IDENTIFYING PRIMARY CARE RELEVANT RISK FACTORS TO REDUCE READMISSIONS By DENNY FE GARCIA AGANA 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 2018
2018 Denny Fe Garcia Agana
Dedicated to my parents and Andre
4 ACKNOWLEDGMENTS I would like to thank the University of Florida Department of Community Health and Family Medicine and H. James Free Center for Primary Care Education and Innovation for giving me the opportunity to complete my PhD. The financial support, mentoring, guidance, and teamwork, have propelled my PhD training and my development into an independent researcher in primary care. I would specifically like to thank Dr. Striley (Chair) and Dr. Carek (Co Chair) who have provided ho urs of their limited time. Your constant support and open availability to help me with not only my dissertation but my personal growth, is greatly appreciated. I would also like to thank the rest of my committee members: Dr. Cook, Dr. Cruz Almeida, and Dr. Young. Thank you for the time you have spent working with me on my dissertation as well as future career goals. I would like to thank my friends in Gainesville who have collaborated with me on projects, have helped when I was stuck on certain statistica l methodologies, have been soundboards on educational goals, as well as have been social support in every way possible. I would also like to thank my parents and Andre who have been supportive in all of my endeavors. I would like to thank Sage who has been there to brighten my days when the PhD and dissertation seemed unachievable. I would also like to thank the University of Florida Department of Epi demiology administrative staff and faculty for the training, collaboration and assistance throughout the P hD process. Lastly, I would like to thank the patients of the Department of Community Health and Family Medicine. Without them, this dissertation would not have been possible.
5 TABLE OF CONTENTS page ACKNOWLEDG MENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ .......... 9 LIST OF ABBREVIATIONS ................................ ................................ ........................... 10 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 BACKGROUND AND SIGNIFICANCE ................................ ................................ ... 13 1.1 Epidemiology, Definition, and Importance of Readmissions ............................. 13 1.2 Characterizing Readmission Rates ................................ ................................ ... 14 1.3 Current Readmissions Predictive Tools ................................ ............................ 15 1.4 Importa nce of Social Determinants of Health ................................ .................... 16 1.5 Current Readmissions Initiatives for Minority Members ................................ .... 17 1.6 Gaps in Current Readmissions Measures and Predictive Models .................... 19 1.7 Theoretical Model ................................ ................................ ............................. 23 1.8 Explanation of the OHIS Rea dmissions Theoretical Model ............................... 24 2 DESIGN AND METHODS ................................ ................................ ....................... 29 2.1 Data Samples ................................ ................................ ................................ ... 29 2.2 Sampling and Recruitment ................................ ................................ ................ 29 2.3 Measures for Aims 1& 2 ................................ ................................ .................... 31 2.4 Approach Aim 1: Specified Measures for Aim 1 ................................ ................ 31 2.4.1 Readmissions ................................ ................................ .......................... 31 2.4.2 Variables for Factor Analysis ................................ ................................ ... 31 2.5 Approach Aim 1: Method of Analysis for Aim 1 ................................ ................. 31 2.6 A pproach Aim 2: Specified Measures for Aim 2 ................................ ................ 32 2.6.1 Outcome: Readmissions ................................ ................................ ......... 32 2.6.2 Covariates ................................ ................................ ............................... 33 2.7 Approach Aim 2: Method of Analysis for Aim 2 ................................ ................. 33 2.7.1 Univariate and Bivariate Analyses ................................ ........................... 33 2.7.2 Multino mial Analyses ................................ ................................ ............... 33 2.8 Measures for Aim 3 ................................ ................................ ........................... 33 2.9 Approach Aim 3: Specified Measures for Aim 3 ................................ ................ 33 2.9.1 Outcome: Readmissions ................................ ................................ ......... 33 2.9.2 Covariates ................................ ................................ ............................... 33 2.10 Approach Aim 3: Method of Analysis for Aim 3 ................................ ............... 34 2.10.1 Univariate and Bivariate Analyses ................................ ......................... 34
6 2.10.2 Multinomial Analyses ................................ ................................ ............. 34 2.11 Approach for Enrichi ng Findings with Patient Centered Data ......................... 34 2.12 Approach for Patient Centered Data: Methods of Recruitment and Analysis .. 34 2.13 Approach for Patient Centered Data: Interview Guide ................................ .... 35 3 READMISSIONS WITH PRIMARY CARE RELATED FACTORS .......................... 44 3.1 Bac kground ................................ ................................ ................................ ....... 44 3.2 Methods ................................ ................................ ................................ ............ 45 3.2.1 Study Overview ................................ ................................ ....................... 45 3.2.2 Statistical Analysis ................................ ................................ ................... 46 3.3 Results ................................ ................................ ................................ .............. 47 3.4 Conclusions ................................ ................................ ................................ ...... 48 4 PRIMARY CARE MODIFIABLE FACTORS OF THE 1 2 READMISSIONS POPULATION ................................ ................................ ................................ ......... 59 4.1 Background ................................ ................................ ................................ ....... 59 4.2 Methods ................................ ................................ ................................ ............ 61 4.2.1 Study Overview ................................ ................................ ....................... 61 4.2.2 Outpatient Behavior ................................ ................................ ................. 62 4.2.3 Health Status ................................ ................................ ........................... 62 4.2.4 Inpatient Factors ................................ ................................ ...................... 62 4.2.5 Social Determinants ................................ ................................ ................ 63 4.2.6 Statistical Analysis ................................ ................................ ................... 63 4.3 Results ................................ ................................ ................................ .............. 63 4.4 Conclusions ................................ ................................ ................................ ...... 65 5 IDENTIFYING THE 1 2 READMISSIONS POPULATION: A CROSS SECTIONAL ANALYSIS FROM THE NATIONWIDE READMISSIONS DATABASE ................................ ................................ ................................ ............. 74 5.1 Background ................................ ................................ ................................ ....... 74 5.2 Methods ................................ ................................ ................................ ............ 77 5.2.1 Study Overview ................................ ................................ ....................... 77 5.2.2 Social Determinants ................................ ................................ ................ 78 5.2.3 Health Status Factors ................................ ................................ .............. 79 5.2.4 Inpatient Factors ................................ ................................ ...................... 79 5.2.5 Statistical Analyses ................................ ................................ .................. 79 5.3 Results ................................ ................................ ................................ .............. 79 5.4 Conclusions ................................ ................................ ................................ ...... 82 6 CONCLUSION ................................ ................................ ................................ ........ 93 6.1 Main Fi ndings ................................ ................................ ................................ ... 96 6.2 Strengths and Limitations ................................ ................................ ............... 106 6.3 Future Research ................................ ................................ ............................. 108
7 APPENDIX A QUALITATIVE INTERVIEW GUIDE PAGE 1 ................................ ....................... 110 B QUALITATIVE INTERVIEW GUIDE PAGE 2 ................................ ....................... 111 C QUALITATIVE INTERVIEW GUIDE PAGE 3 ................................ ....................... 112 LIST OF REFERENCES ................................ ................................ ............................. 113 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 120
8 LIST OF TABLES Table page 2 1 Electronic medical record variable name, variable construct, and variable type Aim 1, and variable type in Aim 2 ................................ ............................... 37 2 2 Nationwide Readmissions Database (NRD) variable name and variable type ... 41 3 1 2016 Exploratory Factor Analysis (EFA) variables, hypothesized constructs, and variables types ................................ ................................ ............................. 51 3 2 2016 Exploratory Factor Analysis (EFA) eigenvalues ................................ ......... 54 3 3 2016 Exploratory Factor Analysis (EFA) factor pattern from the Promax rotation, decimals omitted ................................ ................................ ................... 56 3 4 2016 Exploratory Factor Analysis (EFA) factor structure, decimals omitte d ....... 57 4 1 Descriptive analysis of variables by readmissions group (0 readmissions, 1 2 readmissions only, or 3 or more readmissio ns) ................................ .................. 69 4 2 Multivariate analysis of significant variables from univariate analysis, predicting number of readmissions ................................ ................................ ..... 72 5 1 Weighted descriptive analysis of variables by readmissions group (0 readmissions, 1 2 readmissions only, or 3 or more readmissions) .................... 87 5 2 Multivariate analysis of significant variables from univariate analysis, predicting number of readmissions ................................ ................................ ..... 90
9 LIST OF FIGURES Figure page 1 1 Kaiser Family Foundation Social Determinants of Health ................................ ... 26 1 2 HIDI SDS factors included in risk adjusted readmissions rates .......................... 27 1 3 Theoretical model from the PROFILE study ................................ ....................... 27 1 4 Theoretical model: OHIS (Outpatient Behavior, Health Status, Inpatient Factors, Social Determinants) Readmissions ................................ ..................... 28 2 1 Flow cha rt for Aims 1 and 2 ................................ ................................ ................ 35 2 2 Flow chart for Aim 3 ................................ ................................ ............................ 36 3 1 2016 Exploratory Factor Analysis (EFA) scree plot ................................ ............ 55 3 2 2016 Exploratory Factor Analysis (EFA) factors and variables ........................... 58 5 1 Sample selection of the 2013 NRD ................................ ................................ ..... 87
10 LIST OF ABBREVIATIONS ACSC Ambulatory Care Sensitive Conditions AHRQ Agency for Healthcare Research and Quality CFA Confirmatory Factor Analysis CMS Centers for Medicare and Medicaid Services ED Emergency Department EFA Exploratory Factor Analysis EMR Electronic Medical Records ESI Emergency Severity Index HCUP Health Cost Utilization Project HOSPITAL Hemoglobin, discharge from an Oncology service, Sodium level, Procedure during the index admission, Index type of admission (urgent), number of Admissions during the last 12 months, and Length of stay LACE Length of stay, acuity of admission, Charleson comorbidity index, and number of Emergency Department visits in past 6 months LOS Length of stay NRD Nationwide Readmissions Database OHIS Outpatient behavior, Health status, Inpatient factors, and Social determinants PCP Primary Care Physician PROFILE Preventing early unplanned hospital readmission after critical illness SAS Statistical Analysis Software SID State Inpatient Databases
11 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 IDENTIFYING PRIMARY CARE RELEVANT RISK FACTORS TO REDUCE READMISSIONS By Denny Fe Garcia Agana August 2018 Chair: Catherine W. Striley Major: Epidemiology Readmissions are linked to poorer health outcomes for patients Additionally, the associated red uced payments and potential fines from the Inpatient Prospective Payment System (IPPS), are of great concern to health care systems 1 In FY 2017, $528 million was withheld. 1 I n FY 2018, 2,573 hospitals will face penalties 1 Studies have shown that a portion of readmissions is preventable, and that prevention c an occur at the level of primary care. Yet, primary care in reducing or preventing readmissions is not clear. 2 Before identifying th is role it is necessary to characterize patients who contribute most to the accumulation of readmissions. This population is the patients who have readmitted 1 2 times a year not the high utilizers who are relatively few in number T o understand the characteristics of this population, this dissertation aimed to: 1. Develop a readmissions predictive tool for primary care through factor analysis of outpatient behaviors, social determinants, inpati ent factors, and health status constr ucts 2. Characterize 1 2 readmissions population and predict rates of readmissions among a family medicine d epartment patients and 3. Characterize the 1 2 ambulatory care sensitive condition (ASCS) readmissions population using the 2013
12 Nationwide Readm issions Database (NRD). In a ddition, this dissertation included some qualitative interviews of modal patients to inform quantitative findings and to understand the roles of social support, the presence of pain, and reason for readmission throu gh the perspe ctive of patients. Data for Chapters 3 and 4 ca me from an academic Southeast USA Family electronic medical record Data for Chapte r 5 ca me from the Nationwide Readmissions Database (NRD), a nationally representative database fro m the Health Cost Utilization Project (HCUP). Data for the qualitative interviews ca me from recent readmitters of an academic Southeast USA Family Medicine department Factor analysis was used for Chapter 3 Uni variate bivariate, and multivariate analyse s were used for Chapters 4 and 5 SAS software, Version 9.4 was used for Chapter 3 5. NVivo was used to qualitatively analyze the interviews. This dissertation provides a characterization of the 1 2 readmissions per year population from a primary care clinic population and explores the role of outpatient behavior in readmissions
13 CHAPTER 1 B AC KGROUND AND SIGNIFICANCE 1.1 Epidemiology, Definition, and Importance of Readmissions According to C enters for M edicare and S ervices (CMS) data, about 1 in 5 Medicare patients are readmitted into the hospital within 30 days of discharge. The Affordable Care Act (ACA) added a section to the Social Security Act creating a Hospital Readmissions Reduction Program (HRRP) requiring CMS to reduce payments to Inpatient Prospective Payment System (IPPS) hospitals with excess readmissions. 3 In 2011 i n the US, there were approximately 3.3 million adult readmissions with an associated cost of $41.3 billion. 3 For Medicare and M edicaid patients alone, associated hospital costs due to readmissions totaled $5.1 billion. 3 Though not all hospitalizations, i ncluding readmissions, are unavoidable, various studies have shown that there is a proportion of readmissions that is preventable. 4 Analysis of 2005 claims data has shown that about three quarters of readmissions within 30 days were potentially preventable, equating to a n estimated $12 billion in Medicare spending. 4 Other studies have shown strategies that hospitals can enact to prevent unnecessary readmissions, yet this has proven to be difficult and not straightforward. 4 Therefore, efforts continue to reduce readmissions in both the n ational and lo cal arenas A readmission is defined as a re hospitalization occurring within 30 days of a patient being discharged from an initial hospital stay. Hospital systems are financially penalized for excess readmissions. Calculations for these financial penalties include the readmission payment adjustment amount and the readmission adjustment factor. Although these calculations have specific factors and criteria when calculated, hospitals are aimi ng for readmissions rates below the national level. Particularly with the
14 Southeast USA academic hospital at hand, in FY 2017 there was $754,767 lost due to the financial penalty for readmissions. 1 In FY 2018, there is an expected increased loss of $803,462. 1 With financial penalties in place, there has been both a local and national push to find ways to reduce readmissions rates. Hospitalizations are also linked to poor health outcomes. R ates of readmissions are used as a quality of patient care indicator. Readmissions especially affect older adults health due to the decline in func tional status ; this in turn, affect s their quality of life and overall well being. 5 In addition, older adults are especially likely to report insufficient pain management, which may be due to difficulties communicati ng with their doctors among other factors P ain can lead to rehospitalization to try to improve their pain management. 6 The 30 day period after a hospitalization is considered the time when the patient is most vulnerable to a rehospitalization, or a readmission to the hospital, due to factors such as the lack of direction of future care, no familial or similar social support. 7 Factors rel ating to this vulnerability may lead to patient frustration, then to neglect of warning signs, eventually leading patients back to the hospital. This can be a vicious cycle for patients. Not only does this cycle have an emotional and mental impact, it also exposes patients to higher risks of health care associated infections, results in excess mortality and adds higher costs to the economy. 8 1.2 Characterizing Readmission Rates A person is considered to readmit at a high rate if he or she has 3 or more readmissions within 12 months 4 Published re admissions research focuses on patients considered high readmitters and most researchers have analyzed readmissions as one continuous outcome. Based on the age adjusted percent distribution of the 2015
15 National Health Interview, those who were hospitalized 1 2 times in the past 12 months contributed 6.6% to the total versus those who were hospitalized 3 or more times in the past 12 months who contributed only 0.7%. 9 Within the local clinics, we see a similar pattern in readmissions rates those who readmit 1 2 times contribute more to the number of total readmissions, compared to those who readmit 3 or more times. Little research has focused on understanding or characterizing the middle tier population of readmitters: those who have readmitted 1 2 times in a year. Locally, patients readmit mostly at a rate of 1 2 readmissions ra ther than 3 or more times, in the past 12 months. 10 Because there is a distinct middle group, it is highly likely that this group has important characteristics that need to be identified to better unders tand their readmission behavior. Additionally, the distinct 1 2 readmissions group may have characteristics both different and similar to the 3 or more readmissions group and the 0 readmissions group, which need to be identified. 1.3 Current Readmissions Predictive Tools Predictive tools for readmissions have bee n developed based on findings from many hospital systems. Many of these predictive tools focus on factors related to when the patients are in the hospital. One example of a validated predictive tool is the HOSPITAL score, which is an acronym standing for, and includes hemoglobin level, oncology service discharge, sodium level, hospital procedure, index admission type, admissions during the previous year, and length of stay. 11 Another validated predictive score is the LACE index, another acronym, which includes length of stay, acuity of admission, comorbidity of patient, and emergency department use. 12 The third predictive tool is the LACE+ index which is an extension of the validated LACE index, with the addition of age, sex, teaching status of discharge institution, number of urgent
16 admissions in previous year, number of elective admissions in the previous year, case mix group score, and number of days on alternative level of care status. 13 A recent paper compared the performance of 30 day all cause readmissions risk tools among hospitalized primary care patients, with a 14,663 sample size. 14 The predict readmission or death within 30 d ays, were compared by the creation of a receiver operating characteristic curve (AUC). The tools yielded moderate performance with c scores of 0.66 0.68. 14 With th e moderate performances in the available indices and score tools, other predictive models have added variables such as polypharmacy ( i.e. 6 or more medications) and social determinants of health 15 1.4 Importance of Social Determinants of Health Social determinants of health are defined as conditions in which people are born, grow, work, live, and age, and the wider ( i.e. systematic) forces that affect health risks and health outcomes. 16 17 18 19 According to the Kaiser Family Foundation (Figure 1 1), the soci al determinants of health outcomes are comprised of economic stability, neighborhood and physical environment, education, food, community and social context, and health care system. 19 Among minority groups, there are specific social determinants of health considered key factors of hospitalizations and readmissions. Accor ding to Dr. Cara James of the Centers for Medicare and Medicaid Services (CMS) Office of Minority Health, as much as 80% of health disparities are driven by social determinants of health and structural barriers prevent the health care system from addressin g these disparities. 20 21 As stated in the New England Journal of Medicine Catalyst article, the Hospital Industry Data Institute (HIDI) suggests that before penalizing hospitals for excess readmissions, there should be, inste ad, risk adjusted readmissions rates
