<%BANNER%>

Medical Cost Offset Effect and Absenteeism in Longitudinal Samples of Depressed Pulmonary and Cancer Patients

Permanent Link: http://ufdc.ufl.edu/UFE0022410/00001

Material Information

Title: Medical Cost Offset Effect and Absenteeism in Longitudinal Samples of Depressed Pulmonary and Cancer Patients
Physical Description: 1 online resource (129 p.)
Language: english
Creator: Lee, Andrea
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: absenteeism, antidepressant, cancer, cost, depression, health, medical, mental, offset, psychotherapy, pulmonary, treatment, work
Clinical and Health Psychology -- Dissertations, Academic -- UF
Genre: Psychology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: An intervention that reduces or prevents usual costs to the health care system is called a medical cost offset or the cost offset effect. Past research shows a medical cost offset effect; however, more recent work does not. The purpose of this study was to examine the effects of mental health treatment on health care expenditures and absenteeism in samples of depressed patients with or without comorbid pulmonary conditions or cancer, as well as a general depression group. This study attempted to provide a current assessment of the medical cost offset effect from mental health services. Additionally, this study attempted to provide a more comprehensive examination of mental health treatment effects by including workplace absenteeism as an outcome. The research questions examined in this study were (1) whether total, medical, or drug expenditures were higher for individuals undergoing mental health treatment and whether expenditures reduced over time, (2) whether emergency room, inpatient, outpatient, or office-based provider visits were higher for individuals undergoing mental health treatment and whether utilization decreased over time, and (3) whether work absenteeism rates are lower for individuals undergoing mental health treatment. Data were obtained from the Medical Expenditure Panel Survey (MEPS), a nationally representative survey of the US non-institutionalized, civilian population. Results demonstrate that total expenditures were greater for each treatment group and medical expenditures were greater for only the depression general treatment group. Total emergency room visits, inpatient visits, outpatient visits, and office-based provider visits were all greater for the treatment groups than the no treatment group. No treatment group had a significant change in expenditures over time. Mental health treatment only impacted change for inpatient visits for the pulmonary and cancer diagnoses groups. With respect to work absenteeism, mental health treatment was associated with increased absenteeism rates for the general depression group, but the pulmonary and cancer diagnoses groups showed no differece in work absenteeism between mental health treatment and no treatment.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Andrea Lee.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Frank, Robert G.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0022410:00001

Permanent Link: http://ufdc.ufl.edu/UFE0022410/00001

Material Information

Title: Medical Cost Offset Effect and Absenteeism in Longitudinal Samples of Depressed Pulmonary and Cancer Patients
Physical Description: 1 online resource (129 p.)
Language: english
Creator: Lee, Andrea
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: absenteeism, antidepressant, cancer, cost, depression, health, medical, mental, offset, psychotherapy, pulmonary, treatment, work
Clinical and Health Psychology -- Dissertations, Academic -- UF
Genre: Psychology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: An intervention that reduces or prevents usual costs to the health care system is called a medical cost offset or the cost offset effect. Past research shows a medical cost offset effect; however, more recent work does not. The purpose of this study was to examine the effects of mental health treatment on health care expenditures and absenteeism in samples of depressed patients with or without comorbid pulmonary conditions or cancer, as well as a general depression group. This study attempted to provide a current assessment of the medical cost offset effect from mental health services. Additionally, this study attempted to provide a more comprehensive examination of mental health treatment effects by including workplace absenteeism as an outcome. The research questions examined in this study were (1) whether total, medical, or drug expenditures were higher for individuals undergoing mental health treatment and whether expenditures reduced over time, (2) whether emergency room, inpatient, outpatient, or office-based provider visits were higher for individuals undergoing mental health treatment and whether utilization decreased over time, and (3) whether work absenteeism rates are lower for individuals undergoing mental health treatment. Data were obtained from the Medical Expenditure Panel Survey (MEPS), a nationally representative survey of the US non-institutionalized, civilian population. Results demonstrate that total expenditures were greater for each treatment group and medical expenditures were greater for only the depression general treatment group. Total emergency room visits, inpatient visits, outpatient visits, and office-based provider visits were all greater for the treatment groups than the no treatment group. No treatment group had a significant change in expenditures over time. Mental health treatment only impacted change for inpatient visits for the pulmonary and cancer diagnoses groups. With respect to work absenteeism, mental health treatment was associated with increased absenteeism rates for the general depression group, but the pulmonary and cancer diagnoses groups showed no differece in work absenteeism between mental health treatment and no treatment.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Andrea Lee.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Frank, Robert G.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0022410:00001


This item has the following downloads:


Full Text

PAGE 1

1 MEDICAL COST OFFSET EFFECT AND ABSENTEEISM IN LONGITUDINAL SAMPLES OF DEPRESSED PULMONARY AND CANCER PATIENTS By ANDREA M. LEE 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 2009

PAGE 2

2 200 9 Andrea M. Lee

PAGE 3

3 To my mother and father

PAGE 4

4 ACKNOWLEDGMENTS The opportunity to pursue graduate studies and subsequently to embark on completing this dissertation would not be possible without the loving support of my mother and father. They have stood behind me and encouraged me from the inception of the idea that I wanted to pursue a doctoral degree in psychology, even at a school painfully far away from them. I am immeasurably grateful for their pride in me and my work. Their pride and confidence in me, and echoed sentiments from my brothers and grandparents, b ecame the fuel that kept me trucking along through graduate school. I could not forget to acknowledge the mentorship of Bob Frank who made graduate school a minimally stressful process and more importantly, a professional growth experience, by virtue of his encouragement of my independence and the path that was best for me. There could have been no mentor that would have been a better fit for my needs as a budding psychologist. No less important has been the advisement from Jeff Harman who patiently endu red my countless e mail requests for help and my expressed frustrations with the dissertation. Through it all, he has been nothing short of generous with his time, attention, and expertise. Certainly, the generosity of Rus Bauer with taking on the role of co departure from the University of Florida has been tremendously helpful. He has been a valuable addition to my doctoral committee because his big picture approach to my work has made discussions interesting and thought provoking. L doctoral committee cannot go unacknowledged. Her supportive and calming presence during the dissertation process has been indispensable. Past and present individuals f rom the Florida Center for Medicaid and the Uninsured (FCMU) have been vital to keeping me on track, but also helped me to remember laughter and

PAGE 5

5 fun. Heather Steingraber has most certainly taken on a vital role in ensuring that graduate school would not o vertake me and her steady, supportive presence has been invaluable Lorna Chorba, Jianyi Zhang, and Michelle Romano have all made my experience at FCMU positive with their winning personalities and engaging conversation. Zo Swaine, Lisa Chacko, and Nata lie Blevins have each been current or past officemates at different points, and all are considered friends. Jingbo Yu is also a friend, but has taken on a particularly spe cial role in my dissertation because she acted as an invisible doctoral committee me mber to me with her expertise gleaned from her own dissertation and her patience with my constant barrage of questions. Additional people who have been behind the scenes, encouraging me, and supporting me through various stages of graduate school have all been important to my achievements. My friends from the Department of Clinical and Health Psychology entering class of 2004 and friends from Vancouver, B.C. are all unforgettable and appreciated beyond words.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ......................... 10 LIST OF FIGURES ................................ ................................ ................................ ....................... 13 ABSTRACT ................................ ................................ ................................ ................................ ... 14 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 16 Overview ................................ ................................ ................................ ................................ 16 Purpose of the Present S tudy ................................ ................................ ................................ .. 16 2 LITERATURE REVIEW ................................ ................................ ................................ ....... 18 Cost of Depression ................................ ................................ ................................ .................. 18 The Medical Cost Offset Effect ................................ ................................ .............................. 19 Depression Treatment and Absenteeism ................................ ................................ ................ 23 3 STUDY SIGNIF ICANCE AND RESEARCH QUESTIONS ................................ ............... 26 Significance of this study ................................ ................................ ................................ ........ 26 Specific Aims and Hypotheses ................................ ................................ ............................... 27 Aim 1 ................................ ................................ ................................ ............................... 27 Hypothesis 1a ................................ ................................ ................................ ........... 28 Hypothesis 1b ................................ ................................ ................................ ........... 28 Aim 2 ................................ ................................ ................................ ............................... 28 Hypothesis 2a ................................ ................................ ................................ ........... 28 Hypothesis 2b ................................ ................................ ................................ ........... 29 Aim 3 ................................ ................................ ................................ ............................... 29 Hypothesis 3 ................................ ................................ ................................ ............. 29 4 CONCEPTUAL MODELS ................................ ................................ ................................ .... 30 Mental Health Treatment, Health Care Utilization, and Health Care Expenditures .............. 30 Mental Health Treatment and Absenteeism ................................ ................................ ........... 30 5 DATA AND METHODS ................................ ................................ ................................ ....... 33 Data Source ................................ ................................ ................................ ............................. 33 Variables ................................ ................................ ................................ ................................ 34 Dependent Variables ................................ ................................ ................................ ....... 35

PAGE 7

7 Aim 1: Total health care expenditures, medical expenditures, and prescription drug expenditures ................................ ................................ ................................ .. 35 Aim 2: Health care utilization ................................ ................................ .................. 36 Aim 3: Missed days from work ................................ ................................ ................ 36 Inde pendent Variables ................................ ................................ ................................ ..... 37 Mental health treatment (MHtx) ................................ ................................ .............. 37 Depression Groups (Dep) ................................ ................................ ......................... 37 Pulmonary condition (Pulm) & Cancer (Canc) ................................ ........................ 38 Covariates ................................ ................................ ................................ ........................ 39 Age (Age) ................................ ................................ ................................ ................. 39 Sex (Sex) ................................ ................................ ................................ .................. 39 Race (Racex) ................................ ................................ ................................ ............ 40 Hispanic (Hispan) ................................ ................................ ................................ ..... 40 Years of educa tion (Educyear) ................................ ................................ ................. 40 Total individual income (Ttlp) ................................ ................................ ................. 40 ................................ ............... 41 Perceived physical health (Rthlth) and perceived mental health (Mnhlth) .............. 41 The need for help with activities of daily living (Adlhlp) and instrumental activi ties of daily living (Iadlhp) ................................ ................................ .......... 41 Number of medical comorbidities (Comorb) ................................ ........................... 41 Insurance status (Insstat) ................................ ................................ .......................... 42 Statistical Methods ................................ ................................ ................................ .................. 42 Aim 1: Mental Health Treatment and Health Care Expenditures ................................ ... 43 Hypothesis 1a ................................ ................................ ................................ ........... 43 Hypothesis 1b ................................ ................................ ................................ ........... 45 Predictions based on the two or three part equations ................................ ............ 47 Aim 2: Mental Health Treatment and Health Care Utilization ................................ ....... 48 Hypothesis 2a ................................ ................................ ................................ ........... 48 Hypothesis 2b ................................ ................................ ................................ ........... 49 Predictions based on the three equations ................................ ................................ .. 50 Aim 3: Mental Health Treatment and Absenteeism ................................ ........................ 50 Hypothesis 3 ................................ ................................ ................................ ............. 50 Selection bias and s election correction modeling ................................ .................... 50 6 RESULTS ................................ ................................ ................................ ............................... 59 Overview ................................ ................................ ................................ ................................ 59 Description of the Sample ................................ ................................ ................................ ...... 59 Total Expenditures ................................ ................................ ................................ .................. 61 Two Part Model (Y 1 + Y 2 ) ................................ ................................ .............................. 61 Part I ................................ ................................ ................................ ......................... 61 Part II ................................ ................................ ................................ ........................ 62 Bootstrapping prediction ................................ ................................ .......................... 62 Summary ................................ ................................ ................................ .................. 62 Three Part Model (Y 2 Y 1 ) ................................ ................................ ............................ 63 Part I ................................ ................................ ................................ ......................... 63 Part II ................................ ................................ ................................ ........................ 64

PAGE 8

8 Part III ................................ ................................ ................................ ...................... 64 Bootstrapping prediction ................................ ................................ .......................... 65 Summary ................................ ................................ ................................ .................. 65 Medical Expenditures ................................ ................................ ................................ ............. 66 Two Part Mod el (Y 1 + Y 2 ) ................................ ................................ .............................. 66 Part I ................................ ................................ ................................ ......................... 66 Part II ................................ ................................ ................................ ........................ 66 Bootstrapping prediction ................................ ................................ .......................... 67 Summary ................................ ................................ ................................ .................. 67 Three Part Model (Y 2 Y 1 ) ................................ ................................ ............................ 68 Part I ................................ ................................ ................................ ......................... 68 Part II ................................ ................................ ................................ ........................ 69 Part III ................................ ................................ ................................ ...................... 69 Bootstrapping prediction ................................ ................................ .......................... 69 Summary ................................ ................................ ................................ .................. 70 Prescription Drug Expenditures ................................ ................................ .............................. 70 Two Part Model (Y 1 + Y 2 ) ................................ ................................ .............................. 70 Part I ................................ ................................ ................................ ......................... 71 Part II ................................ ................................ ................................ ........................ 71 Bootstrapping prediction ................................ ................................ .......................... 72 Summary ................................ ................................ ................................ .................. 72 Three Part Model (Y 2 Y 1 ) ................................ ................................ ............................ 72 Part I ................................ ................................ ................................ ......................... 73 Part II ................................ ................................ ................................ ........................ 73 Part III ................................ ................................ ................................ ...................... 74 Bootstrapping prediction ................................ ................................ .......................... 74 Summary ................................ ................................ ................................ .................. 75 Emergency Room Visits ................................ ................................ ................................ ......... 75 Negative Binomial Regression (Y 1 + Y 2 ) ................................ ................................ ........ 75 Three Part Model (Y 2 Y 1 ) ................................ ................................ ............................ 76 Part I ................................ ................................ ................................ ......................... 76 Part II ................................ ................................ ................................ ........................ 77 Part III ................................ ................................ ................................ ...................... 77 Bootstrapping prediction ................................ ................................ .......................... 77 Summary ................................ ................................ ................................ .................. 78 Inpatient Visits ................................ ................................ ................................ ........................ 78 Negative Binomial Regression (Y 1 + Y 2 ) ................................ ................................ ........ 78 Three Part Model (Y 2 Y 1 ) ................................ ................................ ............................ 79 Part I ................................ ................................ ................................ ......................... 79 Part II ................................ ................................ ................................ ........................ 80 Part III ................................ ................................ ................................ ...................... 80 Bootstrapping prediction ................................ ................................ .......................... 81 Summary ................................ ................................ ................................ .................. 81 Outpatient Visits ................................ ................................ ................................ ..................... 82 Negative Binomial Regression (Y 1 + Y 2 ) ................................ ................................ ........ 82 Three Part Model (Y 2 Y 1 ) ................................ ................................ ............................ 82

PAGE 9

9 Part I ................................ ................................ ................................ ......................... 83 Part II ................................ ................................ ................................ ........................ 83 Part III ................................ ................................ ................................ ...................... 84 Bootstrapping prediction ................................ ................................ .......................... 84 Summary ................................ ................................ ................................ .................. 84 Office Based Provider Visits ................................ ................................ ................................ .. 85 Negative Binomial Regression (Y 1 + Y 2 ) ................................ ................................ ........ 85 Three Part Model (Y 2 Y 1 ) ................................ ................................ ............................ 86 Part I ................................ ................................ ................................ ......................... 86 Part II ................................ ................................ ................................ ........................ 86 Part III ................................ ................................ ................................ ...................... 87 Bootstrapping prediction ................................ ................................ .......................... 87 Summary ................................ ................................ ................................ .................. 88 Work Absenteeism ................................ ................................ ................................ .................. 88 Negative binomial regression ................................ ................................ .......................... 89 Poisson regression ................................ ................................ ................................ ........... 89 Post Hoc (Sens itivity) Analyses ................................ ................................ ............................. 90 Psychotherapy vs. Antidepressants ................................ ................................ ................. 90 Mild vs. Severe Depression ................................ ................................ ............................. 91 Full Time vs. Part Time Work ................................ ................................ ........................ 92 7 DISCUSSION ................................ ................................ ................................ ....................... 115 Overview ................................ ................................ ................................ ............................... 115 Summary and Interpretation of Findings ................................ ................................ .............. 116 Treatment groups compared to no treatment general group ................................ .......... 116 Medical treatment groups compared to med ical no treatment groups .......................... 119 Study Limitations ................................ ................................ ................................ .................. 120 Implications ................................ ................................ ................................ .......................... 121 LIST OF REFERENCES ................................ ................................ ................................ ............. 125 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 12 9

PAGE 10

10 LIST OF TABLES Table page 5 3 Del values indicating goodness of fit of regression models for expenditures. .................. 54 6 1 Demographic characteristics of study samples by treatment group for expenditures/utilization equations ................................ ................................ ..................... 93 6 2 Central tendency of demographics for sample included in expenditures/utilization equations ................................ ................................ ................................ ............................ 95 6 3 Proportions of each type of c hange in total expenditures by treatment group (weighted) ................................ ................................ ................................ .......................... 95 6 4 L ogit regression predicting probability of experiencing each type of change in expenditures (part I of two part model) ................................ ................................ ............. 96 6 5 Ordinary least squares regression (OLS) or generalized linear model (GLM) predicting amount of change in expenditures given positive change (Part II of two part model) ................................ ................................ ................................ ......................... 96 6 6 Bootstrapping prediction of expenditure change after two part model (OLS) .................. 97 6 7 Proportions of each type of c hange in total expenditures by treatment group (weighted) ................................ ................................ ................................ .......................... 97 6 8 Mult inomial logit regression predicting probability of experiencing each type of change in expenditures (part I of three part model) ................................ ........................... 98 6 9 Generalized linear regression (GLM) predicting amount of change in expenditures given negative or positive change (parts II and III of three part model) ........................... 99 6 10 Bootstrapping prediction of expenditure change after three part model (GLM) ............. 100 6 11 Proportion of each type of c hange in medical expenditures by treatment group (weighted) ................................ ................................ ................................ ........................ 100 6 12 Proportion of each type of c hange in medical expenditures by treatment group (weighted) ................................ ................................ ................................ ........................ 100 6 13 Proportion of each type of c hange in drug expenditures by treatment group (weighted) ................................ ................................ ................................ ........................ 101 6 14 Proportion of each type of c hange in drug expenditures by treatment group (weighted) ................................ ................................ ................................ ........................ 101 6 15 Negative binomial regression predicti ng amount of change in utilization with treatment ................................ ................................ ................................ .......................... 101

PAGE 11

11 6 16 Negative binomial regression predicting amount of visits for samples with and without treatment ................................ ................................ ................................ ............. 102 6 17 Proportion of each type of c hange in emergency room visits by treatment group (weighted) ................................ ................................ ................................ ........................ 102 6 18 Multinomial logit regression predicting probability of experiencing each type of change in utilization (part I of three par t model) ................................ ............................. 103 6 19 Negative binomial regression predicting amount of change in utilization given negative or positive change (parts II and III of three part model) ................................ ... 104 6 20 Bootstrapping prediction of expenditure change after three part model (GLM) ............. 105 6 21 Proportion of c hange in inpatient hospital visits by treatment group (weighted) ............ 105 6 22 Proportion of c hange in outpatient hospital visits by treatment group (weighted) .......... 106 6 23 Proportion of c hange in office based provider visits by treatment group (weighted) ..... 106 6 24 Work sample demographic characteristics of study samples by treatment group ........... 107 6 25 Central tendency of d emographic characteristics for work sample ................................ 108 6 26 Neg ative binomial regression predicting amount of change in work absenteeism rates with treatment ................................ ................................ ................................ .................. 109 6 27 L ogit regression predicting probability of experiencing each type of change in expenditures when expenditures added over time (part I of two part model) ................. 109 6 28 Part II of two part model predicting amount of change in expenditures for medical treated and untreated groups when expenditures added over time ................................ .. 109 6 29 Bootstrapping prediction of expenditure change after two part model (OLS) for medical treated and untreated groups when expenditures added over time ..................... 110 6 30 Part I of t hree part model for medical groups compared to no treated medical groups for difference in expenditures over time ................................ ................................ .......... 110 6 31 Parts II and III of three part model for medical treated groups compared to untreated medical groups for difference in expenditures over time ................................ ................ 111 6 32 Bootstrapping prediction of expenditure change after three part model comparing medical treated and untreated groups for difference in expenditures over time .............. 111 6 33 Negative binomial regression of medical treated and untreated groups predicting amount of change in utilization with treatment when utilization added over time .......... 112

PAGE 12

12 6 34 Part I of three part model for medical treated and untreated groups for difference in utilization over time ................................ ................................ ................................ ......... 112 6 35 Parts II and III of three part model for medical treated and untreated grou ps for difference in utilization over time ................................ ................................ .................... 113 6 36 Three part model bootstrapping prediction for medical treated and untreated gro ups for difference in utilization over time ................................ ................................ .............. 113 6 37 Negative binomial regression for medical treated and untreated groups predicting amount of change in work absenteeism rates with treatment ................................ .......... 114

PAGE 13

13 LIST OF FIGURES Figure page 4 1 Relationship between depression and health care expenditures with mental health treatment. ................................ ................................ ................................ ........................... 32 4 2 Relationship between mental health treatment and work absenteeism. ............................. 32 5 1 Overlapping panel design of MEPS. ................................ ................................ .................. 55 5 2 Kernel density plot for total expenditures ................................ ................................ .......... 56 5 3 Standardized normal probability plot for total expenditures ................................ ............. 57 5 4 Quantiles of residuals against the quantiles of normal distribution plot for total expenditures ................................ ................................ ................................ ....................... 58

PAGE 14

14 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 MEDICAL COST OFFSET EFFECT AND ABSENTEEISM IN LONGITUDINAL SAMPLES OF DEPRESSED PULMONARY AND CANCER PATIENTS By Andrea M. Lee August 2009 Chair: Robert G. Frank Major: Psychology An intervention that reduces or prevents usual costs to the health care system is called a medical cost offset or the cost offset effect. Past research shows a medical cost offset effect; however, more recent work does not. The purpose of this study was to examine the effects of mental health treatment on health care ex penditures and absenteeism in sample s of depressed patients with or without comorbid pulmonary conditions or cancer as well as a general depression group This study attempted to provide a current assessment of the medical cost offset effect from mental health services. Add itionally, this study attempted to provide a more comprehensive examination of mental health treatment effects by including workplace absenteeism as an outcome. T he re search questions examined in this study were (1) whether total, medical, or drug expenditures were higher for individuals undergoing mental health treatme nt and whether expenditures reduced over time, (2) whether emergency room, inpatient, outpatient, or o ffice based provider visits were higher for individuals undergoing mental health treatment and whether utilization decreased over time, and (3) whether work absenteeism rates are lower for individuals undergoing mental health treatment. Data were obtaine d from the Medical Expenditure Panel Survey (MEPS), a nationally representative survey of the US non institutionalized, civilian population. Results demonstrate

PAGE 15

15 that total expenditures were greater for each treatment group and medical expenditures were gr eater for only the depression general treatment group. Total emergency room visits, inpatient visits, outpatient visits, and office based provider visits were all greater for the treatment groups than the no treatment group. No treatment group had a sign ificant change in expenditures over time. Mental health treatment only impacted change for inpatient visits for the pulmonary and cancer diagnoses groups. With respect to work absenteeism, mental health treatment was associated with increased absenteeism rates for the general depression group, but the pulmonary and cancer diagnoses groups showed no differece in work absenteeism between mental health treatment and no treatment.

