissertation studies the effect of HMO coverage on the choice of inpatient or

outpatient surgery. During the past two decades, outpatient surgery has become a

generally acceptable practice for many medical conditions, and is perceived to cost less

than inpatient surgery. Because HMOs are thought to control costs by promoting

enrollees' wellness and preventing health problems while reducing the use of expensive

care, HMOs are expected to prefer outpatient surgeries to inpatient surgeries given that

outpatient surgeries cost less. This dissertation compares utilization patterns of inpatient

and outpatient surgeries according to type of health plan, thus helping to identify

strategies that promote appropriate use of care and reduce healthcare costs.

As national healthcare expenditures continue to grow, controlling costs remains an

important issue for the U.S. health care system. In 1999, national health care

expenditures totaled $1.2 trillion, a 5.6% increase from 1998 (Kramarow et al. 2001).

While the percent of gross domestic product (GDP) spent on healthcare, 13%, remained

the same as in 1998, the 1999 growth rate exceeded the 1998 rate of 4.8%. This growth

has resulted in payers' continued efforts to control healthcare costs.

Managed care has been one of the primary methods used by payers to contain

healthcare costs. Both private payers and public payers offer managed care plans. In

1999, 6.1 million Medicare beneficiaries enrolled in the capitated program called

Medicare+Choice, which was triple the enrollment in 1994 (Centers for Medicare and

Medicaid Services 2001). As of June 1997, 19.5 million Medicaid recipients were









enrolled in managed care plans, including 8.4 million in HMOs and the rest in primary

care case management (PCCM) programs (Fox 2001). For people with employment-

based health insurance, enrollment in managed care plans increased from 27% in 1988 to

92% in 2000 (Kaiser Family Foundation and Health Research Educational Trust 2001).

HMO enrollment rose from 16% of all Americans with employment-based health

insurance in 1988 to 31% in 1996, and has remained stable since then (Kaiser Family

Foundation and Health Research Educational Trust 2001). Nationally, HMO enrollment

was 30% of the population in 2000 (Kramarow et al. 2001).

The efforts to control healthcare expenditures have focused on specific types of

services. During the early 1980s, 40% of national health expenditures went to hospital

care (Kramarow et al. 2001). In 1983, Medicare introduced the Prospective Payment

System (PPS) that paid hospitals a fixed fee based on the specified Diagnostic Related

Group (DRG) regardless of the level of services provided. Inpatient surgery was one of

the services immediately affected by PPS. Because the payment for an inpatient surgery

was often too low to cover the costs, there were incentives to shift to an outpatient

setting.

The 1989 physician payment reforms introduced a schedule of prices for physician

services based on Resource-Based Relative Value Scales (RBRVS) to facilitate control of

physician expenditure growth. The fees for some technically oriented services, such as

certain surgeries, were considered excessive, while services like primary care were

considered to be under-compensated. The reform was intended to ameliorate fee

inequities. Consequently, the introduction of PPS and RBRVS had a major impact on the

use of inpatient vs. outpatient surgery.









Technological advances and patient preferences have also promoted the growth of

outpatient surgery. The development of less invasive surgical techniques enabled some

surgeries to be performed in ambulatory settings. Minimally invasive procedures usually

produce less postoperative pain, smaller scars, and faster recovery. The new anesthetic

agents reduce postoperative nausea, headaches, and drowsiness, and orally administered

analgesics represent another breakthrough that provides convenience and saves costs

while retaining potency. Some surgical procedures, such as tubal ligation, laparoscopic

cholecystectomy, and endoscopy, have become predominantly outpatient procedures.

Patients favored outpatient procedures because of the convenience of recovering at their

own home. Thus, while technology made the shift from inpatient to outpatient settings

possible, patient preferences also have contributed to the increasing use of outpatient

surgeries.

Previous research has focused largely on how HMO and non-HMO health plans

affect utilization, expenditures, and patient outcomes. But few studies have examined

how the type of health plan affects the choice of treatment method, outpatient or inpatient

surgery in particular. The purpose of this dissertation is to assess the effect of the type of

health plan on the choice of an outpatient or inpatient surgery, given a person who is

diagnosed to undergo a surgery that is feasible in either setting. The studied population is

people under age 65 with private health insurance coverage. This dissertation uses the

1997, 1998, and 1999 MEPS data to test the hypothesis that an HMO patient is more

likely to receive an outpatient surgery than a non-HMO patient, holding other factors

constant. Conclusions from the dissertation will be useful to payers, consumers, and






4


policy makers seeking to identify strategies to both improve appropriate use of care and

control healthcare costs.















CHAPTER 2
BACKGROUND

Health insurance is the main source of payment for health care in the U.S. In 1999,

39.1% of personal health care expenditures, totaling $1.06 trillion, was from private

health insurance or other private funding (Kramarow et al. 2001). Therefore, how

insurers pay for care has a major effect on shaping the U.S. healthcare system. This

chapter summarizes the development of the U.S. health insurance system.

Table 2-1. Health Insurance Coverage for Americans under Age 65, 2000

Payer Types of Coverage Percent"
Private 73.9%
Employer-sponsored 68.2%
Individual-purchased 5.7%
Public
Medicaid 10.4%
Medicare 2.2%
Military 2.8%
Uninsured 15.8%
Source: Mills RJ, Health Insurance Coverage:2000 (2001) Current Population Report. Washington, D.C.,
U.S. Census Bureau.

a This column represents the percentage of total population under age 65 in 2000. The estimates by type of
coverage are not mutually exclusive; people can be covered by more than one type of health insurance
during the year.

b Includes CHAMPUS (Comprehensive Health and Medical Plan for Uniformed Services), Tricare,
Veterans, and military health care.

The U.S. Health Insurance System

Most Americans have at least one type of health insurance (Table 2-1). In 2000,

73.9% of Americans under age 65 had private health insurance, with 68.2% obtaining

their coverage from the workplace. About 15% of Americans under age 65 had coverage

from public programs, including Medicaid, Medicare, and military health insurance, but









15.8% were uninsured. Because most of Americans under age 65 are covered by private

health insurance, this dissertation will focus on this sub-population, and will investigate

how the type of plan for these individuals affects the choice between inpatient and

outpatient surgery.

Types of Private Health Plans

Until the 1970s, traditional fee-for-service (FFS) plans were the usual type of

health plan in the U.S. During the past two decades, however, FFS plans have lost

market share, as managed care enrollment accelerated. Between 1988 and 2000, for

example, the percent of employees in FFS plans declined from 73% to 8% (Figure 2-1).





80% -26%
28% ~O POS
60% 38% 41% PPO

40% HMO
EFFS
20%

0%
1988 1993 1996 1998 1999 2000
Year

Source: The Kaiser Family Foundation and Health Research and Educational Trust. (2001). Employer
Health Benefits, 2000 Annual Survey, Kaiser Family Foundation.

Figure 2-1. Employer-Sponsored Health Insurance Coverage by Plan Type, 1988-2000


Managed care represents a relatively new concept that integrates the insurance

function with the delivery of care as a mean of controlling costs. While HMOs remain

the prototype of managed care organizations (MCOs), managed care approaches are not

limited to HMOs or even to MCOs. Rather, managed care techniques have been adopted









industrywide, with traditional FFS plans also using managed care techniques to control

utilization. Preauthorization for hospitalization and a second opinion for surgery are

examples of managed care utilization management used in FFS plans. Consequently, the

definition of type of health plan has become blurred. Yet some fundamental

characteristics can be used to differentiate among types of health plans.

FFS Plans

Under a FFS plan, patients have free choice of physicians and access to whatever

care they request that is covered by the plan (Enthoven 1978). Under the original form of

FFS, physicians have complete autonomy in clinical decisions, and there is little internal

or external assessment, such as utilization review or quality assurance.

The FFS system produces incentives for over-utilization. Physicians are

reimbursed per service provided; consequently, physicians have incentives to provide

more services in order to increase their income. Patients pay only a fraction of the costs

of care while the third party payers (i.e., insurance companies) are responsible for the

major portion of the costs of care. Therefore, the full price has little effect on patients'

demand for care. Because the FFS system does not manage utilization, patients tend to

request more care than they would if they paid full costs, and physicians are likely to

meet patients' demands.

To control health care costs, employers started to offer their employees managed

care plans. As shown in Table 2-2, enrollment in various types of managed care plans

has increased steadily during the past decade. In 1988, 27% of insured workers were

enrolled in managed care plans, but by 2000, enrollment in managed care plans had

grown to 92%. Given that employer-sponsored health insurance covers the majority of










the privately insured population, this dramatic increase illustrates the impact of managed

care on the U.S. healthcare system.

Table 2-2. Employer-Sponsored Health Insurance Coverage by Plan Type, 1988-2000

1988 1993 1996 1998 1999 2000
FFS 73% 46% 27% 14% 9% 8%
Managed Care Plans 27% 54% 73% 86% 91% 92%
HMO 16% 21% 31% 27% 28% 29%
PPO 11% 26% 28% 35% 38% 41%
POS _7% 14% 24% 25% 22%
Source: The Kaiser Family Foundation and Health Research and Educational Trust. Employer Health
Benefits, 2000 Annual Survey.

HMO Plans

The original form of MCOs was the HMO. Figure 2-2 shows that HMO enrollment

rose from 4% of the U.S. population in 1980 to 30% in 2000, totaling 81 million

Americans (Kramarow et al. 2001). The number of HMO plans has also increased. In

1970, there were fewer than 30 HMO plans; by 1980, the number had grown to 235

plans, and by 1990, to 572 plans (Robinson and Steiner 1998; Kramarow et al. 2001).



35%
30%/o 2o.24 g/. 4 30.1% 30.00
25% -~ ~ ~ ~~~~~~~2 ----------------3o.%_, ^'----------
20% -19.4%
15% 8.9%------1. 4
10%
5% 2.8% 4.0%
0%
1976 1980 1985 1990 1995 1996 1997 1998 1999 2000
Year


Source: Kramarow, E., Lentzner, H., Rooks, R. (2001). Health, United States, 2001. Hyattsville, MD,
National Center for Health Statistics.

Figure 2-2. Percent of Americans Enrolled in HMOs, 1976-2000


HMOs promote prevention and encourage early identification of disease, while

adopting approaches to control the use of care, such as limited choice of provider,









financial incentives for providers, gatekeepers, and utilization management. HMO

enrollees see only HMO physicians or physicians contracted with their health plan, but

bear less out-of-pocket costs. Enrollees in gatekeeping plans are required to go to

primary care physicians for the initial contact of care, and must obtain referrals for

specialty care. Providers are paid on a capitation basis or discounted FFS, and are subject

to utilization management. However, not all HMOs adopt the same set of approaches,

and thus variations among HMOs are observed.

Variations also come from the development of different models of HMOs. These

models differ in terms of the relationship between HMOs and physicians. Some

physicians are employed by HMOs or exclusively take care of HMO enrollees, while

others have contractual but not exclusive relationship with HMOs. However, because

data on HMO models were not available, this dissertation focuses on differences between

HMO and non-HMO plans. The following discussion of five HMO models is included to

illustrate the variation among HMOs.

A staff model HMO employs physicians who are recruited based on a selection

process that meets organizational needs. HMO physicians are salaried, and thus do not

bear financial risk. The HMO assumes full financial risks, and usually uses extensive

utilization management to ensure enrollees' access to necessary care without high costs.

A group model HMO has an exclusive arrangement with one or more large medical

practice groups that treat enrollees of a single HMO. The HMO performs the insurance

function, while the medical group provides clinical care. Although the two functions are

separated, the HMO and medical groups work closely together, and thus have a similar

degree of control as in a staff model. Kaiser Permanente HMO in California, consisting










of Kaiser Foundation Health Plan Inc. and Permanente Medical Group, is a typical

example of a group model HMO.

A network model HMO has an "arm's length" relationship with medical groups.

The HMO does not own, control or contract on an exclusive basis with medical groups,

and medical groups treat patients from multiple HMOs. Medical groups treat only HMO

patients, not those covered by non-HMO plans. Compared with staff and group models,

a network HMO has a relatively diffuse management structure.


.6 100% -

.S 80% -
o
(. 60%

S 40%-
0
20%

0%
1976 1980 1985 1990 1995 1996 1997 1998 1999 2000
SIPA 6.6% 18.7% 30.4% 41.6% 39.4% 44.1% 39.9% 42.6% 40.3% 41.3%
O Group 93.4% 81.3% 69.6% 58.4% 26.0% 23.7% 16.5% 18.0% 19.6% 18.9%
SMixed 34.5% 32.2% 43.4% 39.2% 40.1% 39.9%
Year

Source: Kramarow, E., Lentzner, H., Rooks, R. (2001). Health, United States, 2001. Hyattsville, MD,
National Center for Health Statistics.

Note: Group model includes staff, group, and network model types.

Figure 2-3. Privately Insured HMO Enrollment by Model Type from 1976 to 2000


In the early years of managed care, staff, group, and network models predominated

(Figure 2-3). Over the years, individual practice associations (IPAs) and mixed model

HMOs have become the more common models. During the 1980s, the proportion of

privately insured HMO enrollees in staff, group, or network models declined. By 1995,









IPAs and mixed model HMOs had approximately 74% of all private HMO enrollees

(Kramarow et al. 2001).

An IPA is a hybrid between group model HMO and FFS plans. There are two

types of IPA HMOs. One type is formed between an HMO and physician practices. The

HMO contracts directly with solo or small group practices on a non-exclusive basis. The

second type of IPA is formed by solo or small-group practices, and the IPA contracts

with HMOs. Under the second type of IPA, HMOs pay IPAs on a capitation basis, but

IPAs may pay physicians on a modified FFS basis. Therefore, physicians bear little

financial risk but are subject to extensive utilization review. Unlike network model

HMOs that provide care exclusively to HMO patients, physicians of IPA HMOs see both

HMO and non-HMO patients. Due to the contractual relationship between the HMO and

physicians, IPAs can usually be expanded without large capital investment, unlike staff or

group HMOs (Gabel 1997). Perhaps for this reason, enrollment in IPAs increased

considerably between 1976 and 2000 (Figure 2-3).

A mixed model HMO can be the combination of one of the four models and a

contractual arrangement with different provider organizations or networks. In some

cases, a staff model HMO may contract exclusively with large group practices to form a

mixed model HMO. In other cases, a network model HMO may contract, but non-

exclusively, with solo or small group practices. Enrollment in mixed model HMOs has

grown over the past few years. In 2000, approximately 40% of HMO enrollees were in

mixed model HMOs (Figure 2-3).

Regardless of the variation among HMOs, HMOs are considered to have tighter

control over the use of care than non-HMOs. Meanwhile, as newer forms of HMO plans









evolve, physicians face different utilization management and financial incentives. During

the 1970s, staff and group models were the dominant types of HMO. Physicians'

practices were subject to mainly internal utilization review, and the costs of providing

care were under a budget constraint. Newer models of HMOs often contract with

physicians in different locations and enroll more members. Physicians may be paid on a

capitation basis, but many of them receive discounted fee-for-service payments that may

reintroduce the incentives of over-utilization.

As physicians take on more financial responsibility, HMO enrollees may face

restrictions in accessing medical care. Using a gatekeeper is one of the approaches that

HMOs adopt to control utilization while non-HMO managed care plans also may use

gatekeeper to manage patient care. Newer forms of preferred provider organizations

(PPOs) and point-of-service (POS) plans have modified some of their managed care

approaches and incorporated certain traditional FFS plan characteristics to allow

enrollees access to providers outside of the network.

PPOs and POS Plans

PPOs are the fastest growing type of managed care (Table 2-2). In 1988, 11% of

Americans covered by employer-sponsored insurance were enrolled in PPOs. One

decade later, enrollment in PPOs rose to 35%. PPOs contract directly with a network of

providers for services on a discounted fee. Enrollees are not restricted to in-network

providers, but are required to pay a larger out-of-pocket payment when using out-of-

network providers. Because they are paid by discounted FFS, PPO providers do not

usually bear financial risks.