17 accounting for social determinants of health. 21 HIDI included three sociodemograp hic status (SDS) factors in their model: Medicaid status (individual), census tract poverty rate (contextual), and models nested at census tract (contextual). 22 Figure 1 2 describes the conceptual and empirical basis for these three factors. 22 The adjustment for nonclinical sociodemographic st atus (SDS) factors made their readmissions risk prediction model better in comparison to two models: 1. the standard CMS model ( i.e. readmissions model based on CMS hospital level data), and 2. the Yale CORE model ( i.e. readmissions model that adjusts for patient level and provider level factors). 21 22 The SDS enriched model also reduced the variation in estimated readmission performance assessments compared to the standard CMS Yale model specifications, which are used to penalized hospitals under the Hospital Readmissions Reduction Program. 21 22 Additionally, a recent article on PEARLS+ from Academic Medicine stated that societal forces, social determinants, and health outcomes are connected. 23 The acronym PEARLS+ stands for pol itics, economics, environment, ethics, arms, religion, life circumstances, social roles, social structures, and other forces. PEARLS+ are the key societal factors that influence social determinants of health which then shape health outcomes, including rea dmissions. 1.5 Current Readmissions Initiatives for Minority Members There are many readmissions prevention initiatives focused on racial and ethnic minorities such as: 1. a telehealth and home health disease management program by Alterna Care Health in Springfield, Illinois, and 2. Health Connections Initiative in Louisville, Kentucky. 24 25 However, the most notable readmissions initiative for racial and ethnic minorities is Project Re Engineered Discharge (RED) from Boston,
18 Massachusetts. Project RED founded an intervention with twelve dis crete, mutually reinforcing components that have yielded high patient satisfaction and significantly reduced hospitalizations. 26 for and obtain language assistance, 2. make appointments for follow up medical appointments and post discharge tests/labs, 3. plan for the follow up of results from lab tests or studies that are pending at discharge, 4. organize post discharge outpatient s ervices and medical equipment, 5. identify the correct medicines and a plan for the patient to obtain and take them, 6. reconcile the discharge plan with national guidelines, 7. teach a written discharge plan the patient can understand, 8. educate the pati ent discharge plan, 10. review with the patient what to do if a problem arises, 11. expedite transmission of the discharge summary to clinicians accepting care of the pa tient, and 26 Within these 12 components, there are specific tasks that must be done in order to complete the component. 26 The Project RED website provides various sources/tools for others to use in ReEngineered Dis on the roles of culture, language, and health literacy in readmissions, as well as step by step instructions and responsibilities one has to take to assist a minority patient. 26 It is important to note that Project RED requires a coordinated multidisciplinary team including hospital care team, community providers, and caregivers. 26 They also have virtual patient advocates whose dialogue can be tailored for each patient. 26 Hospitals
19 outside of Boston have succes sfully implemented Project RED; hospitals in Texas are examples. 27 In a randomized controlled trial that compared the systems level Project RED intervention and normal standard of care intervention, patients who received the Project RED intervention had a 30 percent lower readmission rate within 30 days of discharge compare to their counterparts who solely received standard care. 28 Also, patients who received services from Project RED were more likely to be able to identify their index discharge diagnosis, and follow up with their PCP. 28 These patients also reported feeling mor e prepared for discharge. 28 1.6 Gaps in Current Readmissions Measures and Predictive Models Though there h ave been many successful initiatives and predictive models in readmissions, there are still many gaps in the literature. The proportion of readmissions that are preventable, by definition, are related to readmissions that have primary diagnoses codes that are for ambulatory care sensitive conditions (ACSC). As first defined by Billings and colleagues, ACSC s are a set of conditions with diagnoses for which timely and effective outpatient care can help to reduce the risks of hospitalizations by preventing acu te illness and managing chronic conditions. 29 ACSC related hospitalizations can be targeted to prevent subsequent readmissions. However, to our knowledge, no predictive readmissions models have been made to only predict the ACSC related readmissions. Another un der behavior or activity on readmissions. The Agency for Healthcare Research and Quality (AHRQ) and Centers for Medicare and Medicaid Services (CMS) hav e suggested that effective and timely primary care can prevent readmissions 2 30 31 AHRQ has recognized that primary care settings have been mentioned i n various readmissions
20 projects; however, evidence based literature on the role of primary care in preventing readmissions is scarce. 30 There is no literature to our knowledge that measures the Examples of interactions with primary care ( i.e. outpatient behavior) are attending clinic appointments, miss ing clinic appointments, cancell ing clinic appointments, and phone encounters with the clinic. In a preliminary, unpublished analysis of readmissions in Southeast USA Family M edicine clinics, results showed there were significantly increased odds of having 5 6 (OR: 1.58, CI: 1.18 2.11) or >6 (OR: 1.44, CI: 1.09 1.91) readmissions for each additional clinic no show clinic appointment, compared to those with 1 2 readmissions. 10 In addition, there were significantly increased odds of having 3 4 (OR: 1.16, CI: 1.05 1.29) or >6 (OR: 1.29, CI: 1.15 1.43) readmissions for each additional cancelled clinic appointment, compared to those with 1 2 readmissions. 10 This finding provides support for stu Hospital wide discussions at a Southeast USA academic hospital (Readmissions Meetings, oral communication, 2017), have also concluded that outpatient behavioral change can be a potent ial intervention for reducing readmissions. Leaders at these hospital wide discussions stated that once patients have been hospitalized, it is already too late to prevent other hospitalizations. Instead, monitoring patients at an outpatient setting is the key to preventing hospitalizations and readmissions. Another preliminary, unpublished study that incorporated community health workers at a Southeast USA Family Medicine clinic revealed many reoccurring themes for reasons for cancellation or no shows of cl inic appointments, which would result in hospitalizations or readmissions
21 of some of the patients. 32 People in h ospital wide discussions at a Southeast USA academic hospital also mention variables that affect readmissions and are not captured in the medical recor ds. Some of the mentioned medications, transportation issues, and continuity of information between inpatient provider and outpatient provider. The Southeast USA academic hospital has enacted a variety of interventions geare d towards identified factors that appear to affect readmissions within its system. For example, inpatient teams in various departments make sure that hospitalized patients have a scheduled outpatient appointment before leaving the hospital. Additionally, t he Department of Community Health and Family Medicine enacted the allows the recipient to go straight to a primary care outpatient clinic without scheduling an appointm ent. Though individual departments are doing what they can to help reduce readmissions there are still gaps in these varying interventions. With the fragmentation of care between inpatient (hospital) and outpatient (primary care clinic), many of the implemented interventions still fall short in terms of preventing hospitalizations and readmi ssions. In addition, the role of factors related to minority status have been shown to be predictive of negative health outcomes including readmissions. Particularly, ethnic and racial minorities are disproportionately impacted by congestive heart failure (CHF) and cardiovascular risk factors. 33 This translates to a higher readmission and hospitalization rate for CHF in Black and Hispanic patients compared to their non Hispanic W hite cou nterparts. 34 35 36 In a study t hat aimed to determine whether B lack patients have
22 hi gher odds of readmissions than W hite patients f or 3 common conditions ( i.e. acute myocardial infarction, congestive heart failure and pneumonia), B lack patients were more likel y to be readmitted compared to W hite patients, a gap that was related to race and site where care was received. 36 In addition, p atients discharged from minority serving hospitals had 23% higher odds of readmitting compared to patients discharged from nonminority serving hospitals. 36 One study assessed the risk of Medicare readmissions in racial and ethnic groups. 38 The researchers were particularly interested in these racial and ethnic groups because hospitals serving them may be disproportionately penalized. 38 Their study found that Black Medicare patients with pneumonia, heart failure, and acute myocardial infarction (AMI) were more likely to be readmitted than their White counterparts. 38 They also found that Hispanics had significantly increased odds of readmission for AMI compared to their White counterparts. 38 This is consistent w ith other literature, showing that race is an important determinant of readmissions among those with chronic conditions. 36 As previously stated, s o cial determinants of health are defined as conditions in which people are born, grow, work, live, and age, and the wider ( i.e. systematic) forces that affect health risks and health outcomes. 16 17 18 19 These social determinants of health can be related to race or ethnicity, because evidence shows that there are disparities in healthcare equity and quality in racial and ethnic groups. Inequities arise because of root causes intraperson al, interpersonal, institutional, systemic mechanism and allocation of power 39 Health insurance status is an example o f one of the inequities that are commonly planned to be addressed, but such simple solutions do not address disparities across e thnicity, race, geography, and socioeconomic status (S ES ).
23 Similarly, race alone is not the cause of inequities, but is often a marker for those inequities. In the 2010 National Healthcare Quality and Disparities Reports, Blacks and American Indians/Alaska Natives received worse care than their White counterparts for 40% of the quality measures. 40 Conversely, Asians received worse care than their White counterparts for 20% of the quality measures. 40 And at the highest level of disparity, Hispanics received worse care than their White counterparts for 60% of the quality measures. 40 Thus insurance status, race and ethnicity may be markers of SDH although they are insufficient to account for disparities. Each may contribute independently to rates of readmissions. The role of sex is unclear in its association with readmissions Some analysis by sex has shown that men are at higher risk of readmissions (e.g. Medicaid men have increased risk of all cause hospital readmissions). 41 Other analysis has findings in the opposite direction, with women having a higher risk for readmission 30 42 43 In a study looking at acute myocardial infarction read missions in younger patients, female sex was significantly associated with 30 day readmission with a hazard ratio of 1.22. 42 A second study on the role of sex conducted in an Italian Department of Medicine found that females had 1.41 times the likelihood of readmitting, after controlling for other covariates such as the HOSPITAL score and ERA index. 43 1.7 Theoretical Model Based on their review of the literature, a team in Scotland has created a model of readmissions risk model for intensive care unit (ICU) patients (Figure 1 3). 44 A strength of this model is that it is based on a mixed methods study with the inclusion of patient perspectives. 44 Though the PROFILE study readmissions model fills many gaps that other readmissions models lack, there are still other important factors missing. As
24 previously stated, outpatient primary care behaviors are important to include. Therefore, the PROFILE study readmissions model, along with factors in Figure 1 1, were combined to come up with the proposed theoretical model, OHIS Readmissions ( O utpatien t Behavior, H ealth Status, I npatient Factors, S ocial Determinants), of this dissertation seen in Figure 1 4. 1.8 Explanation of the OHIS Readmissions Theoretical Model Based on the readmissions literature that was reviewed, the OHIS Readmissions Theoretica l Model (Figure 1 4) was created and informs this dissertation. The two larger circles describe the greater forces of the hospital system and community that affect overall readmissions in patients. These greater forces also influence the four constructs th at directly impact readmissions: outpatient behavior, health status, inpatient factors, and social determinants. The double headed curved arrows on the top and bottom of the diagram show that the greater forces and proximal factors influence each other. R elevant outpatient behaviors include attended clinic appointments, missed clinic appointments, cancelled clinic appointments, and telephone encounters between visits. e provider. This is an important construct to include in understanding readmissions, because primary care providers are the first point of care that can prevent ambulatory conditions from becoming larger issues. As previously stated, in an unpublished anal ysis of readmissions in Southeast USA Family Medicine clinics, results showed that for every increase in number of cancelled clinic appointments or no shows, there was an associated increased odds for readmissions. 10 It is also important to include this construct in the theoretical model because this could be an area of intervention (e.g. primary care providers can use extra resources such as a nurse case manager).
25 Health status is another important construct that needs to be adjusted for, because status is important in creating effective interventions. Health status can be defined using factors such as weathering, functional status, high risk medications (as a proxy fo r through which lower socioeconomic status creates health disparities especially amongst racial and ethnic minorities. 45 46 A recent study has identified functional status as being associated with 30 day potentially preventable hospital readmissions. 47 Another recent study has foun d that adverse drug events accounted for 13% of 30 day preventable readmissions. 48 Pain, a factor that was also explored qualitative ly in this dissertation, can also contribute to health status. An interaction between pain score, pain medication use, and older age, was found to be a risk factor for unplanned readmissions. 49 Inpatient factors, such as length of stay, di sposition of the patient ( i.e. where the patient was discharged to), ESI level ( i.e. acuity level), and Charlson Comorbidity Index ( i.e. weighted index that predicts risk of death within 1 year for patients with comorbidities) are important in readmissions Many of these factors are part of the validated HOSPITAL score, LACE index, and LACE+ index. 12 13 14 In transition of care, it is important to assure that patients are discharged to the appropriate location to prevent subsequent admi ssions to the hospital. 7 The Agency for Healthcare Research because it is a major factor in preventing readmissions. 50
26 The constructs of social determinants of health and outpatient behavior were hypothesized to contribute most to readmissions. As previously stated above in the to not only readmissions, but all health outcomes. Many readmissions initiatives for racial and et hnic minorities incorporate the impact of social determinants. 16 17 18 19 Finally, as noted above, outpatient behavior is important. As shown in our unpublished readmissions analyses, number of cancelled clinic appointments or no shows are important in predicting increased odds for readmissions and have been included in this model 10 Figure 1 1. Kaiser Family Foundation Social Determinants of Health
27 Figure 1 2. HIDI SDS factors included in risk adjusted readmissions rates Figure 1 3. Theoretical m odel from the PROFILE study
28 Figure 1 4. Theoretical m odel: OHIS (Outpatient Behavior, Health Status, Inpatient Factors, Social Determinants) Readmission s
29 CHAPTER 2 DESIGN AND METHODS 2.1 Data Samples The sample for the analyses of Aims 1 and 2 of this dissertation comes from an academic Southeast USA Family Medicine department. Data for the department are collected daily in EPIC medical records at all relevant clinical sites. A combination of physician s, physician assistants, and nursing staff record all patient information in the EPIC electronic medical record. D ata necessary for this dissertation was requested by the Quality Analyst of the department and anonymized before it was shared for analyses. The sample for the analysis of Aim 3 is from the 2013 Nationwide Readmissions Database (NRD) Healthcare Cost and Utilization Project (HCUP), Agency for Healthcare Research and Quality. The NRD is a part of the family of databases for the Healthcare Cost a nd Utilization Project (HCUP). 51 HCUP is sponsored by the Agency of Healthcare Research and Quality (AHRQ), a branch of the U.S. Department of Health and Human Services. 52 The NRD, first released in 2016, currently had data covering 2013 at the time of the data request for th is dissertation 51 It includes national readmissions rates for all payers and the uninsured. The NRD includes data from approximately 15 million discharges each year (unweighted) and approximately 35 million discharges each year (weighted). 51 The NRD was purchased by the d epartment for use. All data was used according to UF IRB approved protocols. 2.2 Sampling and Recruitment Most patients who readmit, at any rate, in the academic Southeast USA family medicine department 2015 2016 cohort tended t o be middle aged with a mean of 55
30 years old (SD: 15.9 years), to be fem ale (55%), non Hispanic (96%), W hite (59%), and have Medicare insurance (47%). Patients included in the sample of this dissertation were current patients of the academic Southeast USA f amily medicine clinics, aged 18 to 64 years old, and had been hospitalized during the period from Jan 1, 2015 to December 31, 2016. Because this is a retrospec tive chart review, patients did not need to be consented to be included in the analyses of Aims 1 and 2. The flow char t for sampling for Aims 1 and 2 is s hown in Figure 2 1. Utilizing data from the Southeast USA family medicine clinics allowed us to retrieve data that usually would not be available in national databases. Therefore, it provided a more complete picture of factors related to readmissions in primary care patients. Using this local sample also provided the opportunity to interview a few Findings from t he qualitative portion are reported in the conclusion Chapter to help provide more nuanced understanding of the quantitative findings The nationwide readmissions database (NRD) provides a statistically generalizable US patient sample in which to test Aim 3. As previously noted, t he NRD is drawn from the Health Cost Utilization Project (HCUP) State Inpatient databases (SID) to create nati onally representative readmission rates. 51 It includes 22 geographically dispersed states with verifiable patient linkage numbers : Alaska, Arizona, Arkansas, California, Colorado, Connecticut, District of Columbia, Florida, Georgia, Hawaii, Illinois, Indiana, Iowa, Kans as, Kentucky, Louisiana, Maine, Maryland, Massachusetts, Michigan, Minnesota, Mississippi, Missouri, Montana, Nebraska, Nevada, New Hampshire, New Jersey, New Mexico, New York, North Carolina, North Dakota, Ohio,
31 Oklahoma, Oregon, Pennsy lvania, Rhode Island, South Carolina, South Dakota, Tennessee, Texas, Utah, Vermont, Virginia, Washington, West Virginia, Wisconsin, Wyoming 51 These states account for 51.2% of the total US population and 49.3% of all US hospitalizations. 51 In the NRD, most patients were Medicare insured and the top 4 principal diagnoses for readmission were acute myocardial infarction, congestive heart failure, chronic obstructive pulmonary disease, and pneumonia. 53 Sampling for Aim 3 is shown in th e flowchart provided in Figure 2 2. 2.3 Measures for Aims 1& 2 Table 2 1 lists the variables that were analyzed in Aims 1 and 2. The table provides variable name from the electronic medical record, construct, and variable type for each of the a ims. 2.4 Approach Aim 1: Specified Measures for Aim 1 2.4.1 Readmissions For this analysis, characteristics of the patients who readmit in the Southeast US A academic hospital and belong to the affiliated family medicine clinics were analyzed 2.4.2 Variables for Factor Analysis Variables included in the exploratory factor a nalysis (EFA) and confirmatory factor analysis (CFA) for Aim 1 are seen in Table 2 1. No other variables were entered into the analysis. 2.5 Approach Aim 1: Method of Analysis fo r Aim 1 Data used in the EFA was the 2016 readmissions data set from the Southea st USA Family Medicine clinics. Results of EFA in Aim 1 identified factors that could be used to measure readmissions. To achieve this, the four proposed constructs
32 (outpatien t behaviors, social determinants, inpatient factors, and health status) and their respective variables were analyzed to see the correlations between each of the variables. For correlations between continuous variables, polyc h oric correlations, sample sizes and standard deviations were obtained using the PROC CORR command. For correlations between binary and ordinal variables, tetrachoric or polychoric correlations, sample sizes, and standard deviation s were obtained using the PROC FREQ command. For correla tions between binary or ordinal variables with continuous variables, polyserial correlations, sample sizes, and standard deviation s were obtained using the PROC CORR command. Once correlations, sample sizes, and standard deviation s were obtained, they wer e entered into the EFA using the PROC FACTOR command. The principal factor method was used to extract the factors. The scree plot demonstrated how many factors were to be retained. A n obli que (promax) rotation was chosen because components were thought to be correlated. 54 After the number of factors were obtained, a CFA was conducted using the PROC CALIS command. Data used for the CFA was the 2015 readmissions data set from the same Southea st USA Family Medicine clinics. All analysis was completed using SAS software, Version 9.4 2.6 Approach Aim 2: Specified Measures for Aim 2 2.6.1 Outcome: Readmissions For Aim 2, number of readmissions within the last 12 months was the outcome predicted for (1 2 readmissions). Readmissions was categorized as: 0 ( i.e. admitted to the hospital, but no readmission), 1 2, and 3 or more.