PAGE 16

16 CHAPTER 1 INTRODUCTION Overview The health care system in the United States is in a state of fiscal crisis. Total health care expenditures are estimated to be $2.16 trillion in 2006, and are projecte d to rise to over $4 trillion by 2015 (Borger et al., 2006). There are many reasons for the rise in health care expenditures, one of which is the increase in the number of individuals with chronic diseases. With the aging of the population, chronic diseases have risen dramat ically in recent years (World Health Organization [WHO], 2006). By 2040, almost 160 million people will have a chronic condition (The Robert Wood Johnson Foundation, 1996). In 1995, the cost of medical care for Americans with chronic conditions was $470 billion and by 2040, the cost is projected to be as high as $864 billion. In a report by the Agency for Healthcare Research and Quality (AHRQ), the five most expensive types of chronic conditions in 2000 and 2004 were cardiac conditions, trauma related di sorders, cancer, pulmonary conditions, and mental disorders (Soni, 2007). In addition to costing the health care system, chronic conditions result in economic losses in the workplace (Lamb et al., 2006). Depression is a particularly debilitating condi tion that, when comorbid with medical conditions, increases health care costs and workplace economic losses (Himmelhoch et al., 2004; Druss, Rosenheck, & Sledge, 2000). A question then becomes whether treating depression can help reduce the financial burd en of the condition. The pr esent study examine d whether treating depression could result in reductions in health care expenditures and negative work outcomes. Purpose of the Pr esent Study The purpose of the present study wa s to examine the effects of me ntal health treatment on health care expenditures and absenteeism in a sample of depressed patients with or without

PAGE 17

17 comorbid pulmonary conditions or canc er. This study attempted to provide a current assessment of the medical cost offset effect from mental health services and to provide solutions to limitations apparent in previous studies. Additionally, this study attempted to provide a more comprehensive examination of mental health treatment effects by including workplace absenteeism as an outcome.

PAGE 18

18 CHAPTER 2 LITERATURE REVIEW Cost of Depression Estimates of the lifetime prevalence of depression from the National Comorbidity Survey of the US population aged 15 54 is reported to be 15 percent for major depression and 10 percent for subthreshold depression (i.e., 2 4 depression symptoms present) (Kessler et al., 1997). The National Epidemiologic Survey of Alcoholism and Related Conditions (NESARC), which collected data in 2001 and 2002 on the civilian, noninstitutionalized US population aged 18 years and older, reported that the lifetime and 12 month estimates of major depressive disorder were 13 percent and 5 percent, respectively (Hasin et al., 2005). A World Bank study that estimated current and projected patterns of mortality and disability from disease and injury for all regions of the world reported that depression ranked as the fourth highest disabling condition in 1990 and it was projected that depression will rank second by 2020 (Murray & Lopez, 1996). Depression has an impact on heal th care expenditures. It is estimated that health care costs tend to be approximately 50 percent higher for those with depression compared to those without depression (Himmelhoch et al., 2004). One study that examined the accounting records of a large he alth maintenance organization (HMO) revealed that patients diagnosed with depression had higher annual health care costs and higher costs for all categories of care (e.g., primary care, medical specialty, medical inpatient, pharmacy, and laboratory) than p atients without depression (Simon, Von Korff, & Barlow, 1995). In a study of 46,000 employed persons, in which factors predicting increases in medical costs were examined, depression accounted for the largest increase in medical costs (Goetzel et al., 199 8). Not only does depression increase medical utilization, it is associated with work impairment. Major depression was found to be only one of seven conditions related to reduced

PAGE 19

19 performance at work (Wang et al., 2004). In an analysis of the economic c osts of depression that included direct medical costs, lost wages from suicide, and workplace productivity, it was estimated that depression costs totaled $83.1 billion in 2000. Workplace costs accounted for 62 percent of this total (Greenberg et al., 199 8). Studies demonstrated that depression reduces task focus and productivity (Wang et al., 2004) and employees with depression had greater job performance deficits than employees with a chronic medical condition (Adler et al., 2006). Not only does depre ssion impact performance on the job, depression can result in missed work days, or absenteeism. An analysis of two national surveys revealed that workers who were depressed had between 1.5 and 3.2 more days absent in a month than workers without depressio n (Kessler et al., 1999). The salary equivalent disability costs of these absences ranged from $182 to $395 for each depressed employee. Workers with depression have demonstrated more absenteeism than employees with chronic medical conditions (Lerner et al., 2004). However, the combination of depression and chronic medical conditions has an even greater impact on work impairment. It is estimated that employees with depression and comorbid conditions cost 1.7 times more for employers than those with only depression or a medical condition (Druss, Rosenheck, & Sledge, 2000). The Medical Cost Offset Effect An intervention that reduces or prevents usual costs to the health care system is termed a medical cost offset effect or cost offset effect. Numerous s tudies have attempted to ascertain a medical cost offset effect of mental health care in patients with psychological conditions. One of the first offset studies was conducted by Follette and Cummings (1968). The medical records of 152 randomly selected a dults who sought psychological services were examined. Data on their health services utilization were collected one year prior to the beginning of psychological treatment, as well as five years following treatment. Comparing the data to a group matched f or

PAGE 20

20 age, sex, socioeconomic status, and medical utilization rates who had not received psychological treatment, it was found that this comparison group had higher health care utilization rates over time, in addition to a reduction in health care utilization for the group receiving psychological treatment. Following the Follette and Cummings (1968) study, a series of cost offset studies were conducted. In 1984, Mumford and colleagues (1984) published a study that described two analyses of the medical cost offset effect. One analysis was conducted on Blue Cross Blue Shield Federal Employee Plan claims from 1974 to 1978, and the other analysis was conducted on 58 published studies. Both the Blue Cross Blue Shield analysis and the literature review showed th at the medical cost offset effect was more pronounced for older persons and the largest cost offsets were from a reduction in inpatient days. Of the 58 studies reviewed in the second analysis, 85 percent of the studies found a cost offset effect. In ano ther meta analysis of 91 studies from 1967 to 1997, 90 percent of the studies reported a reduction in medical utilization following mental health interventions (Chiles, Lambert, & Hatch, 1999). Twenty eight articles reported dollar savings and 31 percent of these studies reported savings even after taking into account the cost of mental health treatment. Overall, a savings of about 20 to 30 percent was reported across the articles. The effect was most evident for behavioral medicine and psychoeducational interventions. Hunsley (2003) examined this meta analysis and reported that it would take 2,694 studies averaging null effects to conclude that results were due to sampling bias. Despite the evidence supporting the cost offset effect, several studies pro vide evidence against the effect. The Medical Outcomes Study involved 22,000 outpatients who were screened for several chronic conditions, including depression, and these patients were followed over the

PAGE 21

21 course of four y ears (Wells et al., 1996). The stud y examined the health and cost outcomes for depressed individuals who received appropriate mental health treatment, either counseling or medication, according to clinical practice guidelines. The study produced no evidence of reduced inpatient or outpatie nt services. The researchers observed that effective mental health care that improves patient functioning tends to be more expensive than no mental health treatment or inadequate mental health treatment. The Fort Bragg Evaluation Project involved data col lection of children and their families over seven occasions to evaluate the effectiveness of comprehensive mental health services to children and adolescents (Bickman, 1996). The Fort Bragg Evaluation Project was designed to provide evidence for a continu um model of care and improve the quality of mental health care for children and adolescents in a comprehensive mental health system. The Fort Bragg study offered outpatient therapy, day treatment, in home counseling, therapeutic foster homes, specialized group homes, 24 hour crisis management services, and acute hospitalization to those in the intervention group. The comparison group consisted of families who were responsible for coordinating their own care and did not have access to Fort Bragg services. The study findings revealed that mental health expenditures were much higher for children who received comprehensive care at Fort Bragg and this rise in cost was not offset by cost savings elsewhere. analysis have not produced promising results for the medical cost offset effect. In two related studies examining the medical utilization of patients with or without psychological symptoms, it was found that depressed and anxious patients who saw a mental health provider had significantly more medical visits, emergency room visits, and medical outpatient visits than patients with depression or anxiety who had not seen mental health providers (Carbone et al., 2000). There were no significant differences in

PAGE 22

22 medical costs between patients seeing mental health providers and those who had not. However, both studies did not control for illness severity or comorbid medical or psychiatric conditions. The patients in both studies had a relatively young median age and one of the two studies had a small sample size. These factors may have made medical cost offset effects more difficult to demonstrate. In a two year longitudinal study comparing adults who had major depression who had remitted, improved but not rem itted, or remained depressed, there were no significant differences in total health services cost among the groups in year one (Simon et al., 2000). However, cost savings for patients with improved outcomes in year two showed a reduction that was marginal ly significant, suggesting that the medical cost offset effect may become apparent over longer periods. The pattern of cost differences for each group was similar for the categories of cost examined in the study (specialty mental health care, outpatient v isits, and prescriptions). The study sample was derived from HMO clinics, which has implications for cost offset effects. Because managed care restrictions reduce length of treatment and introduces a ceiling on the amount of money spent on medical care, it becomes more difficult to demonstrate the medical cost offset effect because there is less money to be saved (Otto, 1999). Another study of patients receiving treatment under managed care demonstrated that patients who received mental health treatment h ad the highest proportion of medical services for a mental disorder and a greater proportion of pharmacy claims for all medications (Azocar et al., 2003). There were no significant differences in medical expenditures before, during, or after mental health treatment in this sample. In summary, past meta analyse s demonstrate that the medical cost offset effect was evident with greatest cost savings from inpatient days. Other studies published after the meta analyses

PAGE 23

23 reveal no evidence of a medical cost of fset effect in the form of either no difference between treated and untreated groups or higher expenditures for treated groups. Depression Treatment and Absenteeism From a financial standpoint, it is in the best interest of both employees and employers for employee absenteeism rates (i.e., number of days of missed work) to be as low as possible. Given the demonstrated negative impact of depression on absenteeism, employers would likely want to know whether absenteeism could be reduced when depressed employ ees are treated for their depression. In a study examining factors that predict absenteeism in depressed patients, it was found that rates of absenteeism were indeed greater for untreated depressed workers than for treated workers (Souetre, Lozet, & Cimar osti, 1996). One of the first studies examining whether reduced absenteeism as a result of depression treatment offset the cost of the treatment, interviewed workers with depression over the course of a year and examined provider and insurance records (Zhang et al., 1999). The researchers calculated lost earnings from self reported missed work days and hourly wage rates. They determined health care costs from charges recorded on billing and insurance records. The study found that the effect of depres sion treatment on net economic cost was non significant. That is, the cost of depression treatment was not an additional economic burden to employers or employees. The cost of treatment was fully offset by the savings from reduced absenteeism. More rec ent studies have examined the benefits of depression treatment on absenteeism using randomized controlled trials. One such study involved 12 community primary care practices, in which patients were randomized to enhanced or usual care (Rost, Smith, & Dick inson, 2004). The physicians and care managers in the enhanced care condition were trained in guideline concordant pharmacotherapy or psychotherapy. Enhanced care involved psychoeducation, homework assignments, regular follow up, adherence, and adjustmen t of

PAGE 24

24 treatment if symptoms were not improving. Participants in the usual care condition did not receive ongoing intervention from the enhanced care treatment team, but were not prevented from seeking care on their own. Over the course of two years, clini cians in the enhanced care condition provided depression management to patients. Absenteeism rates were evaluated at baseline, 6, 12, 18, and 24 months. Results showed that patients in the enhanced care condition had 22.8 percent lower absenteeism rates (or 10.6 days) than usual care patients over two years. It was estimated that the reduction in absenteeism from the intervention provided an annual economic benefit of $619 per full time employee. Another randomized controlled trial involved a national sa mple of workers (Lo Sasso, Rost, & Beck, 2006). Participants were randomized to enhanced or usual care conditions, similar to the study design by Rost and colleagues (2004). It was demonstrated that when costs of intervention and treatment were taken int o account, the economic net benefit to employers was $30 per worker in the first year and the net savings increased to $257 per worker in the second year of the study. This suggests that employers investing in treatment for their depressed workers may exp erience increased benefits over time. The effects of mental health treatment for work outcomes ar e not unanimously positive. A 1.5 year randomized controlled trial examined the effects of a mental health intervention on sick leave duration, and mental and physical health status compared to usual care at three, six, and 18 months (Brouwers et al., 2006). The mental health intervention consisted of five individual 50 minute sessions over 10 weeks and was based on a pro blem solving approach to depression treatment. The treatment was administered by a trained social worker using a treatment manual in which three phases were described. The first stage involved acknowledging the problem, the second stage involved making a list of problems and developing problem solving strategies, and

PAGE 25

25 the final stage focused on implementing the strategi es identified in step two. The study found that there were no differences in any of the outcomes between the experimental group and the us ual care group. Although participants in the intervention group reported high satisfaction with treatment, the results suggest that the problem solving approach used in the study was not effective for improving depression outcomes. Taken together, the aforementioned studies that utilized enhanced (guideline based) care with medication and/or psychotherapy were shown to reduce absenteeism. The study that did not show a difference used a different approach and it subsequently did not show an effect on absenteeism or depression symptom improvement. Thus, a lthough not reported by these studies, the reduction in absenteeism in the successful studies were likely a result of an improvement in Further studies will need to be done to add to the small body of existing literature.

PAGE 26

26 CHAPTER 3 STUDY SIGNIFICANCE A ND RESEARCH QUESTION S Significance of this study There are several limitations in the research literature on the medical cost offset effect First, when studies compare the costs of treated and untreated patients, there may be a selection bias in which samples are not comparable (Sturm, 2001). That is, patients who received treatment may have different characteristics than patients who did not receive treatment. If there is limi ted patient information in the data, the selection bias is particularly pronounced. This is particularly problematic with administrative datasets which are computerized data collected for administrative purposes, such as data f r om insurance companies I n this study the use of a la rge comprehensive dataset allowed for greater control of these potentially confounding variables, such as health status and comorbid medical conditions. Second, cost offsets have traditionally been referred to as a general ph enomenon that applies to all medical populations. Past medical c ost offset research has not teased apart which medical populations benefit from psychological interventions with respect to reducing medical cost When researchers aggregate diverse populati ons in medical cost offset research, real cost offset effects may be masked. That is, certain vulnerable populations that are high utilizers of medical services may overshadow the medical cost offset effects of other populations (Simon & Katzelni ck, 1997) Thus, cost offset research must begin to focus on specific groups of patients because these patients may demonstrate a particular pattern of medical utilization based on common patient needs of the group. This study wa s a preliminary effort to identify specific cost offset effects in particular populations. In addition to comparing all depressed persons who are treated and untreated, addi tional populations of interest we re pulmon ary and cancer patients who have comorbid depression. These populations w ere chosen because they were among the top

PAGE 27

27 five most expensive conditions in recent years (Soni, 2007). This study utilize d a dataset drawn from a nationally representative sample of the US noninstitutionalized, civilian population. This a llow ed for greater generalizability of the study findings. Third, rapid changes in healthcare financing and spending patterns necessitate frequent review of cost offset effects reflecting current pricing in pharmacological and medical treatments (Hunsley, 2003 ). Very few studies of the medical cost offset effect have been published after analysis. This study update s the literature by using data from 2000 to 200 5 ( converting cost to 2005 inflation rates), which is more indicative of current economic trends. This study addressed the aforementioned limitations of the previous medical cost offset research. In addition, it offer ed insights into absenteeism. Culminating data suggests that treating workers with depression can reduce abs enteeism rates. Few studies, however, have examined the effects of depression treatment in workers with comorbid chronic medical conditions. This population is important because the combination of depression with other medical illnesses tend to be more c ostly to employers than depression alone (Druss, Rosenheck, & Sledge, 2000). Including absenteeism as an outcome variable for the present medical cost offset study allow ed for a more complete study of the effects of depression treatment on individuals wit h depression and a chronic medical condition. To date, few studies examine both medical cost offset and absenteeism in the same study. Specific Aims and Hypotheses Aim 1 To examine the relationship between mental health treatment and health care expendit ures in sample s of depressed patients with or without comorbid pulmonary conditions or cancer and a general depression group

PAGE 28

28 Hypothesis 1a Patients receiving mental health treatment (psychotherapy and/or medication) will have lower total health care expe nditures, lower medical expenditures and the same or higher prescription drug expenditures than those who have a pulmonary diagnosis, a cancer diagnosis, and general depression without mental health treatment. Hypothesis 1b Patients receiving mental heal th treatment will have a negative change (decrease in expenditures over time) in total health care expenditures and medical expenditures from ti me 1 (the end of the first year of the study) to time 2 ( the end of the second year of the study ) for the pulmonary and cancer d iagnose s group s, while the general depression group will have a positive or no change in expenditures For presc ription drug expenditures, it i s expected that patients receiving mental health treatment w ill have a positive change in expenditures from time 1 to time 2 than patients who had not received mental health treatment. Aim 2 To examine the relationship between mental health treatment and health care utilization in sample s of depressed patients with or without comorbid pulmonary conditions or cancer and a general depression group Hypothesis 2a Pulmonary diagnosis and cancer p atients receiving mental health treatment will have fe wer office based provider visits, outpatient hospital visits, inpatient nights, and emergency room vi sits than patients who do not receiv e mental health treatment wh ile the general depression group will have a positive ( increase ) or no change in utilization

PAGE 29

29 Hypothesis 2b Patients receiving mental health treatment will have a negative change ( decrease ) in office based provider visits, outpatient hospital visits, inpatient nights, and emergency room visits from t ime 1 to time 2 compared to the pulmonary and cancer diagnosis group, wh ile the general depression group will have a positive ( increase ) or no c hange in utilization Aim 3 To examine the relationship between mental health treatment and absenteeism in a sample of depressed patients with or without comorbid pulmonary conditions or cancer and a general depression group Hypothesis 3 Patients receiv ing ment al health treatment will have fe wer total work days missed than those who did not receive mental health treatment for each group (general depression, pulmonary diagnosis group, and the cancer diagnosis group)

PAGE 30

30 CHAPTER 4 CONCEPTUAL MODELS This study was based on two separate conceptual mode ls. The first model related mental health treatment to health care utilization and health care expenditu res and the second model related mental health treatment to absenteeism. Mental Health Treatment, H ealth Care Utilization, and Health Care Expenditures Without mental health treatment, depression is expected to increase health care expenditures through an incr eased need for medical services The conceptual framework used in this study posits that ment al health treatment for depression would result in reduced need for medical services, health care utilization, and in turn, reduced health care expenditures (Figure 4 1 ) Proposed mechanisms for how mental health care reduces health care utilization have been identified as the information and decision support pathway, psychophysiological pathway, behavior change pathway, social support pathway, the undiagnosed psychiatric problem pathway, and the somatization pathway (Friedman et al., 1995). However, exam ining these pathways is beyond the scope of this study This study examine d the relationship between mental health treatment and health care expenditures (total medical and prescription drug expenditures ) in aim one. The relationship between mental h ealth treatment and health care utilization (office based provider visits, outpatient hospital visits, inpatient nights, and emergency room visits) was examined in aim two of the study. Mental Health Treatment and Absenteeism The presence of depression im the number of days missed from work. The conceptual model for mental health treatment and absenteeism proposes that individuals with depression who receive treatment would demonstrate

PAGE 31

31 a reduction i n missed days from work. The reduction in missed work days is a result of an improve ment in mental health status. However, m ental health status also influences seeking mental health treatment (Pincus et al., 2001) creating a bi directional relationship (Figure 4 2 ) For example, anhedonia, or a reduction in interest in previously pleasurable activities, may affect This study examine d the relationship between mental health treatment and number of days missed from work (absenteeism) in aim three.