In 2000, 22% of Americans with employer-sponsored health insurance were

enrolled in POS plans (Table 2-2). POS plans are a hybrid of HMOs and indemnity plans









that pay HMO providers on a capitation basis, but pay providers in the indemnity plans

on a FFS basis. Enrollees decide whether to use HMO providers or out-of-network

providers at the time of a medical event. If enrollees self-refer to a specialist or an out-

of-network provider, they are responsible for a larger share of the costs. However, a

recent study found that most POS enrollees did not exercise the POS option (Forrest et al.

2001).

Because studies of newer forms of MCOs are limited, discussion of MCO

performance in the following section will focus on HMOs. Findings related to specific

managed care approaches, such as using a gatekeeper, pre-authorization, and financial

incentives, are also summarized. Studies using data after 1980 are more relevant to the

current situation, so the literature review in the next chapter on the effect of managed care

includes only studies using data collected after 1980.














CHAPTER 3
LITERATURE REVIEW

This chapter reviews the literature on the performance of HMOs vs. non-HMOs

during the past two decades. The focus is on the comparison between HMOs and non-

HMOs because the newer forms of managed care (i.e., PPOs and POS plans) have much

looser utilization management than HMOs, and thus the effect of managed care may be

diluted in PPOs and POS. The following literature review will summarize the findings

on utilization, patient outcome and quality, and on how type of health plan affects choice

of treatment method, with particular attention to the choice between inpatient and

outpatient surgery for a given medical condition.

Utilization

Because HMOs are thought to control costs through controlling utilization, many

studies have examined utilization patterns between HMOs and non-HMO plans. Studies

has assessed different types of utilization, including inpatient care, specialty care, non-

physician practitioner care, preventive care, and expensive care. This section summarizes

the findings.

Inpatient Care

HMO enrollees use less inpatient care than non-HMO enrollees. The Medical

Outcome Study examined utilization and outcomes of patients with chronic conditions,

including hypertension, diabetes mellitus, recent myocardial infarction, and congestive

heart failure, between 1986 and 1990. The findings indicated that chronically ill patients

enrolled in prepaid plans, which consisted of group model HMOs and IPA HMOs, used









41% less inpatient care than patients in FFS plans (Greenfield et al. 1992). Another study

of Medicaid elderly beneficiaries found that patients in capitated plans were 11.2% less

likely to have an inpatient visit than patients in traditional Medicaid (Lurie et al. 1994).

Miller and Luft (1997) reviewed articles published between 1993 and 1997, and

concluded that HMO enrollees use less inpatient care.

However, HMOs do not decrease the use of inpatient care across the board. The

recent findings of the Medical Outcome Study showed that, betweenl986 and 1990,

patients of IPA and group HMOs, who had better physical functioning, experienced a

slightly higher rate of hospitalization than FFS patients. On the other hand, sicker HMO

patients had lower hospitalization rates than FFS patients with comparable clinical

characteristics (Nelson et al. 1998). Based on data collected between 1987 and 1989,

Pearson et al. (1994) found that low- to mid-risk patients of a staff-model HMO were

more likely to be hospitalized than FFS patients with a comparable level of risk. These

findings suggest that HMOs treat patients aggressively during the early stages of illness

to prevent more expensive care later on.

Primary Care vs. Specialty Care

HMOs tend to use more primary care physician services while reducing specialist

care. Specialty care is more expensive than primary care, usually between 1.5 to 2.0

times as costly as primary care (Kongstvedt 2001). HMOs often use a gatekeeper to

constrain enrollees' access to specialists. Although having a gatekeeper may reduce the

use of specialty care, it may not have a substantial effect on overall costs. Based on the

1996 MEPS, one study found that total per capital annual health expenditures for children

in gatekeeping plans were approximately 8 dollars less than for those in indemnity plans

(Pati et al. 2003).









In California, to reduce the use of specialty care, some medical practices

encouraged primary care physicians to retain patients who otherwise required referral to

specialists for simple procedures or examinations. These medical groups rewarded

primary care physicians by offering a higher fee for services that are in the border

between primary and specialty care, such as well-women examinations, suturing and

wound treatment, and drainage of abscesses and cysts (Robinson 1999).

Nine large national surveys, including the RAND Health Insurance Experiment, the

Medical Outcome Study, and the 1996-97 Community Tracking Survey reported that

HMO enrollees used less specialty care (Hellinger 1998; Tu et al.1999). Not

surprisingly, HMO enrollees are usually less satisfied with their access to specialty care

than non-HMO enrollees (Hellinger 1998; Robinson and Steiner 1998; Tu et al. 1999).

Non-Physician Practitioner Care

Studies also have found that HMOs enrollees used more non-physician practitioner

care. Findings from the 1996-97 Community Tracking Survey showed that HMO

enrollees in general used more ambulatory visits, including non-physician practitioner

visits, than patients in non-HMO plans (Tu et al. 1999). Claims data from 1995 and 1996

showed that, in mental health care settings, managed care patients were more likely to be

treated by non-physician providers, such as psychologists, psychiatric nurses, and

psychiatric social workers (Sturm 1997).

Preventive Care

Previous studies found that HMO enrollees used significantly more cancer

screening. A study based on the 1987 National Health Interview Survey showed that

HMO enrollees received more cancer screening tests, including Pap smears,

mammography, breast physical examinations, digital rectal examinations, and blood stool









tests (Bernstein et al. 1991). Another study of Medicare recipients between 1983 and

1986 found that HMO enrollees of staff/group HMOs and IPAs were more likely to have

tonometry, mammography, pelvic examinations, rectal examinations, and fecal occult

blood tests than FFS enrollees (Retchin and Brown 1990). Similar findings were also

reported among enrollees of a network HMO (Udvarhelyi et al. 1991). The 1996-97

Community Tracking Survey found that HMO enrollees were more likely to have

mammography and flu shots than non-HMO enrollees (Tu et al. 1999). Therefore, HMO

enrollees tend to use more preventive care, cancer screenings in particular.

However, a recent study using the 1996-97 Community Tracking Survey found no

significant difference in the use of preventive care between HMO and non-HMO

enrollees (Reschovsky et al. 2000). Using the 1987 National Medical Expenditure

Survey (NMES) and the 1992 National Health Interview Survey (NHIS), another study

compared the difference in the use of preventive care between HMO and non-HMO

enrollees in 1987 and 1992, including blood pressure checks, pap smears, breast

examinations, and mammograms, among female non-elderly HMO enrollees (Weinick et

al. 1998). Weinick found that, in 1987, HMO enrollees used more preventive care than

non-HMO enrollees, but, by 1992, there was no significant difference in the use of

preventive care. These findings suggest that HMOs may be losing their edge of

providing more preventive care.

Expensive Procedures

HMOs enrollees also tend to use fewer costly tests and procedures than FFS

enrollees, particularly when there are alternative approaches available. HMO enrollees

had a significantly lower rate of Caesarian section than FFS enrollees, with the ratio

ranging from 0.68 to 0.97 (Stafford 1990; McCloskey et al. 1992; Tussing and









Wojtowycz 1994). Based on 45,425 births in California, one study further examined the

likelihood of vaginal birth for women who had a previous Caesarian section, and found

that in 1986 enrollees of group model HMOs were 23% more likely to have vaginal birth

than FFS enrollees, but there was no significant difference between IPA and FFS

enrollees (Stafford 1991).

Cardiovascular procedures are another example of different utilization patterns

between HMO and non-HMO plans. Based on data from the 1994-95 National Registry

of Myocardial Infarction, FFS patients under age 65 were more likely to undergo

angiography than HMO patients of the same age (Sada et al. 1998). HMO patients who

enrolled in the National Registry of Myocardial Infarction 2 (NRMI 2) also used less

coronary arteriography, catheter-based revascularization and coronary artery bypass

surgery than FFS patients (Canto et al. 2000). The finding regarding utilization of

angiography was also reported for Medicare patients. Data from the 1994-95

Cooperative Cardiovascular Project of the Health Care Financing Administration

(HCFA) showed that Medicare patients enrolled in HMO plans (Medicare+Choice) were

less likely to undergo coronary angiography after acute myocardial infarction. The

difference between patients in Medicare+Choice and traditional FFS Medicare plans

persisted even when patients were initially admitted to hospitals without an angiography

facility (Guadagnoli et al. 2000).

Summary: Utilization

HMO enrollees use less specialty care, and less expensive tests and procedures.

HMO enrollees use less inpatient care, but there are exceptions that appear to reflect

HMOs' strategies in treating mild conditions more aggressively than FFS plans.

Likewise, HMOs use more early detection and early treatment, although some studies









showed that non-HMO plans might be catching up with HMOs in the use of preventive

care.

Patient Outcomes and Quality of Care

Given that HMO enrollees use less expensive care, a subsequent question is

whether HMO enrollees also receive lower quality of care and have worse outcomes.

Patient outcomes have been measured in terms of mortality rate, survival rate, health

status, and physical health status. Quality of care is measured in various ways, including

process of care, continuity of care, and patient satisfaction.

Mortality

Several studies have found that the outcome of HMO patients with cardiovascular

conditions did not differ from that of FFS patients although the processes of care varied.

HMO patients were more likely to be treated by cardiologists than FFS patients, but FFS

enrollees were more likely to receive vascular catheterization than HMO enrollees.

However, there was no difference in mortality rates due to cardiovascular conditions.

Based on the data from the National Registry of Myocardial Infraction between June

1994 and October 1995, the mortality of myocardial infraction patients enrolling into

HMOs was not significantly different from the mortality of those in FFS plans (Sada et

al. 1998). Similar findings were reported for patients admitted to hospitals participating

in the Global Unstable Angina Registry and Treatment Evaluation Registry during 1995

and 1996 (Every et al. 1998).

Survival Rate

Survival rates for breast cancer and prostate cancer differ between HMO and FFS

patients. Based on the population-based breast cancer registry of Orange County,

California, between 1984 and 1990, the outcome of patients in HMO hospitals was worse









than patients in community hospitals or teaching hospitals (Lee-Feldstein and Feldstein

1994). Another study of prostate cancer patients who were diagnosed between 1985 and

1992 showed that the ten-year survival rate was worse for patients in group staff HMO

plans than patients in FFS plans (Potosky et al. 1999).

Physical and Mental Health

Longitudinal studies of the outcomes of chronically ill patients show no significant

differences between FFS and HMO patients. The Medical Outcomes Study (MOS)

evaluated the outcomes of patients with non-insulin-dependent diabetes mellitus,

hypertension, recent acute myocardial infarction, congestive heart failure, or depression

disorder. The study sampled patients who made office visits to physicians of family

medicine, internal medicine, cardiology, or endocrinology in three major U.S. cities

during the period of time that interviews were conducted. In a four-year follow-up

between 1986 and 1990, and in a ten-year follow-up evaluation, there was no difference

in physical and mental health between HMO and FFS patients (Greenfield et al. 1995).

Another study of rheumatoid arthritis patients also reported no difference in

physical health status between HMO and FFS patients. The Rheumatoid Arthritis Study

was conducted between 1982 and 1994 in northern California, and enrolled participants

in 1982-83 and in 1989 (Yelin et al. 1996). The study assessed the outcome of

rheumatoid arthritis patients for up to 11 years, and found no significant difference in

physical health and several physical function measures between patients of prepaid group

practices and FFS plans. However, based on the Medical Outcome Study, the elderly in

Medicare HMOs and the poor chronically ill HMO enrollees experienced worse physical

health than those in FFS plans (Ware et al. 1996).









Patient Satisfaction

Patient satisfaction of care also differs between HMO and FFS enrollees. Some

studies have found that HMO enrollees are less satisfied with their care than FFS

enrollees (Brown et al. 1993; Adler 1995; Davis et al. 1995). However, another study, a

1993 survey of three large employers, found that HMO enrollees of prepaid group

practices and IPAs were more satisfied with their plans than FFS enrollees (Allen et al.

1994). Safran examined HMO enrollee's satisfaction between 1986 and 1990. He found

that, compared to FFS enrollees, HMO enrollees of IPA and group-model HMOs were

satisfied with financial access and coordination of care, but were not satisfied with

number of physician visits, and physicians' interpersonal and technical skills (Safran et

al. 1994). The same findings were also reported for Medicare patients (Brown et al.

1993; Adler 1995). In general, it appears that HMO enrollees are less satisfied with their

access to care and more satisfied with the financial aspects of the care.

The Shift from Inpatient to Outpatient Surgery

As outpatient surgery has become generally accepted, more surgical procedures

have been performed in outpatient settings. In 1983, 24% of surgeries were performed in

the outpatient department of community hospitals (American Hospital Association 1987).

By 1996, more than half of the surgical procedures in the U.S. were on an outpatient

basis (Detmer and Gelijns 1994; Owings and Kozak 1998). Medicare has attributed this

trend to the introduction of the Prospective Payment System (PPS) and Resource-based

Relative Values (RBRVs). Under the new payment systems, some inpatient surgeries

were paid at a lower rate that provided an incentive to shift inpatient procedures to

outpatient settings. Private payers took up this trend and soon after several studies

reported that surgical outcomes were similar in both settings while the charges incurred









were lower in outpatient settings (Davis and Detmer 1972; Stephenson 1985). During the

1970s, only 35% of payers covered ambulatory surgeries; by the 1980s, 96% included

ambulatory surgeries in their coverage (Detmer and Buchanan-Davidson 1989).

Therefore, the payment system played an important role in promoting ambulatory

surgery.

Utilization

Two studies reported that HMO enrollees were more likely to receive outpatient,

rather than inpatient surgery. Between 1983 and 1993, the growth of hospital

expenditures in California was less rapid in areas with high HMO penetration than in

areas with low HMO penetration; one of the factors accounting for these differences was

that HMOs substituted outpatient for inpatient surgery, including hysterectomies,

coronary artery bypass grafting (CABG), cholecystectomies, and inguinal hernia repair

(Robinson 1996; Trauner and Chesnutt 1996). A recent study based on seven years (

1990-1996) of data from the Healthcare Cost and Utilization Project found that the

likelihood of having an outpatient mastectomy was higher for HMO breast cancer

patients than for non-HMO plan patients (Case et al. 2001). However, more studies are

still needed to provide evidence regarding the effect of HMO coverage on the choice

between inpatient and outpatient surgery.

Outcomes

Many studies have compared the outcomes of specific surgical procedures that

were done in either inpatient or outpatient settings, but none has further compared the

outcomes between HMO and non-HMO patients. Among children with comparable

health status and family support who underwent a tonsillectomy, those with outpatient

procedures had no higher risk of postoperative bleeding than children with inpatient









procedures (Guida and Mattucci 1990; Reiner et al. 1990; Truy et al. 1994; Rakover et al.

1997). A randomized clinical trial was undertaken between 1993 and 1996 to study the

outcome of cataract surgery (Castells et al. 2001). All participating patients with health

status suitable for outpatient cataract surgery were randomly assigned to either an

outpatient or inpatient setting. Although patients with ambulatory surgery experienced

higher complication rates within 24 hours after the surgery, the study found no difference

in visual acuity and postoperative complications four months after the surgery. Other

studies compared the complication rates between outpatient and inpatient catheterization,

but, probably because the number of cases was too small, did not find a statistically

significant difference (Block et al. 1988; Skinner and Adams 1996).

Patient satisfaction has been found to be higher for patients and their families with

ambulatory procedures than those with inpatient procedures. For those who underwent

ambulatory cardiac catheterization, both patients and their families were more satisfied

with the process and recovery (Kern et al. 1990; Lee et al. 1990; Seckler and Held 1990).

Patients having ambulatory catheterization were more satisfied with the convenience,

continuity, and technical aspect of the care, as well as interpersonal communication.

Costs

Studies have reported that outpatient surgery costs less than inpatient surgery. In a

randomized clinical trial, an outpatient cataract surgery cost $1, 001 while an inpatient

surgery costs $1,218 (Castells et al. 2001). On the other hand, if patients stayed in the

hospital overnight only, outpatient cardiac catheterization costs were similar to the costs

of an inpatient procedure. Once catheterized patients stayed more than one night in the

hospital, the costs of an inpatient procedure were higher than that of an outpatient









procedure (Lee et al. 1990). Based on these findings, inpatient surgeries generally cost

more than outpatient surgeries.