33 2.6.2 Covariates Variables that were included in the analysis are in Table 2 1. 2.7 Approach Aim 2: Method of Analysis for Aim 2 Univariate bivariate, and multivariate analyses for this dissertation was conducted using SAS software, Version 9.4 2.7.1 Univariate and Bivariate Analyses Univariate analyses were conducted in order to understand the distribution of each variable. Bivariate analyses, to assess the relationships with the outcom e ( i.e. number of readmissions), were conducted using Chi square tests for categorical predictor variables, and ANOVAs for continuous predictor var iables. 2.7.2 Multinomial Analyses Once significant relationships between variables were identified from the bivariate analyses, these variables were entered into an adjusted multinomial logistic regression. 2.8 Measures for Aim 3 Table 2 2 lists the va riables that were analyzed in Aim 3. Details include variable name from the Nationwide Readmissions Database (NRD) and variable type. 2.9 Approach Aim 3: Specified Measures for Aim 3 2.9.1 Outcome: Readmissions For Aim 3, number of readmissions within the last 12 months was the outcome predicted for (1 2 readmissions). Number of r eadmissions were categorized as: 0, 1 2, and 3 or more. 2.9.2 Covariates Variables included in the analysis are listed in Table 2 2
34 2.10 Approach Aim 3: Method of Analysis for Aim 3 Univariate bivariate, and multivariate analyses for this dissertation was conducted using SAS software, Version 9.4 2.10.1 Univariate and Bivariate Analyses Univariate analyses were conducted in order to understand the distribution of each variable. Bivariate analyses, to assess the relationships with the outcom e ( i.e. number of readmissions), were conducted using Chi square tests for categorical predictor variables, and ANOVAs for continuous predictor variables. 2.10.2 Multinomial Analyses Once significant relationships between variables were identified from the bivariate analyses, these variables were entered into an adjusted multinomial logistic regression. 2.11 Approach for Enriching Findings w ith Patient Centered Data To enrich our findings, we set out to interview patients to better understand our findings from their perspectives. To do so the moda l patient was defined using data from Aims 1 and 2. The objective of this qualitative exploration wa s specifically to understand the importance of social support, presence of pain, and reason for readmission, from the perspective of recently readmitted patients 2.12 Approach for Patient Centered Data: Method s of Recruitment and Analysis Once the modal patient was defined, a convenience sample of patients meeting this pattern of characteristics were in vited to participate in a qualitative on e on one exploratory interview, by current faculty working on the inpatient service. The faculty mem ber asked the patient if it was okay to be contacted by me, and if the answer was yes they were asked to fill out an authorization form with their phone number and the
35 best time to call. Between 4 10 days after discharge, the patient was called and a qual itative interview was c onducted or scheduled. Interviews were conducted over the phone Interviews were recorded While i nterviews were to have continued until saturation of themes were reached, and we were supposed to interview a maximum o f 12 patients, time ran out The i nterview style used was a narrative approach, to allow patients to freely tell their story the more recent hospitalization and readmission rather than relying on interviewer driven questions. All qualitative analysis was completed using the NVivo qualitative analysis program. 2.13 Approach for Patient Centered Data: Interview Guide The interview length was 45 60 minutes. An interview guide was created to assist in the questions that needed to be asked, as well as provided probes for each of the questions. The intervi ew guide is included in Appendices A B, and C Figure 2 1. Flow chart for Aims 1 and 2
36 Figure 2 2. Flow chart for Aim 3
37 Table 2 1. Electronic medical record variable name, variable construct, and variable type Aim 1, and variable type in Aim 2 Variable Name Construct Variable Type Aim 1 Variable Type Aim 2 cancelled appointments outpatient behaviors continuous range: 1 6 continuous range: 1 6 missed appointments outpatient behaviors continuous range: 1 3 continuous range: 1 3 made appointments outpatient behaviors continuous range: 1 6 continuous range: 1 6 phone encounters outpatient behaviors continuous range: 1 10 continuous range: 1 10 active medications health status continuous range: 0 10 continuous range: 0 10 active problems health status continuous range: 0 10 continuous range: 0 10 age health status ordinal 18 24 25 44 45 64 ordinal 18 24 25 44 45 64 albumin # health status ordinal less than 3.4 normal greater than 5.4 ordinal less than 3.4 normal greater than 5.4 systolic blood pressure # health status ordinal 0 119 120 129 130 139 140 and above ordinal 0 119 120 129 130 139 140 and above Charleson Comorbidity Index health status continuous range: 0 1 6 continuous range: 0 1 6 chronic pain present health status (not included) categorical no yes psychological problem present health status (not included) categorical no yes
38 Table 2 1. Continued Variable Name Construct Variable Type Aim 1 Variable Type Aim 2 on high risk medications health status categorical no yes (not included) number of high risk medications health status ordinal 0 1 2 3 ordinal 0 1 2 3 substance use health status categorical no yes categorical no yes acuity (ESI level) inpatient factors (not included) categorical non urgent urgent location of admission inpatient factors (not included) categorical home assisted o ther disposition inpatient factors categorical home assisted categorical home assisted planned readmission o ther ambulatory care sensitive condition inpatient factors (not included) categorical no yes length of stay inpatient factors continuous range: 1 14 continuous range: 1 14 Morse fall score inpatient factors continuous range: 0 1 25 continuous range: 0 1 25 gait/transferring score inpatient factors continuous range: 0 20 continuous range: 0 20
39 Table 2 1. Continued Variable Name Construct Variable Type Aim 1 Variable Type Aim 2 level of assistance needed inpatient factors categorical independent dependent categorical independent dependent ethnicity social determinants (not included) categorical not Hispanic or Latino Hispanic or Latino patient refused insurance social determinants categorical public private categorical public private self pay health literacy social determinants (not included) categorical adequate limited marital status social determinants categorical not married married categorical not married married physical abuse social determinants (not included) categorical denies denies, provider concerned unable to assess yes, past yes, present verbal abuse social determinants (not included) categorical denies denies, provider concerned unable to assess yes, past yes, present race social determinants categorical majority minority categorical majority minority sex social determinants categorical female male categorical female male
40 Table 2 1. Continued Variable Name Construct Variable Type Aim 1 Variable Type Aim 2 someone living with patient social determinants categorical not alone alone categorical not alone alone someone helping with daily needs social determinants (not included) categorical not alone none zip code social determinants (not included) categorical West (higher SES) East (lower SES) high risk medications: anticoagulants, diabetic, opioid containing # most recent lab of that year
41 Table 2 2. Nationwide Readmissions Database (NRD) variable name and variable type Variable Name Variable Type age continuous range: 18 64 number of chronic conditions continuous 0 1 2 3 4 5 6 or more alcohol abuse comorbidity categorical no yes deficiency a nemia comorbidity categorical no yes conge stive heart failure comorbidity categorical no yes chroni c pulmonary disease comorbidity categorical no yes depression comorbidity categorical no yes diabetes uncomplicated comorbidity categorical no yes diabetes with ch ronic complications comorbidity categorical no yes drug abuse comorbidity categorical no yes hypertension comorbidity categorical no yes
42 Table 2 2. Continued Variable Name Variable Type obesity comorbidity categorical no yes psychoses comorbidity categorical no yes discharge month continuous range: 1 12 length of stay continuous range: 0 14 admission day c ategorical not weekend weekend disposition of patient categorical routine transfer to short term hospital skilled nursing facility, intermediate care, and another type of facility Home Health c are A gainst medical advice sex categorical male female insurance categorical no charge other private public self pay
43 Table 2 2 Continued Variable Name Variable Type patient location categorical central (>= 1 million) fringe (outside central) metro area 1 (250,000 999,999) metro area 2 (50,000 249,999) micropolitan not metropolitan or micropolitan median income categorical $1 $37,999 $38,000 $47,999 $48,000 $63,999 $64,000 or more
44 CHAPTER 3 READMISSIONS WITH PRIMARY CARE RELATED FACTORS 3.1 Background As recently reported by the Agency for Healthcare Research and Quality (AHRQ), the role of primary care in reducing or preventing readmissions is not clear. 2 AHRQ, among other hospitals and agencies, have used evidence based research to Engineered Discharge (RED) and Hospital Guide to Reducing Medicaid Readmissions. 2 These current readmissions reductions initiatives heavily focus on factors within the inpatient setting and proper tran sitioning of care ( i.e. disposition of the patient following hospitalization). Yet, the next step of care ( i.e. health and preventing rehospitalization has been understudied. Identifying the role of primary care in preventing readmissions is needed. In 2011 in the US, there were approximately 3.3 million adult readmissions with an associated cost of $41.3 billion. 3 The associated cost of readmissions is not the only concern, but the reduction in reduced payments and potential fines from the Inpatient Pro spective Payment System (IPPS) are of great conce rn to all health care systems 3 In the 2017 fiscal year, $528 million was withheld. According to the anal ysis of data released by the Centers for Medicare and Medicaid Services (CMS), $ 564 million will be withheld from payments to health care systems for fiscal year 2018, an increase from the amount withheld in 2017. 1 Additionally, 2,573 hospitals will face penalties in 2018 1 Given the high costs to health care systems of withheld re imbursements and penalties, other factors affecting readmissions must be identified and studied There are
45 a few readmissions predictive tools such as the HOSPITAL score, LACE index, LACE+ index, and the Readmission Risk Score 11 12 These scores include inpatient factors (e.g. length of stay, ac uity, and emergency department use) and health status factors (e.g. comorbidity of patient and age). The LACE index and the other scores are widely used, yet still have weaknesses. A study comparing the 4 tools and their abilities in predicting scores of 0.66 0.68. 14 Conclusions in readmissions studies h ave pointed to the importance of social determinants of health and their effect on readmissions. Social determinants of health are defined as the conditions in which people are born, grow, work, live, and age, and the wider forces that affect health risks and health outcomes. 16 17 18 19 There have been suggestions of risk adjusting readmission rates for social determinants of health, before penalizing hospitals for excess readmissions. 21 In order to better understand the roles of social determinants and primary care (e.g. outpatient behavior such as missed primary care appointments) we created the outpatient behavior, health status, inpatient factors, and social determinants of health (OHIS ) theoretical model (Figure 1 4 ). This theoretical model is an adaptation of a theoretical model for preventing early unplanned hospital readmission after critical illness (PROFILE). 44 With the OHIS theoretical model informing this study, we aimed to create a readmissions predictive tool relevant for primary care. 3.2 Methods 3. 2.1 Study Overview The present study was a retrospective cross sectional design and was approved by the University of Florida Institutional Review Board. Participants included in this analysis were patients who were hospitalized in calendar years 2015 and 2016 (January
46 1, 2015 December 31, 2016) at a large Southeast United States academic hospital and were also active patients of the associated family medicine outpatient practices. Patients also needed to be aged 18 64 years to be included into the analysi s. created for each patient. Index admission was defined as the first admission that qualified a patie nt to have readmissions. After the index admissions were identified, all other hospitalizations were deleted. Variables used in this study belonged to one of the constructs from the theoretical OHIS model. Table 3 1 lists all of the variables. 3.2.2 Statistical Analysis In order to reduce the number of variables, exploratory factor analysis (EFA) was done. First, correlations between the 23 variables of the 20 16 readmissions data set were ru n (N=567). Correlations between continuous variables were polyc h oric correlations. Correlations betwe en binary and ordinal variables were tetrachoric or polychoric. Correlations between binary or ordinal variables with continuous variables were polyserial. The obtained correlations were entered into the EFA using the PROC FACTOR command. The principal factor method was used to extract the fa ctors. The scree plot demonstrated how many factors were to be retained. Then an oblique (promax) rotation was done due to the many binary variables. The oblique (promax) rotation was also done since the components were thought to be correlated. 54 After the number of factors and variables were obtained, a confirmatory factor analysis (CFA) was conducted using the PROC CALIS command. Data used for the CFA was the 2015 readmissions data set f rom the same Southeast USA Family Medicine clinics. All analysis was completed using SAS software, Version 9.4
47 3.3 Results As part of the EFA, Table 3 2 depicts the preliminary eigenvalues. This shows an 8 factor model based on Kaiser criterion of eigenvalues greater than 1.0. The scree test, Figure 3 1 suggested again 8 factors. However, when items were considered by factor structure, 4 factors were meaningful and had a clear latent or underlying meanin g. Factors 6, 7, and 8 yielded factors with only 1 2 item s While they were potentially important, the fact that they were comprised of 2 items showed that we were not retaining meaningful sets of variables as per our model, but were adding small amounts to the variance explained through these unique factors. Also, a solution is less satisfactory has less than 3 items loaded. 54 The 5 th factor in the 5 factor solution had items that did not share conceptual meaning. 54 Instead, we made the decision to retain 4 meaningful factors of groups of variables. For further tests, these factors were retained from the Promax rotation. When interpreting the Promax rotated factor pattern, an item was said to load on a factor if the loadin g was 0.50 or greater. Table 3 3 depic ts the factor pattern. Table 3 4 depicts the factor structure. Items that loaded on each of the factors are seen in Figure 3 2. Six items were found to load on factor 1, which was labeled as inpatient factors. Four items loaded on factor 2, which was labeled as health status. Three items loaded on factor 3, which was labeled as outpatient behavior. Three items loaded o n factor 4, which was labeled as social determinants. The CFA showed that model fit was not strong. The fit statistics yielded: a comparative fit index (CFI) of 0.8099 (CFI criterion >0.95), root mean square error of approximation (RMSEA) of 0.1176 (RMSE A criterion <0.06), and a standardized root mean square residual (SRMR) of 0.1246 (SRMR criterion <0.08)
48 3.4 Conclusions Using a dataset from a large academic medical center, we used exploratory factor analysis to attempt to reduce the variables needed for our theoretical model, and to confirm the factor structure of readmissions proposed in our model. We were able to reduc e variables to our 4 proposed factors: outpatient behavior, health status, inpatient factors, and social determinants. This helps provide evidence for the structure of our proposed model. However, taking these factors into another year of data through the CFA did not confirm this factor structure. This is the first study, to our knowledge, that incorporated outpatient behavior variables as well as additional social determinants of health variables in a readmissions model. I tems that loaded on specific f actors were of great interest to the field For factor 1 (inpatient factors), albumin was significant. It may be considered as a test to be regularly collected in the inpatient setting. Albumin is a circulating protein necessary to maintain fluid balance. 55 Lower levels of albumin may be a sign of malnutrition, liver disease, or inflammatory disease. 56 These can be signs of persistent stress on the body, which can have negative impact on health. Also in factor 1, all items regarding mobility loaded (i.e. Morse fall score, gait/transferring score, and level of assistance needed). Mobility right after disc harge has been shown to be a marker of 30 day readmissions in older patients. 57 Th is shows evidence that it may be worthwhile to include th ese measures when collecting data for monitoring readmissions In factor 2, whether a pati ent was on high risk medications (yes/no) and the number of high risk medications the patient used was significant Readmissions studies have concluded that polypharmacy predict readmissions better than other readmissions
49 risk models. 15 The US Department of Health and Human Services Office of Disease Prevention and Health Promotion released the adverse drug events action plan which include high risk medications ( i.e. diabetic, opioid, anticoagulant) as major players. 58 Therefore, the role of high risk medications in readmissions needs to be further studie d, because readmissions may be due to adverse drug events. These events may be due to patient behavior and/or to underlying social determinants. The findings from t his study encourage further investigation of outpatient behaviors and social determinants of health variables and their relationship to may be an area that contributes to readmissions and may need to be m onitored. For example, if a patient cancels many appointments to his or her primary care doctor, ext ra resources may need to be put on this patient to get him or her into the clin i c. Patient navigators, nurse case managers, social workers and community health workers have all shown their importance in assisting patients in reducing readmissions. 32 In addition, c linics may need to cons ider monitor ing the social support that a patient has at home, because this may be indicative of whether or not a person is hospitalized. There were limitations to our study. First, the data used comes from a single academic center, which will not be representative of all academic centers in the US. Second, while we were able to include many important variables, including a history of cancel ling clinic appointments, we did not have a measure for disease self management or adherence to prescription or other physician instructions that might have affected readmissions. Our measures of social determinants and barriers to following physician ins tructions were also not as direct as we would have liked. Having transportation,
50 income, adverse home conditions (lack of electricity or other water, lack of refrigeration), and health literacy might all be important and were not included in the data used for this study. Third, most of our variables were dichotomous, which required statistical estimation of correlations. Finally, the interpretation of factor s is somewhat subjective, and we may have erred in the factors retained and tested in the CFA In t he future, researchers should continue to try to provide the field with models predicting readmissions, especially among those with 1 2 readmissions, the bulk of the population readmitting. While we used an informative data set to conduct this analysis, i t would be best to test predictive models in national datasets. This will require greater efforts to collect relevant data on inpatient admissions. While there are efforts across the states to ensure more social determinants are collected (e.g. Hunger Vi tal Sign) and to act on these health related problems, the national databases do not collect such information at this time. 59 For next steps, we plan to repeat this study using data from a larger health care system or, preferably, health systems. A bigger and more representative sample will provide a better test of the model.