PAGE 32

32 Figure 4 1 Relationship between depression and health care expenditures with mental health treatment. Figure 4 2 Relationship between mental health treatment and work absenteeism. Depression Mental Health Status Work Absenteeism Menta l Health Treatment Mental Health Treatment Medical Service Utilization Depression Need for Medical Services Health Care Expenditures

PAGE 33

33 CHAPTER 5 DATA AND METHODS Data Source Data was obtained from a public database, the Medical Expenditure Panel Survey (MEPS), which is a nationally representative survey of the US non institutionalized, civilian population, sponsored by the Agency for Healthcare Research and Quality (AHRQ). The MEPS b egan in 1996 and consists of a set of large medical providers and employers. The MEPS is a rich source of information on health services utilization, costs and payments of health services, and health insurance information of respondents. The MEPS is composed of two main components: the Household Component (HC) and the Insurance Component (IC). The IC is a separate survey that provides data on employer based health insurance. Due to the aims of this study only data from the HC will be used. The HC provides data from a sample of families and individuals across the country drawn from a National Healt h Interview Survey. During the household interviews, MEPS collects information from each person in the household on demographic characteristics, health conditions, health status, use of medical care services, charges and payments, access to care, satisfac tion with care, health insurance coverage, income, and employment. Data in the HC is supplemented by such as whether or not respondents used psychotherapy or psycho tropics. Psychological conditions can be identified in MEPS. The HC interview consists of survey questionnaires covering specific topics that are administered by interviewers. Each section of the survey consists of a series of computer

PAGE 34

34 assisted persona l interview (CAPI) computer screens with questions, interviewing instructions, and skip patterns based on specific topics. The MEPS HC has a panel design in which each panel of households is interviewed five times during a two year period. During the sec ond year of the original panel, a new sample is drawn to create a new panel. Thus, two separate panels are interviewed in the same year, which makes for an overlappin g sampling design and increases the number of individuals interviewed each point in time (see Figure 5 1) The five panel design of related to cost of care and employment. This study combine d MEPS data from the years 2000 to 200 5 to assess the effect o f mental health treatment on health care expenditures, health services utilization, and work absenteeism. The MEPS oversamples some groups (e.g., minority groups), while undersampling others, and survey participants may either provide a partial response or no response. These factors reduce the national representation of the data. In order to account for these factors in the sample design, a sample weight for each case has been developed in MEPS to incorporate into the estimation processes. These sample weights were included in the analysis process in order to maintain the national representation of the survey Variables This section describes the variables that were used in the regression analyses of this study Dependent variables, independent variabl es, and covariates are desc ribed. R egression analysis uses variables with numerical values. Many of the variables used in this study are continuous (such as age) but some of the variables are categorical in nature (such as race) Since categorical vari ables do not have numerical values they will be dummy coded in order to be used in the regression analyses.

PAGE 35

35 Dependent Variables Aim 1: To tal health care expenditures, medical expenditures and prescription drug expenditures Health care expenditures were divided into three variables: total health care expenditures, medical expenditures and prescription drug expenditures E xpenditures in MEPS are defined as the sum of direct payments for care provided during the year, including out of pocket payments and payments by private insurance, Medicaid, Medicare, and other sources. N ot included in MEPS total expenditures are p ayments for over the counter drugs and for alternative care services as well as i ndirect payments not rel ated to specific medical even ts Total expenditures were defined as total payments for all health care services included in MEPS (outpatient department visits, office based medical provider visits, prescribed medicines, hospital inpatient visits, emergency room visits, home health, d ental visits, and other medical expenses) and expenditures for psychological services were included in the total. The variable for total expenditures (totexp) will be obtained from the MEPS HC full year consolidated data file from the years 2000 to 200 5 Medical expenditures were defined as total payments for all health care services associated with medical conditions only. Put another way, any medical expense associated with a psychological condition were excluded from the calculation of the medical ex penditure variable. The variable for medical expenditures was constructed by examining the MEPS HC events data files. All payments associated with a psychological condition identified by ICD 9 codes were identified and excluded from the total. The expen ditures of every event that were not associated with a psychological condition were summed to create the variable for medical expenditures (medexp).

PAGE 36

36 Prescription drug expenditures were defined as the sum of all amounts paid out of pocket and by third par ty payers for each prescription drug purchased. This variabl e was obtained from the MEPS HC full year consolidated data files. When combining all f ive years of data (2000 to 200 5 ), total expenditures, medical expenditures and drug expenditures from 2000, 2001, 2002, 2003 and 2004 data were inflated to 200 5 dollars using the consumer price index (CPI) for medical care (BLS, 2000 200 5 ). Aim 2: Health care utilization Health services utilization was defined using four separate variables: total number of hospital outpatient visits (optotv), total hospital inpatient nights at discharge (ipngtd), total number of all emergency room visits (ertot), and total number of office based provider visits (obtotv). Each of these variables were obtained from the MEPS H C full year consolidated data file. Aim 3: Missed days from work For analyses of absenteeism, the sample used were those employed over the course of data collection and only the working age population (18 65) were included in the analyses. The variables indicating employment status (EMPST31, EMPST42, EMPST53) were obtained from the MEPS full year consolidated data file. The variable representing absenteeism were obtained from two variables in the MEPS HC full ye were combined for each person to create the variable for absenteeism ( work ).

PAGE 37

37 Independent Variables Mental health treatment (MHt x) Mental health treatment was defined in this study as psychotherapy and/or antidepressant medications. Respondents who received psychotherapy were identified from two MEPS HC events files: office based medical provider visit file and outpatient visit f ile. In the office based medical provider visit file, the type of care a patient received wa s coded under several variables Respondents will be considered to have undergone psychotherapy if the visit category was psychotropic medications was determined. If particular anti depressant drugs are present under see Table 5 2 ), respondents were coded as taking antidepressant medications for depression. Antidepressants included classes of antidepressants, such as selective serotonin reuptake inhibitors and monoamine oxidase inhibitors, and anti depressant drugs that are prescribed for depression. This variable ( MHtx ) will be used in all five hypotheses. Depression Groups (Dep) The sample of depressed individuals and those with pulmonary conditions or cancer were identified using the MEPS H C medical conditions data files. The medical conditions file codes each self reported medical condition the individual experiences during the year. In MEPS, m edical conditions are coded using the International Classification of Diseases, Ninth Revision ( ICD 9). In order to preserve respondent confidentiality, the ICD 9 codes are collapsed from fully specified codes to 3 digit code categories. Less than 10 percent of codes are collapsed further by combining two or more three digit codes. The ICD 9 codes are also aggregated into clinically meaningful categories called classification codes (CC) that describe groupings of

PAGE 38

38 similar conditions. This is done with the Clinical Classification Software and the groupings are based on the clinical significance of c ategories, accurate reporting from respondents, and the frequency of the reported condition. For this study depression was identified using ICD 9 code 311. Although ICD 9 code 296 corresponds to depression, it also includes individuals with bipolar dis order. Past research in which ICD 9 codes 296 and 311 were examined, over 90 percent of respondents had a code of number of patients with ICD 9 code 311 sugges ts that respondents are likely self reporting depression (as opposed to major depression), which then received a code of 311 instead of 296. Thus, ICD 9 code 311 was used to identify respondents with depression and ICD 9 code 296 will be excluded. The re sultant variable Dep and MHtx was combined to identify individuals who have been treated for depression (Deptx), and those who have not been treated for depression (Depntx). Pulmonary condition (Pulm) & Cancer (Canc) Using the methodology of past research identifying spending and service use trends for various medical conditions in MEPS, pulmonary conditions were identified from the MEPS HC medical conditions file using CC 127 134 and cancer was identified using CC 11 45 (Soni, 2007) ( see Table 5 1 ). Once all of the conditions we re identified, pulmonary and cancer patients were coded as treated or not treated (pulmtx, pulmntx, canctx, cancntx). For all five hypotheses, treated and untreated samples were combined and dummy coded in order to allow for ease of comparison across conditions. T he untreated group was the comparison group, and Deptx, Pulmtx, and Canctx were the variables to be examined in each hypothesis.

PAGE 39

39 Covariates Because some populations are at higher risk for poor health outcomes than others and thus, higher health care expenditures, it will be important to control for these differences to compare health outcomes among different patient populations (Iezzoni, 2003). Furthermore, data on the working age population reveal key characterist ics that would need to be controlled for in analyses predicting absenteeism (Haveman & Wolfe, 2000). The following variables that are described were included in the regression models (described and depicted in a later section) to control for differences a mong the study participants. This help ed to reduce selection biases in the study design. Age (Age) O lder persons generally have worse clinical outcomes than younger persons (Iezzoni, 2003); it is therefore important to control for age. This variable was obtained from the MEPS HC ful l year consolidated files. T at the end of the second year of data collection determined age for this study In examining the hypotheses for aims one and two, individuals of all ages were included. Because the hypotheses for aim three addressed work with absenteeism as the dependent variable, only the working age individuals (18 to 65) were included in the analyses, since the outcome variable relates to work. Sex (Sex) Sex is an important contro l variable because men and women face different risks for certain diseases. Among men and women 65 years of age and older, men have higher death rates than women for several chronic conditions (Anderson, 2002). Sex also predicts posthospitalization mortal ity (Keeler et al., 1990). This variable was obtained from the HC full year consolidated data file and will be dummy coded (0 = male, 1 = female). This variable was included in all five hypotheses.

PAGE 40

40 Race (Racex) Racial disparities in health care outcomes was also be taken into account in this study because differences in disease prevalence and mortality exist among the races (Iezzoni, 2003). Furthermore, among the working population, the rate of mental and physical disability for African Americans is alm ost twice than that for whites and Hispanics (Mashaw et al., 1996). The race variable will be obtained from the MEPS HC full year consolidated data file and race will be dummy coded into three separate variables (black, asian, and white ) with other as the comparison variable. These variables will be included in all five hypotheses. Hispanic (Hispan) This variable was included in the analysis because there are differences in disability rates for Hispanics compared to non Hispanics (Mashaw et al., 1996). This variable was obtained from the MEPS HC full year consolidated data file and the variable was dummy coded (0 = non Hispanic, 1 = Hispanic). This variable was included in all five hypotheses. Years of education (Educyear) Because of socioeconomic disparities in health status and outcomes, there was a control for income and education factors (Braveman & Tarimo, 2002). More education is associated with fewer disabilities (Wolfe & Haveman, 1990). This variable was obtained f rom the MEPS HC full year consolidated files and represent ed number of years of education. This variable was included in the analysis of all five hypotheses. Total individual income (Ttlp) As aforementioned, socioeconomic factors was captured with variables representing education and income and was included in the analysis because of socioeconomic health

PAGE 41

41 consolidated data file. This var iable was included in the analyses for aims one and two. This variable represent ed the working age population, income for the family ma y be a stronger consideration for returning to work than individual income (Haveman & Wolfe, 2000). Thus, this variable was included in the analyses for aim three that has absenteeism as the outcome variable. This variable was obtained from the HC full y ear consolidated data file. Perceived physical health (Rthlth) and perceived mental health (Mnhlth) Self perceived mental health status and self perceived physical health status are variables defined in MEPS and these are considered risk factors in health care outcomes (Iezzoni, 2003). Respondents were asked to report their self perceived mental and physical health status on a were in the analyses for all five hypothese s and w ere obtained from the HC full year consolidated data files. The need for help with activities of daily living (Adlhlp) and instrumental activities of daily living (Iadlhp) As aforementioned, physical health status are important risk factors in healt h care outcomes. Activities of daily living (ADL) refer to typical everyday activities, including self care activities necessary for fundamental functioning. In contrast, instrumental activities of daily living (IADL) are those activities that are not f undamental to functioning, but allow one to maintain independence in his or her community (e.g., shopping, meal preparation) These variables were in the analyses for all five hypotheses and were obtained from the HC full year consolidated data files. Number of medical comorbidities (Comorb)

PAGE 42

42 Medical comorbidites were taken into account because patients with comorbidities tend to have higher risks of death, complications, functional impairments, and higher health service use (Iezzoni, 2003). The variab le for comorbidities were determined from the MEPS HC medical conditions file by tallying the number of different ICD 9 codes for each individual. This variable was in the analyses of all five hypotheses. Insurance status (Insstat) Health insurance sta tus was an additional variable that was created in order to control for health service utilization. This was a control variable because it is expected that individuals insured throughout the year would have higher expenditures than those intermittently in sured and uninsured throughout the year. The MEPS HC full year consolidated file will be used to identify patients who were insured (i.e., insured all months of the year), intermittently insured (i.e., at least one month of the year without health insuran ce), and uninsured (i.e., no health insurance for all months of the year). This variable was included in the analyses of aims one and two. Statistical Methods The distribution of health care expenditures is typically left censored and skewed with a larg e proportion of zeroes. Because of these properties, estimation using untransformed ordinary least squares (OLS) produces inefficient results. One common method for handling such data is to run a two part model in which the data are estimated with two me thods. Part one invo l v es the estimation of the probability of zero versus non zero values, and part two involves a separate estimation of the nonzero (positive) values For part one, t he statistical procedure typically employed is the logistic regression for a binary outcome ( non zero/zero or use/no use ) and the second part generally involves the ordinary least squares ( OLS ) regression in which expenditures are log transformed (to accommodate skewness) for the nonzero values ( Duan et al., 1984 ) In this study, the hypotheses that added times 1 and 2 expenditures together as the outcome measure

PAGE 43

43 ( hypotheses 1a and 2a ) used the two part m odel as there were a large proportion of zeroes and a log transformed dependent variable (expenditures) successfully corr ected for skewed data for total expenditures For medical and drug expenditures, a different method was used for part II and will be described in a later section. The two part model did not fit the data in which expenditures from time 2 was subtracted from time 1 (hypotheses 1b and 2b) because the resultant dependent variables had both positive and negative values (for example, when time 2 expenditures was lower than time 1 for an individual this resulted in a dependant variable with negati ve value s) Negative values are unable to be transformed to meet OLS assumptions. Furthermore, gamma models are only for non negative values, which is problematic for the negative values of the dependent variable. An additional theoretical consideration is that the effect of mental health treatment on expenditures would likely be different among those who had increased expenditures than those who had a decrease in expenditures following mental health treatment. Running a single regression combining the two types of change would produce biased results. Thus, a three part model was used to overcome these aforementioned issues. Aim 1: Mental Health Treatment and Health Care Expenditures Hypothesis 1a To test the hypothesis that patients who received me ntal health treatment will have lower total health care expenditures and lower medical expenditures than patients who did not receive treatment, and those with mental health treatment will have higher or no difference in prescription drug expenditures comp ared to those without mental health treatment, a two part model was used. In part one, a logistic regression modeled whether mental health treatment significantly predicted a change from zero expenditures to positive expenditures. This was modeled using:

PAGE 44

44 (Logistic)Y 1+2 0 1 Deptx i 2 Pulmtx i 3 Canctx i 4 Bothtx i 5 Age i 6 Sex i + 7 White i 8 Black i 9 Asian i 10 Other i 11 Hispan i 12 Educyear i 13 Ttlp i 14 Rthlth i + 15 Mnhlth i 16 Comorb i 17 Insstat i 18 Adlhlp i 19 Iadlp i i In part two, an Ordinary Least Squares (OLS) model was used for total expenditures The assumptions of OLS must be met in order to obtain unbiased, consistent, and efficient OLS estimators. The following statistical procedures are accepted econometrics m ethods to check for OLS assumptions (Gujarati, 2003) : The assumption of linearity was examined with the the assumption of normality will be tested with examining plots of the data ( the kernel density plot the standardized normal pro bability plot and the quantiles of residuals against the quantiles of normal distribution plot ), and the Park and Glejser tests test ed for heteroskedasticity. Upon testing the assumptions, it was discovered that t he dependent variable t ransformed into the natural log resulted in the normality and linear assumptions being met for the total expenditure variable For total expenditures, after log transformation, the assumption of li nearity was met according to The Kernel density plot ( Figure 5 2 Figure 5 3 Figure 5 4 ) show that the residuals were close to being normally distributed. The Park and Glejser tests deteced heteroskedasticity (p < 0.01 for both). Because of heteroskedasticity, subgroup smearing estimator s for values of expenditures was employed. The model to be estimated for part two of the two part model estimating mental health treatment effects on total expenditures was as follows and the coefficients of interest were 1 2 and 3 :

PAGE 45

45 Log(Y 1+2 ) 0 1 Deptx i 2 Pulm tx i 3 Canc tx i + 4 Both tx i 5 Age i 6 Sex i + 7 White i 8 Black i + 9 Asian i + 10 Other i + 11 Hispan i + 12 Educyear i + 13 Ttlp i 14 Rthlth i + 15 Mnhlth i 16 Comorb i 17 Insstat i 18 Adlhlp i 19 I adlp i i For medical expenditures and prescription drug expenditures, the OLS assumption of linearity was not achieved (log, square root, cube root, and fourth root transformations did not correct for violations in assumptions). As a result, a generalized linear model (GLM) was used. GLM is a generalization of the OLS regression that does not have distributional assumptions. To specifies the distribution of the mean var iance relationship, whereas the link function specifies the covariate mean relationship. An advantage of GLM is that it can give direct predictions of expenditures without the need to make predictions with smearing estimation ( Manning, Basu, and Mullahy, 2005 ). The generalized Gamma model (GGM) with a log link was shown to fit the data for both medical expenditures and prescription drug expenditures. That is, a modified Park test and a link test showed that the gamma distribution and log link linear rel ationship fit the data best. The model equation used to estimate medical and prescription drug expenditures was as follows and the coefficients of interest were 1 2 and 3 : (Gamma)(Y 1 + Y 2 0 1 Deptx i 2 Pulmtx i 3 Canctx i 4 Bothtx i 5 Age i 6 Sex i 7 White i 8 Black i 9 Asian i 10 Other i 11 Hispan i 12 Educyear i 13 Ttlp i 14 Rthlth i 15 Mnhlth i 16 Comorb i 17 Insstat i 18 Adlhlp i 19 Iadlp i i Hypothesis 1b To examine the hypothesis that patients who received mental health treatment will have lower total health care expenditures and lower medical expenditures in time two compared to

PAGE 46

46 their total and medical expenditures in time one (Y 2 Y 1 ) three part models were used because expenditures had negative values, along with zero and positive values. In part one, due to more than two categories (negative, zero, positive) a decision was made among using multinomial logit, multinomial probit, and ordered logit. In order to aid in the decision, the goodness of fit Del measure developed by Hildebrand and colleagues ( Hildebrand, Liang, and Rosenthal, 1977 ) was calculated. For each expenditure variable, multinomial logit had a marginally higher Del value, which indica tes that it likely has a relatively better fit for the models ( see Table 5 3 ). Despite the Del measure favoring multinomial logit, multinomial probit has an advantage of not relying on the assumption that each category or choice are independent; however, its computationally intensive procedure was unrealistic for the timeline of this study The model estimated for each of the expenditures is as follows: (Multinomial Logit) ( Y 2 1 ) 0 1 Deptx i 2 Pulmtx i 3 Canctx i 4 Bothtx i 5 Age i + 6 Sex i 7 Whit e i 8 Black i 9 Asian i 10 Other i 11 Hispan i 12 Educyear i 13 Ttlp i + 14 Rthlth i 15 Mnhlth i 16 Comorb i 17 Insstat i 18 Adlhlp i 19 Iadlp i i For part s two and three, assumptions for OLS were checked and log, square root, cube root, and fourth root transformations did not correct for violations in assumptions. Therefore, the generalized linear model (GLM) was used. For each expenditure variable for parts t wo and three the generalized Gamma model (GGM) with a log link was shown to fit the data well. That is, a modified Park test and a link test showed that the gamma distribution and log link linear relationship fit the data best. For part two (estimation o f change given negative values), the dependent variables were multipled by 1 in order to create positive values and allow for estimation. T he models to be

PAGE 47

47 estimated for each expenditure variable was as follows and the coefficients of interest were 1 2 and 3 : (Gamma) ( Y 2 Y 1 < 0)*( 1) = 0 1 Deptx i 2 Pulmtx i 3 Canctx i 4 Bothtx i 5 Age i 6 Sex i 7 White i 8 Black i 9 Asian i 10 Other i 11 Hispan i 12 Educyear i 13 Ttlp i + 14 Rthlth i 15 Mnhlth i 16 Comorb i 17 Insstat i 18 Adlhlp i 19 Iadlp i i For part three (estimation of change given positive values), the models to be estimated for each expenditure variable was as follows and the coefficients of interest were 1 2 and 3 : (Gamma) ( Y 2 Y 1 > 0) = 0 1 Deptx i 2 Pulmtx i + 3 Canctx i 4 Bothtx i 5 Age i + 6 Sex i 7 White i 8 Black i 9 Asian i 10 Other i 11 Hispan i 12 Educyear i 13 Ttlp i + 14 Rthlth i 15 Mnhlth i 16 Comorb i 17 Insstat i 18 Adlhlp i 19 Iadlp i i Predictions based on the two or three part equations In order to determine if predictions from the two or three part models we re significant overall ( in its aggregate ), the equations were combined to obtain the estimate s of adjusted change in expenditure s per person for each treatment group ( depression general, pulmonary diagnosis, and cancer diagnosis ) For example, the three part model would be depicted as follows: E( Y k | Dep ) = P(D k =0 | Dep ) E( Y k | Dep D k =0) *( 1) + P(D k =2 | Dep ) E( Y k | Dep D k =2) E( Y k |Pulm ) = P(D k =0 | Pulm ) E( Y k | Pulm D k =0) *( 1) + P(D k =2 | Pulm ) E( Y k | Pulm D k =2) E( Y k | Canc ) = P(D k =0 | Canc ) E( Y k | Canc D k =0) *( 1) + P(D k =2 | Canc ) E( Y k | Canc D k =2) where Y k = Y 2 Y 1 The significance of the combined equations were estimated by obtaining confidence intervals (CI) via the procedure of bootstrapping. Bootstrapping is a non parametric method for tests of means (Kennedy 1998) Bootstrapping draws random samples from the data for the

PAGE 48

48 number of times specified (in this study, 1000 iterations were chosen), generating estimations from each sample and each sample is replaced prior to the next drawing. If the 95% CI contains zero then the difference is not significant at the 0 .05 level. Aim 2: Mental Health Treatment and Health Care Utilization Hypothesis 2a To test the hypothesis that patients who have received mental health treatment will have lower total office based provider visits, outpatient hospital visits, inpatient nights, and emergency room visits than patients who have not received mental health treatment, a count data model will be used instead of OLS because count data tends to violate the homoskedasticity assumption, negative predictions can arise, and coefficie nts may not be meaningful if using OLS. One method is to use a Poisson regression which is a nonlinear analysis that assumes that each observed count is drawn at the same rate. A Poisson regression can be problematic because it assumes that the mean is e qual to the variance, it assumes that the probability of the event is independent of how many times the event occurred previously, and it has difficulty modeling data with a large number of zeroes. Due to the limitations of the Poisson regression, the bes t count model is the negative binomial regression because it does not assume that the incidence rate is the same for all individuals and it is able to fit data with a lot of zeroes. The equations for the models are depicted below with the four separate ou tcomes total number of hospital outpatient visits (optotv), total hospital inpatient nights at discharge (ipngtd), total number of all emergency room visits (ertot), and total number of office based provider visits (obtotv). The coefficients of interest will be 1 2 and 4 : (1) f( Y 1+2 ) = + optotv i ) i + 1) i ]) (1 [1/(1+ i )]) optotvi