Research Questions

Given that HMOs are thought to control utilization in order to reduce costs, HMOs

would be expected to prefer outpatient to inpatient surgeries. This dissertation will assess

the effect of the type of private health insurance plan, HMO or non-HMO, on the choice

between an outpatient and an inpatient surgery. Controlling for patients' medical

condition, are HMO patients more likely to receive outpatient surgery than non-HMO

patients? Specifically, for a patient under age 65 who is diagnosed to undergo a surgical

procedure that is feasible in both outpatient and inpatient settings, does an HMO patient

have a higher likelihood of receiving an outpatient surgery than a non-HMO patient?

When excluding surgeries that are done primarily in one setting with few in the other

setting, does HMO status have a greater impact on these surgeries than on all surgical

procedures generally feasible in either inpatient or outpatient setting? Does a specific

managed care approach, having a gatekeeper, affect the choice of a surgery setting?














CHAPTER 4
CONCEPTUAL FRAMEWORK

This chapter develops a model to provide the foundation for the empirical

specification and subsequent data analysis. While this dissertation assesses the effect of

HMO coverage on the choice of outpatient or inpatient surgery, physicians make the final

decision on the choice of a surgery setting; consequently, this chapter begins with a

discussion of the process of physician decision-making. This chapter first presents a

general model of the process by which physicians make clinical decisions, then focuses

on factors that influence the choice among treatments for a given diagnosis (i.e.,

outpatient or inpatient surgery), which fall into three major categories: physician

characteristics, patient characteristics, and payer characteristics. The chapter then

concludes with a summary of the model that will be used as the basis for the

identification of variables and empirical analysis.

A Two-Stage Model of Physician Decision-Making

A general model of the process of physician decision-making must distinguish

between two stages of treatment choice: diagnosis based on presented symptoms,

followed by choice of treatment method. Wolff (1989) developed a model that

disentangles the relationship between the first stage of diagnosis based on the presented

symptoms and the subsequent stage of choosing suitable treatment (Figure 4-1).

Stage One: Determine a Set of Feasible Treatments

In the first stage, patients initiate the contact with physicians for their medical





























Physician
knowledge
(K)


Stage 1: Medically technical decision-making
Determine a set of feasible treatments


Physician


Stage 2: Treatment selection decision-making
Choose a specific treatment method


Source: Wolff, N., Professional uncertainty and physician medical decision-making in a multiple treatment framework, 1989, Soc. Sci. Med., 28 (2):99-107


Figure 4-1: The Two-Stage Model of Physician Decision-Making









needs. Based on the presented symptoms (SYM), physicians use their medical

knowledge and judgment (K) to diagnose (DIA) the condition:

DIA=K(SYM).

Given a diagnosis, physicians evaluate which treatment methods are medically

appropriate, and which are available and accessible in the community, based on factors

such as capital (C), labor (L), and access constraint (A). This evaluation yields a menu of

feasible treatments (Ri, R2, ---Rn):

(Ri, R2, ---Rn)= f(C,L,AIDIA).

Physicians then assess the efficacy of each available treatment (Ri), by means of a

health status (HSi) production function:

HSi=K(Ri|DIA), where i=1,2,---n.

Under the assumption that physicians are homogeneous producers of medical

decisions, the resulting menu of treatment options and the assessment of each treatment's

efficacy only differ according to the availability and accessibility of resources in the

community.

However, the first stage of decision-making is subject to professional uncertainty

and error that may violate this assumption. Decision-making at this stage relies heavily

on technical factors. Many medical conditions are not well defined, and cannot be

diagnosed or treated unambiguously. Thus, physicians' decisions can be heterogeneous,

based on their educational background, clinical training, expertise, and local acceptable

standard of practices. The menu of feasible treatment options (Ri, R2, ---Rn) and the

assessment of each option's efficacy (HSi) are potentially different among physicians.









Stage Two: Choose a Specific Treatment Method

In Stage Two of the decision-making process, one treatment method is selected

from the menu of options (R1, R2,---,Rn). Now the agency relationship between patient

and physician becomes key. Because patients typically delegate a major portion of

medical decision-making to their physicians, patient and physician have a principal-agent

relationship. If acting as a perfect agent, the physician makes decisions based solely on

what is in the best interest of the patient. The physician knows the patient's biomedical

and psychosocial characteristics. The physician is aware of the patient's budget

constraint and preferences regarding health and other goods and services. And the

physician has better information about treatment costs, resource availability and

accessibility. Under these conditions, a physician acting as a perfect agent seeks to

maximize the patient's utility function (U), which is based on the patient's preferences

regarding health outcomes (HS) and all other goods and services (Z):

Choose Rk from (R1, R2,---,Rn) to maximize the utility function

Max U=U(HS,Z).

In the case of perfect agency, the physician determines the treatment choice that

would have been the patient's choice if the patient had the physician's knowledge and

training.

However, during the decision-making process, physicians face financial incentives

and institutional factors associated with the type of delivery organization. Physicians

may want to maximize their income, or may be subject to certain organizational

constraints governing choice of treatment. On the other hand, patients who have acquired

medical knowledge formally or informally may have a major influence on the physician's

decision-making. When patient and physician preferences diverge, the final outcome









depends on the extent of asymmetric information and the bargaining power between

patient and physician. While the physician is the executor of the final decision, patients,

physicians, and payers all play roles in the final choice of treatment.

Choice of Treatment Methods

For purposes of this dissertation, the focus is on the second stage of treatment

choice. That is, the diagnosis is taken as given, and the question then becomes what

factors influence the choice of treatment method for a particular diagnosis. For example,

given the diagnosis of breast cancer, the choices of treatment are mastectomy and breast-

conserving surgery. There is a trade-off between benefits and risks in the choice of

treatment. Some women prefer breast-conserving surgery followed by a course of

radiation therapy because of cosmetic factors, and are willing to bear the increased risks

of recurrence. Others would rather undergo a mastectomy to reduce the risk of

recurrence, but need to overcome the psychological effect due to the change in

appearance (Potosky et al. 1997; Hadley and Mitchell 1997/98). Thus, a patient's

preference may contribute to the physician's decision-making, as well as physician

factors, such as the financial return associated with a particular treatment.

Professional Uncertainty

A key factor influencing choice of treatment method is professional uncertainty.

For some conditions, sufficient scientific evidence exists to identify one appropriate and

preferred treatment. For example, there is little uncertainty in appendicitis diagnosis and

treatment. Once appendicitis is diagnosed, an appendectomy is the generally accepted

treatment. Similarly, for inguinal hernia, repair procedures are the generally accepted

treatment. In contrast, for diagnoses such as low back pain, breast cancer, and hip

replacement, there is little consensus among physicians as to the best choice of treatment,









and physicians tend to have different opinions on whether a certain procedure is more

valuable than others (Wennberg et al. 1982; Cherkin et al. 1994; Birkmeyer et al. 1998).

Consequently, we observe variations in the choice of treatment.

Professional uncertainty thus affects the choice of treatment method in the

following ways. If there is a generally accepted treatment for a particular diagnosis, the

treatment method is given and there is no choice per se (i.e., at the end of Stage One in

the Wolff model, there is only one feasible treatment method). This dissertation focuses

on cases in which professional uncertainty leads to a set of possible treatment methods

for a particular diagnosis. More specifically, the focus is on diagnoses for which either

inpatient or outpatient surgery may be chosen. For example, Case (2001), using 1996

data from the Healthcare Cost and Utilization Project (HCUP), found that 6.8% of all

breast cancer patients who underwent a mastectomy had an ambulatory surgical

procedure while 93.2% had inpatient surgery.

When professional uncertainty exists, factors such as patient, physician, and payer

characteristics can play an important role on the choice of treatment. Payer

characteristics, such as reimbursement method, can affect physician decision-making.

The study by Case (2001) also found that HMO breast cancer patients were more likely

to receive an outpatient mastectomy than patients with other types of health plan after

controlling for clinical characteristics. Therefore, the next step is to develop a model that

incorporates factors affecting the choice between inpatient and outpatient surgery for a

given diagnosis.

The Conceptual Model I

Figure 4-2 presents a general model of the choice between inpatient and outpatient

surgery (Model I). For individuals facing such a choice, this dissertation assesses the









effect of the type of private health plan on the likelihood of receiving outpatient surgery

when controlling for patient characteristics, physician characteristics, and other payer

characteristics.

As discussed earlier, patient preferences can affect the choice between inpatient and

outpatient surgery (Elit et al. 1996). The effect of patient characteristics on morbidity

and mortality has been extensively studied. These characteristics also may have an

impact on the communication process between physician and patient, on how patients

present their symptoms, and on treatment preferences. Key patient characteristics include

health status, race/ethnicity, age, gender, income, education, and family support.

Diagnosis Requiring Surgery



Treatment Option:
Inpatient or Outpatient Surgery






Patient's Characteristics -- Physician Characteristics -- Payer Characteristics




The Choice of
Outpatient or Inpatient Surgery


Figure 4-2. Conceptual Model I of Choice between Inpatient and Outpatient Surgery

Physician Characteristics

When there is professional uncertainty regarding the preferred treatment method,

physician characteristics can play a major role. Key physician characteristics include

practice style, specialty, years of practice, and practice environment.









Principal-agent relationship

A principal-agent relationship occurs whenever an individual delegates decision-

making authority to another individual (Levinthal 1988; Arrow 1991). In the healthcare

sector, patients often rely on physicians' superior medical knowledge in the choice of

medical treatment. Thus, physicians (the agents) are making decisions for patients (the

principals) (Dranove and White 1987).

Information asymmetries exist in any principal-agent relationship. For example,

physicians have superior knowledge about the treatment choices and the expected

efficacy in improving health status, while only patients can judge how changes in their

health status will affect their well being. Uncertainty arises because patients have no

perfect and costless way to monitor physicians' information and actions. Therefore,

patients rely on other approaches, such as a long-standing principal-agent relationship, to

motivate physicians to act in their best interests. If physicians act as perfect agents, they

will make the treatment choices that the patients would have chosen if they had the full

knowledge and information that their physicians have.

Imperfect agents

The principal-agent relationship raises the possibility that agents will put their self-

interest above that of the principals. In particular, the concern is that physicians could

use their position of superior medical knowledge to recommend decisions that benefit the

physician financially with little or no accompanying health benefit to the patient (Wolff

1989). Because patients generally have little ability to assess the quality of physicians'

decisions, approaches such as a second opinion can induce physicians to become a better

agent (Rochaix 1989).









Certain factors encourage physicians to make treatment choices in the best interests

of their patients. For example, the relationship between patients and primary care

physicians plays an important role (Dranove and White 1987). When a long-term

relationship exists, patients expect their primary care physicians, as sophisticated medical

consumers themselves, to refer them to a qualified specialist. By doing so, primary care

physicians may expect a reward in the form of a better reputation, and attract more new

patients to increase their income.

Besides monetary rewards, non-monetary factors such as professional, legal, or

ethical standards can also compel physicians to make decisions that benefit patients rather

than themselves. During their medical education, physicians are socialized to become

professionals who make decisions based on objective science that focuses on the relevant

aspects of the patient's circumstances (Arrow 1991; Clark et al. 1991). The threats of

malpractice litigation and license withdrawal may also prevent doctors from treating

patients inappropriately (Pontes 1995).

Physician practice patterns

Due to professional uncertainty and the principal-agent relationship, physician

practice patterns may play a role in the choice of treatment method. For some conditions,

such as breast cancer, there is little consensus in the medical community regarding a

preferred choice of treatment method, so physicians typically make the final decision.

According to Wennberg (1982), it is the exercise of clinical judgment under conditions of

uncertainty that produces different practice patterns among physicians.

Practice patterns are shaped by prior medical training, clinical experience and

expertise, as well as personality characteristics, value judgments, and professional or

patient convenience (Burns et al. 1995; Rutchik et al. 2001). Physician specialty, years of









practice, and practice environment (e.g., rural or urban area, geographic region) also

influence practice patterns. Practice styles can change if physicians take actions to

modify their clinical decisions. Some managed care approaches, such as physician

profiling followed by feedback on utilization rates, are especially designed to modify

physician practice patterns. The resulting changes can lead to the development of local

standards of acceptable practice (Wennberg 1984).

The Relationship between Physician and Payer Characteristics

The principal-agent relationship may also lead to medical decisions that adversely

affect payers (i.e., health plans). Even if physicians act as perfect agents for patients, the

decisions made by physicians may not be what payers prefer. Therefore, the challenge

for payers is to draw up a contract ensuring that physicians act in the payer's best

interests.

Like patients, payers also encounter the problem of monitoring physician actions.

Information such as blood pressure, weight, and immunization compliance is readily

available from medical records, but the appropriateness of diagnosis and treatment is

difficult to obtain, especially for conditions that lack consensus among physicians

(Randall 1993). Payers would like to control costs by providing only the appropriate use

of care, but usually do not have an indicator to judge that appropriateness. Thus, payers

often use other approaches, such as a gatekeeper, preauthorization, and utilization review,

to ensure that their best interests are served.

Financial incentives

Payers often use financial incentives to induce desired behaviors. Financial

incentives consist of methods of payment, physician payment amounts, withholds, and

bonuses. Different payment methods encourage different practice styles. Under









capitation, physicians are paid prospectively for services provided to a defined

population, regardless of the level of care used. Thus, capitation encourages a cost-

conscious practice style, but may result in undertreatment, especially for sicker patients.

On the other hand, FFS rewards physicians who are willing to treat sicker patients

because they are paid for services they provide, but may induce physicians to overtreat

patients to increase their own income. Thus, neither FFS nor capitation payment is

without shortcomings.

Innovative payment methods that incorporate both capitation and FFS attempt to

correct financial incentives, but have been found to be costly to develop and administer.

For example, medical groups in California have blended capitation and FFS to induce

desired physician behavior, but the blended payment methods require extensive

administration in development, execution, and negotiation (Robinson 1999). Thus, FFS

and capitation remain the predominant forms of payment for physician services.

While there is limited evidence of the effectiveness of capitation, the percent of

physicians paid by capitation has either remained about the same or increased slightly.

One study indicated that the percent of physicians in office practice receiving capitation

rose from 40% in 1996 to 44% in 1998 (Terry 1999). However, an AMA study found

that the percent of physician revenues derived from capitation remained constant at 24%

between 1996 and 1998 (Kane 1999).

Utilization management

Managed care plans usually have a system of utilization management in place.

Utilization management comprises a range of techniques, such as utilization review,

gatekeeping, and preauthorization. Depending on the local practice environment,

managed care plans in different geographic regions adopt different approaches to manage









utilization (Kongstvedt 2001). While there is no standard set of techniques, managed

care plans that emphasize utilization management generally have lower use and lower

cost of care than others. A 1999 survey of eight major managed care organizations

showed that those investing more in utilization management had lower utilization and

medical costs (Kongstvedt 2000).

Summary: payer characteristics

The payer characteristic of primary interest is enrollment in an HMO or non-HMO

plans. Other payer characteristics that may influence the choice between inpatient and

outpatient surgery are financial incentives, including method of physician payment

(capitation vs. FFS) and payment rates (total payment, physician payment, and patient's

share of payment). Finally, the use of utilization management is a payer characteristic

expected to affect the choice between inpatient and outpatient surgery.

The Final Model

The research question focuses on how a specific payer characteristic, namely HMO

coverage, affects the choice of inpatient or outpatient surgery while controlling for

patient, physician, and other payer characteristics. This dissertation also studies the

effect of a specific managed care approach, gatekeeping, on the choice of surgery setting.

Based on the physician, patient, and payer characteristics identified in the above section,

Figure 4-3 presents Model II of the choice between inpatient and outpatient surgery.

Key patient characteristics include health status, race/ethnicity, age, gender,

income, education, and family supports. Several indicators measure patient's health

status, including self-reported health status, number of conditions associated with a

surgical event, and total charge. Important physician characteristics include specialty,

years of practice, and practice environment (e.g., urban or rural area, and geographic









region). Finally, other payer characteristics, such as physician payment method, payment

rates, and how utilization is managed, also influence the choice between outpatient and

inpatient surgery.