51 Table 3 1 2016 Exploratory Factor Analysis (EFA) variables, hypothesized constructs, and variables types Variable Name Construct Variable Type cancelled appointments outpatient behaviors continuous range: 1 6 missed appointments outpatient behaviors continuous range: 1 3 attended appointments outpatient behaviors continuous range: 1 6 phone encounters outpatient behaviors continuous range: 1 10 active medications health status continuous range: 0 10 active problems health status continuous range: 0 10 age health status ordinal 18 24 25 44 45 64 albumin # health status ordinal less than 3.4 normal greater than 5.4 systolic blood pressure # health status ordinal 0 119 120 129 130 139 140 and above
52 Table 3 1 Continued Variable Name Construct Variable Type Charleson Comorbidity Index health status continuous range: 0 1 6 on high risk medications health status categorical no yes number of high risk medications* health status ordinal 0 1 2 3 substance use health status categorical no yes disposition inpatient factors categorical home assisted length of stay inpatient factors continuous range: 1 14 Morse fall score inpatient factors continuous range: 0 1 25 gait/transferring score inpatient factors continuous range: 0 20 level of assistance needed inpatient factors categorical independent dependent insurance social determinants categorical public private marital status social determinants categorical not married married
53 Table 3 1 Continued Variable Name Construct Variable Type race social determinants categorical majority minority sex social determinants categorical female male someone living with patient social determinants categorical not alone alone high risk medications: anticoagulants, diabetic, opioid containing # most recent lab of that year
54 Table 3 2 2016 Exploratory Factor Analysis (EFA) eigenvalues Eigenvalues of the Reduced Correlation Matrix: Total = 23 Average = 1 Eigenvalue Difference Proportion Cumulative 1 5.39726691 2.96997056 0.2347 0.2347 2 2.42729635 0.53262981 0.1055 0.3402 3 1.89466654 0.24551580 0.0824 0.4226 4 1.64915074 0.11864784 0.0717 0.4943 5 1.53050290 0.25706464 0.0665 0.5608 6 1.27343825 0.05623635 0.0554 0.6162 7 1.21720191 0.18777350 0.0529 0.6691 8 1.02942841 0.06246564 0.0448 0.7139 9 0.96696276 0.15478910 0.0420 0.7559 10 0.81217367 0.11867756 0.0353 0.7912 11 0.69349611 0.02873237 0.0302 0.8214 12 0.66476375 0.07295002 0.0289 0.8503 13 0.59181373 0.03606801 0.0257 0.8760 14 0.55574572 0.06117748 0.0242 0.9002 15 0.49456824 0.09154314 0.0215 0.9217 16 0.40302510 0.07611478 0.0175 0.9392 17 0.32691032 0.03854715 0.0142 0.9534 18 0.28836317 0.02565521 0.0125 0.9659 19 0.26270796 0.01645470 0.0114 0.9774 20 0.24625326 0.06487034 0.0107 0.9881 21 0.18138292 0.06576929 0.0079 0.9960 22 0.11561363 0.13834599 0.0050 1.0010 23 .02273236 0.0010 1.0000
55 Figure 3 1 2016 Exploratory Factor Analysis (EFA) s cree p lot
56 T able 3 3 2016 Exploratory Factor Analysis (EFA) factor pattern from the Promax rotation, decimals o mitted Factor 1 Factor 2 Factor 3 Factor 4 Item 67 3 27 1 albumin 72 1 0 8 disposition 54 19 14 10 length of stay 73 15 14 3 Morse fall score 81 11 5 3 gait/transferring score 54 3 9 12 level of assistance needed 9 56 35 1 active medications 28 63 3 7 Charleson Comorbidity Index 10 97 10 3 on high risk medications 9 95 9 1 number of high risk medications 6 4 84 5 cancelled appointments 13 5 90 7 attended appointments 6 1 80 8 phone encounters 34 20 15 53 insurance 9 11 5 87 marital status 6 4 2 56 someone living with patient
57 Table 3 4 2016 Exploratory Factor Analysis (EFA) factor structure, d ecimals o mitted Factor 1 Factor 2 Factor 3 Factor 4 Item 58 4 9 5 albumin 73 21 22 16 disposition 54 27 8 5 length of stay 73 10 30 6 Morse fall score 79 12 24 6 gait/transferring score 59 21 26 19 level of assistance needed 34 72 59 6 active medications 44 71 34 2 Charleson Comorbidity Index 14 91 25 5 on high risk medications 13 89 24 3 number of high risk medications 19 34 83 1 cancelled appointments 11 25 84 3 attended appointments 30 33 82 13 phone encounters 50 37 36 59 insurance 14 7 1 87 marital status 11 1 1 56 someone living with patient
58 Figure 3 2. 2016 Exploratory Factor Analysis (EFA) f actors and variables
59 CHAPTER 4 PRIMARY CARE MODIFIABLE FACTORS OF THE 1 2 READMISSIONS POPULATION 4.1 Background Readmissions research has been a top priority for the United States because hospitals with excessive readmissions receive reduced payments from the Inpatient Prospective Payment System (IPPS) and can be fined. 1 In 2011 in the US, there were approximately 3.3 million adult readmissions with an associated cost of $41 .3 billion. 3 For Medicare and Medicaid patients alone, associated hospital costs due to readmissions totaled $5.1 billion. 3 Various studies have shown that a proportion of readmissions is prev entable. 4 Three quarter s of readmissions were potentially preventable, equating to an estimated $12 billion in Medicare spending in 2005 claims data 4 This number has increased to a n approximate $17 billion in 2011. 60 M any studies have shown strategies that hospitals have used to prevent unnecessary readmissions, yet the best strategy has been difficult to identify. 4 Therefore, there are opportunities to find solutions to the readmission problem. Current readmissions research focuses on characterizing patients who have readmitted at a high rate, which is 3 or more readmissions within 12 months. 2 Other readmissions research also focuses on readmissions as one continuous outcome. One gap in readmissions research is the middle tier population, those who readmit 1 2 times in 12 months. In our health system, we have found that the majority of patients, who readmit, do so at a rate of 1 2 times in 12 months, rather than 3 or more times in 12 months. Because there is a distinct middle tier group, it is important to understand the characteristics of these patients in order to find points of intervention.
60 There have been numerous attempts to develop predictive tools for readmissions such as the HOSPITAL score, LA CE index, LACE+ index, and the Readmission R isk score. 11 12 These predictive tools utilize health status variables (e.g. comorbidity of patient and age), and inpatient factors (e.g. length of stay, acuity of admissions, and emergency department use). These scores, though widely used, still have weaknesses. concluded moderate performance with c scores of 0.66 0.68. 14 Readmissions studies have described the importance of social determinants of health, and their effect in predicting readmissions. Social determinants of health are defined as conditions in which people are bo rn, grow, work, live, and age, and the wider ( i.e. systematic) forces that affect health risks and health outcomes. 16 17 18 19 The Hospital Industry Data Institute (HIDI) suggests that before penalizing hospitals for excess readmissions, there should be risk adjusted readmissions rates accounting for social determinants of health. 21 A djust ing for nonclinical social determinant factors improved the HIDI readmis sions risk prediction model compar ed to the standard CMS model ( i.e. readmissions model based on CMS hospital level data) and Yale CORE model ( i.e. readmissions model that adjusts for patient level and provider level factors). 21 22 The HIDI model added social determinants measured as Medicaid status, census tract poverty rate, and census tract environment measure. In addition to social determinants of health, other factors can be studied in regards to readmissions. One un der behavior or activity on readmissions. The Agency for Healthcare Research and Quality (AHRQ) and Centers for Medicare and Medicaid Services (CMS) have suggested th at
61 effective and timely primary care can prevent readmissions. 2 30 31 AHRQ has recognized that primary care settings have been mentioned in various readmissions reducing initiatives however, evidence based lit erature on the role of primary care in preventing readmissions is scarce. 31 There is no literature to our knowledge that readmissions. Examples of interactions with primary care ( i.e. outpatient behavior) are made clinic appointments, missed clinic app ointments, cancelled clinic appointments, and phone encounters with the clinic. In order to understand the role of outpatient behavior and other factors such as health status, inpatient factors, and social determinants of health, we created the OHIS theo retical model (Figure 1 4 ). The OHIS model is an adaption of a theoretical model for preventing early unplanned hospital readmission after critical illness (PROFILE). 44 The refore the aims of this study are to 1 ) characterize the 1 2 readmissions population in an academic fa mily medicine department, and 2) identify the role of outpatient behavior in readmissions. 4.2 Methods 4.2.1 Study Overview The p resent study was a retrospective cross sectional design and was approved by the University of Florida Institutional Review Board. Participants included in this analysis were patients who were hospitalized in calendar years 2015 and 2016 (January 1, 2015 D ecember 31, 2016) at a large Southeast United States academic hospital and were also active patients of the associated family medicine outpatient practices. Patients also needed to be aged 18 64 years to be included into the analysis.
62 The number of readmi ssions each patient had in 12 months was recorded (i.e 0 readmissions, 1 2 readmissions, and 3 or more readmissions). Then, the index admission was created. Index admissions are defined as the first admissions that qualify patients to having readmissions. After the index admissions were identified, all other hospitalizations were deleted ; inclusion criteria we consider the first admissions. Thus, if any patients readmitted in both 2015 and 2016, index admission information from 2015 was kept and 2016 infor mation was deleted (N=1,212). 4.2.2 Outpatient Behavior Variables from each of the proposed theoretical model constructs were used. The variables representing outpatient behavior within the same year were cancelled clinic appointments, no show clinic appo intments, attended clinic appointments, and telephone encounters with clinic. 4.2.3 Health Status The variables representing health status were number of active medications, number of active problems, age, recent albumin reading, recent systolic blood pre ssure reading, Charleson comorbidity index (CCI), presence of chronic pain, presence of psychological problem, number of high risk medications taken (types: diabetic, opioid, anticoagulant), and presence of substance use. 4.2.4 Inpatient Factors The variables representing inpatient factors are acuity of admission, location patient was admitted from, location patient was discharged to ( i.e. disposition), ambulatory care sensitive condition, length of stay (LOS), Morse fall score, Gait/transferring scor e, and level of assistance needed.
63 4.2.5 Social Determinants The variables representing social determinants of health included were ethnicity, insurance type, health literacy, marital status, presence of physical abuse, presence of verbal abuse, race, sex, presence of someone living with patient, presence of help with daily activities, and zip code. 4.2.6 Statistical Analysis Power wa s calculated with the 30 observations per predictor rule of thumb The required sample size calculation: 34 predictors 30 observations= 1,020. W ith an N of 1,040, w e believe we were sufficiently powered for the analyses. For each of the 34 variables sele cted, b ivariate analyses were conducted in order to assess the relationships with the outcome ( i.e. number of readmissions). Chi square tests were conducted for categorical variables, and ANOVAs for categorical and continuous variables. Once significant co rrelations were identified from the b ivariate analyses, these variables were entered into an adjusted multinomial logistic regression. The analysis for this paper was generated using SAS software, Version 9.4. 4.3 Results Of the 1,040 people with an admi ssion whose de identified data was includ ed in this study, there were 209 unique people who had 1 2 readmissions in 2015 2016. On average, those with 1 2 readmissions were 45 64 years old, mostly female, mostly non Hispanic White, mostly on high risk medications ( at least 1), not married had public insurance, and lived outside of ivariate analysis variables from each of the constructs were significant with the outcome of number of readmissions, at a p < 0.05 level (Table 4 1). For outpatient behavior, the significant variables were cancelled appointments, no show appointments, attended app ointments,
64 telephone encounters. For health status, the significant variables were active medications, active problems, albumin, systol ic blood pressure, Charleson comorbidity index, presence of psychological problem, and number of high risk medications. For inpatient factors, the significant variables were location of admission, disposition, length of stay, Morse fall score, gait/transfe rring sc ore, level of assistance needed. For social determinants only insurance was significant. The variables significantly related to readmission from the bivariate analyses were entered into the multivariate analysis with the exception of disposition. D isposition was dr opped due to lack of variance. F ew people had planned readmissions (14); almost all went home or to an assisted living status. S ex was added as a control variable in the multivariate analysis. The multivariate analysis, seen in Table 4 2, revealed increased risk effects for the 1 2 readmissions group. For every additional telephone encounter (between patient and clinic) there was a 10% increased likelihood of having 1 2 readmissions compared to 0 readmissions, after controlling for all other covariates. Patients with an albumin of less than 3.4, compared to those with normal albumin levels, were 2 times as likely to have 1 2 readmissions compared to 0 readmissions. For every additional increase i n the Charleson comorbidity index (CCI) there was a n 11% increased likelihood of having 1 2 readmissions compared to 0 readmissions. Patients on 2 and 3 high risk medications, compared to 0 high risk medications, were about 2 times and almost 3 times as l ikel y ( AOR:1.919 and 2.675 respectively ) to have 1 2 readmissions compared to 0 readmissions. Patients with a gait/transferring score if 20, compared to a sc ore of 0, were 2 .3 times as likely (AOR: 2.343 ) to have 1 2 readmissions compared to 0
65 readmissions In addition, for every increase in the number of attended clinic appointments, there was a 10% decreased likelihood of having 1 2 readmissions compared to 0 readmissions. The multivariate analysis, seen in Table 4 2, revealed in creased risk effects for the 3 or more readmissions group. For every additional cancelled clinic appointment and telephone encounter (between patient and clinic), there was about 30% and 20% (AOR: 1.270 and 1.186 respectively ) increased likelihood of having 3 or more readmissions compared to 0 readmissions For every additional increase in the Charleson comorbidity index (CCI) there was a 14% increase d likelihood (AOR=1.142) of having 3 or more readmissions compared to 0 readmissions Patients who were admitted from a nywhere other than home or assiste d were 4 times as likely (AOR=4.042) to be in the group with 3 or more readmissions, compared to 0 readmissions. In addition, for every increase in the number of attended clinic appointments, there was a 23% decrease d like lihood (AOR: 0.773) in of having 3 or more readmissions compared to 0 readmissions. 4.4 Conclusions There were many significant findings in our study. Most of the significant multivariate findings were expected, with the exception of one of the outpatient behavior factors, particularly, the increased risk of 1 2 readmissions for every additional telephone encounter (between patient and clinic). We expected that as telephone encounters to the clinic increased, there would be more intervention by the clinic and therefore a decrease in readmissions. We suspect that the number of telephone encounters may not only be between the patient and the primary care outpatient clinic, but other clinics within the health system (e.g. specialists). Therefore, this variable may
66 Attended clinic appointments significantly reduced the odds of both 1 2 and 3 or more readmissions, which may show the importance that primary care clinics have in preventing patients from rehospit alizations. It should be noted that cancelled clinic appointments overall were statistically significant in the multivariate model, but not for the group with 1 2 readmissions. C ancelled clinic appointments may be related to some other fac tor important to readmissions. This importance should be noted since this may be an area that the primary clinic could intervene in ; further exploration is warranted. Some of the health status factors were significant in predicting 1 2 and 3 or more readm issions. Albumin levels of less than 3.4, compared to normal albumin levels, the higher the odds for 1 2 readmissions. A study showed that low albumin levels were associated with increased short and long term mortality in hospitalized patients. 61 As a readmissions. This is confirmed by the inclusion of CCI in current readmissions predictive scores. 11 12 14 High risk medications also increased the odds of 1 2 readmissions. A study including the interaction between pain, pain medication (e.g. opioid use), and older age, was found to be a risk factor for unplanned readmissions. 49 Other readmissions studies have concluded that polypharmacy and higher CCI scores predict readmissions better than other readmissions risk models. 15 Yet, this is the first study to our knowledge that goes a step further by studying the role of hi gh risk medications ( i.e. diabetic, opioid, anticoagulant) in readmissions, as suggested by the made by the national Office of Disease Prevention and Health Promotion 58
67 Few inpatient factors were significant in the multivariate model Patients with a gait and transferring score of 20 ( i.e. impaired) compared to 0 ( i.e. normal/bedrest/immobile) had h igher odds (AOR: 2.177) of 1 2 readmissions compared to 0 readmissions, after controlling for all other covariates. The gait and transferring score is a part of the Morse Fall scale. 62 This score was created to predict future falls, but is also used to identify risk factors to address before a patient is sent home from the hospital. 62 It should be noted that the gait and transferring score was only significant for predicting 1 2 readmissions, and was not significant in the overall model. Therefore, furthe r study is required. In terms of variables reflecting social determinants, none were significant It was surprising that none of the variables that might reflect underlying abuse or social support were significant to the readmission prediction. It is po ssible that the abuse and social support measured were too proximal to the readmission. Events occurring immediately prior to hospitalization that would be captured through a life event scale may be needed addition to inpatient screening. Though there we re many strengths in this study, such the addition of outpatient behavior, there are limitations to be noted. First, the study is based on one academic family medicine department in the Southeast US. This is a limitation because the findings of this study may not be generalizable to all family medicine departments in the US. Another limitation of this study is the cross sectional study design. Because we are looking at a point in time, we cannot predict readmissions over time. We also cannot see if a patie nt was already readmitting before the study period ( i.e. before January 1,
68 2015) or continues to readmit or passes away after the study period ( i.e. after December 21, 2016). For future directions, we recommend expanding the study to other academic famil y medicine departments to achieve generalizability. In addition, we recommend the addition of more social determinants of health. Unfortunately, during the study period, the hospital just began collecting variables such as: concern about transportation at discharge, concern about transportation at home, and problems of accessing medications (e.g. picking up or affording). These are areas that have been discussed at local readmissions meetings; therefore, their relationship to readmis sions should be further studied.