PAGE 49

49 i = e Hypothesis 2b To te st the hypothesis that patients who received mental health treatment will have lower office based provider visits, outpatient hospital visits, inpatient nights, and emergency room visits in time 2 compared to their office based provider visits, outpatient hospital visits, inpatient nights, and emergency room visits in time 1 (Y 2 Y 1 ), three part models were used because the subtraction of time 1 from time 2 resulted in negative values, along with zero and positive values. Multinomial logits were used for p art one due to better Del values (goodness of fit) For parts two and three, negative binomial regression were used for the reasons stated in hypothesis 2a. For part two, negative values in the dependent variable (utilization) were multiplied by 1 in or der to create positive values for estimation. The coefficients of interest were 1 2 and 3 : (1) f( y i ) = i ) i + 1) i ]) (1 [1/(1+ i )]) yi i = e and y i = [ (time 2 ) (time 1 ) < 0]*( 1) For part three, the negative binomial regression was as follows and the coefficients of interests were 1 2 and 3 : (2) f( y i ) = i ) i + 1) i ]) (1 [1/(1+ i )]) yi i = e and y i = [ (time 2 ) (time 1 ) > 0]

PAGE 50

50 Predictions based on the three equations In order to determine if predictions from the three part models are significant overall (in its aggregate), the equations were combined to obtain the estimate s of adjusted change in utilization per p erson for each treatment group ( depression general, pulmonary diagnosis, and cancer diagnosis ) as was done in hypotheses 1a and 2a. For aim two, this procedure was only done with the three part model (hypothesis 2b) because hypothesis 2a was estimated wi th a single equation. Aim 3: Mental Health Treatment and Absenteeism Hypothesis 3 To test the hypothesis that patients who received mental health treatment will have lower total days missed from work (misstot) than those who did not receive mental health treatment the negative binomial regression will once again be used for the reasons stated in hypothesis 2a. The equation for the model is depicted below and the coefficients of interest will be 1 2 and 3. : (1) f(misstot i ) = + misstot i ) i + 1) i ]) (1 [1/(1+ i )]) misstot i i = e Selection bias and selection correction modeling As depicted in chapter four, the conceptual model for the analyses for aim three illustrates health status is not only affected by treatment, it determines whether or not one seeks mental health treatment in the first place. This introduces a problem with comparing treated versus untreated groups because the two groups may be fundamentally differ ent. This study observes mental health status through the use of the variable

PAGE 51

51 representing self perceived mental health status (mthlth ); however, this undoubtedly does not capture the entirety of mental health status. If mental health status is not fully accounted for, this may produce a spurious correlation between mental health treatment and absenteeism. Furthermore, resultant coeffi cients in the model would not accurately reflect the size of the relationship (coefficients would be biased). In order to correct for the selectivity bias and to attempt to more fully control for mental health status, a regression that accounts for observ able confounders and unobservable confounders by the use of instrumental variables was used (Gujarati, 2003). An instrumental variable estimation relies on identifying one or more variables that affect the independent variable of interest (mental health t reatment), but does not independently affect the dependent variable (absenteeism). That is, the instrumental variable must influence mental health treatment and only affects absenteeism through the treatment. The variable chosen for this is insurance. H aving insurance has been found to relate to treatment seeking (Staniec & Webb, 2007), and insurance is not directly related to absenteeism (Lofland & Frick, 2006).

PAGE 52

52 Table 5 1. Clinical c lassification codes and diagnostic categories Medical c onditio n Classification c ode Clinical c lassification s oftware d iagnosis c ategory Pulmonary conditions 127 Chronic obstructive pulmonary disease and bronchiectasis 128 Asthma 129 Aspiration pneumonitis, food/vomitus 130 Pleurisy, pneumothorax, pulmonary collapse 131 Respiratory failure, insufficiency, arrest (adult) 132 Lung disease due to external agents 133 Other lower respiratory disease 134 Other upper respiratory disease Cancer 11 45 Cancer of head and neck; esophagus; stomach; colon; rectum and anus; liver and intrahepatic bile duct; pancreas; other GI organs, peritoneum; bronchus, lung; other respiratory and intrathoracic; bone and connective tissue; melanomas of skin; other non epit helial cancer of skin; breast; uterus; cervix; ovary; other female genital organs; prostate; testis; other male genital organs; bladder; kidney and renal pelvis; other urinary organs; brain and nervous homa; leukemias; multiple myelom; other and unspecified primary; secondary malignancies; malignant neoplasm without specification of site; neoplasms of unspecified nature or uncertain behavior; maintenance chemotherapy, radiotherapy

PAGE 53

53 Table 5 2. Antidepressant medication n ames. Drug c lass Generic n ame Brand n ame Antidepressant Imipramine Desipramine Amitriptyline Nortriptyline Protriptyline Trimipramine Doxepin Clomipramine Maprotiline Amoxapine Trazodone Fluoxetine Bupropion Sertraline Paroxetine Venlafaxine Nefazodone Fluvoxamine Mirtazapine Citalopram Escitalopram Duloxetine Atomoxetine Phenelzine Tranylcypromine Selegiline Clonozepam Propranolol Atenolol Tofanil Norpramin Elavil Aventyl, Pamelor Vivacil Surmontil Sinequan, Adapin Anafranil Ludiomil Asendin Desyrel Prozac Wellbutrin Zoloft Paxil Effexor Serzone Luvox Remeron Celexa Lexapro Cympbalta Strattera Nardil Parnate Emsam (patch) Klonopin Inderal Tenormin

PAGE 54

54 Table 5 3 Del values indicating goodness of fit of regression models for expenditures. Del Total e xpenditures Multinomial logit 0.0685 Multinomial probit 0.0676 Ordered logit 0.0618 Medical e xpenditures Multinomial logit 0.0770 Multinomial probit 0.0705 Ordered logit 0.0562 Drug e xpenditures Multinomial logit 0.1298 Multinomial probit 0.1252 Ordered logit 0.1094

PAGE 55

55 Figure 5 1. Overlapping panel design of MEPS.

PAGE 56

56 Figure 5 2. Kernel density plot for total expenditures

PAGE 57

57 Figure 5 3. Standardized normal probability plot for total expenditures

PAGE 58

58 Figure 5 4. Quantiles of residuals against the quantiles of normal distribution plot for total expenditures

PAGE 59

59 CHAPTER 6 RESULTS Overview This section describes the overall sample characteristics and the results of chi square tests to determine whether there were differences in individual characteristics among the samples of interest (depression general, pulmonary diagnosis, and cancer diagn osis). Descriptive statistics are followed by the statistical results from each aim. Statistical results are organized by outcome measure: total expenditures, medical expenditures, prescription drug expenditures, emergency room visits, inpatient visits, outpatient visits, office based provider visits, and work absenteeism. The pulmonary and cancer groups are reported both in comparison to an aggregated no treatment general group (Tables 6 1 to 6 26) and to the no treatment pulmonary or cancer groups (Tables 6 27 to 6 37 ). Description of the Sample The sample included 6,028 in dividuals with depression. Within the sample, 4,024 individuals received mental health treatment, while 2,004 participants did not receive mental health treatment. Among those receiving treatment, 1,119 had a comorbid pulmonary diagnosis and 194 individual had a comorbid cancer diagnosis. None of the remaining 2,488 depressed and treated participants had either a cancer or pulmonary comorbidity. The sample consisted of 69.14% females and 30.86% males T he majority of the sample was White (84.52%). The average age of participants was 45.1 3 with 11.59 years of education. The average yearly income was almost $20,000. Table 6 1 presents the percentages of the sample within each category or characteristic, along with chi square values. Table 6 2 provides the averages, standard deviations, and ranges of the sample characteristics with continuous values

PAGE 60

60 For each characteristic, chi square values indicate that there were significa nt differences among the study groups (p<0.000) (Table 6 1) Among the treated groups, individuals most likely to get treated were between the ages of 45 and 64. For the untreated group, the age group most likely to forgo treatment are those between 25 a nd 44. F or f emales treatment seeki ng was highest for depressed females with a comorbid pulmonary condition and they were more likely to seek treatment than forgo treatment. Men were relatively more likely to forgo treatment. With respect to race, White participants in the depression general group were most likely to seek treatment. Among Black participants, the highest proportion of individuals were in the no treatment group. Among Asian participants, treatment seeking was more likely among those with a cancer comorbidity. There was also a relatively high percentage of Hispanic participants in the no treatment group compared to treated Hispanic participants. Among the poor, near poor, and low income participants, the proportion of individuals forgoin g treatment was higher than the treatment groups within the same poverty categories. Middle income and high income participants had the highest percentages of individuals in treatment groups (pulmonary diagnosis and cancer diagnosis, respectively). Insur ed participants were most likely to seek treatment. With regards to self perceived physical and mental health status, those who were rated as having treatment a ment. Furthermore, participants with ADL and IADL needs with a cancer diagnosis were most likely to seek treatment compared to others with similar needs. However, more people without ADL or IADL

PAGE 61

61 needs tend to seek treatment than those with such needs. F inally, participants with 3 comorbid conditions and a pulmonary diagnosis were most likely to seek treatment. In summary, individuals who were most likely to seek treatment were W hite individuals aged 45 64 with high income, and three comorbid conditions. Individuals least likely to seek mental health treatment appear to be children under age 12, males, Asians, those perceived physical and mental health Total Expenditures Two Part Model (Y 1 + Y 2 ) The weighted percentages of participants who had zero vs. positive expenditures for the two part model is represented in Table 6 3. Overall, the percentage of participants who had zero expenditures was 27.31% and the percentage who had positive expenditures was 72.69%. The coefficients, standard errors, and p values of significance tests from the regression are depicted in Table 6 4 (Part I), Table 6 5 presents the coefficients, standard errors, an d p values of significance tests from the regressions in p art II and Table 6 6 presents the bootstrapp ed results Part I A logistic regression was run to predict the probability of having zero or non zero values for each treatment group, relative to the no treatment group. The predicted probabilities of each group to have expenditures greater than zero were 74.74 % for the depression general group 69.05 % for the pulmonary group 66.88 % for the cancer group and 67.54 % for the no treatment group. Holdin g all other variables constant, the probabilities for each treatment group did not differ significant ly from the no treatment group A comparison of the pulmonary treatment group with th e pulmonary no treatment group revealed that the pulmonary treatment group was

PAGE 62

62 less likely to have any expenditures than the no treatment pulmonary group (p=0.001). The cancer treatment group did not differ from the no treatment cancer group (0.060). Part II A log transformed OLS regression was performed on part II of the two part model to determine if there wa s a significant difference in expenditures between treatment and non treatment among those who had positive expenditures during the study period. The m ean predicted expenditures for the depression general, pulmonary, and cancer groups were $ 14,398.59 $ 15,800.25 and $ 18,422.04 respectively. Results demonstrate that each treatment group had significantly higher expenditures than the mean predicted exp enditures of the group that did not have any mental health treatment ($8,897.61) (p =0.000). When the no treatment pulmonary group was compared directly to the treated pulmonary group, the treated group had higher expenditures than the no treatment group ( p=0.000), whereas the cancer treated and untreated groups were not significantly different (p=0.291). Bootstrapping prediction Using each of the two equations from the two part model, the combination of the equations were used to predict the overall amoun t of change in expenditures for each treatment group and to determine whether or not there was a significant overall effect of mental health treatment on expenditures. Bootstrapp ed results show that the overall total expenditures for the treatment groups were significantly higher than the no treatment group and both medical conditions were also higher than the untreated pulmonary or cancer groups Summary T reatment groups were not more likely to have any expenditures than the non treatment group Among those who had expenditures greater than zero, those with mental health treatment

PAGE 63

63 were more likely to experience an increase in expenditures regardless of treatment group. However, the cancer treated group when compared to the untreated cancer group, did not yield significant results. Th e increase in total expenditures held true when aggregating the zero and non zero groups to estimate expenditures (i.e., there was an overall increase in total expenditures for each treatment group). Three Part Model (Y 2 Y 1 ) The weighted percentages of participants who experienced each type of change in total expenditures for the three treatment groups are presented in Table 6 7 Overall, the p e r centages of participants who had a decrease, no change, and increase in total expenditures with mental health treatment were 38.57%, 27.31%, and 34.12%, respectively. Table 6 8 s hows the results from Part I Table 6 9 depicts the results from Parts II and III and Table 6 10 displays the bootstrapped results Part I A mul t i nomial logit regression was used to predict the probability of having a negative change, no change, or positive change in expenditures for each treatment group, relative to the no treatment group. The predicted probabilties for the depression general, pulmonary cancer, and no treatment groups to experience a decrease in expenditures were 37.86 %, 33.80%, 28.38 %, and 43.63%, respectively. The predicted probabilties for the depression general, pulmonary, cancer, and no treatment groups to experience an increase in expenditures were 36.88 %, 35.25 %, 38.55% and 29.38 %, respectively. Holding all other variables constant, compared to the no treatment group, the depression general group was 1.39 times as likely to experience a positive change in expenditures (p=0.000) In contrast, depressed treated participants with a pulmonary diagnosis were 69.6% more likely to experience a negative change (decrease) in expenditures relative to the no treatment general group (p=0.002), as well as in comparison to the no treatment

PAGE 64

64 c ancer group (p=0.000). D epressed treated participants with a cancer diagnosis were 61.3% more likely to experience a negative change in expenditures when compared to the no treatment general group, but not the no treatment cancer group (p=0.072) Part II A GGM regression with a log link function was used to determine the amount of negative change in expenditures given a negative change for each treatment group. The predicted mean expenditures for the depression general, pulmonary, cancer, and no treatment groups were $ 2264.89 $ 1965.70 $ 2416.4 3, and $ 1906.14, respectively Each treatment group (depression general, pulmonary diagnosis, cancer diagnosis) experienced a significant increase in the amount of negative change in expenditures compared to the no treatment group (p=0.000, p=0.026, and p=0.000, respectively) and the cancer group experienced the largest change. However, when the untreated pulmonary or cancer groups were isolated to compare to the pulmonary or cancer treated groups, there were no significant decreases in expenditures (p=0.520 and p=0.526). Part III A GGM regression with a log link function was used to determine the amount of positive change in expenditures given a positive change for each treatment group. The predicted mean expend itures for the depression general, pulmonary, cancer, and no treatment groups were $ 1840.681 $ 1616.279 $ 2270.57 and $ 1115.305 respectively. Controlling for all other variables, each treatment group experienced a significant increase in the amount of p ositive change in expenditures relative to the no treatment group (p=.010, p=0.032, and p=0.013, respectively) with the cancer group experiencing the largest change in expenditures. When the pulmonary and cancer no treated groups were isolated, only the p ulmonary treatement group had a significant increase in expenditures relative to the no treatment pulmonary group (p=0.035).

PAGE 65

65 Bootstrapping prediction Using each of the three equations from the three part model, the combination of the equations were used to predict the overall amount of change in expenditures for each treatment group and to determine whether or not there was a significant overall effect of mental health treatment on expenditure change Bootstrapp ed results show that for each treatment group, there was not a significant change in expenditures when the negative change, no change, and positive change groups were aggregated and compared to the no treatment general group Both the pulmonary and cancer g roups had an overall significant increase in expenditures overall when compared with the no treatment pulmonary or cancer groups. Summary Individuals in the depression general group were more likely to experience an i ncrease in expenditures with treatment, whereas the pulmonary and cancer diagnoses groups were more likely to experience a decrease in total expenditures with treatment T his held true when the pulmonary no treatment group was compared to the pulmonary treatment group When those with negativ e change and positive change were examined separately, each negative change group had a greater decrease in expenditures with treatment and each positive change group had a greater increase in expenditures with treatment. When isolating the treatment grou ps into medical condition, only the pulmonary treatment group with positive expenditures had a significant increase in expenditures. However, when combining the negative, no change, and positive change groups together, there was no overall significant dif ferences in expenditures when comparing to the no treatment general group In t he pulmonary and cancer conditions, however, there was a significant increase in total expenditures when compared with the corresponding no treatment medical condition group

PAGE 66

66 M edical Expenditures Two Part Model (Y 1 + Y 2 ) The weighted percentages of participants who had zero vs. positive expenditures for the two part model is represented in Table 6 1 1 Overall, the percentage of participants who had zero expenditures was 13.12% and the percentage who had positive expenditures was 86.88%. The coefficients, standard errors, and p values of significance tests from the regression are depicted in Table 6 4 for Part I Table 6 5 presents the coefficients, standard errors, and p values of significance tests from the regressions in Parts II, and Table 6 6 shows the results from the boostrapp ed predictions Part I A logistic regression was run to predict the probability of having zero or non zero values for each treatment group, relative to the no treatment group. The predicted probabilities of each group to have expenditures greater than zero were 88.43 %, 86.91 %, 83.37 %, and 73.61 % for the depression general, pulmonary, cancer, and no treatm ent groups, respectively. Holding all other variables constant, the treated depression general group was significantly more likely to experience medical expenditures greater than zero than the no treatment group (p=0.000). The treated group with a pulmon ary diagnosis and t he treated group with a cancer diagnosis did not show a significant difference in the probability to have any expenditu res compared with the untreated general group ( p= 0. 185 and p= 0.495, respectively ) ; however, comparing the pulmonary no treatment group to the pulmonary group revealed that the treated group was less likely to experience expenditures greater than zero (p=0.016) Part II A GGM regression with a log link function was performed on part II of the two pa rt model to determine if there wa s a significant difference in expenditures between treatment and non

PAGE 67

67 treatment among those who had positive expenditures during the study period. The predicted mean expenditures for each group were $9,344.22, $10,323.20, $13,312.50, and $6,291. 30 for the depression general, pulmonary, cancer, and no treatment groups, respectively. Results demonstrate that for those who had some expenditures during the study period, each treatment group had significantly higher medical expenditures than the grou p who did not have any mental health treatment (p=0.000) and this still held true with the pulmonary treated versus pulmonary untreated group (p=0.000), but not for treated versus untreated cancer groups (p=0.291) Bootstrapping prediction Using each of the two equations from the two part model, the combination of the equations were used to predict the overall amount of change in expenditures for each treatment group and to determine whether or not there was a significant overall effect of m ental health treatment on expenditures. Bootstrapp ed results show that for the depression general treatment group, there was an overall significant positive change in expenditures with treatment when the zero and positive value groups were aggregated. Fo r the treatment group with a pulmonary diagnosis or cancer diagnosis, there was an overall increase in expenditures when these groups were compared to the untreated pulmonary and cancer groups. Summary The depression general group w as more likely to have m edical expenditures from mental health treatment T he pulmonary treatment g roup was less likely to have expenditures compared to the no treatment pulmonary group Amongst those with a non zero (positive) expenditures, each treatment group was more likely to have an increase in medical expenditures except the cancer treatment group compared to the no treatment cancer group Each group had an overall significant increase in medical expenditures from mental health treatment.