A Diagnosis Requiring a Surgery



Treatment Options:
Inpatient or Outpatient Surgerv





Patient Characteristics Physician Characteristics Payer Characteristics
-Health Status -Specialty -HMO vs. Non-HMO Plans
-Number of -Years of Practice -Having a Gatekeeper
Condition ....... .Practice Environment -Financial incentive
-Total Charge 4."*.*......:." (Area Characteristics) Physician Payment Method
-Race/ethnicity4 Urban or Rural Area Rate Setting
-Age Geographic Region Total Payment
-Gender Physician Payment
-Income Patient's Share of Payment
-Education
-Family support





Choice of Inpatient or Outpatient Surgery


Figure 4-3. The Conceptual Model II of Choice between Inpatient and Outpatient Surgery


Model II represents a simplified relationship among the three groups of

characteristics, while interaction between groups of characteristics may exist. For

instance, physician's practice environment (urban or rural area, and geographic region) is

included in the final model under physician characteristics, but this characteristic also can

affect payment and charge levels, and some patient characteristics (such as income, race






38


and ethnicity). Therefore, area characteristics may a better way to describe these

environment characteristics.














CHAPTER 5
EMPIRICAL SPECIFICATION

This chapter presents the empirical specification for the subsequent data analysis.

The specific research questions are to assess the effect of HMO enrollment status, and the

effect of gatekeeper plan enrollment on the choice of outpatient or inpatient surgery,

given that a patient is diagnosed to have a surgery that is feasible in either setting.

Chapter 4 presented a conceptual model of treatment choice in which patient, physician,

and payer characteristics jointly determine the choice between inpatient and outpatient

surgery for a given diagnosis (see Figure 4-3).

Based on the conceptual model, the most general form of the empirical

specification is:

(Choice of Outpatient or Inpatient Surgery I Diagnosis)
= f (HMO/non-HMO; patient characteristics; physician characteristics; other payer
characteristics).

Dependent Variable

The dependent variable is the choice of an outpatient (vs. inpatient) surgery when

the choice set for a given diagnosis contains both inpatient and outpatient surgery. An

outpatient surgery is a medical event that occurs in an outpatient setting, and patients are

discharged in the same day. On the other hand, an inpatient surgery is done in an

inpatient setting. However, some patients identified as having an inpatient procedure are

discharged from a hospital without an overnight stay. For example, the 1996 National

Hospital Discharge Survey estimated that 383,000 hospital discharges, approximately 2%

of all inpatient discharges with procedures, were reported with zero night's hospital stay









(Owings 1998). In this dissertation, an inpatient surgical case is defined as an inpatient

event with at least one overnight stay. Otherwise, the surgical case is considered

outpatient.

Surgical procedures for which there is little variation in the choice of inpatient vs.

outpatient setting will be excluded from the analysis. Some surgeries are done

exclusively in inpatient settings due to the intensity and required equipment of the

surgery or post-surgical care. Surgeries requiring intense monitoring and post-surgical

care, such as a coronary artery bypass graft, a partial excision of the large intestine, or a

colostomy, are usually done in inpatient settings. Other surgical procedures, such as a

vasectomy, ligation or stripping of varicose veins, are typically performed on an

outpatient basis (Owings and Kozak 1998). These types of surgeries, in which there is no

variation in setting, are excluded from the analysis.

This dissertation thus focuses on surgical procedures that are performed in both

inpatient and outpatient settings. In 1996, 83 out of 99 two-digit ICD-9 procedure codes

were reported as both inpatient and outpatient procedures (see Appendix A). The 1996

National Health Care Survey (NHCS) estimated that 71.9 million procedures were

performed in the United States, with 44% being done in outpatient settings (Owings and

Kozak 1998). Based on the 1996 NHCS, surgical cases with one of these 83 ICD-9

procedure codes are included in the analysis.

This dissertation further studies the effect of HMO coverage on a subset of the 83

two-digit ICD-9 procedure codes, after excluding surgical procedures done mostly in one

of the settings (outpatient or inpatient) with only a few done in the other setting. This

subset of surgery cases includes only surgeries that had a reported ratio of inpatient to









outpatient cases (I/O ratio) between 0.2 and 5 in the 1996 National Health Care Surveys

(see Appendix A).

Independent Variables

Primary Independent Variable

The primary independent variable is health plan type (PLAN), which has four

possible values: one-HMO coverage, one non-HMO gatekeeper plan coverage, one non-

HMO non-gatekeeper plan coverage, and two-plan coverage. Given a diagnosis for

which both inpatient and outpatient surgery are feasible, it is hypothesized that HMO

patients are more likely to have outpatient surgery, all else equal. Because HMOs are

designed to control costs, HMOs are more likely to manage the use of costly care, such as

inpatient care. Therefore, enrollment in an HMO plan is expected to increase the

likelihood of a patient receiving an outpatient surgery.

This dissertation does not attempt to explore the effect of managed care in general

on the choice of treatment because managed care plans also include alternatives such as

preferred provider organization (PPO) plans. Although PPOs typically manage

utilization, they provide enrollees freedom in choice of provider and are considered to be

a less restrictive form of managed care. PPO plans are considered non-HMO plans in this

analysis. However, this dissertation does examine the effect of one managed care

approach, having a gatekeeper, on the choice of outpatient surgery.

Control Variables: Patient Characteristics

Three patient characteristics (health status, number of conditions, and total charge)

are used to control the severity of a surgery case. Other patient characteristics include

race/ethnicity, age, gender, income, education, and family support. Patient characteristics

control for differences in mortality and morbidity, as well as differences in the









communication process between physicians and patients, presentation of symptoms, and

treatment preferences (Mort et al. 1994).

Health status

A key patient determinant of the choice of treatment is health status. Substantial

evidence indicates that self-rated health is a significant predictor of mortality and

morbidity. A 1971 study of 3,128 randomly selected non-institutional elderly in

Manitoba, Canada, found that self-rated health was significant in predicting mortality

within three years, independent of clinically assessed health status (Mossey and Shapiro

1982). For persons with "poor" self-rated health status, the risk of mortality within three

years (1971-1973) was 2.92 times the risk of mortality for persons with "excellent" self-

rated health status. Similarly, a study of 2,682 males in Finland, aged 42 to 60, reported

that the level of self-rated health was significantly associated with all-cause death, death

from cardiovascular disease, the incidence of myocardial infarction, and the extent of

carotid atherosclerosis (Kaplan et al. 1996).

While most previous studies have used self-rated health status as a baseline

predictor to examine the association with mortality, a recent study found that repeated

observations of self-rated health status over time detect declines of physical health status.

Based on 20 years of data from the National Health and Nutrition Examination Survey-I

Epidemiological Follow-up Study, the study argued that self-rated health status reflects

not only existing illness but also undiagnosed but preclinical conditions (Ferraro and

Kelley-Moore 2001). Therefore, self-rated health status with five levels (poor, fair, good,

very good, and excellent) will be used to control for the severity of a surgical case.









Number of conditions

Number of conditions is associated with the severity of a patient's condition at the

time of surgery. Patients who have complex clinical conditions usually require a surgery

involving a hospital stay (Pepine 1991; Skinner and Adams 1996). Similarly, a

preexisting condition is important in determining the feasibility of an outpatient

tonsillectomy (Truy et al. 1994). In general, the number of conditions reported along

with a surgical event reflects the patient's health status (Hadley and Mitchell 1997/98).

Total charge

Total charge reflects the intensity of services provided. Unlike payment that

depends on the payer's negotiation power, total charge represents the provider's

assessment of the intensity of care needed for a certain condition. Because severe

conditions usually require extensive care and have a high charge, total charge was used to

control for the severity of illness (Manning et al. 1984).

Race and ethnicity

Race and ethnicity are key patient characteristics affecting the use of health care.

Other studies have found that black patients used less care than white patients. Kressin

(2001) reviewed articles published between 1966 and 2000 and found that African

Americans were less likely to have invasive cardiovascular procedures, including cardiac

catheterization, percutaneous transluminal coronary angioplasty (PTCA), and CABG

(Kressin and Petersen 2001). A study of the Northern California Kaiser Permanente

Medical Care Program showed that, among women younger than 60, African Americans

were more likely to be hospitalized for congestive heart failure than white Americans

(Alexander et al. 1995). Based on the 1988 Massachusetts hospital discharge data, blacks

had lower rates of eight procedures, including abdominal aortic aneurysm repair,









appendectomy, and cardiac valve replacement, but higher rates for two procedures,

hysterectomy and prostatectomy (Mort et al. 1994). A study of end-stage renal disease

patients between 1996 and 1997 showed that black patients were less likely to receive

kidney transplantation than whites (Ayanian et al. 1999). One study on claims data from

the 1992 Colorado's fee-for-service Medicaid program found that total annual

expenditures per child with otitis media were higher for white children than for Hispanic

or black children (Bondy et al. 2000).

Other studies have also reported variation in the use of care for non-black

minorities. Hispanics and Asian Americans were more likely to report barriers to health

care than non-Hispanic whites (Phillips et al. 2000). In fact, between 1977 and 1996,

Hispanics had a decreased likelihood of having at least one ambulatory visit (Weinick et

al. 2000). Mexican Americans were less aware of the available treatment of stroke and

less trusting of their health care providers than non-Hispanic whites (Morgenstern et al.

2001). Therefore, race and ethnicity is included in the study to control for possible

effects on the use of health care.

Age

Age is an important determinant of use of health care. Older patients have a higher

likelihood of experiencing preventable hospitalizations (Blustein et al. 1998). Based on

the National Health Care Survey, the number of surgical procedures done in inpatient

settings increases with patient age (Owings and Kozak 1998). Older respondents were

more likely to be screened for cholesterol (Davis et al. 1998). According to the American

College of Cardiology/American Hospital Association (ACC/AHA) guidelines,

outpatient cardiac catheterization is recommended for patients younger than 75 years old

(Pepine et al. 1991).









Between 1980 and 1995, a study reported that the pattern of using inpatient vs.

outpatient surgical procedures varied by age. For children under age 15, substitutions

between inpatient and outpatient surgeries occurred in tonsillectomies, repairs of inguinal

hernia, reductions of fracture and dislocation and operations on muscle, tendon, fascia,

and bursa. For patients aged 15 to 44, substitutions were seen in tubal sterilizations and

reductions of fracture and dislocation, while substitutions for the operations on the

urinary system, cystoscopies, and cholecystectomies were observed among patients aged

45 to 64 and those 65 or older (Kozak et al. 1999). Other studies found that age was not a

significant factor in tonsillectomies. Tonsillectomies for younger children were no longer

considered risky and could be done either in outpatient or inpatient settings (Reiner et al.

1990; Truy et al. 1994; Rakover et al. 1997). Therefore, age may affect the choice of

inpatient vs. outpatient surgery.

Gender

Gender also influences the choice of treatment. Males and females are different

physiologically, and thus vary in the morbidity of different conditions. In one study,

women tended to report more symptoms, used more ambulatory care, and were more

likely to present chronic conditions than men (Clark et al. 1991). In 1996, the ten most

common surgical procedures for males were different from those for females (Owings

and Kozak 1998). While most of these procedures had similar proportions done in

ambulatory settings, males tended to receive diagnostic ultrasounds in ambulatory

settings while females had a smaller percentage of procedures done in ambulatory

settings. Since men use care differently than women, gender can affect the choice of

treatment.









Family income

The effect of low-income on the use of health care remains significant even after

adjusting for other measures of socioeconomic status, such as no insurance or low

education level. Based on 1995 New York State hospital discharge data, patients who

lived in lower income neighborhoods were less likely to be treated with cardiac

catheterization, PTCA, CABG, and any revascularization procedure (Philbin et al. 2000).

Income was a significant determinant of receiving these invasive procedures even for

patients who were admitted into a hospital with on-site CABG and PTCA. Lower

income Medicare patients are more likely to experience a preventable hospitalization.

However, after controlling for patients' demographic factors, socioeconomic status,

history of chronic illness, and health status, low income became insignificant in affecting

the likelihood of preventable hospitalization (Blustein et al. 1998).

Education

Lack of knowledge about health care may discourage the use of care. Studies show

that less-educated Americans tend not to use preventive care or recognize symptoms that

need medical attention (Davis 1998; Gomick 2000). Medicare beneficiaries with lower

education levels tend to experience preventable hospitalization that could have been

managed in ambulatory settings in 1992 (Blustein et al. 1998). In addition, physicians

tend to follow-up intensively on patients with professional careers. Therefore, patients'

level of education may affect the choice of treatment method.

Family support

Social characteristics, such as living with a spouse or being from a two-parent

family, are considered favorable criteria for outpatient surgery without a hospital stay.

For example, children are recommended for outpatient tonsillectomies if they have strong









family support (Rakover et al. 1997). The clinical guideline for ambulatory anorectal

surgery developed by the American Society of Colon and Rectal Surgeons recommends

family support as a criterion for outpatient surgery (The Standards Task Force 1998-99).

Because patients need immediate rest and assistance after an outpatient surgery, those

having family support are more likely to get sufficient post-operation care.

Control Variables: Physician Characteristics

Physician characteristics include physician specialty, years of practice, and practice

environment (Selby et al. 1999). In this study, urban or rural area and geographic region

control for the practice environment although these practice environment variables can

also affect some payer and patient characteristics. Urban or rural area and geographic

region are included to control for area characteristics. Two other characteristics,

physician specialty and years of practice, are unobserved and thus inevitably omitted

from this study.

Urban or rural area

The area where physicians practice can affect their decisions about the choice of

treatment. Physicians practicing in urban areas have different patterns of care than those

in rural areas. For instance, breast cancer patients who were treated in urban areas were

more likely to receive breast-conserving treatment than those treated in rural areas

(Nattinger and Goodwin 1994). Furthermore, hospitals and surgeons' practices tend to be

located in Metropolitan Statistical Areas (MSAs) where sufficient labor and capital

resources are available and where advanced technology is more likely to be adopted for

outpatient surgery. Freestanding surgery centers are usually located in urban areas.

Pauly and Erder (1993) found that patients were more likely to undergo outpatient

surgery when freestanding surgery centers were available. Patients living in an MSA are









likely to have better access to outpatient surgical centers and thus are more likely to

undergo outpatient surgery.

Geographic region

Variations in using different types of treatment are also found among geographic

regions. For example, the use of breast-conserving surgery was higher in Middle Atlantic

states and New England, and lower in East South Central states and the West South

Central states (Nattinger et al. 1992). Birkmeyer et al. (1998) found that some surgeries,

such as lower extremity revascularization, carotid endarterectomy, and back surgery,

were more frequently performed in certain regions of the country than others. Based on

national fee-for-service Medicare claims from 1996 through 1997, the rate of major

amputations per year for individuals without diabetes were higher in the Southern and

Atlantic states, but the pattern was not consistent among diabetes patients (Wrobel et al.

2001). A recent study found that the use of outpatient mastectomy, rather than inpatient

mastectomy, was more prevalent in some states, such as Maryland, Colorado, and

Connecticut (Case et al. 2001). Similarly, the use of outpatient hernia repair also varied

by state. For Medicare patients who had hernia repair in 1987-88, over one third of the

cases were done in an outpatient setting, but the rates of outpatient surgery by state varied

tremendously, ranging from 89.9% in Washington to 6.3% in Georgia (Mitchell and

Harrow 1994). Consequently, geographic region is included to account for regional

variation in the use of outpatient surgery.

Unobserved physician characteristics

Physician specialty and years of practice may affect physicians' decisions regarding

the choice of treatment. Studies have found that patients seeing physicians of different

specialties used different levels of care and incurred different costs (Bartman et al. 1996;









Levetan et al. 1999; Selby et al. 1999). The number of years in practice also influenced

physician referral and prescribing treatment (O'Leary et al. 2000; Rutchik et al. 2001).

Although physician specialty and years of practice may affect the choice between

inpatient and outpatient surgery, they are not included in this study due to data

limitations.

Control Variables: Other Payer Characteristics

In addition to HMO vs. non-HMO coverage, other payer characteristics may

influence the choice of treatment method. In particular, payment rate, payment method,

and utilization management may play a role.

Payment rate

Payment rate is likely to have an important influence on the choice of treatment

method. Total payment for a surgical procedure includes the insurer's payments to

physician and facility, as well as the patient's out-of-pocket payment. Payers set the

rates, and the level of rates can encourage or discourage the use of certain types of

services. For example, a surgical procedure that requires a high out-of-pocket payment

could lead to reduced use of the procedure (Pauly and Erder 1993). Payment rates

included are total payment, physician payment (the sum of insurer and patient payment),

and the patient's share of payment for a surgical procedure.