69 Table 4 1. Descriptive a nalysis of var iables by readmissions group (0 readmissions, 1 2 readmissions only, or 3 or more readmissions) Number of Readmissions 0 N= 779 1 2 N= 209 3 or more N= 52 P value OUTPATIENT BEHAVIOR CANCELLED APPOINTMENTS < 0.0001 NO SHOW APPOINTMENTS 0.8 < 0.0001 ATTENDED APPOINTMENTS 5.3 0.02 TELEPHONE ENCOUNTERS 5.5 7.0 < 0.0001 HEALTH STATUS ACTIVE MEDICATIONS 7.3 < 0.0001 ACTIVE PROBLEMS 8.4 9.0 < 0.0001 AGE 0.27 18 24 30 (3.9 %) 7 (3.4 %) 1 (1.9%) 25 44 209 (26.8 %) 45 (21.5 %) 10 (19.2%) 45 64 540 (69.3 %) 157 (75.1 %) 41 (78.9%) ALBUMIN < 0.0001 Less than 3.4 117 (15.0%) 64 (30.6%) 17 (32.7%) Normal (3.4 5.4) 662 (84.9%) 145 (69.4%) 35 (67.3%) SYSTOLIC BLOOD PRESSURE 0.02 0 119 273 (35.0 %) 82 (39.2 %) 29 (55.8%) 120 129 165 (21.2 %) 46 (22 .0 %) 2 (3.8%) 130 139 152 (19.5%) 35 (16.8 %) 9 (17.3%) 140 and above 189 (24.3 %) 46 (22.0 %) 12 (23.1%) CHARLESON COMORBIDITY INDEX 2.3 < 0.0001 CHRONIC PAIN 0.48 Yes 82 (10.5 %) 27 (12.9 %) 7 (13.5%) PSYCHOLOGICAL PROBLEM 0.05 Yes 399 (51.2 %) 1 0 9 (52.2 %) 35 (67.3%) NUMBER OF HIGH RISK MEDICATIONS < 0.0001 0 284 (36.5 %) 44 (21.0 %) 12 (23.1%) 1 350 (44.9 %) 89 (42.6 %) 26 (50.0%) 2 132 (16.9 %) 66 (31.6 %) 11 (21.1%) 3 13 (1.7 %) 1 0 (4.8 %) 3 (5.8%) SUBSTANCE USE 0.52 Yes 140 (18.0 %) 32 (15.3 %) 10 (19.2%)
70 Table 4 1. Continued Number of Readmissions 0 N= 779 1 2 N= 209 3 or more N= 52 P value INPATIENT FACTORS ACUITY (N=969 ) 1.00 Non urgent 4 (0.5%) 1 (0.5%) 0 (0.0%) Urgent 740 (99.5%) 180 (99.5%) 44 (100.0%) LOCATION OF ADMISSION 0.02 Home 746 (95.8 %) 202 (96.7 %) 45 (86.5%) Assisted 12 (1.5 %) 4 (1.9 %) 2 (3.9%) Other 21 (2.7 %) 3 (1.4%) 5 (9.6%) DISPOSITION < 0.0001 Home 647 (83.1 %) 144 (68.9 %) 35 (67.3%) Assisted 115 (14.8 %) 54 (25.8 %) 16 (30.8%) Planned readmission 8 (1.0 %) 5 (2.4 %) 1 (1.9%) Other 9 (1.1 %) 6 (2.9 %) 0 (0.0%) AMBULATORY CARE SENSITIVE CONDITION (N=1035 ) 0.41 Yes 181 (23.4 %) 48 (23.0 %) 16 (30.8%) LENGTH OF STAY 3.8 4.8 < 0.0001 MORSE FALL SCORE 46.3 51.6 0.0002 GAIT/TRANSFERRING SCORE 3.5 6.0 < 0.0001 LEVEL OF ASSISTANCE NEEDED 0.05 Independent 271 (34.8 %) 58 (27.8 %) 15 (28.8%) Dependent 508 (65.2 %) 151 (72.2 %) 37 (71.2%) SOCIAL DETERMINANTS ETHNICITY 0.82 Not Hispanic or Latino 741 (95.1 %) 198 (94.7 %) 51 (98.1%) Hispanic or Latino 36 (4.6 %) 11 (5.3 %) 1 (1.9%) Patient refused 2 (0.3%) 0 (0.0%) 0 (0.0%) INSURANCE 0.0001 Public 468 (60.1 %) 146 (69.9 %) 45 (86.5%) Private 247 (31.7 %) 51 (24.4 %) 6 (11.5%) Self pay 6 4 (8.2%) 12 (5.7 %) 1 (2.0%) HEALTH LITERACY (N=791 ) 0.71 Adequate 457 (75.4 %) 108 (74.5 %) 28 (70.0%) Limited 149 (24.6 %) 37 (25.5 %) 12 (30.0%) MARITAL STATUS 0.67 Not married 507 (65.1 %) 130 (62.2 %) 36 (69.2%) Married 272 (34.9 %) 79 (37.8 %) 16 (30.8%)
71 Table 4 1. Continued Number of Readmissions 0 N= 779 1 2 N= 209 3 or more N= 52 P value PHYSICAL ABUSE (N=1032 ) 0.18 Denies 726 (93.8 %) 194 (93.7 %) 45 (88.2%) Denies, provider concerned 0 (0.0%) 1 (0.5%) 0 (0.0%) Unable to assess 28 (3.6 %) 8 (3.9 %) 5 (9.8%) Yes, past 19 (2.5 %) 4 (1.9 %) 1 (2.0%) Yes, present 1 (0.1 %) 0 (0.0%) 0 (0.0%) VERBAL ABUSE (N=1031 ) 0.24 Denies 722 (93.4 %) 1 92 (92.7 %) 45 (88.2%) Denies, provider concerned 0 (0.0%) 1 (0.5%) 0 (0.0%) Unable to assess 28 (3.6 %) 8 (3.9 %) 5 (9.8%) Yes, past 21 (2.7 %) 6 (2.9 %) 1 (2.0%) Yes, present 2 (0.3%) 0 (0.0%) 0 (0.0%) RACE 0.37 Majority 416 (53.4 %) 123 (58.9 %) 32 (61.5%) Minority 363 (46.6 %) 86 (41.1 %) 20 (38.5%) SEX 0.06 Female 453 (58.2 %) 107 (51.2 %) 25 (48.1%) SOMEONE LIVING WITH PATIENT (N=1038 ) 0.52 Not alone 629 (80.9 %) 165 (79.3 %) 38 (73.1%) Alone 149 (19.1 %) 43 (20.7 %) 14 (26.9%) SOMEONE HELPING WITH DAILY NEEDS (N=1028 ) 0.91 Not alone 758 (98.3 %) 203 (98.5 %) 50 (98.0%) None 13 (1.7 %) 3 (1.5 %) 1 (2.0%) ZIPCODE 0.34 West (Higher SES) 217 (27.9 %) 56 (26.8 %) 13 (25.0%) East (Lower SES) 292 (37.5 %) 64 (30.6 %) 20 (38.5%) 270 (34.6 %) 89 (42.6 %) 19 (36.5%) Average Denotes that variable has different N
72 Table 4 2. Mu ltivariate a nalysis of significant variables from univariate analysis, predicting number of readmissions Number of Readmissions 0 1 2 3 or more P v alue OUTPATIENT BEHAVIOR CANCELLED APPOINTMENTS REF 1.082 (0.978 1.197 ) 1.270 (1.051 1.534 )* 0.0 2 NO SHOW APPOINTMENTS REF 1.1 71 (0.999 1.373 ) 1.314 (0.982 1.757 ) 0.05 ATTENDED APPOINTMENTS REF 0.930 (0.867 0.99 7) 0.773 (0.682 0.877 ) 0.0001 TELEPHONE ENCOUNTERS REF 1.098 (1.031 1.170 )* 1.186 (1.053 1.335 )* 0.000 9 HEALTH STATUS ACTIVE MEDICATIONS REF 1.024 (0.950 1.104 ) 1.042 (0.891 1.218 ) 0.74 ACTIVE PROBLEMS REF 1.013 (0.928 1.106 ) 1.115 (0.895 1.390 ) 0. 6 1 ALBUMIN 0.0035 Less than 3.4 REF 1.989 (1.326 2.985)* 1.548 (0.745 3.216) Normal (3.4 5.4) REF REF REF SYSTOLIC BLOOD PRESSURE 0.1 9 0 119 REF REF REF 120 129 REF 1. 0 59 (0.678 1.654 ) 0.146 (0.034 0.670 ) 130 139 REF 0.817 (0.506 1.320 ) 0.682 (0.287 1.621 ) 140 and above REF 0.778 (0.499 1.212 ) 0.705 (0.326 1.527 ) CHARLESON COMORBIDITY INDEX REF 1.1 14 (1.038 1. 195 )* 1.1 4 2 (1.009 1.292 )* 0.0036 PSYCHOLOGICAL PROBLEM 0.4 8 Yes 0.915 (0.644 1.300 ) 1.425 (0.720 2.821 ) NUMBER OF HIGH RISK MEDICATIONS 0.0 8 0 REF REF REF 1 REF 1.237 (0.803 1.904 ) 1.467 (0.651 3.304 ) 2 REF 1.919 (1.162 3. 1 67 )* 1.091 (0.403 2.957 ) 3 REF 2.675 (1.016 7. 0 44 )* 3. 0 83 (0.606 15.69 4) INPATIENT FACTORS LOCATION OF ADMISSION 0.1 9 Home REF REF REF Assisted REF 1.431 (0.430 4.757 ) 2.482 (0.475 12.964 ) Other REF 0.754 (0.206 2.759 ) 4.042 (1.071 15.254 )* LENGTH OF STAY REF 1.013 (0.963 1.065 ) 1.080 (0.994 1.174) 0.19 MORSE FALL SCORE REF 0.999 (0.987 1.011 ) 1.012 (0.991 1.033) 0.54 GAIT/TRANSFERRING SCORE 0.07 0 REF REF REF 10 REF 1.280 (0.800 2.050 ) 0.694 (0.292 1.648 ) 20 REF 2.343 (1.176 4.669 )* 0.406 (0.087 1.883 )
73 Table 4 2. Continued Number of Readmissions 0 1 2 3 or more P value LEVEL OF ASSISTANCE NEEDED 0.4 2 Independent REF REF REF Dependent REF 0.875 (0.590 1.299 ) 0.627 (0.298 1.319 ) SOCIAL DETERMINANTS INSURANCE 0.5 4 Public REF REF REF Private REF 1.084 (0.717 1.639 ) 0.594 (0.228 1.544 ) Self pay REF 1.049 (0.526 2.092 ) 0.231 (0.027 1.966 ) SEX 0.0 9 Male REF 1.355 (0.968 1.896) 1.655 (0.880 3.112 ) Denotes Decreased Risk Denotes Increased Risk
74 CHAPTER 5 IDENTIFYING THE 1 2 READMISSIONS POPULATION : A CROSS SECTIONAL ANALYSIS FROM THE NATIONWIDE READMISSIONS DATABASE 5.1 Background Because hospitals with excessive readmissions receive reduced payments from the Inpatient Prospective Payment System (IPPS) and can also be fined, readmission reduction research is a top priority in the United States 3 In 2011, there were approximately 3.3 million adult readmissions in the US with an associated cost of $41. 3 billion 3 For Medicare and Medicaid patients alone, associated hospital costs due to readmissions totaled $5.1 billion. 3 While not all hospitalizations, including readmissions, are avoidable, various studies have shown that some are preventable. 4 For instance, an analysis of 2005 claims data showed that about three quarters of readmissions within 30 days were potentially preventable, equating to an estimated $12 billion in potentially unnecessary Medicare spending. 4 While a growing body of literature is present to assist in d ecreasing readmission rates, several key areas are not as well studied. For instance, a proportion of readmissions that are preventable are related to readmissions that have primary diagnoses codes that are associated with ambulatory care sensitive conditi ons (ACSC), and these a re understudied. ACSCs are a set of conditions with diagnoses for which timely and effective outpatient care can help to reduce the risks of hospitalizations, and therefore readmissions, by preventing acute illness and managing chron ic conditions. 63 In addition, the readmissions literature has predominately focused on characterizing the high readmitting population (i.e. those with 3 or more readmissions) versus others 3 as well as analyzing readmissions as one conti nuous outcome. 4 Little research has focused on understanding or characterizing the 1 2 readmissions per year population.
75 Surprisingly, this group of people contribute most to the total number of readmissions. Thus, to better understand their readmission behavior, researc h is needed to characterize this understudied subgroup with respect to social determinants of health, heath status variables, and inpatient factors, consistent with a biopsychosocial model of health 64 To our knowledge, no predictive readmissions models have been made to predict ACSC related readmissions and to characterize this 1 2 readmissions subgroup. Social determinants are an understudied area in readmissions even though they affect many health outcomes, including readmissions. Social determinants of health are defined as the conditions in which people are born, grow, work, live, and age, and the wider (i.e. systematic) forces that affect health risks and hea lth outcomes. 16 17 18 19 According to a recent article on PEARLS+ from Academic Medicine societal forces, social determinants, and health outcomes are connected. 23 The acronym PEARLS+ stands for politics, economics, environment, ethics, arms, religion, life circumstances, social roles, soc ial structures, and other forces. PEARLS+ are the key societal factors that influence social determinants of health, which then shape health outcomes, including readmissions. Examples of important social determinants include sex, insurance, patient locatio n, and median income. By definition, insurance, where patients live (i.e. patient location), and income, are variables related to all health outcomes, especially readmissions. 21 Health status variables, which have been studied in readmissions, define the baseline health of patients and are important given that those who are sicker tend to have a higher risk for readmissi ons. Health status variables include age, number of chronic health conditions and comorbidities. A s a ge i ncreases, health quality often
76 declines. Older age is thus a proxy for poor health, especially when combined with pain medication use 49 while number of chronic diseases 12 as well as specific comorbidities (e.g. congestive heart failure), 34 35 36 have also been reported to be correlated with readmissions. Inpatient factors are most commonly reported for readmissions research using common validated readmissions scores (i.e. HOSPITAL score, LACE index, and LACE+ index) 11 12 13 These inpatient factors include length of stay of index admission and disposition of the patient (i.e. where the patient was discharged to). For example, the Agency for Healthcar because it is a major factor in preventing readmissions. 50 65 In order to better understand t he roles of the aforementioned factors, we created the outpatient behavior, health status, inpatient factors, and social determinants of health (OHIS) t heoretical model (Figure 1 4). Although there were no variables available to represent outpatient behavior, we proceeded to use the rest of the OHIS theoretical model to inform this study. The aim of this study was to characterize the middle tier ACSC readmissions population (e.g. those who readmit 1 2 times in one year), using factors relating to social determinant, health status, and inpatient factors. We expected that a number of variable s would predict 1 2 readmissions within 1 year. For the social determinants variables females were hypothesized to have higher odds of having 1 2 readmissions compared to 0 readmissions, compared to their male counterparts, when controlling for all other covariates. It was also hypothesized that those with public insurance would have higher odds for 1 2 readmissions compared to 0 readmissions,
77 compared to other insurance types, when controlling for all other covariates. Also, we hypothesized that those wit h lower median incomes would have higher odds for 1 2 readmissions compared to 0 readmissions, compared to those with higher median incomes, when controlling for all other covariates. For the health status variables it was hypothesized that with every yea r increase in age, there would be higher odds of having 1 2 readmissions compared to 0 readmissions, when controlling for all other covariates. It was also hypothesized that with the presence of alcohol abuse comorbidity, drug abuse comorbidity, and compli cated diabetes comorbidity, there would be higher odds of having 1 2 readmissions compared to 0 readmissions, when controlling for all other covariates. For the inpatient factors variables it was hypothesized that for every day increase of length of stay, there would be higher odds of having 1 2 readmissions compared to 0 readmissions, when contro lling for all other covariates. 5.2 Methods 5.2.1 Study Overview The present study used a retrospective cross sectional design and was approved by the University of Florida Institutional Review Board. Data used for this study come from the 2013 Nationwide Readmissions Database (NRD), Healthcare Cost and Utilization Project (HCUP), Agency for Healthcare Research and Quality. It includes hospitalizations and readmis sions from January 1, 2013 to December 31, 2013. The NRD includes data from approximately 15 million discharges each year, representing approximately 35 million discharges each year, when weighted, and it provides nationally representative information on r eadmissions 51 This database includes 22 geographically dispersed states with verifiable patient linkage numbers. 51 These states account for 51.2% of the total US population and 49.3% of all US hospitalizations. 51
78 From the 14,325,172 hospi talizations in the 2013 NRD, the index admissions were abstracted (Figure 5 1) Index admissions were defined as the first admissions that qualify patients to have readmissions. After the index admissions were identified all other hospitalizations were deleted. In addition, for each of the index admissions, only primary diagnoses with ACSC related hospitalizations were kept (N=1,990,023). From this sample, only people aged 18 64 years old, with a known location after dis charge (i.e. disposition), and with a 0 14 days length of stay were included. Patients younger than 18 and older than 64 were deleted because the populations of interest were working age adults. Patients whose length of stays were longer than 14 days were deleted, because upon inspection of the distribution of the data, 95% of the hospitalizations had lengths of stays of 14 days or less, and because these admissions were likely to be substantially different given our within 30 day readmission criteria. If an index admission had any missing data for our key variables of interest, it was deleted. The final sample size was N= 716,216. Finally, we categorized the main outcome variable, number of readmissions into 0 readmissions, 1 2 readmissions, and 3 or more readmissions, within 2013 5.2.2 Social Determinants The social determinants included in the analysis were sex (female or male), insurance (no charge; other; private; public; self pay), patient location (central; fringe; metro area 250,000 999,999; metr o area 50,000 249,999; micropolitan; not metropolitan or micropolitan), and median income ($1 37,999; $38,000 47,999; $48,000 63,999; $64,000 more).