PAGE 68

68 Three Part Model (Y 2 Y 1 ) The weighted percentages of participants who experienced each type of change in total expenditures for the three treatment groups are presented in Table 6 1 2 Overall, the p e r centages of participants who had a decrease, no change, and increase in medical exp enditures with mental health treatment were 45.50 %, 6.80 %, and 47.69 %, respectively. Table 6 8 shows the results from Part I Table 6 9 depicts the results from Parts II and III, and Table 6 10 displays the bootstrapped results. Part I A mulitnomial logit regression was used to predict the probability of having a negative change, no change, or positive change in expenditures for the depression general group and the pulmonary diagnosis group, relative to the no treatment group. Because the cancer diag nosis group had no participants with zero (no change), they were excluded from part I. The predicted probabilties for the depression general, pulmonary, and no treatment groups to experience a decrease in expenditures were 40.86%, 44.57%, and 46.89%, resp ectively. The predicted probabilties for the depression general, pulmonary, and no treatment groups to experience an increase in expenditures were 38.93%, 46.75%, and 43.69%, respectively. Holding all other variables constant, compared to the no treatme nt group, the depression general group was 1. 91 times more likely to experience a negative (decrease) ch ange in expenditures (p=0.000), as well as 2.23 times more likely to experience a positive change (increase) in expenditures (p=0.000) compared to the n o treatment group. Similarly, the pulmonary diagnosis group was simultaneously 5.76 times more likely to experience an negative change (decrease) in expenditures and a 7.43 times more likely to experience a positive change in expenditures (p=0.000) relat ive to the no treatment general group. Furthermore, when the pulmonary group

PAGE 69

69 was compared to the no treatment pulmonary group, there was a significant decrease in expenditures for those who showed a decrease over time (p=0.000). Part II A GGM regression w ith a log link function was used to determine the amount of negative change in expenditures given a negative change for each treatment group (cancer diagnosis was included in this part) The predicted mean expenditures for each treatment group (depression general, pulmonary diagnosis, cancer diagnosis) were $ 2,275.40, $ 1 ,993.68, $ 2 ,855.93, and $ 1 626.54, respectively. Each treatment group experienced a significant increase in the amount of negative change in expenditures compared to the no treatment group (p=0.000, p=0.0 10 and p=0.000, respectively) and the cancer group experienced the largest change. Part III A GGM regression with a log link function was used to determine the amount of positive change in expenditures given a positive change for ea ch treatment group. The predicted mean expenditures for the depression general, pulmonary, cancer, and no treatment groups were $ 2 ,405.45, $ 2 ,540.70, $ 4 ,200.70, and $ 1 ,990.04, respectively. Controlling for all other variables, the cancer diagnosis group experienced a significant increase in the amount of positive change in expenditures relative to the no treatment grou p (p=0.000 ), whereas t he depression general group and the pulmonary diagnosis group did not show a significant increase in positive expend itures (p=0.562 and p=0.450). Bootstrapping prediction Using each of the three equations from the three part model, the combination of the equations were used to predict the overall amount of change in expenditures for each treatment

PAGE 70

70 group and to determi ne whether or not there was a significant overall effect of mental health treatment on expenditures. Bootstrapp ed results show that for each treatment group, there was not a significant change in expenditures when the negative change, no change, and posit ive change groups were aggregated and compared to the no treatment general group However, when the medical groups were compared directly to its corresponding no treatment medical group, both the pulmonary and cancer treatment groups had a significant ove rall increase in expenditures over time. Summary Both the depression general group and the pulmonary diagnosis groups were likely to experience a change (positive or negative) in medical expenditures from mental health treatment. Of the participants who e xperienced a negative change in medical expenditures, each treatment group (depression general, pulmonary, and cancer) experienced a greater decrease in expenditures than no treatment Only the cancer group amongst those with a positive change showed a si gnificant increase in medical expenditures with treatment. Comparing the pulmonary and cancer treatment groups to the corresponding no treatment medical group, there was no significant increase nor decrease in expenditures. The pulmonary and cancer treat ment groups showed an overall increase in medical expenditures with mental health treatment compared to the pulmonary and cancer no treatment groups. Prescription Drug Expenditures Two Part Model (Y 1 + Y 2 ) The weighted percentages of participants who had z ero vs. positive expenditures for the two part model is represented in Table 6 13. Overall, the percentage of participants who had zero expenditures was 11.61% and the percentage who had positive expenditures was 88.39%. The coefficients, standard errors, and p values of significance tests from the regression are

PAGE 71

71 depicted in Table 6 4 for Part I, Table 6 5 presents the coefficients, standard errors, and p values of significance tests from the regressions in Parts II, and Table 6 6 shows the results from th e boostrapp ed predictions Part I A logistic regression was run to predict the probability of having zero or non zero values for each treatment group, relative to the no treatment group. The predicted probability of hav ing any expenditures for the depression general, pulmonary, cancer, and no treatment groups were 92.4%, 89.4%, 85.4%, and 79.9%, respectively. Statistical results indicate that the depression general group and the pulmonary diagnosis group probabilities were significantly di fferent than the no treatment group. That is, h olding all other variables constant, the treated depression general group was 92% more likely to have any drug expenditures than the no treatment group (p=0.000) and t he treated group with a pulmonary diagnos is was 51.9% more likely to have any drug expenditures than the no treatment group (p=0.000). When the pulmonary treatment group was compared to the no treatment pulmonary group, the treated group was significantly less likely to have expenditures greater than zero (p=0.050). Part II A GGM regression with a log link function was performed on part II of the two part model to determine if there is a significant difference in expenditures between treatment and non treatment among those who had positive expend itures during the study period. The predicted mean expenditures for the depression general, pulmonary, cancer, and no treatment groups were $4, 189.64 $ 5 199.13 $ 4 ,126.83, and $ 1 897.3 6, respectively. Statistical r esults demonstrate that for those who ha d some expenditures during the study period, each of the treatment group s had significantly higher medical expenditures than the general group that did not have any mental

PAGE 72

72 health treatment (p=0.000) and the pulmonary group was more likely to have higher dr ug expenditures than the no treatment pulmonary group (p=0.000) Bootstrapping prediction Using each of the two equations from the two part model, the combination of the equations were used to predict the overall amount of change in expenditures for each treatment group and to determine whether or not there was a significant overall effect of mental health treatment on expenditures. Bootstrapp ed results show that for each treatment group comparing to the no treatment general group there was not an overall significant change in drug expenditures when the zero and positive value groups were aggregated. However, when the treated pulmonary or cancer groups were compared directly to the no treatment pulmonary or cancer groups, there was a significan t increase in the treatment groups. Summary Depression general and pulmonary diagnosis groups were most likely to have positive drug expenditures with mental health treatment compared to no treatment. When only those with drug expenditures greater than ze ro were examined, all three treatment groups had a significant increase in expenditures associated with mental health treatment. However, there was not a significant overall difference in drug expenditures for the depression general group, but an increase in drug expenditures for the pulmonary and cancer treatment groups Three Part Model (Y 2 Y 1 ) The weighted percentages of participants who experienced each type of change in drug expenditures for the three treatment groups are presented in Table 6 1 4 Overall, the p e r centages of participants who had a decrease, no change, and increase in drug expenditures with mental health treatment were 44.31 %, 5.29 %, and 50.40 %, respectively. Table 6 8 shows the results

PAGE 73

73 from Part I Table 6 9 depicts the results f rom Parts II and III, and Table 6 10 displays the bootstrapped results. Part I A mulitnomial logit regression was used to predict the probability of having a negative change, no change, or positive change in expenditures for the depression general group an d the pulmonary diagnosis group, relative to the no treatment group. Because the cancer diagnosis group had no participants with zero (no change), they were excluded from part I. The predicted probabilties for the depression general, pulmonary, and no tr eatment groups to experience a decrease in expenditures were 40.43%, 41.51%, and 39.48%, respectively. The predicted probabilties for the depression general, pulmonary, and no treatment groups to experience an increase in expenditures were 54.95%, 56.72% and 31.75%, respectively. The depression general and pulmonary groups were significantly different than the no treatment group. That is, h olding all other variables constant, compared to the no treatment group, the depression general group was 8.83 tim es more likely to experience a negative change (decrease) in expenditures (p=0.000), as well as 14.68 times more likely to experience a positive change (increase) in expenditures (p=0.000) compared to the no treatment group. Similarly, the depressed trea tment group with a pulmonary diagnosis were simultaneously 24.81 ti mes more likely to experience a negative change (decrease) in expenditures and a 41.35 times more likely to experience a positive change (increase) in expenditures (p=0.000) relative to the no treatment group. Part II A GGM regression with a log link function was used to determine the amount of negative change in expenditures given a negative change for each treatment group. The predicted mean expenditures for the depression general, pulmon ary, cancer, and no treatment groups were $

PAGE 74

74 551.18, $ 693.08, $ 429.86, and $ 398.34 respectively. Each treatment group (depression general, pulmonary diagnosis, cancer diagnosis) experienced a significant increase in the amount of negative change in exp enditures compared to the no treatment group (p=0.000, p=0.0 0 0, and p=0.00 1 respectively) and the cancer group experienced the most change. When comparing the no treatment pulmonary group to the treated pulmonary group, there was a significant decrease f or the treated group (p=0.000), whereas the cancer treatment group did not show a significant decrease compared to the no treatment cancer group (p=0.846). Part III A GGM regression with a log link function was used to determine the amount of positive chan ge in expenditures given a positive change for each treatment group. The predicted mean expenditures for the depression general, pulmonary, cancer, and no treatment groups were $ 608.1 6 $ 800.53 $725.88 and $228.35, respectively. Controlling for all oth er variables, among those who experienced a positive change, each treatment group experienced a significant increase in the amount of positive change in expenditures relative to the no treatment general group (p=0. 000 ). Both the pulmonary and cancer treat ment groups showed a significant increase in drug expenditures when compared to the no treatment pulmonary and cancer groups (p =0.001). Bootstrapping prediction Using each of the three equations from the three part model, the combination of the equations were used to predict the overall amount of change in expenditures for each treatment group and to determine whether or not there was a significant overall effect of mental health treatment on expenditures. Bootstrapp ed results show that for each treatment group compared to the no treatment general group there was not a significant change in expenditures when the

PAGE 75

75 negative change, no change, and positive change groups were aggregated. When the pulmonary and cancer groups were compared to the corresponding untreated medical group, there was an overall significant increase for the pulmonary treatment group, whereas there was a significant decrease in the cancer treatment group. Summary Both depression general and pulmonary diagnosis groups were more likely to experience a change (positive or negative) in drug expenditures relative to the no treatment group. Additionally, each group (depression general, pulmonary, and cancer) with positive or negative drug expenditures had increasingly positive or negative cha nges, respectively, associated with mental health treatment with the exception of the cancer treatment group compared to the no treatment cancer group There was a significant overall increase in drug expenditures for the pulmonary and cancer treatment groups compared to the no treatment medical groups. Emergency Room Visits Negative Binomial Regression (Y 1 + Y 2 ) For total emergency room visits, a negative binomial regression was estimated in order to determine if the treated groups had higher emergency room visits than the no treatment group Table 6 15 depicts the incidence rate ratios, standard errors, and p value results. Table 6 16 shows the predicted average number of visits for each group. The depression general group had a mean of 0.72, the pul monary diagnosis group had a mean of 1.11, the cancer diagnosis group had a mean of 0.98 visits, and the no treatment group had 0.64 mean visits. Results indicate that for each treatment group, emergency room visits were more likely relative to the no tre atment group (p=0.000, p=0.000, and p=0.025). The incidence rates indicated that the depression general group was 1.18 times higher, the pulmonary diagnosis group was 1.23 times higher, and the cancer diagnosis group was 1.21 times higher than the no trea tment general group. When

PAGE 76

76 comparing to the no treatment cancer group, the treated cancer group had an incident rate 1.32 times higher than the no treatment cancer group (p=0.050), whereas there was no difference between the treated and untreated pulmonary group (0.991). Three Part Model (Y 2 Y 1 ) The weighted percentages of participants who experienced each type of change in emergency room visits for the three treatment groups are presented in Table 6 1 7 Overall, the p e r centages of participants who had a decrease, no change, and increase in emergency room visits with mental health treatment were 16.92 %, 59.94 %, and 23.14 %, respectively. Table 6 1 8 shows the results from Part I Table 6 1 9 depicts the results from Parts II and III, and Table 6 2 0 display s the bootstrapped results. Part I A mulitnomial logit regression was used to predict the probability of having a negative change, no change, or positive change in visits for the depression general group the pulmonary diagnosis group, and the cancer diagn osis group relative to the no treatment group. The predicted probabilities of having a negative change in emergency room visits were 20.30 %, 18.38 %, 16. 24 %, and 17.03 % for the depression general, pulmonary, cancer, and no treatment general groups, respect ively. The predicted probabilities of having a positive change in emergency room visits for each group were 20.46%, 19.70%, 18.49%, and 17.06% for the respective groups. Statistical results indicate that the depression general group was the only treatment group with significant differences from the no treatment group. That is, h olding all other variables constant, compared to the no treatment group, the depression general group was simultaneously more likely to experience a negative change (decrease) in expenditures (p=0.00 1 ), as well as more likely to experience a positive change (increase) in expenditures (p=0.00 7 ) compared to the no treatment group. Furthermore, the tr eated pulmonary group

PAGE 77

77 compared to the untreated pulmonary group was more likely to experience a decrease in emergency room visits. All other estimates were non significant. Part II A negative binomial regression was used to determine the amount of negat ive change in expenditures given a negative change for each treatment group. The predicted mean number of emergency room visits for each group (depression general, pulmonary, cancer, and no treatment) were 0.30, 0.27, 0.24, and 0.23, respectively. Ea ch treatment group (depression general, pulmonary diagnosis, cancer diagnosis) that experienced a negative change in expenditures were not significantly different from the no treatment general group (p=0.086, p=0.175, and 0.505, respectively). However, th e cancer diagnosis treatment group with a decrease in visits over time had a higher incidence rate than the no treatment cancer group (p=0.028). Part III A negative binomial regression was used to estimate the amount of positive change in expenditures gi ven a positive change for each treatment group. The predicted mean number of visits for each group were 0.28, 0.31, 0.19, and 0.23 for the depression general, pulmonary, cancer, and no treatment groups, respectively. Controlling for all other variables, among those who experienced a positive change, the only treatment group that experienced a s ignificant increase in the amount of positive change in expenditures relative to the no treatment general group was the pulmonary diagnosis group (p=0.0 49 ). The ca ncer treatment group had a significant increase in expenditures compared to the no treatment cancer group (p=0.001). Bootstrapping prediction Using each of the three equations from the three part model, the combination of the equations were used to predi ct the overall amount of change in utilization for each treatment

PAGE 78

78 group and to determine whether or not there was a significant overall effect of mental health treatment on utilization Bootstrapp ed results show that for each treatment group comparing to the general no treatment grouop there was not a significant change in visits when the negative change, no change, and positive change groups were aggregated. However, when the medical groups were compared to the no treatment medical groups, the pulmonary and cancer treatment groups had more emergency room visits overall than the no treatment pulmonary or cancer group. Summary Total emergency room visits are greater with each treated group compared to the untreated general group, but the depression general group was more likely to have a change in emergency room visits (positively or negatively) over time The pulmonary treatment group was more likely to have a decrease in emergency room visits. Of participants experiencing a positive ch ange with mental health treatment, the pulmo nary and cancer diagnose s group s experienced a significant increase with treatment. Nevertheless the overall difference, combining negative, no change, and positive change, was non significant for each treatmen t group when compared with the no treatment general group The medical condition treatment groups did show greater utilization than the no treatment medical groups. Inpatient Visits Negative Binomial Regression (Y 1 + Y 2 ) For total inpatient visits, a nega tive binomial regression was estimated in order to determine if the treated groups were more or less likely to undergo inpatient visits. Table 6 15 depicts the incidence rate ratios, standard errors, and p value results. Table 6 16 shows the predicted av erage number of visits for each group and the depression general group had an average of 2.78, the pulmonary diagnosis group had an average of 3.06, the cancer diagnosis

PAGE 79

79 group had an average of 5.98 visits, and the no treatment group had an average of 7.52 visits. Results indicate that for the depression general and the cancer diagnosis treatment group, inpatient visits were more likely relative to the no treatment group (p=0.024 and p=0.002). The incidence rates indicated that the depression general grou p was 1.40 times higher and the cancer diagnosis group was 2.25 times higher than the no treatment group. However, both t he pulmonary and cancer diagnosis treatment group s did not show s ignificant results for total inpatient visits when compared to the p ulmonary and cancer diagnosis untreated groups (p=0.588 and p=0.550, respectively) Three Part Model (Y 2 Y 1 ) The weighted percentages of participants who experienced each type of change in inpatient visits for the three treatment groups are presented i n Table 6 21. Overall, the percentages of participants who had a decrease, no change, and increase in inpatient visits with mental health treatment were 11.15%, 69.84%, and 19.01%, respectively. Table 6 18 shows the results from Part I Table 6 19 depic ts the results from Parts II and III, and Table 6 20 displays the bootstrapped results. Part I A mulitnomial logit regression was used to predict the probability of having a negative change, no change, or positive change in visits for the depression genera l group, the pulmonary diagnosis group, and the cancer diagnosis group relative to the no treatment group. The predicted probabilities of having a negative change in inpatient visits were 12.53%, 11.71%, 17.83%, and 10.21% for the depression general, pulmonary, cancer, and no treatment groups, respectively. The predicted probabilities of having a positive change in inpatient visits for each group were 14.79%, 19.12%, 25.28%, a nd 12.56% for the respective groups. Statistical results indicate that h olding all other variables constant, compared to the no treatment group, the

PAGE 80

80 depression general group and the cancer diagnosis group were more likely to experience a negative change ( decrease) in utilization (p=0.0 10 and p=0.004, respectively ) The pulmonary diagnosis group and the cancer diagnosis group were more likely to experience a positive change (increase) in utilization (p=0.012 and 0.003, respectively ) compared to the no trea tment group. All other differences were non significant. W hen the pulmonary treatment group was compared to the pulmonary no treatment groups, the results were not significant. The cancer treatment group was more likely to experience an increase in inpa tient visits compared to the no treatment cancer group (p=0.000). Part II A negative binomial regression was used to determine the amount of negative change in utilization given a negative change for each treatment group. The predicted mean number of inpa tient visits for each group (depression general, pulmonary, cancer, and no treatment) were 1.25, 0.75, 1.48, and 1.05, respectively. Each treatment group (depression general, pulmonary diagnosis, cancer diagnosis) that experienced a negative change in utilization was not significantly different than the no treatment group (p=0.0 53 p=0. 347 and 0. 191 respectively) However, the pulmonary treatment group had a slower decrease in expenditures than the no treatment pulmonary group (p=0.023). Part III A negative binomial regression was used to estimate the amount of positive change in utilization given a positive change for each treatment group. The predicted mean number of visits for each group were 1.01, 1.83, 2.34, and 1.12 for the depression genera l, pulmonary, cancer, and no treatment groups, respectively. Controlling for all other variables, among those who experienced a positive change, each treatment group (depression general, pulmonary

PAGE 81

81 diagnosis, cancer diagnosis) that experienced a positive c hange in utilization did not experience a significant change in the amount of positive change (p=0.380, p=0.750, and 0.182, respectively) and this held true when the treated medical groups were compared to the untreated medical groups. Bootstrapping predic tion Using each of the three equations from the three part model, the combination of the equations were used to predict the overall amount of change in utilization for each treatment group and to determine whether or not there was a significant overall eff ect of mental health treatment on utilization Bootstrapping results show that for the depression general treatment group, there was not a significant change in inpatient visits when the negative change, no change, and positive change groups were aggregat ed. However, f or the pulmonary diagnosis group and the cancer diagnosis group, there was an overall increase in inpatient visits compared to the no treatment general group The cancer treatment group had more inpatient visits than the no treatment cancer group. Summary The depression general group and the cancer diagnosis group had an increased incidence of inpatient visits than the untreated group, and the pulmonary diagno sis group approached significance. The c ancer diagnosis group was more likely to experience a ny change (positive or negative) in visits over time, whereas th e pulmonary diagnosis group was more likely to experience an increase in inpatient visits and the depression general group were more likely to experience a decrease in visits from mental health treatment. However, there were no differences in visits amongst the negative or positive change groups compared to the general treatment group, but the pulmona ry treatment group compared to the untreated pulmonary group

PAGE 82

82 showed a lower incidence rate T he overall difference in visits combining each type of change (negative, none, positive) indicate that the pulmonary and cancer diagnoses groups had higher utiliz ation than the no treatment general group. The cancer treatment group also had more inpatient visits than the no treatment cancer group. Outpatient Visits Negative Binomial Regression (Y 1 + Y 2 ) For total outpatient visits, a negative binomial regression w as estimated in order to determine if the treated groups were more or less likely to undergo outpatient visits. Table 6 15 depicts the incidence rate ratios, standard errors, and p value results. Table 6 16 shows the predicted average number of visits fo r each group and the depression general group had an average of 2.00, the pulmonary diagnosis group had an average of 3.00, the cancer diagnosis group had an average of 5.21 visits, and the no treatment group had an average of 4.01 visits. Results indicat e that for each treatment group, outpatient visits were more likely relative to the no treatment group (p=0.000, p=0.043, and p=0.000). The incidence rates indicated that the depression general group was 1.45 times higher, the pulmonary diagnosis group wa s 1.43 times higher, and the cancer diagnosis group was 2.29 times higher than the no treatment group. The pulmonary treatment group had an incidence rate 1.35 times higher than the no treatment pulmonary group (p=0.008). Three Part Model (Y 2 Y 1 ) The we ighted percentages of participants who experienced each type of change in outpatient visits for the three treatment groups are presented in Table 6 2 2 Overall, the percentages of participants who had a decrease, no change, and increase in outpatient visi ts with mental health treatment were 17.90 %, 58.15 %, and 23.95 %, respectively. Table 6 18 shows the

PAGE 83

83 results from Part I Table 6 19 depicts the results from Parts II and III, and Table 6 20 displays the bootstrapped results. Part I A mulitnomial logit regression was used to predict the probability of having a negative change, no change, or positive change in visits for the depression general group, the pulmonary diagnosis group, and the cancer diagnosis group relative to the no treatment group. The pr edicted probabilities of having a negative change in outpatient visits were 19.86 %, 20.96 % 23.86 % and 16.32% for the depression general, pulmonary, cancer, and no treatment groups, respectively. The predicted probabilities of having a positive change in outpatient visits for each group were 20.17 %, 20.50 %, 32.63 %, and 17.53 % for the respective groups. Statistical results indicate that h olding all other variables constant, compared to the no treatment group, each treatment group was more likely to experi ence both positive or negative changes in expenditures when compared to the no treatment group (p=0.000). However, when the pulmonary and cancer groups were compared to its corresponding no treatment groups, results were non significant. Part II A negativ e binomial regression was used to determine the amount of negative change in utilization given a negative change for each treatment group The predicted mean number of outpatient visits for each group (depression general, pulmonary, cancer, and no treatme nt) were 0.63, 0.96, 1.45, and 0.65, respectively. Amongst the individuals experiencing a negative change in utilization, the depression general group was the only group to experience a significant increase in the amount of negative change (p=0.008). The medical groups compared to the untreated medical groups did not show a significant difference.

PAGE 84

84 Part III A negative binomial regression was used to estimate the amount of positive change in utilization given a positive change for each treatment group The predicted mean number of visits for each group were 0.94, 0.69, 0.96, and 0.48 for the depression general, pulmonary, cancer, and no treatment groups, respectively. Amongst the individuals experiencing a positive change in utilization, each treatme nt group experience d a significant increase in the amount of positive change relative to the no treatment group (p=0.000 p=0.000, and p=0.014 ). The medical groups compared to the untreated medical groups did not show a significant difference. Bootstrappi ng prediction Using each of the three equations from the three part model, the combination of the equations were used to predict the overall amount of change in utilization for each treatment group and to determine whether or not there was a significant ov erall effect of mental health treatment on utilization Bootstrapping results show that for each treatment group compared to the no treatment general group there was not a significant change in outpatient visits when the negative change, no change, and p ositive change groups were aggregated. The pulmonary treatment group had fewer outpatient visits than the no treatment pulmonary group and the cancer treatment group had more outpatient visits than the no treatment cancer group. Summary For total outpatient visits, each treatment group experienced a greater incidence of visits with mental health treatment. Each treatment group was also more likely to experience a change (positive or negative) in visits than the no treatment general group except the medical groups compared to the untreated medical groups did not show a significant difference Only the

PAGE 85

85 participants in the depression general group that had either a positive or negative change in visits over time showed a significant increa se or decrease in visits, respectively. However, there was no significant overall change in visits when the negative, no change, and positive change were aggregated when groups were compared to the no treatment general group When the medical condition t reatment groups were compared to their corresponding medical condition no treatment group, the pulmonary treatment group had fewer outpatient visits and the cancer treatment group had more outpatient visits overall. Office Based Provider Visits Negative Binomial Regression (Y 1 + Y 2 ) For total outpatient visits, a negative binomial regression was estimated in order to determine if the treated groups were more or less likely to have any outpatient visits. Table 6 15 depicts the incidence rate ratios, stand ard errors, and p value results. Table 6 16 shows the predicted average number of visits for each group and the depression general group had an average of 23.07, the pulmonary diagnosis group had an average of 31.60, the cancer diagnosis group had an aver age of 30.91 visits, and the no treatment group had an average of 12.60 visits. Results indicate that for each treatment group, office based provider visits were more likely experienced relative to the no treatment group (p=0.000). The incidence rates in dicated that the depression general group was 1.88 times higher, the pulmonary diagnosis group was 2.07 times higher, and the cancer diagnosis group was 1.96 times higher than the no treatment general group. When compared to the corresponding untreated me dical groups, the pulmonary treated group had an incidence rate 1.54 times higher and the cancer treated groups had an incidence rate 1.32 times higher.