Unobserved payer characteristics

Studies show that payment method and utilization management affect the use of

care, but unfortunately these variables are not available for this analysis. One study

found that physician group practices that pay physicians based on capitation tend to use

fewer resources (Kralewski et al. 2000). Utilization review led to reducing the use of

hospital care and surgical procedures (Feldstein et al. 1988; Rosenberg et al. 1995).









Utilization review, physician profiling and clinical guidelines had a significant effect on

reducing the variation in physician practice (Rizzo 1993; Wolff and Schlesinger 1998;

Kralewski et al. 2000).

The Analytic Model

Based on the empirical specification, an analytic model is formulated to incorporate

relevant factors affecting the choice of outpatient surgery:

Choice of (Outpatient Surgery)

= f (PLAN, health status, number of conditions, total charge, race/ethnicity, age,
gender, income, education, family support, MSA, geographic area, total payment,
physician payment, patient out-of-pocket payment)


Table 5-1 lists the independent variables expected to influence the probability of

choosing outpatient rather than inpatient surgery. The next chapter describes the

implementation of the analytic model using a specific dataset.

Table 5-1. Independent Variables Affecting the Probability of Choosing a Surgery
Setting

Variables Characteristics
Type of health plan: HMO vs. non-HMO Payer
Having a gatekeeper Payer
Self-reported health status Patient
Number of conditions Patient
Total charge Patient
Race/ethnicity Patient
Age Patient
Gender Patient
Family income Patient
Education Patient
Family support Patient
MSA Physician
Region Physician
Total payment Payer
Physician payment Payer
Out-of-pocket payment Payer














CHAPTER 6
DATA, VARIABLES AND STATISTICAL ANALYSIS

The data that test the research questions of this dissertation come from a nationally

representative household survey, the MEPS. A description of the MEPS dataset is

followed by a detailed description of variables obtained from the survey. The chapter

concludes with a discussion of statistical methods used for the analysis.

Data

The MEPS is a national probability survey cosponsored by the Agency for

Healthcare Research and Quality (AHRQ) and the National Center for Health Statistics

(NCHS) to address policy issues in financing for health care. Begun in 1996, the MEPS

collects information about health care utilization and expenditures to provide nationally

representative estimates for the U.S. civilian non-institutionalized population. This

dissertation relies on the combined 1997, 1998, and 1999 MEPS data to assess the effect

of HMO coverage on the choice of outpatient or inpatient surgery.

The MEPS consists of four components: the Household Component (HC), the

Medical Provider Component (MPC), the Insurance Component (IC), and the Nursing

Home Component (NHC). The HC is the core survey that determines the samples of the

MPC and part of the IC. As part of the HC and the IC, insurance and employment data

are collected from respondents' employers, unions, and other sources of private health

insurance. The information includes the number and types of private insurance plans

offered, benefits associated with these plans, premium contributions by employers and

employees, eligibility requirements, and employer characteristics. The NHC surveys









nursing homes and persons residing in or admitted to nursing homes at any time during

the survey year. Because the NHC has a different sample design from the HC, which is

the primary data source for this dissertation, the NHC is not addressed here.

Every year, the HC samples a new panel of households from the respondents to the

previous year's National Health Interview Survey (NHIS). Each panel of households is

surveyed through face-to-face interviews five times during a two-year period by using an

overlapping panel design. For example, the first panel was interviewed between 1996

and 1997, while the interviews of the second panel of households began in 1997 and

continued in 1998. Due to the overlapping panel design, after the first year of the MEPS

(1996), two panels of households are interviewed each year. This dissertation pools the

1997, 1998, and 1999 MEPS data to study the effect of health plan type on the choice of

outpatient surgery.

The response rate for the 1997 survey was 74.0%, 67.9% for the 1998 MEPS, and

64.3% for the 1999 survey (MEPS 2001a; MEPS 2001b; MEPS 2002a). The 1997 MEPS

dataset has 36,340 respondents, with 22,953 respondents in the 1998 MEPS and 23,565

individuals in the 1999 survey.

Sample Design and Sample Weights

Because the MEPS interviews a portion of the previous year's NHIS sample, the

MEPS uses the same complex sampling design as the NHIS. In the sampling design for

the 1995 to 2004 NHIS, the nation is partitioned into 1,995 primary sampling units

(PSUs) that are counties or groups of adjacent counties, and 52 of the largest

metropolitan areas are assigned to 52 PSUs (Botman et al. 2000). PSUs are further

partitioned into 237 design strata. PSUs are selected from each stratum, and households

are sampled from each PSU based on known probabilities. Because of the MEPS'









overlapping sampling design, when pooling three years' MEPS data, as this dissertation

does, some adjustment to the assigned PSUs and strata is required to obtain national

estimates (MEPS 2003a).

As a nationally representative survey, each case in the MEPS data represents a

group of Americans that share similar characteristics used to sample from the population.

A sample weight for each case is developed to incorporate in the estimation processes.

These sample weights are constructed to account for sample design, including unequal

probability sampling of the population (i.e., oversampling minority groups), as well as

nonresponse rates and partial responses from some survey participants. This dissertation

uses sample weights, adjusted by pooled strata and PSUs, to estimate variances and test

hypotheses.

MEPS Data Collection

The MEPS collects data beginning with household interviews (Figure 6-1).

Households are asked about demographics, employment, income, health status, health

insurance, utilization of care, and payment. Healthcare utilization data include hospital

stays, other hospital care, office-based physician care, other medical provider care, dental

services, home health, prescribed medications, medical equipment and supplies, and

alternative care. Detailed data for each healthcare encounter are collected, including type

of practitioner, time spent with provider, type of care, medical conditions, main surgical

procedures, charges and/or payments.

When a household reports any use of care, the MEPS subsequently surveys their

medical providers, which becomes the MPC of the MEPS. Medical providers include

office-based physicians, doctors of osteopathy, other practitioners practicing under

physician supervision, hospitals that provide inpatient, outpatient, and emergency room









care, home health care agencies, long term care institutions, and pharmacies. Information

from the MPC is used to verify respondent-reported medical events and collect additional

information. In particular, the MPC validates respondent-reported payments or charges

and types of services received.



The Household Component (HC)




The Medical Provider Component (MPC)




Hospital inpatient Outpatient Medical
stay visits office visits Home Health
Visits Prescription

drug
+ I' FR vi it +
Dental visits

Figure 6-1. The MEPS Data Collection

In the MPC, providers are interviewed based on a predetermined coverage:

* 100% of the providers of Medicaid recipients;

* 75% of the providers of managed care enrollees; and

* 25% of the providers of the remaining HC respondents who are either enrolled in
non-managed care plans or other public plans.

The data of the MPC are collected through telephone or mailed questionnaires with

a telephone follow-up for nonresponses. Combining information from household

respondents and medical providers, the data on medical events are categorized into

Hospital Inpatient Stay, Outpatient Department Visits, Office-Based Provider,

Emergency Room, Dental, Other Medical, Home Health, and Prescription Medication.









Medical conditions and type of medical procedures come from respondents' reports, and

are verified through the MPC survey. Professional coders then code medical conditions

and procedures into a fully specified ICD-9-CM condition or procedure code. Due to

confidentiality restrictions, the ICD-9-CM codes are further collapsed into 3-digit

condition codes or 2-digit procedure codes.

Construction of Dataset for Analysis

In order to assess the effect of HMO coverage on the choice of outpatient or

inpatient surgery, the dataset includes only surgeries that were performed in both the

inpatient and outpatient settings. The dataset constructed for this dissertation comes from

16 MEPS public use data files and three National Health Interview Survey public used

data files, including:

* The 1997, 1998, and 1999 Hospital Inpatient Stays Files (HC-016DF, HC-026DF,
and HC-033DF) (MEPS 2001c; MEPS 2001d; MEPS 2002b): containing
information about types of surgical procedure and medical condition, expenditure,
days of hospital stay, and type of payer.

* The 1997, 1998, and 1999 Outpatient Department Visits Files (HC-016F, HC-
026F, and HC-033F) (MEPS 2001e; MEPS 2001f; MEPS 2002c): containing
information about types of surgical procedure and medical condition, expenditure,
and type of payer.

* The 1997, 1998, and 1999 Consolidated Data Files (HC-020, HC-028, and HC-
038) (MEPS 2001a; MEPS 2001b; MEPS 2002a): providing demographic
information.

* The 1997, 1998, 1999 Person Round Plan Files (HC-047) (MEPS 2003b):
providing detailed information on health plan coverage, including number of plans
covering a specific period of time, type of health plan, and managed care
information.

* The 1998 and 1999 Medical Conditions Files (HC-027 and HC-037) (MEPS
2001g; MEPS 2002d): linking to surgical events in the 1998 and in the 1999
Hospital Inpatient Stays Files and Outpatient Department Visits Files. Number of
medical conditions associated with a surgical event is used as a control independent
variable, COND. The 1997 data files do not need this linkage because COND is
provided in the public use files, named NUMCOND.









* Three linking files (96HIS_97MEPX.DAT, NHMEP98X.DAT, and
NHMEP99X.DAT) (MEPS 2001h; MEPS 2002e; MEPS 2002f): linking survey
information reported by the same respondents in the MEPS and the previous year's
NHIS; these files are necessary because each year a new MEPS-HC panel is drawn
from the previous year's NHIS sample. 96HIS_97MEPX.DAT links the 1997
MEPS with the 1996 NHIS, and NHMEP98X.DAT links between the 1998 MEPS
dataset and the 1997 NHIS public use dataset. NHMEP99X.DAT links the 1999
MEPS with the 1998 NHIS.

* The 1996 NHIS, the 1997 NHIS, and the 1998 NHIS (DA2658.DAT,
DA2954.PERSON, and DA3107p.DAT) (U.S. Dept. of Health and Human
Services 1998a; U.S. Dept. of Health and Human Services 1998b; U.S. Dept. of
Health and Human Services 1999): providing health status information reported by
the MEPS respondents who also participated in the previous year's NHIS and had a
surgical event during the period of the first MEPS interview.

* The 1996 to 1999 Pool Estimates Files (H36.ssp) (MEPS 2003a): containing
adjusted strata and PSU for pooling multiple years' MEPS data.

Figure 6-2 shows a step-by-step data construction, and Appendix B contains the

corresponding SAS programming:

1. Step 1: Take the 1997, 1998, and 1999 MEPS Hospital Inpatient Stays files (HC-
016D, HC-026D, and HC-033D), keep only scheduled surgical cases, i.e., exclude
cases that "operation or surgical procedure" is not the main reason for
hospitalization (in the MEPS dataset, RSNINHOS not equal to "1"); take the 1997,
1998, and 1999 MEPS Outpatient Department Visits files (HC-016F, HC-026F,
and HC-033F), exclude cases that indicate no surgical procedure performed on the
visit (in the MEPS dataset, SURGPROC="0");

2. Step 2: Merge the above two files with the 1997, 1998, and 1999 MEPS Full Year
Consolidated Data files (HC-020, HC-028, and HC-038), keep only cases for age 0
to 64, and add types and number of health plan coverage in the time when a surgery
occurred by linking the constructed data files with person round plan files (HC-
047).

3. Step 3: Refine the dataset by selecting surgical procedures that are reported as both
inpatient and outpatient surgeries. This step is first based on a variable in the
MEPS datasets: SURGPROC in Hospital Stay File, and SURGNAME in
Outpatient Visit File. If SURGPROC or SURGNAME do not provide specific type
of surgery, then the selection process is based on the variables of two-digit ICD-9
procedure codes (IPPRO1X and IPPRO2X in Hospital Stay File, and OPPRO1X in
Outpatient Visit File).

a. First, choose cases having one of six main surgical procedures
(arthroscopic surgery, cardiac catheterization, tonsillectomy, dilation and









curettage (D&C), cataract surgery, and pacemaker) in the MEPS variable
(SURGPROC or SURGNAME).

b. Second, match the remaining cases that have valid ICD-9 procedure codes
(from 01 to 99) with the list of ICD-9 procedure codes reported as both
inpatient and outpatient cases in the 1996 National Health Care Survey
(Table 6-1).

4. Step 4: Exclude cases with non-private insurer payers.

5. Link cases from Step 4 with the 1998 and 1999 medical condition files (HC-027 and
HC-037) to get the number of conditions associated with the surgical events in the
1998 and 1999 MEPS.

6. Then, merge the data file with the 1996-99 pooled estimate file (HC-036) to obtain
the reassigned strata and PSU after pooling the 1997, 1998, and 1999 MEPS data.

This dissertation uses the 1996 National Health Care Survey as the reference for

constructing the dataset because the 1996 National Health Care Survey contains the most

recent and complete national information on inpatient and outpatient surgeries. However,

since new technology could be developed over the time, it is possible that certain

inpatient surgeries in 1996 might have become mostly outpatient surgeries in 1997, 1998,

or 1999. One component of the National Health Care Surveys, the National Hospital

Discharge Survey, collects inpatient data every year. Thus the 1997, 1998, and 1999

inpatient surgical cases in the National Hospital Discharge Surveys are used to verify that

surgical procedures that had inpatient and outpatient cases in 1996 were still reported as

inpatient surgery in 1997, 1998, and 1999 (Owings and Lawrence 1999; Popovic and

Kozak 2000; Popovic 2001).

Table 6-1 shows the number of cases of surgeries that were reported as inpatient

and outpatient surgery. Ratios of inpatient to outpatient cases (I/O ratio) are included for

both the MEPS data and the data from the 1996 National Health Care Surveys. While











The 1997 MEPS Dataset The 1998 MEPS Dataset


Inpatient Outpatient Inpatient Outpatient Inpatient Outpatient
(3,710 cases) (16,035 cases) (2,588 cases) (10,470cases) (2,420 cases) (9,551 cases)
I I I I I I


Step 1.
Surgery is the Surgery is Surgery is the Surgery is
main reason of performed during main reason of performed
hospitalization. this visit, hospitalization. during this visit.


1,176 cases 1,555 cases 821 cases 1,032 cases


Surgery is the
main reason of
hospitalization.


Surgery is
performed during
this visit.


1,012 cases


Step 2.
Merging with the Full Year Consolidated Data files which provide demographic and
socioeconomic status information, keeping only cases of age 0 to 64


768 cases 1,110 cases 536 cases 763 cases 490 cases 687 cases


Step 3.
Select procedures that were reported as inpatient and outpatient procedures

i l 1 1
230 cases 344 cases 151 cases 244 cases 119 cases 181 cases


Step 4.
Keeping cases whose surgical expenses were paid by private insurance companies only


125 cases 233 cases 67 cases 181 cases 68 cases 140 cases


260 Inpatient Surgery Cases
554 Outpatient Surgery Cases


Figure 6-2. Dataset Construction


The 1999 MEIS Dataset









the National Health Care Surveys collect data from providers, the MEPS interviews

households from the population. Because of different sources for the information, the I/O

ratios reported in the MEPS data differ from those in the 1996 National Health Care

Survey. In 1996, only 11.4% of the population under age 65 had an outpatient visit or

inpatient discharge with procedures (Owing and Kozak 1998). Given that the MEPS

sampled a household based on characteristics other than type of surgery, it is likely that

people having certain types of surgeries are not sampled into MEPS. Therefore, the 1996

National Health Care Surveys serve as an external reference to select procedures for this

dissertation. A subset of data containing only cases of surgical procedures with an I/O

ratio between 0.2 and 5 is also constructed to test one of the research questions.

The resulting dataset contains 814 cases representing 9,595,656 surgical cases in

the nation between 1997 and 1999, including 3,333,464 surgical cases in 1997; 3,466,624

cases in 1998; and 2,795,568 in 1999. The dataset requires an additional modification to

the original inpatient vs. outpatient coding. Because this dissertation defines inpatient

surgery as a surgery with at least one night inpatient stay, cases from Hospital Inpatient

Stay files with zero night stays are recorded as outpatient surgical cases. Table 6-1 shows

the unweighted number of cases by type of surgical procedures in the final dataset.