79 5.2.3 Health Status Factors The health status factors included in the analysis were age (18 64), number of chronic conditions (0 6), as well as a number of AHRQ comorbidity measures (yes/no) such as alcohol abuse, deficiency anemia, congestive heart failure, chronic pulmonary disease, depression, diabetes uncomplicated, diabetes with chronic complications, dru g abuse, hypertension, obesity, and psychoses 5.2.4 Inpatient Factor s The inpatient factors included in the analysis were discharge month, length of stay (0 14), admission day (weekend or nonweekend), and disposition (routine; transfer to short term hos pital; skilled nursing, intermediate care, other facility; Home Health Care; against medical advice) 5.2.5 Statistical Analyses For each of the 21 variables selected, bivariate analyses were conducted in order to assess the relationships with the outcome variable, number of readmissions. Chi square tests were conducted for categorical variables, and ANOVAs were conducted for continuous variables. Once significant correlations were identified from the bivariate analyses, these variables were entered into a n adjusted multinomial logistic regression. The analysis for this paper was generated using SAS software, Version 9.4 5.3 Results Of the 1,658,963 (unweighted=716,216) people with an admission whose de identified data was included in this study, 276,523 (unweighted=120,303) people had 1 2 readmissions in 2013. On average, those with 1 2 readmissions were 49 years old, mostly female, had 6 or more chronic conditions, had public insurance, lived in central locations (i.e. areas with >1 million people), and had low incomes (i.e. $1 $37,999). For
80 the bivariate analysis with the outcome of readmissions, all of the tested variables were sig nificant (Table 5 1). For the multivariate analysis there were both decreased and increased risk effects after controlling for all other covariates, as seen in Table 5 2. For every year increase in age, there was a decreased odds of having 1 2 readmissions (AOR: 0.992), compared to 0 readmissions. Females had less likelihood of having 1 2 readmissions than males (AOR: 0.957), compared to 0 readmissions. Some of the comorbidities such as hypertension (AOR: 0.972) and obesity (AOR: 0.768), decreased odds of 1 2 readmissions, compared to 0 readmissions. Being discharged during a month later in the year versus earlier in the year decreased odds of having 1 2 readmissions (AOR: 0.974), compared to 0 readmissions. Insurance type mattered; any insurance other than public lower odds of having 1 2 readmissions (No charge AOR: 0.659, Other AOR: 0.735, Private AOR: 0.623, Self pay AOR: 0.597), compared to 0 readmissions. Living in a less populated area compared to central populated locations, reduced odds of 1 2 readmis sions (Fringe AOR: 0.983, Metro 1 AOR: 0.957, Metro 2 AOR: 0.968, Micropolitan AOR: 0.863, Not metropolitan or micropolitan AOR: 0.856), compared to 0 readmissions. Patients with median incomes $48,000 or more, compared to median incomes of $1 $37,999, h ad slightly lower odds of having 1 2 readmissions (AORs: 0.967) compared to 0 readmissions. Having a median income of $64,000 or more, compared to median incomes of $1 $37,999, reduced odds of having 1 2 readmissions (AOR: 0.937) compared to 0 readmissio ns. As number of chronic conditions increased, the odds of having 1 2 (1 AOR: 1.589, 2 AOR: 2.133, 3 AOR: 2.543, 4 AOR: 2.960, 5 AOR: 3.361, 6 or more AOR:
81 4.154) increased in a linear manner compared to 0 readmissions, after controlling for all other co variates. Many of the comorbidities such as anemia (1 2 readmissions AOR: 1.286), congestive heart failure (1 2 readmissions AOR: 1.240), depression (1 2 readmissions AOR: 1.037), diabetes uncomplicated (only 1 2 readmissions AOR: 1.036), diabetes complica ted (1 2 readmissions AOR: 1.065), drug abuse (1 2 readmissions AOR: 1.192), and psychoses (1 2 readmissions AOR: 1.116), independently increased risk of readmission. For every day increase in length of stay, there were increased odds for having 1 2 (AOR: 1.043) compared to 0 readmissions, after controlling for all other covariates. Patients who were admitted on the weekend, compared to those admitted on weekdays, had a slightly higher odds of having 1 2 (AOR: 1.037), compared to 0 readmissions, after contr olling for all other covariates. Those who were sent to all other facilities compared to routine disposition, had higher odds of having 1 2 readmissions (Transfer to short term hospital AOR: 1.304, Skilled nursing AOR: 1.101, Home Health Care AOR: 1.353, A gainst medical advice AOR: 2.143) compared to 0 readmissions, after controlling for all other covariates. For the 3 or more readmissions group, for every year increase in age, there were lower odds of 3 or more readmissions compared to 0 readmissions (AOR : 0.964). Those who were obese had increased odds of 3 or more compared to 0 readmissions than those without obesity (AOR: 0.595). As the discharge month went later in the year, there was a decreased odds of having 3 or more readmissions (AOR: 0.960), comp ared to 0 readmissions. Females had decreased odds of 3 or more readmissions than men (AOR: 0.862) compared to 0 readmissions. Other factors that decreased the likelihood of 3 or more readmissions included having insurance types other than public, having a
82 median income of $64,000 or more, and living in a less populated rather than central, populated areas. Number of chronic conditions increased the likelihood of 3 or more readmissions, compared to 0 readmissions. Those with 6 or more chronic conditions had 16.4 times higher odds of 3 or more readmissions compared to 0 readmissions. A number of comorbidities independently increased odds of 3 or more readmissions compared to 0 readmissions (alcohol abuse, deficiency anemia, congestive heart failure, depressio n, diabetes with chronic complications, drug abuse, hypertension, and psychoses). In addition, length of stay increased odds of 3 or more readmissions compared to 0 readmissions (AOR: 1.015 per day increase) while those admitted on the weekend versus weekd ay admission had increased odds of 3 or more readmissions. Leaving the hospital against medical advice versus routine disposition also increased odds of 3 or more readmissions compared to 0 readmissions (AOR: 3.885). 5.4 Conclusions We set out to predict t he largest group of readmissions: 1 to 2 a year. To do so, we compared to 0 readmissions. We also predicted the high readmissions, those 3 and above in a year. Our model, in general, was predictive. For instance, alcohol abuse, drug abuse, and diabetes a lone increased risk for 1 to 2 readmissions; number of chronic conditions also increased the risk. Longer lengths of stay increased risk for being in both the 1 2 and 3 or more readmissions groups. Similarly, the social determinants of having public insura nce and lower median income increased the risk of being in one of the readmissions group over having another type of insurance and a higher median income. However, some results were not as hypothesized. Every year increase in age, for example, results in lower odds of 1 2 and 3 or more readmissions. We suspected that
83 as patients age, more health conditions would arise and complicate overall health for people, and result in more hospitalizations and readmissions. This was hypothesized because readmissions especially affect older adults due to the decline in functional status affecting their quality of life and overall well being 5 One likely explanation for this opposite finding may be related to the fact that our sample was comprised mostly of younger and middle aged adults. T here is an increasing view that middle age is a pivotal period in the life course in terms of balancing growth and decline, linking earlier and later periods of life, and bridging younger and older generations. Thus, in this age range, increasing age may b e associated with an increased focus on health promotion and well being, which can have a significant positive impact 66 Perhaps there is a protective effect to a certain age, but then a switch from a negative slope to a positive one, once adults cross the threshold of being older adults (i.e. 65 ye ars and older) 67 Future research in ambulatory care related readmissions among older samples are urgently needed. Of particular interest, females had lower odds of readmissions compared to their male counterparts. Previous literature has not been clear on the role of sex. Some analyses have shown that men are at higher risk of readmissions for all cause hospital readmissions among non older adult patients (e.g. Medicaid men have increased risk of all cause hospital readmissions) 41 Other studies, such as one in an Italian Department of Medicine, have shown that women have higher risk of readmissions 43 Another unexpected finding was that obesity reduced re admissions, that it was protective. It would be expected that when a comorbidity, such as obesity, is present,
84 expect more hospitalizations and readmissions. However, we found the opposite. This observed protective effect is called the obesity paradox 68 A recent study published in JAMA Cardiology has mentioned that the obesity paradox is an ar tificial phenomenon due to lead time bias with the use of cross sectional data 69 In this recent article, the researchers used a life course perspective, which eliminated the lead time bias and showed that the obesity paradox di d not hold true 69 In addition, it should be noted that BMI, which is used to measure obesity, is a crude and flawed anthropometric biomarker that does not take into consideration other important factors such as fat mass, nutrit ional status, cardiorespiratory fitness, distribution of body fat, and other factors that sed risk of type 2 diabetes, cardiovascular disease and death, even after controlling for BMI 70 71 Therefore, given the limitations of BMI as a measure of obesity, its effect on readmissions may not be clear 69 The most surprising finding of this stu dy is the role of disposition to a skilled nursing, intermediate care, or other facility. For the 1 2 readmissions group, there were increased odds of readmission, while the 3 or more readmissions group there decreased odds of readmission as seen in Table 5 2 There may be several explanations accounting for our findings. First, it is possible that sending the 1 2 readmission patients to skilled facilities negatively impacted their health. We can speculate that being away from their house and family can in deed worsen, instead of improve their health conditions. Alternatively, location of disposition may not be as relevant to readmissions as social support and the support of health maintenance are
85 (e.g. if they did not have family at home or appropriate supp ort to follow medical advice, getting medications, etc.). In general, our findings support the importance of having skilled healthcare personnel and settings to keep patients out of the hospital for those patients who tend to be sicker and have more hospit alizations. However, this appears not be the same for the 1 2 readmissions patients. More research is needed to further elucidate the differences in patient subgroups as well as the potential characteristics explaining such differences. For effective trans ition of care and prevention of subsequent hospitalizations, guidelines recommend assurance of appropriate disposition, e.g. to a skilled nursing facility 50 65 In transition of care, it is important to assure that patients are discharged to the appropriate location to prevent subsequent admissions to the hospital 65 The Agency for Healthcare Research and Quality Readmissions, specifically requires readmissions 50 Though there were many strengths in this study, such as the large sample size and the focus on ACSC related readmissions, there are limitations to be noted. First, we were limited in what variables we could analyze to predict rea dmissions. The NRD is a nationwide database, and the database distributer, HCUP, has explained that sensitive variables were excluded because they could be patient identifiers. This is a limitation because these types of variables may be able to be more pr edictive of ACSC related readmissions, than the de identified variables included. Another limitation of this study is the inability to know whether the actual index admission occurred in 2012 and if additional readmissions occurred in 2014. This is due
86 to the fact that hospitalizations included in this dataset were strictly between January 1, 2013 December 31, 2013. For example, if a patient had an index admission on December 25, 2012, and readmitted on January 10, 2013, the readmission would be identifie d as the index admission. This type of coding was another protective feature made by data distributor, HCUP, to protect patients by making un linkable hospitalizations from past and future years. Therefore, we may have lost some patients in our analyses an d we were not able to follow their behavior over time. We also do not know if patients passed away in, for instance, 2014, due to readmissions in 2013. For future directions, we recommend expanding the age range to include the older adults, to see the true role of age in readmissions. We suspect that once people reach the threshold of older adults, there will be a change in the role of age. In addition, we recommend the use of other readmissions data sources to cover areas addressing additional social deter minants or other factors relating to health behavior before hospitalizations (e.g. number of attended primary care clinic appointments). This may include the incorporation of local electronic medical record (EMR) data that collects more specific data. The use of EMR data would increase the variability in the type of data analyzed. We also suggest incorporating a mixed methods approach (i.e. qualitative data). EMR data is limited to how healthcare providers record information in the health record, therefore, additional information collected from both patients and providers can fill in these gaps.
87 Figure 5 1. Sample selection of the 2013 NRD Table 5 1. Weighted d escriptive a nalysis of var iables by readmissions group (0 readmissions, 1 2 readmissions only, or 3 or more readmissions) Number of Readmissions 0 N= 1,373,395 1 2 N= 276,523 3 or more N= 4,038 P value AGE <0.0001 # CHRONIC COND <0.0001 0 40,817 (3.0%) 2,100 (0.8%) 23 (0.3%) 1 94,940 (6.9%) 7,976 (2.9%) 239 (2.6%) 2 133,780 (9.7%) 15,231 (5.5%) 517 (5.7%) 3 159,467 (11.6%) 22,441 (8.1%) 774 (8.6%) 4 168,465 (12.3%) 28,326 (10.2%) 1,012 (11.2%) 5 164,933 (12.0%) 32,675 (11.8%) 1,151 (12.7%) 6 or more 610,993 (44.5%) 167,776 (60.7%) 5,328 (58.9%) COMORBIDITIES ALCOHOL ABUSE <0.0001 Yes 79,475 (5.8%) 18,741 (6.8%) 885 (9.8%)
88 Table 5 1. Continued Number of Readmissions 0 N= 1,373,395 1 2 N= 276,523 3 or more N= 4,038 P value DEFICIENCY ANEMIA <0.0001 Yes 249,860 (18.2%) 73,197 (26.5%) 2,760 (30.5%) CONGESTIVE HEART FAILURE <0.0001 Yes 106,766 (7.8%) 34,306 (12.4%) 1,059 (11.7%) CHRONIC PULMONARY DISEASE <0.0001 Yes 260,417 (19.0%) 62,318 (22.5%) 1,716 (19.0%) DEPRESSION <0.0001 Yes 195,624 (14.2%) 47,655(17.2%) 1,657 (18.3%) DIABETES UNCOMPLICATED <0.0001 Yes 267,806 (19.5%) 61,168 (22.1%) 1,673 (18.5%) DIABETES WITH CHRONIC COMPLICATIONS <0.0001 Yes 116,050 (8.4%) 34,183 (12.4%) 1,128 (12.5%) DRUG ABUSE <0.0001 Yes 99,814 (7.3%) 28,490 (10.3%) 1,632 (18.0%) HYPERTENSION <0.0001 Yes 686,511 (50.0%) 153,619 (55.6%) 5,286 (58.4%) OBESITY <0.0001 Yes 280,197 (20.4%) 55,471 (20.1%) 1,376 (15.2%) PSYCHOSES <0.0001 Yes 103,732 (7.6%) 29,177 (10.6%) 1,228 (13.6%) DISCHARGE MONTH mid <0.0001 LENGTH OF STAY <0.0001
89 Table 5 1. Continued Number of Readmissions 0 N= 1,373,395 1 2 N= 276,523 3 or more N= 4,038 P value ADMISSION DAY 0.0009 Weekend 328,794 (23.9%) 67,016 (24.2%) 2,403 (26.6%) DISPOSITION <0.0001 Routine 1,125,574 (82.0%) 196,353 (71.0%) 6,433 (71.1%) Transfer to short term hospital 8,827 (0.6%) 2,180 (0.8%) 46 (0.5%) Skilled nursing, intermediate care, other facility 70,671 (5.1%) 20,840 (7.5%) 385 (4.3%) Home Health Care 132,576 (9.7%) 43,239 (15.6%) 1074 (11.9%) Against medical advice 35,747 (2.6%) 13,911 (5.0%) 1106 (12.2%) SEX <0.0001 Female 733,974 (53.4%) 142,066 (51.4%) 4,328 (47.9%) INSURANCE <0.0001 No charge 23,602 (1.7%) 3,341 (1.2%) 177 (2.0%) Other 68,499 (5.0%) 11,328 (4.1%) 379 (4.2%) Private 418,506 (30.5%) 55,277 (20.0%) 1,132 (12.5%) Public 693,349 (50.5%) 185,681 (67.1%) 6,588 (72.8%) Self pay 169,439 (12.3%) 20,897 (7.6%) 768 (8.5%) PATIENT LOCATION <0.0001 Central (>=1 million) 383,823 (27.9%) 83,033 (30.0%) 3,699 (40.9%) Fringe (outside Central) 324,254 (23.6%) 63,543 (23.0%) 2,007 (22.2%) Metro area 1 (250,000 999,999) 258,422 (18.8%) 53,274 (19.3%) 1,658 (18.3%) Metro area 2 (50,000 249,999) 134,349 (9.8%) 27,398(9.9%) 713 (7.9%) Micropolitan 157,645 (11.5%) 28,656 (10.4%) 577 (6.4%) Not metropolitan or micropolitan 114,902 (8.4%) 20,619 (7.4%) 389 (4.3%) MEDIAN INCOME <0.0001 $1 $37,999 522,818 (38.1%) 111,853 (40.4%) 3,784 (41.8%) $38,000 $47,999 377,798 (27.5%) 76,016 (27.5%) 2,425 (26.8%) $48,000 $63,999 287,018 (20.9%) 55,276 (20.0%) 1,855 (20.5%) $64,000 more 185,762 (13.5%) 33,378 (12.1%) 980 (10.8%)
90 Table 5 2. Mu ltivariate a nalysis of significant variables from univariate analysis, predicting number of readmissions Number of Readmissions 0 1 2 3 or more P v alue AGE REF 0.992 (0.991 0.993) 0.964 (0.961 0.967) <0.0001 # CHRONIC COND <0.0001 0 REF REF REF 1 REF 1.589 (1.475 1.712)* 4.611 (2.416 8.800)* 2 REF 2.133 (1.987 2.290)* 7.165 (3.806 13.491)* 3 REF 2.543 (2.371 2.727)* 9.224 (4.915 17.308)* 4 REF 2.960 (2.761 3.174)* 11.583 (6.178 21.714)* 5 REF 3.361 (3.134 3.605)* 13.720 (7.317 25.727)* 6 or more REF 4.154 (3.876 4.452)* 16.493 (8.805 30.894)* COMORBIDITY ALCOHOL ABUSE 0.0007 Yes REF 0.990 (0.964 1.016) 1.229 (1.101 1.371)* DEFICIENCY ANEMIA <0.0001 Yes REF 1.286 (1.266 1.305)* 1.579 (1.470 1.695)* CONGESTIVE HEART FAILURE <0.0001 Yes REF 1.240 (1.214 1.266)* 1.363 (1.231 1.508)* CHRONIC PULMONARY DISEASE 0.1415 Yes REF 1.010 (0.994 1.026) 0.940 (0.867 1.019) DEPRESSION 0.0001 Yes REF 1.037 (1.018 1.056)* 1.149 (1.054 1.253)* DIABETES UNCOMPLICATED <0.0001 Yes REF 1.036 (1.019 1.054)* 1.027 (0.943 1.119) DIABETES WITH CHRONIC COMPLICATIONS <0.0001 Yes REF 1.065 (1.043 1.088)* 1.180 (1.065 1.307)*
91 Table 5 2. Continued Number of Readmissions 0 1 2 3 or more P v alue DRUG ABUSE <0.0001 Yes REF 1.192 (1.166 1.218)* 1.574 (1.446 1.714)* HYPERTENSION <0.0001 Yes REF 0.972 (0.958 0.986) 1.222 (1.139 1.311)* OBESITY <0.0001 Yes REF 0.768 (0.756 0.781) 0.595 (0.544 0.651) PSYCHOSES <0.0001 Yes REF 1.116 (1.092 1.141)* 1.344 (1.222 1.477)* DISCHARGE MONTH REF 0.974 (0.972 0.975) 0.960 (0.951 0.968) <0.0001 LENGTH OF STAY REF 1.043 (1.040 1.045)* 1.015 (1.003 1.028)* <0.0001 ADMISSION DAY <0.0001 Weekend REF 1.037 (1.022 1.053)* 1.091 (1.017 1.171)* DISPOSITION <0.0001 Routine REF REF REF Transfer to short term hospital REF 1.304 (1.210 1.404)* 1.090 (0.721 1.647) Skilled nursing, intermediate care, other facility REF 1.101 (1.072 1.130)* 0.727 (0.622 0.849) Home Health Care REF 1.353 (1.327 1.379)* 1.090 (0.983 1.210) Against medical advice REF 2.143 (2.077 2.211)* 3.885 (3.518 4.290)* SEX <0.0001 Female REF 0.957 (0.945 0.970) 0.862 (0.808 0.920)
92 Table 5 2. Continued Number of Readmissions 0 1 2 3 or more P v alue INSURANCE <0.0001 No charge REF 0.659(0.622 0.6 98) 0.694 (0.541 0.890) Other REF 0.735 (0.714 0.758) 0.659 (0.568 0.763) Private REF 0.623 (0.613 0.633) 0.382 (0.346 0.422) Public REF REF REF Self pay REF 0.597 (0.583 0.612) 0.541 (0.482 0.607) PATIENT LOCATION <0.0001 Central (>=1 million) REF REF REF Fringe (outside Central) REF 0.983 (0.966 1.001) 0.790 (0.726 0.860) Metro area 1 (250,000 999,999) REF 0.957 (0.940 0.975) 0.742 (0.681 0.809) Metro area 2 (50,000 249,999) REF 0.968 (0.946 0.991) 0.625 ( 0.554 0.705) Micropolitan REF 0.863 (0.842 0.884) 0.443 (0.383 0.512) Not metropolitan or micropolitan REF 0.856 (0.832 0.881) 0.458 (0.386 0.543) MEDIAN INCOME <0.0001 $1 $37,999 REF REF REF $38,000 $47,999 REF 0.987 (0.971 1.003) 0.996 (0.921 1.078) $48,000 $63,999 REF 0.967 (0.950 0.984) 0.977 (0.897 1.064) $64,000 more REF 0.937 (0.917 0.958) 0.863 (0.775 0.962) Denotes Decreased Risk Denotes Increased Risk
93 CHAPTER 6 CONCLUSION Readmissions in the United States is a top priority because hospitals with excessive readmissions receive reduced payments from the Inpatient Prospective Payment System (IPPS) and may incur penalties. 3 In the 2017 fiscal year, $528 million was not given to hospitals. According to the analysis of Centers for Medicare and Medicaid Services (CMS) data, $564 million will be withheld for fiscal yea r 2018, which is an increase from 2017. 1 To add to the fisca l consequences facing hospitals due to readmissions, 2,573 hospitals will face penalties in fiscal year 2018. 1 Rehospitalizations put patients at additional risk for poor health outcomes. The 30 days after a hospitalization is considered to be the most vulnerable period for a rehospitalization, or readmission, due to factors such as lack of familial or similar social support and unresolved pain and slow healing. 7 Patients may get frus trated, and may neglect warning signs, eventually leading these patients back to the hospital. This cycle of readmissions can reoccur and can have emotional and mental impacts on patients. In addition to these discomforts, being in the hospital exposes pat ients to higher risks of health care associated infections, excess mortality and higher costs to the economy. 8 Also, when patients readmit to the hospital, there are implications related to quality of care. One implication may be that the quality of care during the index hospitaliz ation was not appropriate. For instance, future care plans may have been insufficient. 7 With the current urgency to study readmissions, there are studies and initiatives focused on readmissions, but there are many gaps in the current research. First, readmissions predictive tools have been validated and are widely used such as the HOSPITAL score, LACE index, LACE+ index, and the readmission risk score. 11 12