PAGE 86

86 Three Part Model (Y 2 Y 1 ) The weighted percentages of participants who experienced each type of change in office based provider visits for the three treatment groups are presented in Table 6 2 3 Overall, the percentages of participants who had a decrease, no change, and increase in emergency room visits with mental health treatment were 41.61 %, 10.77 %, an d 47.63 %, respectively. Table 6 18 shows the results from Part I Table 6 19 depicts the results from Parts II and III, and Table 6 20 displays the bootstrapped results. Part I A mulitnomial logit regression was used to predict the probability of having a negative change, no change, or positive change in visits for the depression general group, the pulmonary diagnosis group, and the cancer diagnosis group relative to the no treatment group. The predicted probabilities of having a negative change in office based provider visits were 43.10%, 40.49%, 39.57%, and 46.32% for the depression general, pulmonary, cancer, and no treatment groups, respectively. The predicted probabilities of having a positive change in emergency room visits for each group were 48.96 %, 51.02%, 55.13%, and 36.90% for the respective groups. Statistical results indicate that h olding all other variables constant, compared to the no treatment group, each treatment group was less likely to experience both positive and negative changes in e xpenditures when compared to the no treatment group (p=0.000). When the no treatment pulmonary group was compared to the treated pulmonary group, there was a decreased likelyhood of an increase in office based provider visits over time (p=0.014). Part II A negative binomial regression was used to determine the amount of negative change in utilization given a negative change for each treatment group. The predicted mean number of office based provider visits for each group (depression general, pulmonary, ca ncer, and no

PAGE 87

87 treatment) were 3.68, 3.62, 4.26, and 3.33, respectively. Amongst the individuals experiencing a negative change in utilization, each treatment group experienced a significant increase in the amount of negative change (p=0.011, p=0.003, a nd p=0.000) and this held true when the cancer treatment group was compared to the untreated cancer group (p=0.046) The incidence rates indicated that the depression general group was 1.24 times more likely to experience an office based provider visits, the pulmonary diagnosis group was 1.47 times more likely, and the cancer diagnosis group was 1.43 times more likely to experience an office based provider visit relative to the no treatment general group Part III A negative binomial regression was used to estimate the amount of positive change in expenditures given a positive change for each treatment group. The predicted mean number of visits for each group were 4.09, 4.54, 4.93, and 2.45 for the depression general, pulmonary, cancer, and no treatment groups, respectively. Amongst the individuals experiencing a negative change in utilization, each treatment group experienced a significant increase in the amount of negative change (p=0.000, p=0.000, and p=0.014). The incidence rates indicated that the depression general group was 1.35 times more likely to experience an office based provider visits, the pulmonary diagnosis group was 1.48 times more likely, and the cancer diagnosis group was 1.56 times more likely to experience an office based provider v isit relative to the no treatment group. The pulmonary treatment group had a higher incidence of office based provider visits than the no treatment pulmonary group p=(0.040). Bootstrapping prediction Using each of the three equations from the three part m odel, the combination of the equations were used to predict the overall amount of change in utilization for each treatment

PAGE 88

88 group and to determine whether or not there was a significant overall effect of mental health treatment on utilization Bootstrapp ed results show that for each treatment group, there was not a significant change in office based provider visit s when the negative change, no change, and positive change groups were aggregated. Summary For total office based provider visits, each treatment group had a greater number of visits associated with mental health treatment. A n examination of a difference in visits over time revealed that each group was less likely to experience a change over time (positive or negative) with mental health treatment. Amongst those who had a positive or negative change, they were shown to experience an increased respective positive or negative change with treatment. However, combining the negative, no change, and positive changes together reveal that there was not a significant overall difference in office based provider visits within the groups. Work Absenteeism Table 6 24 and Table 6 25 shows the individual characteristics of the work sample. The work sample consisted of depressed individuals aged 18 65 who were em ployed during each round of the study timeframe. The total sample consisted of 3, 930 participants, 1,399 did not receive mental health treatment and 2,531 received mental health treatment. Males consisted of 31.17 % of the sample and females constituted 68. 83 % of the sample. The average age of the work sample was 43.61 years and the majority were White ( 88.24 %). Chi square re sults indicate that participant groups were significantly different on each study characteristic examined in this study with the exception that ADL and IADL needs were similar across the treatment and non treatment groups Participants most likely to seek treatment were female, white, high income, and insured i ndividuals aged 45 64 The highest proportion of individuals who sought treatment

PAGE 89

89 rated their physical conditions. Table 6 26 depicts the statistical results on the work sample as discussed next. Negative binomial regression A negative binomial regression was used to determine whether those treated for depression in the work sample were more likely to show a decrease in work absenteeism. The regression was weighted, but did not include a selection correction that would be needed to correct for any fundame ntal differences between treated vs. untreated participants. The uncorrected regression showed that treated depression general participants were more likely to experience an increase in work absenteeism rates than the no treatment group (p=0.0 47 ). Partic ipants in the treated pulmonary diagnosis group and the treated cancer diagnosis group did not show significant differences from the no treatment general group (p=0. 166 and p=0. 434 respectively), and this result was consistent when the pulmonary and cance r treatment groups were compared to the untreated pulmonary and cancer groups. Poisson regression In order to correct for the selection bias as discussed in Chapter 5 (page 52), a n ) was used to examine whether those treated for depression in the work sample were more likely to show a decrease in work absenteeism. The instrumental variable poisson regression was used because other options (e.g., heckman two step) relied on the assum ption of a normal distribution of error terms, which did not fit the data. A poisson regression is commonly used for count data and when a negative bi nomial instrumental regression i s unavailable. Unfortunately, the do not al low for weighting of the data, so the results do not provide a nationally representative estimate of the US non institutionalized, civilian population. Nevertheless, the results corroborate the uncorrected negative binomial regression when groups were com pared to the no

PAGE 90

90 treatment general group and the no treatment medical groups That is, the depression general group was most likely to experience an increase in work absenteeism rates (p=0.004) whereas the pulmonary diagnosis and cancer diagnosis groups d id not yield significant results (p=0.107 and p=0.324, respectively) Post Hoc (Sensitivity) Analyses Psychotherapy vs. Antidepressants The main analyses combined psychotherapy and antidepressants in order to define mental health treatment. In order to determine if there was a significant difference between psychotherapy and antidepressant therapy on total and medical expenditures, further analyses were done. Psychotherapy and antidepressant therapy (treatment type) were separated for each of the treatment groups (depression general, pulmonary, and cancer) and compared with the no treatment group For total expenditures, it appears that the main results in which the depression general group had higher expenditures with treatment and more likely to show an increase in total expenditures over time are due to both psychotherapy and antidepressant medication. However, the increased reduction in total expenditures over time for the depression general group was due to an tidepressant medication. For the pulmonary diagnosis group, the decreased likelihood of experiencing a reduction in expenditures was due to medication. For the cancer diagnosis group, antidepressant medication appears to be the driver of increased cost w hen expenditures are totalled across time periods, whereas individuals with cancer experiencing a reduction in expenditures over time was less likely with antidepressants, but more likely with psychotherapy. For medical expenditures, antidepressant medicat ion was the culprit of significant results for the depression general group. That is, individuals in the depression general group taking antidepressant medication were more likely to have any expenditures, show a change in

PAGE 91

91 expenditures over time. For the pulmonary diagnosis group, psychotherapy played the prominent role. Despite medication increasing the likelihood of change over time for pulmonary patients, amongst individuals who had any medical expenditures or experienced a reduction, psychotherapy ac ted to reduce the amount of negative change and increase medical expenditures. For the cancer diagnosis group, antidepressant medication had the effect of minimizing the magnitude of reduction and increasing medical expenditures. Mild vs. Severe Depressio n MEPS included a measure screening for depression, the Patient Health Questionnaire (PHQ 2), in its 2004 and 2005 survey. The PHQ d the two (Kroenki, Spitzer, & Williams, 2003). resulting in a score from 0 6. On the PHQ 2 scale, a 3 or higher on the PHQ 2 is considered depression. In order to determine if differing levels of depression severity may have an impact on results, the sample with PHQ 2 data was used. A 3 or 4 on the PHQ 2 was considered moderate depression and a 5 or 6 on the PHQ 2 was considered severe depression for this analysis. Due to inadequate sample size, the pulmonary and cancer groups were not included in the analysis. Only the depression general group was examined. Post hoc results indicate that the only difference between the moderate and severe groups was that amongst those who experienced a reduction in total expenditures over time, the severe depression group had a greater reduction in expenditures th an the no treatment group (p=0.043 ).

PAGE 92

92 Full Time vs. Part Time Work In this study, full time and part time workers were not differentiated. Thus, for this post hoc analysis, it was determined whether there was a significant difference between full and part time workers. A negative binomial regression revealed that there were no significant differences between full and part time workers, except for individuals with cancer. Results indicate that treated cancer patients working part time had a greater inc idence rate of work absenteeism than individuals who did not receive mental health treatment (p=0.02).

PAGE 93

93 Table 6 1. Demographic characteristics of study samples by treatment group for expenditures/utilization equations Variables Depression General Treatment ( N= 2488 ) % Pulmonary Comorbid Treatment ( N= 1199 ) % Cancer Comorbid Treatment ( N= 194 ) % No Treatment (N=2004) % Total ( N= 6028 ) % Chi2 test p value Age 0 5 0.12 0 0 0.45 0.22 0.000 6 12 1.77 2.25 0 2.69 2.09 13 17 5.27 4.59 1.03 3.89 4.43 18 24 6.67 5.09 3.61 13.07 8.23 25 44 33.72 30.94 21.65 39.12 34.17 45 64 38.87 41.70 44.85 29.49 36.60 65+ 13.59 15.43 28.35 11.28 14.27 Gender Female 69.65 74.90 69.07 65.02 69.14 0.000 Male 30.35 25.10 30.93 34.98 30.86 Race W hite 87.02 83.90 85.57 81.34 84.52 0.000 Black 8.72 9.92 7.73 12.92 10.32 Asian 1.57 1.08 2.06 2.00 1.59 Other 2.69 5.10 4.64 3.74 3.57 Ethnicity Hispanic 14.67 11.93 9.28 25.75 17.47 0.000 Non Hispanic 85.33 88.07 90.72 74.25 82.53 Poverty c ategory Poor/ n egative 20.94 22.94 20.62 24.60 22.61 0.000 Near p oor 6.31 5.75 4.12 7.09 6.44 Low i ncome 15.27 15.01 13.92 18.76 16.36 Middle i ncome 27.65 30.78 24.74 28.19 28.35 High i ncome 29.82 25.52 36.60 21.36 26.24 Insurance s tatus 0.000 Uninsured 9.69 5.42 7.22 20.01 12.08 Intermittent 14.23 10.93 13.92 19.76 15.39 Insured 76.09 83.65 78.87 60.23 72.53 Physical h ealth s tatus 0.000 Excellent 10.13 5.75 3.61 13.12 9.89 Very g ood 27.48 20.18 23.20 24.15 23.04 Good 33.20 31.19 26.80 33.88 32.78 Fair 19.17 22.94 23.71 18.26 19.79 Poor 10.81 17.60 13.40 7.49 11.60

PAGE 94

94 Table 6 1. Continued Variables Depression General Treatment ( N= 2488 ) % Pulmonary Comorbid Treatment ( N= 1199 ) % Cancer Comorbid Treatment ( N= 194 ) % No Treatment (N=2004) % Total ( N= 6028 ) % Chi2 test p value Mental health status Excellent 9.53 9.51 6.19 15.02 11.18 0.000 Very g ood 20.70 18.68 19.59 21.76 20.49 Good 36.94 36.78 38.14 36.58 36.78 Fair 20.50 23.02 18.56 17.81 20.14 Poor 10.09 9.67 8.25 5.74 8.49 Comorbidity c ount 1 55.99 0 0 41.57 36.93 0.000 2 44.01 41.37 57.22 41.87 42.15 3 0 58.63 42.78 15.82 19.34 4 0 0 0 0.75 1.58 Need help with IADL Yes 10.81 12.93 15.46 6.54 10.22 0.000 No 89.19 87.07 84.54 93.46 89.78 Need help with ADL Yes 5.79 7.34 10.82 3.24 5.64 0.000 No 94.21 92.66 89.18 96.76 94.36

PAGE 95

95 Table 6 2 Central tendency of demographics for sample included in expenditures/utilization equations Variables Depression g eneral t reatment ( N= 2488 ) % Pulmonary c omorbid t reatment ( N= 1199 ) % Cancer c omorbid t reatment ( N= 194 ) % No t reatment (N=2004) % Total ( N= 6028 ) % Age Average 45.50 47.01 55.05 41.57 45.13 Standard d eviation 17.79 17.40 16.55 18.11 18.10 Minimum 5 6 13 4 4 Maximum 85 85 85 85 85 Years of e ducation Average 11.78 11.67 12.64 11.14 11.59 Standard d eviation 3.55 3.59 3.20 3.64 3.59 Minimum 0 0 0 0 0 Maximum 17 17 17 17 17 Income Average 21232.42 19691.56 22375 18107.94 19963.55 Standard d eviation 25038.11 22830.40 28156.22 21865.90 23698.17 Minimum 40337 333 28961 90050 90050 Maximum 182836 200722 171953 163684 200722 Comorbidity c ount Average 1.44 2.59 2.43 1.76 1.86 Standard d eviation 0.50 0.49 0.50 0.74 0.78 Minimum 1 2 2 1 1 Maximum 2 3 3 4 4 Table 6 3 Proportions of each t ype of c hange in total expenditures by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Zero e xpenditures 25.72 27.52 32.44 27.31 Non z ero e xpenditures 74.28 72.48 68.56 72.69 Total 100.00 100.00 100.00 100.00

PAGE 96

96 Table 6 4 L ogit regression predicting probability of experiencing each type of change in expenditures ( part I of two part m odel ) Relative risk of having positive change vs. no change Coef. s.e. p value Total e xpenditures No treatment Reference Depression general 0.114 0.066 0.084 Pulmonary diagnosis 0.155 0.079 0.051 Cancer diagnosis 0.132 0.146 0.364 Medical e xpenditures No treatment Reference Depression general 0.288** 0.078 0.000 Pulmonary diagnosis Cancer diagnosis 0.130 0.187 0.098 0.274 0.185 0.495 Drug e xpenditures No treatment Reference Depression general 0.920** 0.107 0.000 Pulmonary diagnosis 0.519** 0.128 0.000 Cancer diagnosis 0.116 0.269 0.666 **Significant at 0.05 level Table 6 5 Ordinary least squares regression ( OLS ) or generalized linear model (GLM) predicting amount of change in expenditures given positive change ( Part II of two part model ) Amount of c hange Coef. s.e. p value Total e xpenditures No treatment Reference Depression general 0.785** 0.069 0.000 Pulmonary diagnosis 0.845** 0.050 0.000 Cancer diagnosis 0.985** 0.119 0.000 Medical e xpenditures No treatment Reference Depression general 0.300** 0.066 0.000 Pulmonary diagnosis 0.400** 0.052 0.000 Cancer diagnosis 0.654** 0.087 0.000 Drug e xpenditures No treatment Reference Depression general 0.792** 0.074 0.000 Pulmonary diagnosis 1.008** 0.058 0.000 Cancer diagnosis 0.777** 0.136 0.000 **Significant at 0.05 level

PAGE 97

97 Table 6 6 Bootstrapping prediction of expenditure change after t wo part model ( OLS ) N Mean S.D. 95%CI Total e xpenditures Depression general 1000 12235.52** 10536.79 (12165.27, 15104.86) Pulmonary diagnosis 1000 10657.36** 9250.19 (10168.78, 12472.32) Cancer diagnosis 1000 12673.82** 10992.70 (10496.44, 16412.93) Medical e xpenditures Depression general 1000 9344.22** 6922.44 (8398.14, 10365.60) Pulmonary diagnosis 1000 10323.20 7647.69 (9440.85, 11605.71) Cancer diagnosis 1000 13312.50 9862.24 (10879.77, 16254.00) Drug e xpenditures Depression general 1000 4189.64 2926.54 (3936.48, 4478.54) Pulmonary diagnosis 1000 5199.13 3631.69 (4789.87, 5666.10) Cancer diagnosis 1000 4126.83 2882.66 (3558.48, 4718.46) **Significant at 0.05 level Table 6 7 Proportions of each t ype of c hange in total expenditures by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Decreased 37.66 35.45 29.38 38.57 No change 25.72 27.52 31.44 27.31 Increased 36.62 37.03 39.18 34.12 Total 100.00 100.00 100.00 100.00

PAGE 98

98 Table 6 8 Multinomial logit regression predicting probability of experiencing each type of change in expenditures ( part I of three part model) Relative risk of having negative change vs. no change Relative risk of having positive change vs. no change RRR s.e. p value RRR s.e. p value Total e xpenditures No treatment Reference Depression general 0.9 41 0. 079 0.4 69 1.391** 0.104 0.000 Pulmonary diagnosis 0.696** 0.062 0.000 1.099 0.093 0.263 Cancer diagnosis 0.613** 0.094 0.002 1.276 0.242 0.200 Medical e xpenditures No treatment Reference Depression general 1.911** 0.188 0.000 2.228** 0.179 0.000 Pulmonary diagnosis 5.755** 1.505 0.000 7.434** 2.167 0.000 Drug e xpenditures No treatment Reference Depression general 8.829** 0.774 0.000 14.675** 0.918 0.000 Pulmonary diagnosis 24.810** 8.084 0.000 41.352** 13.262 0.000 **Significant at 0.05 level

PAGE 99

99 Table 6 9 Generalized linear regression (GLM) predicting amount of change in expenditures given negative or positive change ( parts II and III of three part model) Amount of negative c hange Amount of positive change Coef. s.e. p value Coef. s.e. p value Total e xpenditures No treatment Reference Depression general 0.378** 0.081 0.000 0.299** 0.115 0.010 Pulmonary diagnosis 0.272** 0.121 0.026 0.239** 0.111 0.032 Cancer diagnosis 0.628** 0.127 0.000 0.544** 0.218 0.013 Medical e xpenditures No treatment Reference Depression general 0.388** 0.109 0.000 0.070 0.120 0.562 Pulmonary diagnosis 0.286** 0.110 0.010 0.053 0.118 0.45 Cancer diagnosis 0.619** 0.091 0.000 0.558 ** 0.143 0.000 Drug e xpenditures No treatment Reference Depression general 0.539** 0.0739 0.000 0.608** 0.063 0.000 Pulmonary diagnosis 0.823** 0.053 0.000 0.834** 0.064 0.000 Cancer diagnosis 0.401** 0.115 0.001 0.695** 0.103 0.000 **Significant at 0.05 level

PAGE 100

100 Table 6 10 Bootstrapping prediction of expenditure change after three part model ( GLM ) N Mean S.D. 95%CI Total e xpenditures Depression general 1000 46.21 1161.41 ( 487.20, 967.06)) Pulmonary diagnosis 1000 265.00 1086.23 ( 391.34, 925.66) Cancer diagnosis 1000 979.70 1788.33 ( 1088.26, 2746.94) Medical e xpenditures Depression general 1000 130.04 1103.48 ( 2107.51, 684.36) Pulmonary diagnosis 1000 547.03 1247.29 ( 1040.33, 1742.27) Cancer diagnosis 1000 1344.78 194.24 ( 1825.03, 7134.67) Drug e xpenditures Depression general 1000 56.97** 336.14 ( 501.202, 56.364) Pulmonary diagnosis 1000 107.45** 439.18 ( 970.617, 28.874) Cancer diagnosis 1000 296.01 403.31 ( 917.411, 330.244) **Significant at 0.05 level Table 6 11 Proportion of each t ype of c hange in medical expenditures by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Zero e xpenditures 12.74 6.51 9.79 13.12 Non z ero e xpenditures 87.26 93.49 90.21 86.88 Total 100.00 100.00 100.00 100.00 Table 6 12 Proportion of each t ype of c hange in medical expenditures by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Decreased 46.38 43.12 44.85 45.50 No change 6.15 0.50 0 6.80 Increased 47.47 56.38 55.15 47.69 Total 100.00 100.00 100.00 100.00

PAGE 101

101 Table 6 13 Proportion of each t ype of c hange in drug expenditures by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Zero e xpenditures 7.92 6.17 9.79 11.61 Non z ero e xpenditures 92.08 93.83 90.21 88.39 Total 100.00 100.00 100.00 100.00 Table 6 14 Proportion of each t ype of c hange in drug expenditures by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Decreased 42.04 41.62 40.21 44.31 No change 1.33 0.17 0 5.29 Increased 56.63 58.22 59.79 50.40 Total 100.00 100.00 100.00 100.00 Table 6 15 Negative binomial regression predicting amount of change in utilization with treatment Negative binomial results IRR s.e. p value Emergency r oom v isits No treatment Reference Depression general 1.174** 0.030 0.000 Pulmonary diagnosis 1.229** 0.056 0.000 Cancer diagnosis 1.213** 0.104 0.025 Inpatient v isits No treatment Reference Depression general 1.402** 0.209 0.024 Pulmonary diagnosis 1.312 0.190 0.062 Cancer diagnosis 2.250** 0.588 0.002 Outpatient v isits No treatment Reference Depression general 1.452** 0.136 0.000 Pulmonary diagnosis 1.427** 0.250 0.043 Cancer diagnosis 2.293** 0.491 0.000 Office b ased p rovider v isits No treatment Reference Depression general 1.882** 0.108 0.000 Pulmonary diagnosis 2.068** 0.082 0.000 Cancer diagnosis 1.959** 0.109 0.000 **Significant at 0.05 level