A Subset of the Constructed Dataset

A subset of the final constructed dataset is prepared to include only cases of

surgical procedures that had ratios of inpatient to outpatient cases (I/O ratios) between 0.2

and 5 in the 1996 National Health Care Survey (Table 6-2). Thus, cases excluded from

the subset are surgeries that were mostly done in one of the settings (i.e., inpatient or

outpatient setting). This subset contains 391 cases, including 203 inpatient cases and 188

outpatient cases.










Table 6-1. Number of Cases by Type of Procedure


The 1997, 1998, 1999 MEPS 1996
SURGERY (based on a variable in the (Private Insured) NHCS
MEPS data) TOTAL Inpatient Outpatient I/Ob I/Ob
Arthroscopic surgery 157 18 139 0.13 0.04
Cardiac Catheterization 70 43 27 1.59 2.50
Dilation and Curettage 79 6 73 0.08 0.16
Tonsillectomy 66 14 52 0.27 0.09
Cataract Surgery 35 0 35 0 0.04
Pacemaker 6 5 1 5.00 6.5
Two-Digit ICD-9 Procedure Codec and the Description
03 Spinal cord and canal operations 3 3 0 # 1.20
06 Thyroid and parathyroid operations 7 4 3 1.33 1.85
08 Eyelids operations 2 0 2 0 0.17
16 Orbit/eyeball operations 5 1 4 0.25 0.87
18 External ear operations 2 0 2 0 0.27
19 Middle ear reconstructure 1 1 0 # 0.07
20 Other mid and inner ear operations 19 0 19 0 0.05
21 Operations on nose 7 0 7 0 0.11
22 Nasal sinus operations 4 0 4 0 0.07
23 Tooth removal & restoration 1 0 1 0 0.17
27 Other mouth and face operations 1 0 1 0 0.49
28 Tonsil and adenoid operations 5 1 4 0.25 0.09
29 Operations on pharynx 1 0 1 0 0.40
31 Larynx trachea operations 3 0 3 0 0.81
33 Other operations on lung, bronchus 1 0 1 0 1.63
36 Operations on heart vessels 4 4 0 # 21.00
37 Other heart and pericardium operations 2 2 0 # 2.70
38 Vessel incision, excision, and 2 0 2 0 5.11
39 Other operations on vessels 4 3 1 3.00 6.94
40 Lymphatic system operations 1 0 1 0 1.05
44 Other gastric operations 4 1 3 0.33 1.61
45 Intestine incision, excision, 16 2 14 0.31
anastomosis
48 Rectal & perirectal operations 1 1 0 # 0.51
49 Operations on anus 4 0 4 0 0.21
51 Biliary tract operations 61 32 29 1.10 1.22
53 Repair of hernia 19 5 14 0.36 0.21
54 Other abdomen region operations 3 1 2 0.50 0.97
56 Operations on ureter 2 0 2 0 0.93
57 Urinary bladder operations 7 6 1 6.00 0.47
59 Other urinary tract operations 2 2 0 # 1.45
60 Prostate & seminal vesicle operations 3 3 0 # 1.69
63 Operations on Sperm cord, epididymis, 6 0 6 0 0.05
vas deferens
64 Operations on penis 2 0 2 0 0.25









Table 6-1. Continued


The 1997, 1998, 1999 MEPS 1996
(Private Insured) NHCS
Two-Digit ICD-9 Procedure Codes' andPrivate Insu
the Description TOTAL Inpatient Outpatient I/Ob I/Ob
65 Operations on ovary 3 2 1 2.00 3.84
66 Fallopian tube operations 12 2 10 0.20 1.07
67 Operations on cervix 4 0 4 0 0.11
68 Other uterine incision and excision 54 52 2 26 1.73
69 Other uterus and support operations 6 0 6 0 0.20
70 Vagina and cul-de-sac operations 2 2 0 # 1.90
76 Facial bone and joint operations 2 1 1 1 2.04
77 Incision, excision, division bone 16 1 15 0.07 0.48
78 Other bone operations except face 8 4 4 1 0.73
80 Incision, excision joint 7 6 1 6.00 0.24
81 Joint repair 37 19 18 1.06 1.04
83 Other muscle, tendon, fascia, 3 2 1 2 0.57
bursa operations
84 Other musculoskeletal procedures 9 4 5 0.80 3.13
85 Operations on the breast 15 4 11 0.40 0.18
86 Skin & subcutaneous operations 13 0 13 0 0.65
97 Replace & remove devices 2 2 0 # 0.49
Total cases based on ICD-9 401 174 227 0.77
GRAND TOTAL 814 260 554 0.47_
Note: # These procedures have zero outpatient cases, so the ratio of inpatient to outpatient cases is invalid.

aThese six surgeries are based on a MEPS' variable, SURGNAME in Outpatient Department Visit data
files and SURGPROC in Hospital Inpatient Stay files.

b I/O represents the ratios of inpatient to outpatient cases.

' These cases did not have information in "SURGNAME" or "SURGPROC", but had reported information
in "OPPRO1X" (the MEPS Outpatient Department Visit data files) or "IPPRO1X" and/or "IPPRO2X" (the
MEPS Hospital Inpatient Stay data files).










Table 6-2. Number of the Subset of Cases by Type of Procedure

The 1997, 1998, 1999 MEPS 1996
SURGERY (based on a variable in the TOTAL Inpatient Outpatient I/Ob NHCS
MEPSa) I/Ob
Cardiac Catheterization 70 43 27 1.59 2.50
Two-Digit ICD-9 Procedure Code and the Description
03 Spinal cord and canal operations 3 3 0 # 1.20
06 Thyroid, parathyroid operations 7 4 3 1.33 1.85
16 Orbit, eyeball operations 5 1 4 0.25 0.87
18 External ear operations 2 0 2 0 0.27
29 Operations on pharynx 1 0 1 0 0.40
31 Larynx, trachea, neck operations 3 0 3 0 0.81
33 Other operations on lung, bronchus 1 0 1 0 1.63
37 Other heart, pericardium operations 2 2 0 # 2.70
40 Lymphatic system operations 1 0 1 0 1.05
44 Other gastric operations 4 1 3 0.33 1.61
45 Intestine incision, excision, 16 2 14 0.14 0.31
anastomosis
48 Rectal & perirectal operations 1 1 0 # 0.51
49 Operations on anus 4 0 4 0 0.21
51 Biliary tract operations 61 32 29 1.10 1.22
53 Repair of hernia 19 5 14 0.36 0.21
54 Other abdomen region operations 3 1 2 0.50 0.97
57 Urinary bladder operations 7 6 1 6.00 0.47
59 Other urinary tract operations 2 2 0 # 1.45
60 Prostate, seminal vesicle operations 3 3 0 # 1.69
64 Operations on penis 2 0 2 0 0.25
65 Operations on ovary 3 2 1 2.00 3.84
66 Fallopian tube operations 12 2 10 0.20 1.07
68 Other uterine incision, excision 54 52 2 26 1.73
69 Other uterus, support operations 6 0 6 0 0.20
70 Vagina & cul-de-sac operations 2 2 0 # 1.90
76 Facial bone & joint operations 2 1 1 1 2.04
77 Incision, excision, division bone 16 1 15 0.07 0.48
78 Other hone operations excent face 8 4 4 1 0.73
80 Incision, excision joint 7 6 1 6.00 0.24
81 Joint repair 37 19 18 1.06 1.04
83 Other muscle, tendon, fascia, bursa 3 2 1 2 0.57
operations
84 Other musculoskeletal procedures 9 4 5 0.80 3.13
86 Skin & subcutaneous operations 13 0 13 0 0.65
97 Replace & remove devices 2 2 0 # 0.49
Total cases based on ICD-9 321 160 161 0.77
1 GRAND TOTAL 391 203 188 0.47 _
Note: # These procedures have zero outpatient cases, so the ratio of inpatient to outpatient cases is invalid.

a This surgery was from a MEPS's variable, SURGNAME in Outpatient Department Visit data files and
SURGPROC in Hospital Inpatient Stay files.
b I/O represents the ratios of inpatient to outpatient cases. All cases in this Table had surgeries with an I/O
ratio between 0.2 and 5.0 in the 1996 National Health Care Surveys.













Table 6-3. Independent Variables by Type of Characteristic, Conceptual Basis, and Questions from the MEPS

Type of
Variable C terc Conceptual Basis Questions in the MEPS Questionnaires and Possible Answers
Characteristic


HMO status increases the
likelihood of having an
outpatient surgery.

Since HMOs tend to
manage care by reducing
expensive care, such as
inpatient surgeries, when
there are less costly
alternatives, i.e.,
outpatient surgeries,
HMO patients may be
more likely to have a
surgery on an outpatient
setting than those in non-
HMO plans.


Payer


PLAN
(Type of
Health
Plan)


If the person is covered by private insurance and answered one of the following
questions using the term "HMO," HMO is set to "yes."

1. From which of the sources on this card did anyone in the family purchase health
insurance? (HX23)
Possible Answers:
1: From a group or association; 2: From a health insurance purchasing alliance; 3:
Directly through a school; 4: Directly from an insurance agent; 5: Directly from
insurance company; 6: Directly from an HMO; 7: From a union; 8: From anyone's
previous employer (cobra); 9: From anyone's previous employer (not cobra); 10:
From spouse's/deceased spouse's previous employer; 11: From some other
employer; 12: Under plan of someone not living here; 91: Other source; -7:
Refuse; -8: Don't know

2. {You mentioned that (PERSON) {(are/is)/(were/was)} self-employed and had
health insurance through that business.} Which category on this card comes closest
to {the main/another} way (PERSON) (purchase/purchases) this insurance?
(HX03)
Possible Answers:
1: From a professional association; 2: From a small business group; 3: From a
union; 4: From a health insurance purchasing alliance; 5: Directly from an
insurance agent; 6: Directly from insurance company; 7: Directly from an HMO; 8:
From a previous employer; 9: From a previous employer (cobra); 91: Other; -7:
Refuse; -8: Don't know.

3. What is the name of the insurance company or HMO from which
(POLICYHOLDER) receives hospital and physician benefits? (HX51)
Possible Answer:
1: Insurance company; 2: HMO; 3: Self-insured company.













Table 6-3. Continued


Type of
Variable Typeof Conceptual Basis Questions in the MEPS Questionnaires and Possible Answers
SCharacteristic


4. From which of the sources on this card did anyone in the family purchase health
insurance?
Possible Answers:
1: From a group or association; 2: From a health insurance purchasing alliance; 3:
Directly through a school; 4: Directly from an insurance agent; 5: Directly from
insurance company; 6: Directly from an HMO; 7: From a union; 8: From anyone's
previous employer (cobra); 9: From anyone's previous employer (not cobra); 10:
From spouse's/deceased spouse's previous employer; 11: From some other
employer; 12: Under plan of someone not living here; 91: Other source; -7: Refuse;
-8. Don't know.

5. What is the name of the {other} insurance company or HMO for
(POLICYHOLDER)'s (ESTABLISHMENT) insurance? (HX54)
Possible Answers:
1: Insurance company; 2: HMO; 3: Self-insured company.

6. If the person answered yes to the following question (MC01):
Now I will ask you a few questions about how (POLICYHOLDER)'s health
insurance through (ESTABLISHMENT) works for non-emergency care. We are
interested in knowing if (POLICYHOLDER)'s (ESTABLISHMENT) plan is an
HMO, that is, a Health Maintenance Organization. With an HMO, you must
generally receive care from HMO physicians. For other doctors, the expense is not
covered unless you were referred by the HMO or there was a medical emergency. Is
(POLICYHOLDER)'s (INSURER NAME) an HMO?
Possible Answers:
Yes, No


Payer


PLAN
(Type of
Health
Plan-
Continued)


(Continued)













Table 6-3. Continued


Type of
Variable Type of Conceptual Basis Questions in the MEPS Questionnaires and Possible Answers
Characteristic
PLAN Payer Gatekeeper, as a managed {(Do/Does)/As of (END DATE), did} (POLICYHOLDER)'s insurance plan
(Type of care approach, could reduce require (POLICYHOLDER) to sign up with a certain primary care doctor, group
Health the use of inpatient surgery of doctors, or a certain clinic which (POLICYHOLDER) must go to for all of
Plan- (POLICYHOLDER)'s routine care? (MC02)
Continued) Possible Answers:
1: Yes; 2: No; -7: Refuse; -8: Don't know
HLTH Patient Patients with worse pre- In general, compared to other people of (PERSON)'s age, would you say that
(Self- surgery health tend to have (PERSON)'s health is excellent, very good, good, fair, or poor (EO1)
reported an inpatient surgery. Possible Answers:
Health 1: Excellent; 2: Very good; 3: Good; 4: Fair; 5: Poor; -9: Not ascertained; -8:
Status) Don't know; -7: Refused; -3: No Data in round; -1: Inapplicable.
COND Patient The more conditions A constructed variable based on what respondents reported on their conditions,
(Number of associated with a surgical which are then coded by professional coders into ICD-9 Codes. COND is the
conditions) event, the higher the level counts of the ICD-9 codes.
of severity tends to be.
Thus, an inpatient surgery Was this hospital stay related to any specific health condition or were any
tends to be chosen. conditions discovered during this hospital stay? (HS03)
Possible Answers:
Yes, No
What conditions were discovered or led (PERSON) to enter the hospital? (HS04)
TCH (Total Patient The level of charge A constructed variable from a series of charge/payment questions. Please refer
Charge) represents the intensity of to payment information at the end of this Table since this variable is imputed
the procedure. High from a series of charge/payment questions.
charges are set for
procedures requiring
extensive care.













Table 6-3. Continued


Type of
Variable Type of Conceptual Basis Questions in the MEPS Questionnaires and Possible Answers
Characteristic
AGE Patient Young patients are What is (READ NAME BELOW)'s date of birth? Enter MM/DD/YYYY.
more likely to have an
outpatient surgery.
GENDER Patient Male patients tend to I have (PERSON) recorded as (READ GENDER BELOW). Is that correct?
have outpatient Possible Answers:
surgeries. 1: Male; 2: Female
RACE Patient People of different Please look at this card and tell me the group which best describes (PERSON)'s
(Race/ races/ethnicities may racial background. (RE101)
Ethnicity) use health care Possible Answers:
differently due to the 1: American Indian; 2: Aleut, Eskimo; 3: Asian Or Pacific Islander; 4: Black; 5:
differences in White; 6: Other; -7: Refused; -8: Don't know
morbidity and Which group represents (PERSON)'s main national origin or ancestry (RE99)
mortality, presentation Possible Answers:
of symptoms, and 1: Puerto Rican; 2: Cuban; 3: Mexican, Mexican-American, Mexicano, Chicano; 4:
communication with Other Latin American; 5: Other Spanish; 91: Other; -7: Refused; -8: Don't Know
physicians.
INCOME Patient Affordability of a A constructed variable based on all income sources reported by respondents. The
(Annual surgery can affect the following are several key questions asked:
family decision of surgery How much money did (READ NAME(S) ABOVE) receive from wages or salary,
income as a settings tips, commissions, or bonuses? (IN 18)
percent of How much did (READ NAME(S) ABOVE) receive in interest from savings
Federal accounts, bonds, NOW accounts, money market accounts, or similar types of
Poverty investments? (IN19)
Level) How much money did (READ NAME(S) ABOVE) receive from alimony? (IN22)
How much money was (READ NAME(S) ABOVE)'s net gain or net loss from the
sale of property or other assets, including the sale of (his/her/their) home, if it was
taxable? (IN24)
Looking at this card, which range best estimates how much money was received
(from Social Security and equivalent tier 1 Railroad Retirement benefits)? (IN32)













Table 6-3. Continued


Type of
Variable Cha ter c Conceptual Basis Questions in the MEPS Questionnaires and Possible Answers
Characteristic
EDU Patient Lack of knowledge What is the highest grade or year of regular school (PERSON) ever completed?
(Education) about health care may (RE103)
discourage the use of Possible Answers:
certain types of care 0-17 Years.
SUPPORT Patient Family support For adults, this is a constructed variable based on the interview roster of household
(Family increases the members and the following question:
support) likelihood of {(Are/Is) (PERSON) now/As of December 31, 1996, (were/was) (PERSON)}
outpatient surgery. married, widowed, divorced, separated, or never married? (RE97)
Possible Answers:
1: Married; 2: Widowed; 3: Divorced; 4: Separated; 5: Never Married.
SUPPORT is set to "1"-married and living with spouse; otherwise, "0".
For children, if both father and mother ID is indicated, i.e. mother and father live
in the household, SUPPORT is set to "1"-From 2-Parent Family; otherwise, "0."
MSA (Living Physician Physicians practicing Assigned when sampled
inside MSA/ in MSAs have better
non-MSA) access to ambulatory
surgical center, than
those in non-MSAs,
thus are more likely to
prescribe an outpatient
surgery.
REGION Physician Physicians practicing Assigned When sampled
(Geographic in some regions of the
Region) country tend to use
more outpatient
surgeries than those in
other regions.