94 These scores include inpatient factors (e.g. length of stay, acuity, and emergency department use) and health status factors (e.g. comorbidity of patient and age). Yet these predictive tools still have weaknesses. A study comparing the 4 tools and their abilitie scores of 0.66 0.68. 14 The interpretation of these scores is that the models did not include some of the important predictors of rehospitalization. Next, the readmissions literature describes the importance of appropriate transition of care (i.e. disposition) to prevent readmissions to the hospital. 50 65 A step order to prevent readmissions. Yet, the role that primary care plays in p reventing the Agency for Healthcare Research and Quality (AHRQ). 2 AHRQ stated that though there are readmissions tools and initiatives, they heavily focus on factors in the inpatient setting not on the clinical setting the patient used before and after the inpatient episodes. Another gap in the readmissions literature is identifying factors related to preventable readmissions. Some readmissions analyses have reported that a proportion of readmissions are, indeed, preventable. 4 Preventable readmissions or hospitalizations, can be related to ambulator y care sensitive (ACS) diagnoses. Ambulatory care sensitive conditions (ACSC) are defined as conditions with diagnoses for which effective and timely outpatient care can help to reduce the risks of hospitalizations, and therefore readmissions, by preventin g acute illness and managing chronic conditions. 63 Clearly, ACSCs are related to ambulatory care modifiable factors.
95 What happens in the clinic may be most important, especially as it related to the ACSCs. The last gap in the readmissions literature is the characterization of the 1 2 readmissions per yea r population. Studies of readmissions have predominantly focused on characterizing the high readmitting population (3 or more readmissions per year) compared to others, as well as analyzing readmissions as one continuous outcome. 4 Surprisingly, this 1 2 readmissions per year group contribute most to the total number of readmissions but have generally been ignored. A prominent prevention epidemiologist has called for recognition that it is not the outliers who drive trends in public health, but rather the people who are mo re central to the distribution of outcomes. 72 This dissertation followed the wisdom to explicate modifiable factors affecting readmissions. In order to add to the readmissions literature, this dissertation attempted to address these important gaps. To address t hese important gaps, we created the OHIS theoretical model (Figure 1 4) to help inform all 3 studies. While there are many gaps in the readmissions literature, the principal focus of this dissertation was to identify primary care relevant factors in readmi ssions. To do so, first, we created a readmissions predictive tool for primary care using data from a local sample of family medicine department patients. Second, we characterized the 1 2 readmissions population among a local sample of family medicine depa rtment patients. Third, we characterized the 1 2 ambulatory care sensitive readmissions population among a nationally representative sample, using the Nationwide Readmissions Database (NRD). In addition, to inform our quantitative findings of this disserta tion, a few qualitative interviews were conducted to help enrich our understanding of readmissions with the patient perspective.
96 6.1 Main Findings The results of Chapter 3 (creation of a readmissions predictive tool for primary care), showed that our OHIS theoretical model was correct in the assignment of items to certain factors. The items that were chosen for their respective constructs or factors (as was successful in reducing the number of items from 23 to 16. Our findings are unique because this is the first study to our knowledge that attempts to create a re admissions predictive tool for primary care. Also, this is the first study to our knowledge that incorporates outpatient behavior (e.g. cancelled clinic appointments) and detailed social determinants (e.g. whether someone lives at home with the patient) in readmissions research. This EFA and subsequent CFA were first steps. While the CFA yielded a poor fit to a second sample of data, our results point to the importance of continuing to investigate the role that outpatient behavior and detailed social determ inants play in readmissions. In Chapter 4 (characterization of the 1 2 readmissions population among a local sample of family medicine department patients), we tested whether the variables of the OHIS theoretical model predicted 1 2 readmissions. In term s of outpatient behavior, we found that number of phone encounters was significant in predicting 1 2 readmissions. However, increased phone encounters predicted increased odds for 1 2 readmissions, compared to 0 readmissions, which was opposite than hypoth esized. We suspected that the phone encounters were not just between the patient and primary clinic, but also between patient and other clinics within the same health system. We also found that cancelled clinic appointments were significant in predicting 3 or more readmissions. Therefore, this finding provides evidence that these behaviors may be useful in clinics to
97 potentially prevent patients from becoming high readmitters (i.e. 3 or more readmissions). Another novel finding was being on 2 or 3 high risk medications (i.e. diabetic, opioid, anticoagulant) was predictive of increased odds of 1 2 readmissions, compared to 0 readmissions. This is the first study to our knowledge that investigates the role of high risk medications and their relationship to rea dmissions. We also tested whether gait/transferring score predicted 1 2 readmissions, which resulted in those with impaired gait compared to normal gait had increased odds of 1 2 readmissions compared to 0 readmissions. Therefore, outpatient behaviors, num ber of high risk medications, and gait/transferring score may be areas that can be monitored and could be the focus of interventions to prevent patients from having the first readmission. Chapter 5 yielded generalizable findings, because it was an analys is of a national sample of hospitalizations and readmissions. We analyzed the factors predicting ACSC related readmissions, and many of them were predictive of 1 2 readmissions. One surprising finding was for every year increase in age, there was a lower o dds of 1 2 readmissions, compared to 0 readmissions. This was opposite of what we hypothesized; however, this finding may be due to the younger and middle aged adults included in the sample (18 64 years of age). We suspect that had we replicated the study in a sample enriched with in older aged adults, we would have found an increased odds of 1 2 readmissions. We also found that the presence of obesity comorbidity had a protective effect for 1 2 readmissions. This falls in line with the obesity paradox, tho ugh a recent study published in JAMA Cardiology mentioned that the obesity paradox is an artificial phenomenon due to lead time bias with the use of cross sectional data. 69 The most surprising finding of this analysis was the ro le of disposition to a skilled nursing,
98 intermediate care, or other facility. For the 1 2 readmissions group, there were increased odds of readmission based on this disposition, while in the 3 or more readmissions group there were decreased odds of readmis sion. We suspect that sending the 1 2 readmissions patients to skilled nursing facilities can negatively impact their health, because being away from home may worsen conditions rather than improve conditions, and because they may be exposed to additional h ospital infectious agents, or may have another detrimental exposure. However, these patients also have been more likely to readmit under this disposition because of their home situation. They may have been unable to be discharged home due to lack of resour ces to sustain their health in that environment (e.g. no power when need oxygen delivered) or due to lack of caregiving or support in their homes. More research is needed to understand the role of disposition and its relationship to risk of readmission. Interestingly, Chapters 4 and 5, though using different data sources, yielded an important finding: there are distinct differences between the 0 readmissions group vs. 1 2 readmissions group, and the 0 readmissions group vs. the 3 or more readmissions grou p. This dissertation sought out to identify characteristics of the 1 2 readmissions group and posed that this distinct group may have different characteristics compared to the other readmissions groups. Since different characteristics were identified, furt her study on these characteristics are needed because these may reveal points of intervention. To provide an additional enriched understanding to this exploration of the factors relevant to rehospitalization of the 1 2 readmission group, I conducted a fe w qualitative interviews of patients whose characteristics matched our readmission group. Many
99 themes emerged from the interview such as the role of social support, the relationship with healthcare, trust in primary care doctor, patient doctor communicati on, scheduling, presence of pain, normalcy of conditions/pain, and the desire to find a solution. One participant mentioned throughout the interview the importance of social support from his wife, daughter, and church. For instance, he stated that the chur ch was the site of The paramedic at church called the having help getting around the house or errands, the participant re vealed that he and his wife lived with their daughter and her family. Another participant spoke about the help she receives when she is not feeling well or needs help with anything She stated, church family is more than my real family. I only have one child, my son, who liv es in One participant also spoke about his relationship with healthcare. At first, he described the in the conversation, when we asked about his one wish about his readmission, h e revealed that he actually had a negative experience during his stay at the hospital.
100 e of trust in primary care doctor emerged. When talking about his vein, he referred to his follow parti cipant expressed his preference and trust in his primary care doctor over the issues. Particularly, here is where the theme of patient doctor communication emerged. The Another participant talked about how he felt that the people are n ot doing their jobs. They are not taking care of patients. I think, sometimes, I He also stated that he hated going to the hospital but follows the going to the hospital. But every time for the last three months, the clinic ca lls me and says you need to go to the hospital, you need to go to the hospital. Quite naturally, with my health problems, they are going to keep me. I know every time that I go to the hospital, they are going to keep me. Another participant reported havin g a lot of trust in
101 thought he was just being kind of mean you know B ut as I looked into it I realized that he was just protecting me as well as himself. S o you know, he wants wha t's best for me. When ut the whole goal of the doctors a nd myself is to get me mobile again and you know healthy S o I understand that they're helping me and they want to help me They want what's best for me. I switched all my doctors now, all of them I switched over to Shands so I don't have a doctor here and a doctor there I go to East Side to the clinic now ; I have m y general doctors there M y orthopedist is with you guys my neurosurgeon is with you Everybody's with you and the continuity of care is so much better because everybody knows what they are they're all the same page going on They can go in the computer pull up everything that anybody else has e ver done for me and look at it and know you know h I see here they're doing this just it's a good feeling most recent primary care doctor visit, revealed scheduling issues at the clinic. When ask ed if they scheduled a follow up appointment to figure out what was going on with the blood in the urine, he revealed, scheduled. (Muffled) My wife was at the desk and the 2 nd Another participant revealed that he does not like how he does not get to see the same doctor each time he goes to the clinic, due to th time that I go to the clinic
102 Anoth er participant stated that she preferred going to the primary care clinic, however mentioned issues about times the clinics were open. She ell I prefer go into primary care if I can get to them and if they are open. I prefer to, first, go there before I go to the hospital. But I know myself pretty well and I The theme of pain also emerged in the conversation s. First, one participant mentioned his index hospitalization reason being ches t pains that he felt while in condition/pain emerged. He described that Also when asked whether he thought that the readmission could have been prevented by listening to warning signs, such as his body not feeling right. He answered with a Another participant spoke about his arthritis and identified that could be done for it. He stated through physical therapy an
103 Another participant talked about her pain, and how pain was the I'm in so muc h pain that going to the clinic is really not going to help I have to go to the ER. You know, going to the clin i c and sitting there when I was in excruciating pain is just One of the participant s seemed to blame the quality of care for his readmission time, I had diarrhea for a week. And th rowing up for almost a week. And I had been to Though it was clear that even with some issues with the communication with doctors, the participant still trusts in his primary care doctor, and is hopeful that they would find out what is wrong with him. Another participant also talked about the quality of care he received in the hospital. Yes, yes, yes. The doctors are teaching the students. They let my, uh, water, let me gain over 15 pounds in five days. I was really upset about it. I said here you are teaching these students but yall come in here every clinic is on it. So what are y all going to do? Let me sit in the hospital and gain all of this weight? Another participant talked about the excellent care she received from her doctor and app reciated the hon I believe he can fix me or at least where I can have some quality of life. This is the first step. You know, when we were going into it he told me, and I like th at he's honest with me XXX
104 I'm no t happy with what I did I'm not happy with the outcome I wasn't able to do what I actually he felt like he let me down because he did not do what we had both talked about doing Almost all doctor s The themes that emerged from the qualitative findings relate back to the OHIS theoretical model. Outpatient behaviors was represented through the themes of clinic scheduling and relationship with healt hcare. Health status was represented by the theme of pain. Inpatient factors was represented by the theme of quality of care. Social determinants was represented by the theme of social support. These themes are also interrelated to one another, particularl y the inpatient and outpatient factors, which show how the community and hospital settings interact with one another, as the larger forces contributing to the outpatient behaviors, health status, inpatient factors, and social determinants constructs. Over all, this dissertation provided evidence that the components of the OHIS theoretical model predict readmissions. This is the first investigation incorporating outpatient behaviors, specific health status factors (e.g. number of high risk medications and albumin), specific inpatient factors (e.g. gait/transferring score), and detailed social determinants of health (e.g. whether someone lives with the patient at home, in the same predictive model. Given our findings, primary care clinics mi ght benefit from monitoring health. Factors related to outpatient behavior and some social determinants may be points of intervention, to prevent any readmissions. Chapter 5 provi ded novel evidence for the 1 2 ACSC readmissions group. This is the first study to our knowledge to solely
105 look at ACSC related readmissions as well as characterize the 1 2 readmissions population, in a nationally representative sample. When reflecting on the findings of the quantitative and qualitative studies, there are some overlap and important lessons learned. First, social support came through in the interviews. We also saw that this factor was relevant in the findings of Chapter 3, with marital statu s and someone living with patient as factors that hung together in the EFA. However it should be noted, that whether social support had a positive or negative effect on whether a participant was rehospitalized or not, needs to be further explored. Next, th e relationship with healthcare theme (particularly the subtheme of trust in primary care doctor) and scheduling theme inform the quantitative findings of Chapters 3 and 4. Specifically, Chapter appointme nts, and phone encounters hung together. We hypothesize that because the participant described trust in his primary care doctor and his preference for seeing his primary care doctor (over going to the hospital), it is likely that appointments to the primar y care clinic would be made and kept. Here, the theme of scheduling is important to note, because scheduling and communication barriers with the primary care clinic can be points of intervention for preventing readmissions. Lastly, the presence of pain so mewhat explained the findings from all 3 analyses. For Chapter 3, number of high risk medications was significant in the EFA. One of the medications included in this categorization of high risk medications was opioids, which are used for pain. Also in Chapter 3, the Morse Fall score, gait/transferring score, and level assistance needed, all hung together and highly loaded on a factor in the EFA. These mobility items are related to functioning and can also be related to pain. Patients
106 who confront pain may h ave reduced mobility. In Chapter 5, an increasing number of chronic conditions and a variety of comorbidities were significant in predicting 1 2 and 3 or more readmissions. These variables can also be related to pain, because many symptoms of these comorbi dities can be associated with pain. Additionally, as chronic comorbidities increase, it is also likely that pain increases. This can be a function of less ability to exercise or can be directly related to the illnesses. The OHIS theoretical model proved to be helpful in guiding our 3 aims. The main construct in the model that specifically proved to be an area of high importance was outpatient behavior, a factor that has been neglected in other models. While our findings will need to be replicated, the fi ndings from each of the 3 aims, and from the patient interviews, leant support to the importance of the inclusion of outpatient behavior when modeling readmissions. Future research can continue to refine the OHIS model in predicting readmissions among prim ary care patients. The findings of this dissertation, through the OHIS model, were able to shed access to medication and transportation issues need to be recorded and res olved to make some impact on readmissions. Most importantly, through the OHIS model, the importance of bridging outpatient (i.e. outpatient behaviors construct) and inpatient (i.e. inpatient factors construct) are important in making an impact on readmissi ons. F ragmentation between hospital and community care need to be reduced. 6.2 Strengths and Limitations This dissertation is the first to incorporate outpatient behaviors, specific health status factors (e.g. number of high risk medications), specific in patient factors (e.g. gait/transferring score), and detailed social determinants of health (e.g. whether
107 someone lives with the patient at home) when analyzing readmissions. In addition, it is the first study to solely look at ACSC related readmissions as well as characterize the 1 2 readmissions population, in a nationally representative sample. Also, due to the local component of this dissertation, we were able to add a few qualitative interviews. This is a strength because we were able to better inform o ur quantitative findings. Though this dissertation had many strengths there were weaknesses that must be noted. First, the data used in Chapters 3 and 4, were from 1 academic family medicine department in the Southeast US. This is a limitation because the findings of this study may not be generalizable to all family medicine departments in the US. Also, the designs of the studies were cross sectional. For Chapter 4, we could not predict readmissions over time. For example, we could not analyze whether a pa tient was already readmitting before the study period (before January 1, 2015) or continued to readmit or passed away after the study period (after December 31, 2016). This was also a limitation in the Chapter 5 data. The distributor of the NRD made the da ta un linkable from year to year, as a protective feature. For example, if a patient had an index admission on December 25, 2012, and readmitted on January 10, 2013, the readmission would be identified as the index admission. We may have lost some patients in our analyses and we were not able to follow their behavior over time. Another limitation in Chapter The NRD excluded sensitive variables that could be used to identify patients. However, th ese variables may be more predictive of readmissions than the variables provided in the NRD.