PAGE 102

102 Table 6 16 Negative binomial regression predicting amount of visits for samples with and without treatment Predicted v isits for t reatment s ample Predicted v isits for s ample without t reatment N Mean SD N Mean SD Emergency r oom v isits Depression general 2425 0.720 0.349 3418 0.845 0.492 Pulmonary diagnosis 1171 1.107 0.526 4672 0.716 0.382 Cancer diagnosis 176 0.980 0.513 5667 0.787 0.439 Inpatient v isits Depression general 2425 2.784 3.582 3418 2.661 4.373 Pulmonary diagnosis 1171 3.058 3.662 4672 2.625 4.154 Cancer diagnosis 176 5.975 7.461 5667 2.611 3.869 Outpatient v isits Depression general 2425 2.001 1.156 3418 2.219 1.884 Pulmonary diagnosis 1171 2.998 1.609 4672 1.911 1.555 Cancer diagnosis 176 5.205 2.714 5667 2.033 1.481 Office b ased p rovider v isits Depression general 2425 23.072 7.473 3418 20.415 13.185 10.140 Pulmonary diagnosis 1171 31.600 9.709 4672 18.990 Cancer diagnosis 176 30.907 8.945 5667 21.226 11.190 Table 6 17 Proportion of each t ype of c hange in emergency room visits by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Decreased 17.44 18.02 20.10 16.92 No change 60.25 53.96 54.12 59.94 Increased 22.31 28.02 25.77 23.14 Total 100.00 100.00 100.00 100.00

PAGE 103

103 Table 6 18 Multinomial logit regression predicting probability of experiencing each type of change in utilization ( part I of three part model ) Relative risk of having negative change vs. no change Relative risk of having positive change vs. no change Coef. s.e. p value Coef. s.e. p value Emergency r oom v isits No treatment Reference Depression general 0.265** 0.078 0.001 0.259** 0.096 0.007 Pulmonary diagnosis 0.072 0.111 0.517 0.189 0.117 0.106 Cancer diagnosis 0.261 0.156 0.094 0.049 0.171 0.773 Inpatient v isits No treatment Reference Depression general 0.220** 0.085 0.010 0.267 0.137 0.053 Pulmonary diagnosis Cancer diagnosis 0.320 0.545** 0.170 0.188 0.061 0.004 0.494** 0.675** 0.196 0.223 0.012 0.003 Outpatient v isits No treatment Reference Depression general 0.314** 0.068 0.000 0.311** 0.058 0.000 Pulmonary diagnosis 0.391** 0.105 0.000 0.320** 0.084 0.000 Cancer diagnosis 0.906** 0.196 0.000 1.005** 0.190 0.000 Office b ased p rovider v isits No treatment Reference Depression general 0.346** 0.051 0.000 1.095** 0.113 0.000 Pulmonary diagnosis 0.474** 0.096 0.000 1.066** 0.198 0.000 Cancer diagnosis 0.518** 0.127 0.000 1.743** 0.250 0.000 **Significant at 0.05 level

PAGE 104

104 Table 6 19 Negative binomial regression predicting amount of change in utilization given negative or positive change ( parts II and III of three part model) Amount of negative c hange Amount of positive change IRR s.e. p value IRR s.e. p value Emergency r oom v isits No treatment Reference Depression general 0.915 0.047 0.086 1.008 0.043 0.842 Pulmonary diagnosis 0.961 0.028 0.175 1.083** 0.044 0.049 Cancer diagnosis 1.033 0.050 0.505 0.906 0.083 0.283 Inpatient v isits No treatment Reference Depression general 1.201 0.113 0.053 0.854 0.153 0.380 Pulmonary diagnosis 0.824 0.169 0.347 0.951 0.150 0.750 Cancer diagnosis 0.830 0.118 0.191 1.524 0.481 0.182 Outpatient v isits No treatment Reference Depression general 0.706** 0.917 0.008 1.554** 0.154 0.000 Pulmonary diagnosis 0.945 0.251 0.830 1.245 0.174 0.117 Cancer diagnosis 1.230 0.276 0.358 1.144 0.131 0.242 Office b ased p rovider v isits No treatment Reference Depression general 1.243** 0.105 0.011 1.352** 0.088 0.000 Pulmonary diagnosis 1.465** 0.184 0.003 1.476** 0.067 0.000 Cancer diagnosis 1.426** 0.137 0.000 1.564** 0.283 0.014 **Significant at 0.05 level

PAGE 105

105 Table 6 20 Bootstrapping prediction of expenditure change after three part model ( GLM ) N Mean S.D. 95%CI Emergency r oom v isits Depression general 1000 .0796 0.137 ( 0.009, 0.114) Pulmonary diagnosis 1000 0.0328 0.129 ( 0.035, 0.134) Cancer diagnosis 1000 0.069 0.135 ( 0.218, 0.079) Inpatient v isits Depression general 1000 0.453 1.136 ( 0.546, 0.354) Pulmonary diagnosis 1000 0.733** 0.941 (0.084, 1.324) Cancer diagnosis 1000 2.526** 2.418 (0.354, 5.461) Outpatient v isits Depression general 1000 0.271 0.327 ( 0.033, 0.567) Pulmonary diagnosis 1000 0.232 0.234 ( 0.617, 0.167) Cancer diagnosis 1000 0.423 0.341 ( 1.493, 0.646) Office b ased p rovider v isits Depression general 1000 0.452 1.475 ( 0.257, 1.068) Pulmonary diagnosis 1000 0.827 1.670 ( 0.226, 1.869) Cancer diagnosis 1000 0.578 1.823 ( 2.625, 3.753) **Significant at 0.05 level Table 6 21 Proportion of c hange in inpatient hospital visits by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Decreased 11.21 13.01 16.49 11.15 No change 70.18 64.30 55.15 69.84 Increased 18.61 22.69 28.35 19.01 Total 100.00 100.00 100.00 100.00

PAGE 106

106 Table 6 22 Proportion of c hange in outpatient hospital visits by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Decreased 17.64 23.35 25.77 17.90 No change 58.32 49.29 32.47 58.15 Increased 24.04 27.36 41.75 23.95 Total 100.00 100.00 100.00 100.00 Table 6 23 Proportion of c hange in office based provider visits by treatment group (weighted) Depression g eneral ( % ) Pulmonary d iagnosis ( % ) Cancer d iagnosis ( % ) Total ( % ) Decreased 41.40 39.12 39.69 41.61 No change 7.68 6.26 3.61 10.77 Increased 50.92 54.63 56.70 47.63 Total 100.00 100.00 100.00 100.00

PAGE 107

107 Table 6 24. Work sample d emographic characteristics of study samples by treatment group Variables Depression g eneral t reatment ( N= 1 703 ) % Pulmonary c omorbid t reatment ( N= 681 ) % Cancer c omorbid t reatment ( N= 98 ) % No t reatment (N=1 399 ) % Total ( N= 3930 ) % Chi2 test p value Age 18 24 4.99 3.38 4.08 9.86 6.36 0.000 25 44 45.63 46.99 25.51 51.89 47.33 45 64 48.85 48.90 66.33 37.67 45.65 65 0.53 0.73 4.08 0.57 0.66 Gender Female 71.34 77.53 69.39 61.33 68.83 0.000 Male 28.66 22.47 30.61 38.67 31.17 Race W hite 91.60 87.67 89.80 84.06 88.24 0.00 0 Black 5.05 6.61 0 10.08 6.97 Asian 1.29 1.17 6.12 2.14 1.68 Other 2.06 4.55 4.08 3.72 3.11 Ethnicity Hispanic 10.16 8.52 6.12 24.37 14.76 0.000 Non Hispanic 89.84 91.48 93.88 75.63 85.24 Poverty c ategory Poor/ n egative 4.87 5.29 1.02 10.44 6.77 0.000 Near p oor 4.29 1.76 0 4.79 3.92 Low i ncome 10.16 9.84 17.35 18.16 13.26 Middle i ncome 32.06 37.74 23.47 34.60 33.79 High i ncome 48.62 45.37 58.16 32.02 42.26 Insurance s tatus 0.000 Uninsured 7.81 5.43 10.20 19.87 11.76 Intermittent 9.28 6.90 5.10 12.65 9.95 Insured 82.91 87.67 84.69 67.48 78.30 Physical h ealth s tatus 0.000 Excellent 14.86 9.25 10.20 14.80 13.61 Very g ood 33.59 31.86 43.88 30.38 32.16 Good 36.76 42.00 27.55 38.46 38.45 Fair 10.63 13.07 14.29 12.79 11.93 Poor 4.17 3.82 4.08 3.57 3.84 Mental h ealth s tatus Excellent 13.68 13.07 13.27 17.87 14.94 0.0 16 Very g ood 27.95 26.14 28.57 26.59 27.10 Good 42.10 43.61 41.84 39.31 41.55 Fair 12.98 15.42 14.29 13.30 13.59 Poor 3.29 1.76 2.04 2.93 2.82

PAGE 108

108 Table 6 24. Continued Variables Depression general treatment ( N= 1703 ) % Pulmonary comorbid treatment ( N= 681 ) % Cancer comorbid treatment ( N= 98 ) % No treatment (N=1399) % Total ( N= 3930) % Chi2 test p value Comorbidity count 1 54.26 0 0 43.46 38.98 0.000 2 45.74 36.71 55.10 41.67 42.39 3 0 63.29 44.90 14.30 17.71 4 0 0 0 0.57 0.92 Need help with IADL Yes 1.70 1.91 3.06 1.07 1.53 0. 264 No 98.30 98.09 96.94 98.93 98.47 Need help with ADL Yes 1.06 0.59 1.02 0.50 0.76 0. 413 No 98.94 99.41 98.98 99.50 99.24 Table 6 25. Central tendency of d emographic characteristics for work sample Variables Depression g eneral t reatment ( N= 1703 ) % Pulmonary c omorbid t reatment ( N= 681 ) % Cancer c omorbid t reatment ( N= 98 ) % No t reatment (N=1399) % Total ( N= 3930) % Age Average 43.61 43.96 48.68 40.67 42.83 Standard d eviation 10.68 10.26 10.99 11.50 11.07 Minimum 18 18 21 18 18 Maximum 68 65 65 65 65 Years of e ducation Average 13.33 13.53 14.35 12.64 13.16 Standard d eviation 2.70 2.67 2.19 2.96 2.81 Minimum 0 0 9 0 0 Maximum 17 17 17 17 17 Income Average 39388.03 37807.12 41485.37 32315.13 36711.43 Standard d eviation 29491.63 25318.92 36145.71 26431.24 28086.65 Minimum 19010 333 28961 90050 90050 Maximum 182836 171953 145489 163684 182836 Comorbidity c ount Average 1.46 2.63 2.45 1.72 1.81 Standard d eviation 0.50 0.48 0.50 0.72 0.75 Minimum 1 2 2 1 1 Maximum 2 3 3 4 4

PAGE 109

109 Table 6 2 6 Negative binomial regression predicting amount of change in work absenteeism rates with treatment Negative binomial regression, weighted, no selection correction Poisson regression, unweighted, with selection correction IRR s.e. p value Coef. s.e. p value No treatment Reference Depression general 1.344 0.170 0.021 0.277 0.105 0.009 Pulmonary diagnosis 1.101 0.143 0.461 0.072 0.117 0.538 Cancer diagnosis 1.384 0.379 0.237 0.296 0.285 0.300 Table 6 27. L ogit regression predicting probability of experiencing each type of change in expenditures when expenditures added over time ( part I of two part model ) Relative risk of having positive change vs. no change Coef. s.e. p value Total Expenditures Pulmonary diagnosis 0.340** 0.102 0.001 Cancer diagnosis 0.587* 0.306 0.060 Medical Expenditures Pulmonary diagnosis Cancer diagnosis 0.488** 0.619 0.199 0.328 0.016 0.064 Drug Expenditures Pulmonary diagnosis 0.429** 0.217 0.050 Cancer diagnosis 0.619 0.328 0.064 **Significant at 0.05 level Table 6 28. Part II of two part model predicting amount of change in expenditures for medical treated and untreated groups when expenditures added over time Amount of change Coef. s.e. p value Total Expenditures Pulmonary diagnosis 0.402** 0.069 0.000 Cancer diagnosis 0.117 0.110 0.291 Medical Expenditures Pulmonary diagnosis 0.198** 0.073 0.008 Cancer diagnosis 0.007 0.136 0.962 Drug Expenditures Pulmonary diagnosis 0.685** 0.059 0.000 Cancer diagnosis 0.344 0.223 0.129 **Significant at 0.05 level

PAGE 110

110 Table 6 29. Bootstrapping prediction of expenditure change after two part model (OLS) for medical treated and untreated groups when expenditures added over time N Mean S.D. 95%CI Total Expenditures Pulmonary diagnosis 1000 18654.66** 11388.29 (13865.27, 14082.20) Cancer diagnosis 1000 34157.13** 14562.34 (17892.00, 18255.92) Medical Expenditures Pulmonary diagnosis 1000 19028.22** 13642.63 (11133.90, 11307.05) Cancer diagnosis 1000 32122.92** 16567.99 (2809.19, 3045.90) Drug Expenditures Pulmonary diagnosis 1000 5603.02** 3264.87 (1000.22, 1258.18) Cancer diagnosis 1000 7552.97** 2807.70 (1022.96, 1190.62) **Significant at 0.05 level Table 6 30. Part I of t hree part model for medical groups compared to no treated medical groups for difference in expenditures over time Relative risk of having negative change vs. no change Relative risk of having positive change vs. no change RRR s.e. p value RRR s.e. p value Total Expenditures Pulmonary diagnosis 0.630** 0.071 0.000 0.812 0.101 0.097 Cancer diagnosis 0.499 0.189 0.072 0.612 0.179 0.099 Medical Expenditures Pulmonary diagnosis 5.16e 07** 3.91e 07 0.000 6.49e 07** 0.041 0.000 Drug Expenditures Pulmonary diagnosis No participants in no change group to make estimation **Significant at 0.05 level

PAGE 111

111 Table 6 31. Parts II and III of three part model for medical treated groups compared to untreated medical groups for difference in expenditures over time Amount of negative change Amount of positive change Coef. s.e. p value Coef. s.e. p value Total expenditures Pulmonary diagnosis 0.088 0.136 0.520 0.260** 0.122 0.035 Cancer diagnosis 0.196 0.308 0.526 0.081 0.177 0.650 Medical expenditures Pulmonary diagnosis 0.044 0.140 0.756 0.175 0.118 0.142 Cancer diagnosis 0.039 0.187 0.835 0.115 0.203 0.574 Drug expenditures Pulmonary diagnosis 0.561** 0.055 0.000 0.482** 0.075 0.000 Cancer diagnosis 0.045 0.229 0.846 0.634** 0.179 0.001 **Significant at 0.05 level Table 6 32. Bootstrapping prediction of expenditure change after three part model comparing medical treated and untreated groups for difference in expenditures over time N Mean S.D. 95%CI Total Expenditures Pulmonary diagnosis 1000 385.55** 1359.14 ( 160.97, 32.38) Cancer diagnosis 1000 1750.35** 5698.06 ( 4000.71, 2839.37) Medical Expenditures Pulmonary diagnosis 1000 813.34** 2162.07 ( 1497.90, 1257.72) Cancer diagnosis 1000 3548.98** 10413.89 (278.57, 741.64) Drug Expenditures Pulmonary diagnosis 1000 56.17** 56.17 ( 898.05, 818.74) Cancer diagnosis 1000 123.28** 781.84 ( 940.98, 870.18) **Significant at 0.05 level

PAGE 112

112 Table 6 33. Negative binomial regression of medical treated and untreated groups predicting amount of change in utilization with treatment when utilization added over time Negative binomial results IRR s.e. p value Emergency Room Visits Pulmonary diagnosis 0.999 0.072 0.991 Cancer diagnosis 1.324** 0.186 0.050 Inpatient Visits Pulmonary diagnosis 1.133 0.261 0.588 Cancer diagnosis 1.133 0.235 0.550 Outpatient Visits Pulmonary diagnosis 1.351** 0.151 0.008 Cancer diagnosis 1.159 0.493 0.729 Office Based Provider Visits Pulmonary diagnosis 1.539** 0.077 0.000 Cancer diagnosis 1.321** 0.130 0.006 **Significant at 0.05 level Table 6 34. P art I of three part model for medical treated and untreated groups for difference in utilization over time Relative risk of having negative change vs. no change Relative risk of having positive change vs. no change Coef. s.e. p value Coef. s.e. p value Emergency Room Visits Pulmonary diagnosis 0.394** 0.200 0.049 0.309 0.181 0.087 Cancer diagnosis 0.021 0.352 0.953 0.152 0.349 0.663 Inpatient Visits Pulmonary diagnosis Cancer diagnosis 0.106 0. 458 0.222 0. 472 0.631 0. 332 0.222 1.577 ** 0.190 0. 403 0.244 0. 000 Outpatient Visits Pulmonary diagnosis 0.242 0.178 0.175 0.265 0.176 0.133 Cancer diagnosis 0.243 0.389 0.532 0.170 0.355 0.631 Office Based Provider Visits Pulmonary diagnosis 0.249 0.148 0.091 0.616** 0.251 0.014 Cancer diagnosis 0.067 0.311 0.829 0.762 0.744 0.306 **Significant at 0.05 level

PAGE 113

113 Table 6 35. Parts II and III of three part model for medical treated and untreated groups for difference in utilization over time Amount of negative change Amount of positive change IRR s.e. p value IRR s.e. p value Emergency Room Visits Pulmonary diagnosis 0.9 86 0.0 95 0. 881 1.050 0.088 0.559 Cancer diagnosis 1.276** 0.142 0.028 1.509** 0.181 0.001 Inpatient Visits Pulmonary diagnosis 0.551** 0.145 0.023 1.386 0.274 0.100 Cancer diagnosis 0. 924 0.1 8 1 0. 685 1. 072 0. 352 0. 833 Outpatient Visits Pulmonary diagnosis 1.080 0.214 0.699 1.300 0.253 0.178 Cancer diagnosis 0.865 0.394 0.750 1.411 0.276 0.078 Office Based Provider Visits Pulmonary diagnosis 1.229 0.147 0.086 1.232** 0.125 0.040 Cancer diagnosis 1.472** 0.285 0.046 1.192 0.227 0.356 **Significant at 0.05 level Table 6 36. Three part model bootstrapping prediction for medical treated and untreated groups for difference in utilization over time N Mean S.D. 95%CI Emergency Room Visits Pulmonary diagnosis 1000 0.161 ** 0.281 ( 0.190, 0.182 ) Cancer diagnosis 1000 0.109** 0.501 ( 1.548, 1.137) Inpatient Visits Pulmonary diagnosis 1000 0.451 2.044 ( 0.047, 0.042) Cancer diagnosis 1000 2.157 ** 8.136 ( 1.405, 1.051 ) Outpatient Visits Pulmonary diagnosis 1000 0.114** 0.791 ( 0.561, 0.494) Cancer diagnosis 1000 0.015 ** 4.133 ( 1.701, 1.448 ) Office Based Provider Visits Pulmonary diagnosis 1000 1.258 2.411 ( 0.015, 0.142) Cancer diagnosis 1000 0.647** 5.578 ( 1.119, 0.740) **Significant at 0.05 level

PAGE 114

114 Table 6 37. Negative binomial regression for medical treated and untreated groups predicting amount of change in work absenteeism rates with treatment Negative binomial regression, weighted, no selection correction Poisson regression, unweighted, with selection correction IRR s.e. p value Coef. s.e. p value Pulmonary diagnosis 1.006 0.146 0.966 0.005 0.115 0.966 Cancer diagnosis 1.070 0.361 0.843 0.102 0.291 0.726

PAGE 115

115 CHAPTER 7 DISCUSSION Overview This study hypothesized a decrease in total and medical expenditures for individuals with a comorbid medical condition (pulmonary or cancer) who sought mental health treatment. It was also hypothesized that the depression general group that did not have a specific comorbidity of a pulmonary or cancer diagnosis (but may have other comorbidities) would show either an increa se in expenditures or no change in expenditures over time. In this study, it was suggested that isolating individuals with expensive chr onic medical conditions would reveal a medi cal cost offset effect that is otherwise mask ed when individuals with varying risk factors are pooled together. The medical cost offset hypothesis that mental health treatment can reduce or prevent usual cost to the health care system would be supported if total or medical expenditures are reduced for groups who seek mental health treatment. Analysis of m edical expenditures is critical to the analysis of medical co st offset. Medical expenditures excludes the cost of psychological care. Observing a reduction in medical expenditures demonstrates a medical cost offset effect. P rescription drug expenditures were examined because increased care is thought to increase cost and mental health care was not exclu ded from this possibility. It was hypothesized that there would be either no difference or an increase in prescription drug expenditures with mental health treatment. This study also examined whether mental health treatment affects health care utilization Examination of health care utilization in light of total or medical expenditures should shed some light on which factors drive increased or decreased cost. In t his study it was hypothesized that there would be a reduction in utilization as a medical cost offset effect could be observed through several different medical services.