Table 6-3. Continued


Type of
Variable Ch ter c Conceptual Basis Questions in the MEPS Questionnaires and Possible Answers
Characteristic
SPECIALTY Physician Physician training Not available due to data limitation
(Physician's affects clinical
Specialty) decision-making.
YEARS OF Physician Year of practicing Not available due to data limitation
PRACTICE reflects physician's
(Number of training, and thus
years affects clinical
practicing) decision
Physician Payer How a physician is Not available due to data limitation
Payment paid produces
Method different incentives.
(Capitation, Capitation
discount FFS, encourages a cost-
or FFS) conscious practice
style, while FFS
rewards productivity.
PTPAY Payer Higher patient out-of- A constructed variable from a series of charge/payment questions. Please refer to
(Out-of- pocket payment for a payment information at the end of this Table since this variable is imputed from a
pocket surgery (inpatient or series of charge/payment questions.
payment) outpatient)
encourages patients
to express the
concern of
affordability.












Table 6-3. Continued


Type of
Variable T e of Conceptual Basis Questions in the MEPS Questionnaires and Possible Answers
S Characteristic


Higher payment for
inpatient surgery
encourages doing a
surgery in inpatient
settings.


Higher physician
payment for a surgery
(inpatient or
outpatient)
discourages doing a
surgery in a
particular setting.


TOTAL
(Total
Payment)


Payer


PAYMD
(Physician
Payment)


A constructed variable from a series of charge/payment questions, and supplemented
with survey results from the Medical Provider Survey. The followings are some of
the key questions:
To whom was the bill sent?

How much was the total charge for (PERSON)'s stay at (HOSPITAL) that began on
(ADMIT DATE)/(PERSON)'s visit to (PROVIDER) on (VISIT DATE)/the last
purchase of {NAME OF PRESCRIBED MEDICINE...} for (PERSON)/the services
for (FLAT FEE GROUP) for (PERSON)/the {OME ITEM GROUP NAME} used
by (PERSON) since (START DATE)/ services received at home from (PROVIDER)
during (MONTH) for (PERSON)/(PROVIDER)'s services as part of the visit made
on (VISIT DATE)}? (CP09)

Is this a situation in which (PERSON) (are/is) required to pay a certain set amount
each time {(PERSON) (visit/visits) (PROVIDER) regardless of what happens during
the visit/(PERSON) (receive/receives) services of this type}? (CP10)
How much of the { {AMT TOT CH}/total charge} did anyone in the family pay for
(PERSON)'s stay at (HOSPITAL) that began on (ADMIT DATE)/ (PERSON)'s
visit to (PROVIDER) on (VISIT DATE)/the last purchase of {NAME OF
PRESCRIBED MEDICINE...} for (PERSON)/the services for (FLAT FEE GROUP)
for (PERSON)/the {OME ITEM GROUP NAME} used by (PERSON) since
(START DATE)/services received at home from (PROVIDER) during (MONTH)
for (PERSON) /(PROVIDER)'s services as part of the visit made on (VISIT
DATE)}? (CP11)

How much did (SOURCE) pay? (CP13)
Has any source reimbursed or paid back anything to (PERSON) (or anyone in the
family) for the amount paid 'out-of-pocket'? That is, has any source reimbursed any
of the {$/% FAMILY PAID} paid? (CP14)


Payer













Table 6-4. Description of Variables


Variable Label Type Description
Dependent Variable
INOUT Outpatient Surgery Nominal Two classes: outpatient or inpatient
Independent Variables
PLAN Type of health plan Nominal Four classes: One-HMO, One non-HMO gatekeeper, one non-HMO non-
gatekeeper, Two-plan
HLTH Self-reported Health Status Ordinal Five levels: poor, fair, good, very good, excellent
COND Number of conditions Ordinal Three levels: 0-1, 2, 3-4 conditions
TCH Total Charge Continuous Total charge
RACE Race/Ethnicity Nominal Two classes: Non-Hispanic White,
Non-white or Hispanics
AGE Age Continuous 0 to 64 years old
GENDER Gender Nominal Two classes: Male, Female
INCOME Annual family income as % FPLa Ordinal Five levels: Less than 100% FPL, 100-124%FPL,
125-199% FPL, 200-399% FPL, 400% FPL or more
EDU Education Ordinal Three levels of education: less than high school, high school, some college or
higher
SUPPORT Family support Nominal Two classes: "1"-"Living with spouse" (adults), or
"from a two-parent family" children)";
"0"-"not living with spouse" (adults), or
"from a one-parent family" (children).
MSA MSA Nominal Two classes: MSA, Non-MSA
REGION Region Nominal Four classes: Northeast, Midwest, South, West
TOTAL Total Payment Continuous Sum of payer share of facility payment and physician payment, and patient's
share of payment
PAYMD Physician Payment Continuous Physician payment from payer and from patient
PTPAY Out-of-pocket payment Continuous Patient share of facility and physician payment
YEAR Year Nominal Three classes: 1997, 1998, and 1999
a FPL: Federal Poverty Level.









Variables

This section describes variables used in the analysis. Some variables come directly

from the MEPS dataset, such as age, income, and health status, while others are

constructed from two or more MEPS variables, such as family support, race/ethnicity,

and payment. Table 6-3 shows the original questions in the MEPS used to derive

independent variables. Based on the conceptual model of treatment choice developed in

Chapter 4 (see Figure 4-3), each variable in Table 6-3 is specified as one of the patient,

physician, and payer characteristics that jointly determine the choice between inpatient

and outpatient surgery. Table 6-3 presents the characteristics, the conceptual basis, and

the origins of each variable. Table 6-4 further describes the final formats of the variables

in the constructed dataset. Table 6-5 contains descriptive statistics for discrete variables,

and Table 6-6 provides descriptive statistics for continuous variables.

Dependent Variable

The dependent variable, INOUT, is dichotomous (INOUT=1 for outpatient surgery,

INOUT=0 for inpatient surgery). An inpatient surgery was defined as a surgical

procedure performed during a patient hospital stay of one or more nights, while an

outpatient surgery was performed in an outpatient setting or in an inpatient setting with

zero nights stay. In the dataset, 68.1% of the cases were outpatient surgeries while 31.9%

were inpatient surgical procedures (Table 6-5).

Primary Independent Variables

To assess the effect of HMO enrollment status on the choice of surgery setting,

PLAN was the primary independent variable. PLAN had four possible values:

1. One HMO: people with one health insurance plan, which was an HMO;









2. One non-HMO gatekeeper plan: people with one health insurance plan, which was a
non-HMO gatekeeper health plan;

3. One non-HMO non-gatekeeper plan: people with one health insurance plan, which
was a non-HMO non-gatekeeper health plan; and

4. Two-plan coverage: those covered by two health plans.

The MEPS identified an individual's HMO enrollment status through the questions

listed in Table 6-3. When a respondent was not enrolled in an HMO, or was not certain

about their health plan type, he/she was classified as a non-HMO enrollee. Among all the

surgical cases in the constructed dataset, 46.3% were covered by one HMO health plan,

6.0% were enrollees of a non-HMO gatekeeper plan, and 38.9% were enrolled in a non-

HMO, non-gatekeeper plan. Figure 6-3 shows the breakdown of the 8.7% of respondents

having two-plan coverage (71 cases). The majority of respondents with two health plans

(54 cases) were covered by at least one non-HMO non-gatekeeper plan.


Two HMOs
n=13

One HMO & One Non-HMO Gatekeeper plan
n=4


One HMO & One Non-HMO Non-Gatekeeper plan
n=30


One Non-HMO Gatekeeper & One Non-HMO Non-
Gatekeeper plan or
Two Non-HMO Non-Gatekeeper plans
n=24


Figure 6-3. Types of Health Plan Coverage for Cases with Two-plan Coverage


Covered by Two Health
Plans
n=71









Table 6-5. Descriptive Statistics for Discrete Variables (Unweighted n=814; Weighted
n=9,595,657)

Level/Class Weighted" Unweighted
Level/Class
Frequency % Frequency %
INOUT Outpatient 6,771,182 70.6 554 68.1
Inpatient 2,824,475 29.4 260 31.9
PLAN One-HMO 4,406,708 45.9 377 46.3
One non-HMO Gatekeeper 577,144 6.0 49 6.0
One non-HMO non-
Onenon-H non- 3,671,750 38.3 317 38.9
Gatekeeper Plan
Two plans 940,056 9.8 71 8.7
HLTH Poor 324,875 3.4 32 3.9
Fair 1,045,211 10.9 105 12.9
Good 2,606,597 27.2 222 27.3
Very good 3,056,292 31.9 258 31.8
Excellent 2,552,196 26.6 195 24.0
COND 0 45,606 0.5 2 0.3
1 8,535,207 89.9 728 89.4
2 853,355 8.9 68 8.4
3-4 161,490 1.7 16 2.0
RACE Non-Hispanic White 8,258,534 86.1 644 79.1
Non-White or Hispanic 1,337,123 13.9 170 20.9
GENDER Male 4,187,401 43.6 313 38.5
Female 5,408,257 56.4 501 61.5
INCOME Less than 100% FPL 339,691 3.5 36 4.2
100-124% FPL 294,664 3.1 31 3.8
125-199% FPL 797,722 8.3 83 10.2
200-399% FPL 3,278,588 34.2 298 36.6
400% FPL or More 4,884,992 50.9 366 45.0
EDU Less than High School 682,436 7.2 79 9.8
High school 4,776,133 50.1 397 49.1
Some college or higher 4,069,169 42.7 333 41.2
SUPPORT Living with spouse/ two 6,970,556 72.6 588 72.2
No spouse/single parent 2,625,101 27.4 226 27.8
MSA MSA 7,865,969 82.0 633 77.8
Non-MSA 1,729,688 18.0 181 22.2
REGION Northeast 1,942,467 20.2 166 20.4
Midwest 2,614,114 27.2 205 25.2
South 3,613,755 37.7 314 38.6
West 1,425,322 14.9 129 15.9
YEAR 1997 3,333,464 34.7 358 44.0
1998 3,466,624 36.1 248 30.5
1999 2,795,568 29.1 208 25.6
a Number of weighted cases is the product of number unweighted cases and the corresponding sample
weight of each unweighted case.









Table 6-6. Descriptive Statistics for Continuous Variables (Unweighted n=814; Weighted
n=9,595,657)

Weighted Unweighted
Standard Standard
Mean Mean
Error Error
AGE 37.84 0.76 38.82 0.58
TCH (Total Charge) 6,456.13 249.47 6,808.49 245.13
TOTAL (Total Payment) 3,800.08 159.76 4,094.96 167.18
PAYMD (Physician Payment) 994.95 51.34 1,013.58 47.57
PTPAY 151.91 16.57 154.57 12.70
(Patient Out-of-pocket Payment)
Weighted data are based on information of each unweighted case and its corresponding sample weight.
The means and standard errors of this column are the results of 9,595,657 weighted cases.

Cases that did not report health plan information (23 cases) were included in the

non-HMO non-gatekeeper plan category. Table 6-7 and Table 6-8 compare cases

reporting a non-HMO non-gatekeeper plan with those that did not provide health plan

information. Most of the characteristics did not differ significantly between the two

groups, although three variables were significantly different, including percentage of

high-income cases (income 400% FPL or higher), total payment, and patient payment.

Because the two groups of surgery cases had similar characteristics, cases with no plan

information were included in the non-HMO non-gatekeeper category.

When excluding surgeries performed primarily in either an inpatient or outpatient

setting, as defined in Table 6-2, the subset of cases had a similar distribution across types

of health plan coverage (Table 6-9). Over 48% were HMO enrollees, while 36.3% had

non-HMO non-gatekeeper health plan coverage. Only 3.8% indicated that they were

enrolled in a non-HMO gatekeeping plan, and 11.3% were covered by two plans.









Table 6-7. Comparing Cases with a Non-HMO Non- Gatekeeper Plan and Cases with No
Self-Reported Plan Information (Nominal and Ordinal Variables)

Self-Reported No Plan
No Plan
Non-HMO
Non- O Information
Non-Gatekeeper Plan (n
(n=294)
Variables Level/Class Percenta
INOUT Outpatient 70.7 73.9
Inpatient 29.3 26.1
p value 0.7480
COND 0-1 91.5 91.3
2 7.5 4.4
3-4 1.0 4.4
p value 0.5255
HLTH Poor 3.1 8.7
Fair 14.0 13.0
Good 27.3 30.4
Very good 33.5 17.4
Excellent 22.2 30.4
p value 0.3607
Excellent/ Yes 55.4 47.8
Very Good No 44.6 52.2
Health p value 0.4808
Fair/Poor Yes 17.0 21.7
Health No 83.0 78.3
p value 0.5648
RACE Non-Hispanic White 82.7 78.3
Non-White or Hispanic 17.3 21.7
p value 0.5957
GENDER Male 39.1 39.1
Female 60.9 60.9
p value 0.9989
SUPPORT Living with spouse/ two 74.8 60.9
parents
No spouse/single parent 25.2 39.1
p value 0.1441
MSA MSA 66.0 78.3
Non-MSA 34.0 21.7
p value 0.2300
REGION Northeast 16.7 13.0
Midwest 34.4 21.7
South 34.4 47.8
West 14.6 17.4
P value 0.4969









Table 6-7. Continued


Self-Reported No Plan
No Plan
Non-HMO
Non- O Information
Non-Gatekeeper Plan (n
(n=294)
Variables Level/Class Percent
INCOME Less than 100% FPL 3.7 4.4
100-124% FPL 6.5 8.7
125-199% FPL 10.2 8.7
200-399% FPL 36.1 56.5
400% FPL or More 43.5 21.7
p value 0.2869
High Income Yes p43.5 21.7
(400% FPL or No 56.5 78.3
more) p value 0.0425
EDU Some High School or 11.0 17.4
High School Graduate 46.7 39.1
Some College or 42.3 43.5
p value 0.5974
Some High Yes p11.0 17.4
School or No (High School Graduate 89.0 82.6
Less p value 0.3450
YEAR 1997 44.6 26.1
1998 32.0 34.8
1999 23.5 39.1
p value 0.1467
a Percent is based on unweighted number of cases.









Table 6-8. Comparing Cases with a Non-HMO Non-Gatekeeper Plan and Cases with No
Self-Reported Plan Information (Continuous Variables)

Self-Reported
Non-HMO No Plan
Non-Gatekeeper Information (n=23)
Plan (n=294)
Standard Standard p
Mean Mean
Error Error value
AGE 39.58 1.01 36.30 3.33 0.3471
TCH (Total Charge) 6,982.87 420.22 5,814.33 906.02 0.2429
TOTAL (Total Payment) 4,812.24 338.02 3,185.44 692.88 0.0356
FACILITY (Facility Payment) 3,381.75 299.73 2,255.85 613.56 0.1002
PAYMD (Physician Payment) 1,161.20 86.47 875.78 212.79 0.2149
PTPAY (Patient Out-of- 269.29 28.85 53.81 34.41 0.0000
pocket Payment)


Table 6-9. Cases for the Subset of the Dataset by Health Plan Coverage

Weighted Unweighted
Cases Percent Cases Percent
One HMO plan 2,166,381 49.5 190 48.6
One non-HMO gatekeeper plan 154,471 3.5 15 3.8
One non-HMO non-gatekeeper plan 1,510,203 34.5 142 36.3
Two plans 545,084 12.5 44 11.3
TOTAL 4,376,139 100.0 391 100.0
Note: This subset of data includes only cases of surgical procedures having an I/O ratio between 0.2 and
5.0 in the 1996 National Health Care Surveys. Number of weighted cases is the product of the number of
unweighted cases and the corresponding sample weight of each unweighted case.


Control Variables: Patient Characteristics

For the purpose of this analysis, self-perceived health status (HLTH) represents

pre-surgery health in general. In the MEPS, self-perceived health status was reported in

each interview. Thus, the health status reported in the interview prior to the surgical

procedure was used. For example, if a respondent had a surgery during the second round

of the interview (Round Two), the health status reported in the previous round (Round

One) was used. Almost 56.0% of the surgical cases indicated that their health status was

"excellent" or "very good," while 27.3% reported "good" health status. Only a small









percent (3.9%) reported "poor" health status, with 12.9% reporting "fair" health status

(Table 6-5).

For surgical cases occurring during the first round of the MEPS interview, the pre-

surgery health status came from the previous year's NHIS by linking the MEPS file with

the 1996, 1997, and 1998 NHIS data. There were 63 cases in the 1997 MEPS, 51 cases

in the 1998 MEPS, and 49 cases in the 1999 MEPS having health status (HLTH)

information from the previous year's NHIS data files.

Number of conditions (COND) was the count of ICD-9 condition codes reported in

a surgical event. COND was taken directly from a variable (NUMCOND) in the 1997

MEPS dataset, but for 1998 and 1999 MEPS data, COND was the count of medical

conditions that could be linked to the surgical event. Most cases (89.4%) had one

condition, while 10.4% had two to four conditions associated with the surgical event

(Table 6-5).

Total charge (TCH) was the sum of facility, physician, and patient charge, and was

taken directly from the MEPS dataset. Total charge was used as an indication of the

intensity of the service rendered, and was intended to control for the severity of the

surgery. The mean charge per case was $6,808.49 (Table 6-6).

Patients' race (RACE) was a constructed variable based on two MEPS variables,

RACEX and RACETHNX. Over 79.0% of the surgical cases in the constructed dataset

were non-Hispanic Whites. Due to the small number of cases, Black, Hispanic,

American Indian, Aleut, Eskimo, Asian, and Pacific Islander were grouped into one

category, non-White or Hispanic (RACE=0), which comprised almost 21.0% of all cases

(Table 6-5).









Gender (GENDER) is either male or female. A majority of the cases were female

(61.5%), while 38.5% were male (Table 6-5).

Two socioeconomic status variables, family annual income and education were

based on two MEPS variables, POVCAT and EDUCYR. Family annual income

(INCOME) was reported as the percent of federal poverty level (FPL), which took into

account family size (Federal Register 2001). INCOME had five levels: less than 100%

FPL, 100% to 124%, 125% to199%, 200% to 399%, and 400% FPL or more. A small

proportion of cases (8.0%) were respondents from families with less than 125% FPL,

while 45.0% were from families with incomes of at least 400% FPL. Over 10.0% of the

cases were from families with incomes between 125 and 199% FPL, and 36.6% were in

the range of 200% to 399% FPL income (Table 6-5).

Education (EDU) was based on the MEPS variable EDUCYR. For people younger

than 18 and not married, EDU was taken from mother's education level, or father's in the

case of single-parent (father) family. Almost half of the cases (49.1%) were high school

graduates, and 41.2% had at least some college education, while 9.8% did not graduate

from high school (Table 6-5).

Family support (SUPPORT) was derived from SPOUSEIN in the MEPS dataset.

For adults, a person had a value of "1" if married and living with spouse. Children who

came from a two-parent family, i.e., mother's and father's ID were valid (MOMPID and

DADPID) in the MEPS data, also have a value of "l." Otherwise, SUPPORT was

assigned "0." Over 72.0% of the cases had a value of"1" (Table 6-5).









The variable AGE is based on the reported age (AGE97X, AGE98X, and AGE99X

in the MEPS dataset). Age is a continuous variable. The mean AGE is 38.8 years (Table

6-6).

Control Variables: Physician Characteristics

Two geographic variables, MSA and REGION, represent physician practice

environment. Both variables were taken directly from the MEPS dataset. Almost 78.0%

of the surgical cases were people residing in an MSA (Table 6-5). A large proportion of

respondents (38.6%) lived in the South, with 20.4% living in the Northeast and 25.2% in

the Midwest. Only 15.9% lived in the West (Table 6-5).

Control Variables: Other Payer Characteristics

Other than enrollment in an HMO, payer characteristics included three payment

rates: total payment (TOTAL), physician payment (PAYMD), and patient out-of-pocket

(PTPAY). TOTAL was the sum of the payment by a private insurer for a surgical event

and the patient's out-of-pocket payment for both facility and physician. As shown in

Table 6-6, based on unweighted samples, mean TOTAL was $4,094.96, mean PAYMD

was $1,013.58, and mean PTPAY was $154.57.

Other Control Variable

Because the datasets used in this analysis were combined from three years data,

YEAR, i.e. 1997, 1998, and 1999, was included to account for other changes over time

(Table 6-5). Over 30.0% of cases were from the 1998 MEPS, 44.0% were from the 1997

data, and 25.6% were from the 1999 MEPS data.

Statistical Analysis

This section describes the statistical analysis, based on the conceptual framework

(Chapter 4) and the empirical specification (Chapter 5). The unit of observation is a









privately insured non-elderly person who had a surgical procedure that could be done in

either an outpatient or inpatient setting. For patients having such a surgical procedure,

the models test the following hypotheses:

* HMO patients were more likely to have outpatient surgery than non-HMO patients,
when controlling for patient, physician, and payer characteristics.

* Patients enrolled in a gatekeeper plan were more likely to have outpatient surgery
than patients with a non-gatekeeper plan.

* When excluding surgeries that are done primarily in one setting and only occasionally
done in the other setting, HMO status has a stronger effect on the likelihood of having
an outpatient surgery than for surgeries in general.

The analysis uses logistic regression because the dependent variable, the choice of

outpatient surgery or inpatient surgery, has two outcomes. In this analysis, outpatient

surgery is set to "1" while inpatient surgery is "O". For binary dependent variables,

logistic regression predicts the probability of one outcome [ z(x)) ] assuming that the

cumulative density function of the error term is logistic. The specification of the logit

regression is:


logit[(x,)]=log( )=a+yfl~, +e, (6-1)
1 -7(x,) 11

where the probability of INOUT=1 is z(x,) = e- (6-2)

a is the constant, ; is the error term, and Pi is the parameter of the ith independent
variable ( x ).

To test the effect of HMO enrollment status ( /3o) and the effect of gatekeeper

enrollment status (GATE), the adjusted Wald statistic is used to test the null hypothesis

that / equals to zero. It is expected that the null hypothesis will be rejected. In other









words, the expectation is that the effect of HMO enrollment status will be significant and

positive.

Sample Size

Sample size is an important issue in statistical analysis. Based on maximum

likelihood estimation, f, is consistent, efficient and asymptotically normal if sample size

is large enough. Small sample size can lead to two problems (Peduzzi et al. 1996). First,

estimated f, 's can be biased. Second, sample variance of the estimated f, can be too

large, resulting in false insignificance. To avoid these problems, the general guidelines

for sample size are:

* At least 10 events per variable (EPV) (Peduzzi et al. 1996; Long 1997). For example,
the EPV of the dataset used in this dissertation is computed as:
E: the smaller number of events between outpatient surgical events (554
cases) and inpatient surgical events (260 cases);
V: number of independent variables, (x,, x -, x 2);

E 260
EPV = =12.4
V 21


* Total sample size should be at least 100 (Long 1997; Cohen et al. 1999). However, if
there is little variation in the dependent variable, e.g., almost all outcomes are "1," a
larger total sample size is needed.


* If the independent variables are highly collinear, a large sample size is required for
better estimation.


Since the EPV is greater than 10, total sample size is 814, and the number of events for

each value of the dependent variable, i.e., "1" and "0," have some variation, (554 and 260

respectively), the sample size is sufficient.









Analytical Issues

Other analytical issues include possible selection bias, unobserved factors, and

collinearity between independent variables.

Self-selection

Self-selection may confound the effect of type of health plan on the likelihood of

receiving an outpatient surgery (Figure 6-4). People with certain characteristics, such as

being younger or healthier, may prefer HMO plans to non-HMO plan, and younger or

healthier patients may be more likely to receive an outpatient surgery. In such a

situation, the estimated coefficient for HMO will be biased upward. A two-stage

estimation method will be used to account for self-selecting into an HMO (or non-HMO).

The estimator from the first stage predicts the probability of choosing a health plan given

a person's characteristics. The first stage estimator will then be incorporated into the

error term of the second stage equation, which is the equation estimating the probability

of having an outpatient surgery.


Patients self-select
into HMO


HMO ... Having an outpatient surgery
Biased toward

Figure 6-4. The Possible Self-Selection Effect.

Unobserved variables

Three relevant variables: physician specialty, payment method (capitation vs. FFS),

and the extent of utilization management are not available for this analysis. As shown in

Figure 6-5, HMOs may use different specialties than non-HMO plans, which would have

an unobservable effect "b." In turn, physicians of different specialties may make









different choices between outpatient and inpatient surgery, and this "c" effect is also not

observed. Having an HMO as the payer has effect "a" on the likelihood of receiving

outpatient surgery. Given that both "b" and "c" are not observed, "a" can still be unbiased

in terms of measuring the total effect, but partial effects, "b" and "c," will not be

identified. This omission may result in false insignificance of the primary independent

variable (i.e., having HMO as the payer) (Greene 2000).


Figure 6-5. The Effect of Unobserved Variable: Surgeon's Specialty


Figure 6-6. The Effect of Unobserved Variables: Payment Method and Utilization
Management


A similar situation holds for the unobserved payment method and the extent of

utilization management (Figure 6-6). As before, the effect of having HMO coverage on


Surgeon's specialty

..HMO Outpatient surgery
HMO a Outpatient surgery


....yPayment Method... -
...,... b """ ...,

HMO a- Outpatient surgery



Utilization management

i ..."...... .............
HMO -- Outpatient surgery









the likelihood of receiving outpatient surgery is the total effect, "a," but the partial effects

"b" and "c" will be missed. Unfortunately, these variables are not available for the

analysis, and therefore are omitted from the analysis.

Collinearity between independent variables

In this study, multicollinearity is another important issue. Independent variables

may be highly correlated, i.e., HLTH and CON, INCOME and EDU. For example,

previous studies have found that self-perceived health status and clinically measured

health status have a parallel predictive power (Mossey and Shapiro 1982; Idler and

Benyamini 1997). Correlation between each pair of independent variables will be

examined. If collinearity is found to be a significant problem, the final model may omit

certain variables.

Statistical software

STATA version 7.0 is used for the statistical analysis. STATA's survey estimates

can appropriately adjust for complex sample design and yield robust variance estimates

(Cohen 1997; Cohen et al. 1999).

Plan of analysis

All analyses are performed on both the constructed dataset (all 814 surgical cases)

and a subset of cases (391 cases) as defined in Table 6-2. First, inpatient/outpatient

surgery cases are tabulated for each of the four possible values of the primary

independent variable, PLAN. STATA uses the usual Pearson X2 statistic for two-way

tables to assess whether the distribution of inpatient/outpatient surgery cases depends on

health plan type. To account for the survey design, the Pearson X2 statistic is transformed

into an F statistic with non-integer degrees of freedom using a second-order Rao and

Scott correction (StataCorp 2001).









Means and standard errors are calculated for charge and payment variables,

including total charge, total payment, physician payment, and patient out-of-pocket

payment. Statistical significance of the difference between health plan types in the same

surgery setting (i.e., within outpatient settings, or within inpatient setting) is tested.

Univariate analyses are performed to estimate the magnitude and significance of

the one-to-one correlation between one independent variable and the dependent variable,

having an outpatient surgery. One logistic regression model is first fitted for each

independent variable (univariate analysis), and estimates a coefficient (3) for the

independent variable as shown in formula 6-1.

Based on the preliminary findings, multivariate regression fits multiple variables

into the model to estimate the effects of health plan type on the likelihood of having an

outpatient surgery, holding other factors constant. Statistical significance of the

coefficients is based on an adjusted Wald test, and p-values are indicated. Odds ratios

(ORs) are reported.














CHAPTER 7
FINDINGS

This dissertation examines the effect of HMO and gatekeeper health plan coverage

on the choice between outpatient and inpatient surgery. Using a constructed dataset from

the pooled 1997, 1998, and 1999 MEPS public use data files, analyses are performed to

answer the research questions:

1. For a patient under age 65 who is diagnosed to undergo a surgical procedure that is
feasible in either outpatient or inpatient settings, after controlling for severity, does an
HMO patient have a higher likelihood of receiving an outpatient surgery (vs.
inpatient) than a non-HMO patient?

2. As a widely used managed care approach, does using a gatekeeper affect the choice
of outpatient or inpatient surgery? Are patients with gatekeeper plans more likely to
have an outpatient surgery?

3. When excluding surgeries that are done primarily in one setting and only occasionally
done in the other setting, does HMO status have a stronger effect on the likelihood of
having an outpatient surgery than for surgeries in general?

To study the third research question, a subset of the data was constructed by

excluding surgeries done primarily in one setting (inpatient or outpatient). The subset of

data contains cases of surgeries having reported inpatient cases between 0.2 and 5.0 times

of that of outpatient cases (I/O ratios between 0.2 and 5.0) in the 1996 National Health

Care Surveys (see Table 6-2).

Contingency Tables

The results come from several types of analyses. First, the distribution of nominal

variables is tabulated in two-way tables by type of health plan (Tables 7-1 through 7-4).

Three-way tables then present the distribution of selected control variables for each









surgery setting by health plan type (Table 7-5 and Table 7-6). Next, mean charges and

payments for outpatient and inpatient surgery by health plan type are analyzed (Table 7-

7 and Table 7-8). Univariate regression analyses assess the correlation between each

dependent variable and each independent variable (Table 7-9 and Table 7-10). Finally,

results are presented from multivariable regression analyses (Tables 7-11 through 7-17).

Two-way Tables: Surgery Setting by Health Plan Type

Table 7-1. Weighted Number (Percent) of Outpatient and Inpatient Surgical Cases by
Health Plan Coverage (Based on Unweighted n=814)

Number of Cases (Percent)
Outpatient Surgery Inpatient Surgery Total
One HMO Plan 3,094,861 (70.2%) 1,311,847 (29.8%) 4,406,708(100.0%)
One Non-HMO
OtekeeperO 445,918 (77.3%) 131,226 (22.7%) 577,144(100.0%)
Gatekeeper Plan
One Non-HMO
OneNon-HMGa O P 2,627,267 (71.5%) 1,044,483 (28.5%) 3,671,750(100.0%)
Non-Gatekeeper Plan
Two Plans 603,135 (64.2%) 336,921 (35.8%) 940,056(100.0%)
p value 0.5532
One Plan 6,168,046 (71.3%) 2,487,556 (28.7%) 8,655,602(100.0%)
Two Plans 603,135 (64.2%) 336,921 (35.8%) 940,056(100.0%)
p value 0.2669
Note: Weighted number of cases were the product of unweighted number of cases and sample weights.

Table 7-2. Unweighted Number (Percent) of Outpatient and Inpatient Surgical Cases by
Health Plan Coverage (Based on Unweighted n=814)

Number of Cases (Percent)
Outpatient Surgery Inpatient Surgery Total
One-HMO Plan 249 (66.0%) 128 (34.0%) 377 (100.0%)
One Non-HMO
OneNon-HO 36 (73.5%) 13 (26.5%) 49 (100.0%)
Gatekeeper Plan
One Non-HMO
OneNon O 225 (71.0%) 92 (29.0%) 317 (100.0%)
Non-Gatekeeper Plan
Two Plans 44 (62.0%) 27 (38.0%) 71 (100.0%)
p value 0.2831
One Plan 510 (68.6%) 233 (31.4%) 743 (100.0%)
Two Plans 44 (62.0%) 27 (38.0%) 71 (100.0%)
p value 0.2502