PAGE 116

116 This study adopted a broad perspective on the potential benefits of mental health treatment. Work absenteeism was also examined to add a perspective beyond the health care system and provide a broader view of societal and personal cost as a result of missed days from work. The following discussion will summarize the findings of this study Summary and Interpretation of Findings Treatment groups compared to no treatment gene ral group Total expenditu res for each treatment group were higher overall compared to the no treatment general group Medical expenditures were higher for only the depression general group. These results suggested that in the depression general group, mo re care simply cost ed more. T he higher medical expenditures suggest that aggregating individuals with various comorbiditie s into one group to examine medical cost offset result s in a lack of an offset effect. For the pulmonary and cancer treatment groups mental health treatment added cost to the total expenditures, but when the cost of psychological care was partialed out, medical expenditures remained the same, suggesting that psychological care at minimum does not increase the use of other medical serv ices for these groups These results are contrary to a medical cost offset effect, supporting the notion that additional care simply costs more. Oth er r esults indicated when expenditures were added over time, the depression general g roup was more likely to have medical or drug expenditures, and the pulmonary group w as also more likely to have drug expenditures. The results for drug expenditures confirm the study hypothesis that there would be no change or an increase in drug expenditures with mental heal th treatment. With rising prices of pharmaceuticals and increased drug use in the population ( Scherer, 2004 ), this is not surprising. Furthermore, each group (depression general, pulmonary, and cancer) that had total, medical, and drug expenditures great er than zero had significantly higher expenditures than the no treatment group. Thus, those who utilize services tend to have

PAGE 117

117 rising cost with mental health treatment. Post hoc analyses reveal that the main culprit of rising total cost seemed to be antid epressant medication, as opposed to psychotherapy. Once again, this is likely due to the pricing and amount of medication use When the change in expendi tures over two time periods was examined, the results revealed that each treatment group exhibited a c hange (positive, negative, or both) with mental health treatment, indicating a notable impact of mental health treatment on expenditure trends over time Isolating only those with negative (decrease) or positive (increase) change, each treatment group wit h a negative change showed a significant decrease in total, medical, and drug expenditures over time compared to no treatment and there was a significant increase for each group with a positive change in expenditures, except for the depression general and pulmonary groups with respect to medical expenditures. In general those who have a change in expenditures over time tend to show a larger magnitude of change when mental health care is added to the mix of medical services. It may be the case that these individuals have characteristics or unobserved factors that drive this change over time and the seeking of mental health treatment may simply be a reflection of these factors. The exception to mental health treatment impact ing change was for the depression general and pulmonary treatment groups that show ed an increase in medical expenditures over time Taking into account the result that individuals with depression typically have higher expenditures, it appears that for the depression genera l and pulmonary groups, mental health treatment may be preventing an increase in medical expenditures. This conjecture is strengthened by the overall results (adding negative, positive, and no change groups together) indicating that there is no significan t change in total, medical, or drug expenditures over time, positive or negative, with mental health treatment. T he study time

PAGE 118

118 frame and examining change with only two time points may have prevented the detection of an overall change. One consistent find ing is that treated cancer patients who had a reduction in total, medical, or drug expenditures over time had the greatest reduction in expenditures compared to the other treated groups. This suggests that particular medical groups may show differential b enefit from mental health treatment and further suggests that teasing apart different medical conditions may be beneficial to future cost offset studies. Breaki ng down type of medical service total utilization of emergency room visits, inpatient visits, outpatient visits, and office based provider visits tends to be higher for each treatment group than the no treat ment group. In the context of the study result that total expenditures are greater with mental health treatment, it appears that each of thes e utilization measur es contribute to the increase. When examining change in emergency room visits over time, mental health treatment had an impact on change for each treatment group for each measure of utilization, except for emergency room visits. That is, the pulmonary and cancer groups had no change in emergency room visits over time However emergency room visits also did not show an overall difference in visits for those receiving mental health treatment. It appears that despite no difference in overall change in expenditures the treated groups with a pulmonary or cancer diagn osis showed significant change in inpatient hospital stays overall. This is not surprising because these comorbid medical conditions are among the top five most expensive medical conditions ( Soni, 2007 ) and their medical needs may require inpatient services. For the remaining health care utilization measures, outpatient hospital visits and office based provider visits, there was no overall difference in visit counts over t ime Because mental health treatment, particularly psychotherapy, is typically in an outpatient or office based setting,

PAGE 119

119 the lack of an increase in these measures may suggest a potential (but statistically unobserved) medical cost offset effect. Finally with respect to work absenteeism, it appears that the treated aggregated depression general group had more days missed from work than their non treated counterparts and those with a comorbid pulmonary or cancer diagnosis. This suggests that not only may aggregating depressed individuals into a group mask a medical cost offset effect and even show an increase in total expenditures, there may be a corresponding effect o n work absenteeism. That is, collapsing everyone who is depressed into one group regard less of other medical or psychological conditions may inadvertently create an impression that there is no societal benefit (from a cost perspective) of mental health treatment. It is noteworthy that the expensive medically comorbid groups with a pulmonary or cancer diagnosis showed no significant increase in medical expenditures nor work absenteeism counts. It may be that the benefits of mental health treatment on health care expenditures and work related outcomes are most robust for those with chronic me dical conditions. Medical treatment groups compared to medical no treatment groups When the pulmonary and cancer treatment groups were compared to the pulmonary and cancer no treatment groups, overall results argue against a medical cost offset effect. Fo r both medical groups, every category of expenditures (total, medical, and drug) was higher overall than the corresponding medical no treatment group when expenditures were totalled over time. When change in expenditures was examined, each type of expendi tures showed a significant increase over time for the treatment groups compared to the no treatment medical groups with the exception of the cancer treatment group that showed a significant decrease in drug expendit ures relative to the no treatment cancer group

PAGE 120

120 In terms of utilization added over time, there was a significant increase in utilization for emergency room visits for the cancer treated group and there was a significant overall increase in emergency visits o ver time for the cancer treatment group compared to the no treatment cancer group which suggests that individuals with cancer who seek out mental health treatment may also be in more advanced stages with their cancer. Other results show an increase in ou tpatient visits for the pulmonary treated group relative to the no treatment pulmonary group and an increase office based provider visits for both medical treated groups compared to the no treatment medical groups This result is consistent with the fact that mental health treatment is most often in these settings. When change in utilization was examined over time, the cancer treatment group had higher utilization rates for each utilization measure compared to the no treatment cancer group. Thus, it appe ars that the increase in total expenditures with treatment for the cancer group was due to an increase in emergency room, inpatient, outpatient, and office based provider visits. For the pulmonary treatment group, only emergency room visits were higher th an the no treatment pulmonary group, whereas outpatient visits were lower for the treatment group than the no treatment group. There was no difference in inpatient or office based provider visits between the pulmonary treatment and no treatment groups. T he pulmonary group utilization results suggest that an increase in emergency room visits is driving the increased cost and mental health treatment may be shifting care away from outpatient settings. Study Limitations Despite the longitudinal nature of this study, the two and a half year timeframe may not be sufficient to examine real change over time. Simon and Katzelnick (1997) suggested a cost offset effect may require several y ears before it can be observed Furthermore, the length of

PAGE 121

121 mental health tre atment prior to participation in this study could not be determined. It is possible Second, the study results that showed different patterns among those who experienced a negative no change, or positive change in expenditures suggest that the groups may have fundamental difference s not captured by this study For example, negative change individuals tend ed to have an increased negative change with mental health treatmen t, whereas individuals with positive change tend to have an increased positive change with treatment. This finding might also suggest differences in the quality of treatment received. The quality of mental treatment could not be determined in this study Quality of treatment could help determine the relationship between treatment effectiveness and health care expenditures or utilization. Third, identifying individuals using ICD 9 codes has l imited reliability and validity, particularly due to participa nts reporting their own diagnoses. On a related note, the severity of depression could not be determined with accuracy in this study which reduced precision in the analysis and may have masked any nuances among differing degrees of severity. Lastly, the re was a reduction in precision of the analysis because antidepressant medication was collapsed with psychotherapy to create the mental health treatment variable. Implications Mental health treatment and mental health disorders have historically carried a social stigma. With the biomedical model dominating health care, mental health is traditionally underemphasized in health care. The medical cost offset effect, if demonstrated, would be a strong argument for increased attention to and treatment of mental health disorders. Particularly with the demonstrated cost and impairment of depression comparable to other medical disorders (e.g., Murray & Lopez, 1996) mental health care is increasingly given more importance in recent years. Despite the diffi culty of demonstrating a medical cost offset effect,

PAGE 122

122 the data quantifying the negative impact of depression on a personal and societal level cannot be ignored. The medical cost offset effect is a particularly compelling concept for mental health profession als The medical cost offset effect provides a way to quantify the value of mental health care. If the medical cost offset effect were a robust and consistent phenomenon, psychologists and other mental health professionals could use the effect to justify payment for services to third party payers and pol icymakers. The cost offset also provides a compelling argument for the integration of phsyical and mental health care. However, this study along with most recent published studies have not supported the existence of a medical cost offset effect. The medical cost offset effect may have dwindled with the institution of cost control efforts during the last decade of the 20 th century. The managed care era spanned from approximately 1980 to 200 0; however, t he use of health maintenance organizations increased dramatically in the 1990s (Robinson, 2004). During this time, h ealth care spending was reduced by reductions in fees and payment rates These changes reduced the opportunity for cost savings from other methods. Despite the difficulties with demonstrating a medical cost offset effect, it remains a noteworthy phenomenon to examine because of consistent increases in the cost of medical care and a potentially changing political climate in the near future. As this study demonstrated with differential results for individuals with comorbid medical conditions and the aggregated general depression group future studies may benefit from increasing the specificity of the populations under study. That is, the me dical cost offset effect is likely to be the most applicable when one can determine the specific comorbid medical conditions that interact to drive costs. The challenge of defining the nature of this interaction is great. In this study, h owever, when the

PAGE 123

123 medical groups were compared to the no treatment medical groups (instead of the general no treatment group), the medical offset did not matreialize suggesting a medical cost offset effect for expensive medical groups may not be meaningful. In fact, this study strengthens the idea that more care simply costs more and individuals with a chronic medical condition who seek mental health care are no exception to this. The search for a medical cost offset may focus outcome researchers on the wrong issue. The m edical cost offset justifies treatmen t based on cost savings alone Treatment should be validated on the effectiveness of the intervention, improved quality of life, and functioning. There are serious implications for health care if each treatment must d emonstrate a cost offset in order to be deemed valuable. If insurance companies or policymakers focus solely on the bottom line, the quality of health care will likely suffer. Thus, it is important to continually keep patient quality of life as an import ant outcome and marker of success. Cos t effectiveness analyses c omparing interventions in terms of a cost to effectiveness ratio, and cost benefit analysis, which estimates value by weighing the benefits of a treatment to the cost of the treatment (Santer re & Neun, 2007) is more in line with the goal of quality health care. Future studies should focus more on effectiveness and benefit to the patient. In this study, medication was a significant driver of health care expenditures. The proportion of depressed individuals on antidepressant medication in the sample was higher than the proportion undergoing psychotherapy. Previous research has demonstrated that the combination of psychotherapy and antidepressant medication is superior to either psychotherapy or medication al one (Freedland et al., 2004). This combined approach was not most frequent treatment approach among individuals in this study

PAGE 124

124 T he reduction in work days was also examined in this study Future studies should expand the societal benefit analysis even further to include areas such as cost to other public programs, the losses and difficulties experienced by family of individuals needing mental health care, lost income, and dimi n i shed worforce productivity when at work. Recent studies have failed to demonstrate the medical cost offset effect Despite this failure, work to estimate medical cost offset effects should not be abandonned On the contrary, it is important to recognize the magnitude of change in the financing health services has far outpaced research examining the interaction among clinical services and cost. More efforts to better understand these relationships are neede d to update the literature and examine trends over time. Such efforts must also give weight to the assessment of quality of care and how quality compares to the cost of care.

PAGE 125

125 L IST OF R EFERENCES Adler, D. A., McLaughlin, T. J., Rogers, W. H., Chang, H., La pitsky, L., & Lerner, D. (2006). Job performance deficits due to depression. American Journal of Psychiatry, 163 1569 1576. Anderson, R. N. (2002). Deaths: Leading causes for 2000. National vital statistics report, 50 (16). Hyattsville, MD: National Cen ter for Health Statistics. Azocar, F., McCarter, L. M., Cuffel, B. J., & Croghan, T. W. (2003). Patterns of medical resource and psychotropic medicine use among adult depressed managed behavioral health patients. The Journal of Behavioral Health Services & Research, 31 (1), 26 37. Beckstead, J. W. (2003). Attitudes accentuate attributes in social judgment. Journal of Social Psychology, 143 185 201. Journal of th e Mental Health Administration, 23 7 15. Borger C., Smith, S., Truffer, C., Keehan, S., Sisko, A., Poisal, A., & Clemens, M. K (2006). Health spending projections through 2015: changes on the horizon. Health Affairs Web Exclusive 25 (2), 61 73. Brave man, P., & Tarimo, E. (2002). Social inequalities in health within countries: Not only an issue for affluent nations. Social Science and Medicine, 54 (11), 1621 1635. Carbone, L. A., Orav, E. J., Fricchione, G. L., & Borus, J. F. (2000). Psychiatric sympt oms and medical utilization in primary care patients. Psychosomatics, 41 (6), 512 518. Chiles, J. A., Lambert, M. J., & Hatch, A. L. (1999). The impact of psychological interventions on medical cost offset: A meta analytic review. Clinical Psychology: Science and Practice, 6 (2), 204 220. Druss, B. G., Rosenheck, R. A., & Sledge, W. H. (2000). Health and disability costs of depressive illness in a major U.S. corporation. American Journal of Psychiatry, 157 (8), 1274 1278. Duan, N., Manning, W. G., Morri s, C. N., & Newhouse, J. P. (1984). Choosing between the sample selection model and the multi part model. Journal of Business & Economic Statistics, 2 (3), 283 289. Follette, W. T., & Cummings, N. A. (1968). Psychiatric services and medical utilization in a prepaid health plan setting. Medical Care, 5 25 35. Freedland, K. E., Carney, R. M., & Skala, J. A. (2004). Depression and mood disorders. In L. J. Haas (Ed.) Handbook of Primary Care Psychology (5 19). New York, NY: Oxford University Press.

PAGE 126

126 Friedman, R., Sobel, D., Myers, P., Caudill, M., & Benson, H. (1995). Behavioral medicine, clinical health psychology, and cost offset. Health Psychology, 14 (6), 509 518. Goetzel, R. Z., Anderson, D. R., Whitmer, W., et al. (1998). The relationship between modifia ble health risks and health care expenditures: An analysis of the multi employer HERO health risk and cost database. Journal of Occupational Environmental Medicine, 40 (10), 843 854. Gujarati, D. (2003). Basic Econometrics (4 th Ed.). New York: McGraw Hill. Hasin, D. S., Goodwin, R. D., Stinson, F. S., & Grant, B. F. (2005). Epidemiology of major depressive disorder. Archives of General Psychiatry, 62 1097 1106. Haveman, R., & Wolfe, B. (2000). The economics of disability and disability policy. In A. J. Culyer & J. P. Newhouse (Eds.). Handbook of Health Economics, Volume 1 (995 1051) Amsterdam, The Netherlands: Elsevier Science B.V. Hildebrand, D., Laing, M., & Rosenthal, A. (1977). Prediction Analysis of Cross Classifications. New York: John Wiley. Hi mmelhoch, S., Weller, W. E., Wu, A. W., et al. (2004). Chronic medical illness, depression and use of acute medical services among Medicare beneficiaries. Medical Care,42 (6), 512 521. Hunsley, J. (2003). Cost effectiveness and medical cost offset considerations in psychological service provision. Canadian Psychology, 44 (1), 61 73. Iezzoni, L. I. (2003). Risk adjustment for measuring health care outcomes 3 rd edition. Chicago, IL: Health Administration Press. Keeler, E. B., Kahn, K. L., Draper, D. Sherwood, M. J., Rubenstein, L. V., Reinisch, E. J., Kosecoff, J., & Brook, R. H. (1990). Charges in sicknes at admission following introduction of the prospective payment system. Journal of the American Medical Association, 264 (15), 1962 1968. Kessler, R. C., Barber, C., Birnbaum, H. G., Frank, R. G., Greenberg, P. E., Rose, R. M., Simon, G. E., & Wang, P. (1999). Depression in the workplace: Effects on short term disability. Health Affairs, 18 (5), 163 171. Kessler, R. C., Zhao, S., Blazer, D. G., & S wartz, M. (1997). Prevalence, correlates, and course of minor depression and major depression in the national comorbidity survey. Journal of Affective Disorders, 45 19 30. Kroenke, K., Spitzer, R. L., & Williams, J. B. W. (2003). The patient health ques tionnaire 2: Validity of a two item depression screener. Medical Care, 41 (11), 1284 1292.

PAGE 127

127 Lamb, C. E., Ratner, R. H., Johnson, C. E., Ambegaonkar, A. J., Joshi, A. V., Day, D., Sampson, N., & Eng, B. (2006). Economic impact of workplace productivity lo sses due to allergic rhinitis compared with select medical conditions in the United States from an employer perspective. Current Medical Research and Opinion, 22 (6), 1203 1210. nary and cardiac patients with depression or anxiety. Lerner, D., Adler, D. A., Chang, H., Lapitsky, L., Hood, M. Y., Perissinotto, C., Reed, J., McLaughlin, T. J., Berndt, E. R., & Rogers, W. H. (2004). Unemployment, job retention, and productivity loss among employees with depression. Psychiatric Services, 55 (12), 1371 1378. Lofland, J. H., & Frick, K. D. (2006). Effect of health insurance on workplace absenteeism in the U.S. workforce. Journal of Occupational Environmental Medicine, 48 13 21. Lo Sasso, A. T., Rost, K., & Beck, A. (2006). Modeling the impact of enhanced depression treatment on workplace functioning and costs: A cost benefit approach. Medical Care, 44 (4), 352 358. Manning, W. G., Basu, A., & Mullahy, J. (2005). Generalized mode ling approaches to risk adjustment of skewed outcomes data. Journal of Health Economics, 24 (3), 465 488. Mashaw, J., Reno, V., Burkhauser, R., & Berkowitz, M. (Eds.). (1996). Disability, Work, and Cash Benefits. Kalamazoo, MI: Upjohn Institute for Employ ment Research. Mumford, E., Schlesinger, H. J., Glass, G. V., Patrick, C., & Cuerdon, T. (1984). A new look at evidence about reduced cost of medical utilization following mental health treatment. American Journal of Psychiatry, 141 1145 1158. Murray, C ., & Lopez, A. (Eds.) (1996). The Global Burden of Disease. Cambridge, MA: World Health Organization, Harvard University Press. Otto, M. W. (1999). Psychological interventions in the age of managed care: A commentary on medical cost offsets. Clinical Psyc hology: Science and Practice, 6 (2), 239 241. Pincus, H. A., Pechura, C. M., Elinson, L., & Pettit, A. R. (2001). Depression in primary care: Linking clinical and systems strategies. General Hospital Psychiatry, 23 311 318. Robert Wood Johnson Foundatio n (1996). Chronic Care in America: A 21st Century Challenge Princeton, New Jersey. Robinson, J. C. (2004). Reinvention of health insurnace in the consumer era. Journal of the American Medical Association, 291 (15), 1880 1886.

PAGE 128

128 Rost, K., Smith, J. L., & Dickinson, M. (2004). The effect of improving primary care depression management on employee absenteeism and productivity: A randomized trial. Medical Care, 42 (12), 1202 1210. Santerre, R. E., & Neun, S. P. (2007) Health Economics: Theories, Insights, a nd Industry Studies, 4 th edition. Mason, OH: South Western Scherer, F. M. (2004). The pharmaceutical industry Prices and progress. New England Jouirnal of Medicine, 351 (9), 927 932. Simon, G. E., & Katzelnick, D. J. (1997). Depression, use of medica l services and cost offset effects. Journal o f Psychosomatic Research, 42 (4), 333 344. Simon, G. E., Revicki, D., Heiligenstein, J., Grothaus, L., VonKorff, M., Katon, W. J., & Hylan, T. R. (2000). Recovery from depression, work productivity, and health care costs among primary care patients. General Hospital Psychiatry, 22 153 162. Simon, G. E., Von Korff, M., & Barlow, W. (1995). Health care costs of primary are patients with recognized depression. Archives of General Psychiatry, 52 (10), 850 856. Soni, A. (2007). The five most costly conditions, 2000 and 2004: Estimates for the U.S. civilian noninstitutionalized population. Statistical Brief #167. Agency for Healthcare Research and Quality, Rockville, MD. Souetre, E., Lozet, H., & Cimarosti, I. (1997). Predicting factors for absenteeism in patients with major depressive disorders. European Journal of Epidemiology, 13 87 93. Staniec, J. F. O., & Webb, N. J. Utilization of infertility services: How much does money matter? Health Services Resea rch, 42 (3), 971 989. Sturm, R. (2001). Economic grand rounds: The myth of medical cost offset. Psychiatric Services, 52 738 740. Wang, P. S., Beck, A. L., Berglund, P., McKenas, D. K., Pronk, N. P., Simon, G. E., & Kessler, R. C. (2004). Effects of ma jor depression on moment in time work performance. American Journal of Psychiatry, 161 1885 1891. Wells, K. B., Sturm, R., Sherbourne, C. D., & Meredith, L. S. (1996). Caring for Depression. Cambridge, MA: Harvard University Press. Wolfe, B. L., & Haveman, R. (1990). Trends in the prevalence of work disability from 1962 to 1984, and their correlates. Milbank Quarterly, 68 (1), 53 80. Zhang, M., Rost, K. M., Fortney, J. C., & Smith, G. R. (1999). A community study of depression treatment and employm ent earnings. Psychiatric Services, 50 (9), 1209 1213.

PAGE 129

BIOGRAPHICAL SKETCH Andrea Lee, born in Vancouver, British Columbia, Canada, had a desire to become a psychologist at the age of 14. She itched to graduate high school in order to study psychology. She earned Burnaby, British Columbia in 2003. Over the years, she refined her interest in psychology to encompass clinical psychology and health psychology Her overwhelming desire to help others on an individual level was eventually supplemented by a desire to make a difference on a population level. This interest in population level changes was the beginning of her desire for knowledge in health policy. The obvious next step for her was to pursue a doctoral degree in psychology. She was delighted to discover that not only could she be trained superbly in the areas of clinical psychology and health psychology at the University of Florida, she would also be able to gain valuable exposure to health policy. Thus, the Department of Clinical and Health Psychology at the University of Florida became the obvious choice for her graduate studies. Since her matriculation at the University of Florida in 2004, she has gained the knowledge and skills that will allow her to attempt to make the kind of difference she has dreamed of since her teenage years. Prior to graduating with her doctoral degree, she will be heading to the University of Manitoba to complete a yea r long pre future and career, she hopes to positively contribute to the already outstanding reputation of the linical and Health Psychology.