108 6.3 Future Research Near future research to be done is the completion of 12 qualitative interviews. These 12 interviews will be analyzed as a whole to better inform the current quantitative findings as well as future quantitative research. Interpretation of our results is limited because of the local data we used; future research could extend the sample to other academic family medicine de partments and, eventually to other types of primary care departments such as general internal medicine and pediatrics. This will help ensure the findings are generalizable. We also suggest the addition of collection of more specific social determinants of health data such as concerns about transportation and problems of accessing medications. While we intuitively know that such barriers can be associated with readmissions, hospital intakes do not routinely gather such important individual barriers to diseas e and illness self management. We also suggest that future research look at subpopulations of adults, specifically those 65 years of age and older (i.e. older adults). We suspect that once people reach a certain age threshold, the predictors of readmissio ns may change. Another future direction could be a longitudinal analysis of patients who readmit. With the limitations of the time frame, we could not see the true effect that the tested variables had on the outcome of readmissions. In addition, we sugges t additional examination of outpatient behaviors and their relationship to readmissions. Once the role these variables are better understood, simple interventions can be done at a local clinic setting, to see whether monitoring and affecting these outpatie nt behaviors indeed reduce the odds of readmissions. A pilot study on this type of intervention can help elucidate the role primary care has in keeping patients healthy. It can also elucidate the role primary care has in preventing readmissions, a gap that has yet to be filled. Based on these pilot studies, a multisite
109 intervention through primary care could then be implemented. The public health benefit of such an intervention would be large. The decrease in or end of negative consequences for public and a cademic hospital systems would also be an important positive consequence of such work. Lastly, it is important and suggested to further bridge the communication gap between the inpatient and outpatient setting. If further research identifies the role of th e outpatient clinic, but information from the inpatient side does not communicate information to the outpatient clinic, then there would be no to very minimal impact on readmissions. For example, inpatient teams are informed of patients who have readmitted a week later after the events have happened. Though it is important to retrospectively discuss these patients to try to avoid the readmissions, a more effective approach would be to prevent the readmissions from happening. The only way that this would be achievable would be by repairing the fragmentation of care between the outpatient and inpatie nt settings. This bridging of fragmentation is visually depicted in the OHIS model of this dissertation. We suggest that the OHIS theoretical model be further s tud ied and refined, because it includes both outpatient and inpatient related variables as well as variables which identify bridges between the settings. We suggest that the OHIS theoretical model be used to help inform intervention studies because it can ser ve as a reminder to continue to increase communication between the two fragments of care: the outpatient and inpatient settings.
110 APPENDIX A QUALITATIVE INTERVIEW GUIDE PAGE 1
111 APPENDIX B QUALITATIVE INTERVIEW GUIDE PAGE 2
112 APPENDIX C QU ALITATIVE INTERVIEW GUIDE PAGE 3
113 LIST OF REFERENCES 1. 2,573 hospitals will face readmission penalties this year. Is yours one of them? Advisory Board. http://www.advisory.com/daily briefing/2017/08/07/hospital penalties. Published August 7, 2017. Accessed May 8, 2018. 2. Lefkowitz D. The Re Engineered Visit or Primary Care (AHRQ REV). Federal Register. https://www.federalregister.gov/documents/2017/02/1 3/2017 02893/agency information collection activities proposed collection comment request. Published February 13, 2017. Accessed May 5, 2018. 3. Hines A, Barrett M, Jiang H, Steiner C. Conditions With the Largest Number of Adult Hospital Readmissions by P ayer, 2011. HCUP Statistical Brief #172. https://www.hcup us.ahrq.gov/reports/statbriefs/sb172 Conditions Readmissions Payer.pdf. Published April 2014. Accessed May 1, 2018. 4. James J. Health Policy Brief: Medicare Hospital Readmissions Reduction Program Health Affairs. http://healthaffairs.org/healthpolicybriefs/brief_pdfs/healthpolicybrief_102.pdf. Published November 12, 2013. Accessed January 6, 2018. 5. Squires D, Anderson C. U.S. Health Care from a Global Perspective: Spending, Use of Services, Pri ces, and Health in 13 Countries. Issue Brief (Commonwealth Fund) 2015;15:1 15. 6. Horgas A, Yoon S, Grall M. Pain Management. In: Evidence Based Geriatric Nursing Protocols for Best Practice 4th ed. New York (NY): Springer Publishing Company; 2012:246 2 67. 7. Nelson J. Transitional care can reduce hospital readmissions. American Nurse Today April 2015. https://www.americannursetoday.com/transitional care can reduce hospital readmissions/. Accessed January 3, 2018. 8. Sydnor ERM, Perl TM. Hospital Epid emiology and Infection Control in Acute Care Settings. Clin Microbiol Rev 2011;24(1):141 173. doi:10.1128/CMR.00027 10 9. Summary Health Statistics: National Health Interview Survey, 2015. Centers for Disease Control and Prevention. https://ftp.cdc.gov/p ub/Health_Statistics/NCHS/NHIS/SHS/2015_SHS_Table_P 10.pdf. Accessed June 23, 2017. 10. Agana D. Characteristics of UF Health Family Medicine Patients Frequently Readmitted to the Hospital in 2016. Poster presented at the: American Public Health Associatio n Annual Meeting; November 6, 2017; Atlanta, GA. 11. Donz JD, Williams MV, Robinson EJ, et al. International Validity of the HOSPITAL Score to Predict 30 Day Potentially Avoidable Hospital Readmissions. JAMA Intern Med 2016;176(4):496 502. doi:10.1001/ja mainternmed.2015.8462
114 12. Walraven C van, Dhalla IA, Bell C, et al. Derivation and validation of an index to predict early death or unplanned readmission after discharge from hospital to the community. CMAJ 2010;182(6):551 557. doi:10.1503/cmaj.091117 13. van Walraven C, Wong J, Forster AJ. LACE+ index: extension of a validated index to predict early death or urgent readmission after hospital discharge using administrative data. Open Med 2012;6(3):e80 e90. 14. Garrison GM, Robelia PM, Pecina JL, Dawson NL Comparing performance of 30 day readmission risk classifiers among hospitalized primary care patients. J Eval Clin Pract 2017;23(3):524 529. doi:10.1111/jep.12656 15. Logue E, Smucker W, Regan C. Admission Data Predict High Hospital Readmission Risk. J Am Board Fam Med 2016;29(1):50 59. doi:10.3122/jabfm.2016.01.150127 16. Social Determinants of Health. HealthyPeople.gov. https://www.healthypeople.gov/2020/topics objectives/topic/social determinants of health. Accessed August 3, 2017. 17. Social Determi nants of Health | CDC. https://www.cdc.gov/socialdeterminants/. Published February 15, 2018. Accessed August 3, 2017. 18. WHO | Social determinants of health. WHO. http://www.who.int/social_determinants/en/. Accessed August 3, 2017. 19. Artiga S, May 10 EH P, 2018. Beyond Health Care: The Role of Social Determinants in Promoting Health and Health Equity. The Henry J Kaiser Family Foundation May 2018. https://www.kff.org/disparities policy/issue brief/beyond health care the role of social determinants in pro moting health and health equity/. Accessed June 1, 2018. 20. Podcast: Office of Minority Health builds business case for achieving health equity Modern Healthcare. http://www.modernhealthcare.com/article/20160423/PODCAST/304239941. Accessed July 10, 2017. 21. Before Penalizing Hospitals, Account for the Social Determinants of Health. NEJM Catalyst. https://catalyst.nejm.org/penalizing hospitals account social determinants of health/. Published October 24, 2016. Accessed August 3, 2017. 22. Reidhead M. Including sociodemographic factors in risk adusted readmissions measure. HIDI HealthStats. https://www.mhanet.com/mhaimages/SociodemFactos_HealthStats_0216.pdf. Published February 2016. Accessed August 10, 2017. 23. Ventres W, Kravitz JD, Dharamsi S. PEAR LS+: Connecting Societal Forces, Social Determinants, and Health Outcomes. Acad Med 2018;93(1):143. doi:10.1097/ACM.0000000000002012
115 24. Peterson Sgro K. Reducing Acute Care Hospitalization and Emergent Care Use Through Home Health Disease Management: One Home Healthcare Now 2007;25(10):622. doi:10.1097/01.NHH.0000298930.45717.29 25. Health Connections Initiative An Evidence Based Community Care Transitions Intervention Program Providing Assistance to a Low Income Patient Populatio n. https://mghdisparitiessolutions.files.wordpress.com/2015/09/kentuckyone health_hci intervention sign.pdf. Accessed July 12, 2017. 26. Components of Re Engineered Discharge (RED). https://www.bu.edu/fammed/projectred/components.html. Accessed July 12, 20 17. 27. Markley J, Andow V, Sabharwal K, Wang Z, Fennell E, Dusek R. A Project to Reengineer Discharges Reduces 30 day Readmission Rates. Ajn, American Journal of Nursing 2013;113(7):55 64. doi:10.1097/01.NAJ.0000431922.47547.eb 28. Jack BW, Chetty VK, An thony D, et al. A Reengineered Hospital Discharge Program to Decrease Rehospitalization. Ann Intern Med 2009;150(3):178 187. 29. Freund T, Campbell SM, Geissler S, et al. Strategies for Reducing Potentially Avoidable Hospitalizations for Ambulatory Care S ensitive Conditions. Ann Fam Med 2013;11(4):363 370. doi:10.1370/afm.1498 30. Betancourt J, Tan McGrory A, Kenst B. Guide to Preventing Readmissions among Racially and Ethnically Diverse Medicare Beneficiaries. September 2015:30. 31. AHRQ Initiative to Assess Role of Primary Care in Preventing Readmissions. AJMC. http://www.ajmc.com/newsroom/ahrq initiative to assess role of primary care in preventing readmissions. Accessed May 5, 2018. 32. Striley C. Lessons learned intervening through CHWs in a family medicine practice to reduce emergency and hospital care. Oral presented at the: Primary Care Innovations Conference; March 10, 2017; Gainesville, FL. 33. Bahrami H, Kronmal R, Bluemke DA, et al. Differences in the incidence of congestive heart failure by e thnicity: the multi ethnic study of atherosclerosis. Arch Intern Med 2008;168(19):2138 2145. doi:10.1001/archinte.168.19.2138 34. Vivo RP, Krim SR, Cevik C, Witteles RM. Heart Failure in Hispanics. Journal of the American College of Cardiology 2009;53(14 ):1167 1175. doi:10.1016/j.jacc.2008.12.037 35. Rodriguez F, Joynt KE, Lpez L, Saldaa F, Jha AK. Readmission rates for Hispanic Medicare beneficiaries with heart failure and acute myocardial infarction. Am Heart J 2011;162(2):254 261.e3. doi:10.1016/j.a hj.2011.05.009
116 36. Brown DW, Haldeman GA, Croft JB, Giles WH, Mensah GA. Racial or ethnic differences in hospitalization for heart failure among elderly adults: Medicare, 1990 to 2000. American Heart Journal 2005;150(3):448 454. doi:10.1016/j.ahj.2004.11 .010 37. Joynt KE, Orav EJ, Jha AK. Thirty day readmission rates for Medicare beneficiaries by race and site of care. JAMA 2011;305(7):675 681. doi:10.1001/jama.2011.123 38. McHugh MD, Carthon JMB, Kang XL. Medicare readmissions policies and racial and et hnic health disparities: a cautionary tale. Policy Polit Nurs Pract 2010;11(4):309 316. doi:10.1177/1527154411398490 39. Medicine NA of S Engineering, and, Division H and M, Practice B on PH and PH, States C on C BS to PHE in the U. Communities in Action: Pathways to Health Equity Washington, DC: National Academies Press; 2017. 40. Agency for Healthcare Reasearch and Quality. Disparities in Healthcare Quality Among Racial and Ethnic Groups | AHRQ Archive. https://archive.ahrq.gov/research/findings/nhqrdr/ nhqrdr11/minority.html. Accessed June 27, 2017. 41. Jiang HJ, Wier LM. All Cause Hospital Readmissions among Non Elderly Medicaid Patients, 2007: Statistical Brief #89. In: Healthcare Cost and Utilization Project (HCUP) Statistical Briefs Rockville (MD): Agency for Healthcare Research and Quality (US); 2006. http://www.ncbi.nlm.nih.gov/books/NBK53601/. Accessed April 24, 2017. 42. Dreyer RP, Ranasinghe I, Wang Y, et al. Sex Differences in the Rate, Timing, and Principal Diagnoses of 30 Day Readmissions in Younger Patients with Acute Myocardial Infarction. Circulation 2015;132(3):158 166. doi:10.1161/CIRCULATIONAHA.114.014776 43. De Giorgi A, Boari B, Tiseo R, et al. Hospital readmissions to internal medicine departments: a higher risk for females? Eur Rev Med Pharmacol Sci 2016;20(21):4557 4564. 44. Walsh TS, Salisbury L, Donaghy E, et al. PReventing early unplanned hOspital readmission aFter critical ILlnEss (PROFILE): protocol and analysis framework for a mixed methods study. BMJ Open 2016;6(6):e012590. doi:10.1136/bmjopen 2016 012590 45. Szanton SL, Gill JM, Allen JK. Allostatic Load: A Mechanism of Socioeconomic Health Disparities? Biol Res Nurs 2005;7(1):7 15. doi:10.1177/1099800405278216 46. Age Patterns of Allostatic Load Scores Among Blacks and Whites in the United States. Am J Public Health 2006;96(5):826 833. doi:10.2105/AJPH.2004.060749
117 47. Middleton A, Graham JE, Ottenbacher KJ. Functional Status Is Associated With 30 Day Potentially Preventable Hospital Readmissions After Inpatient Rehabilitation Among Aged Medicare Fee for Service Beneficiaries. Arch Phys Med Rehabil 2018;99(6):1067 1076. doi:10.1016/j.apmr.2017.05.001 48. Dalleur O, Beeler PE, Schnipper JL, Donz J. 30 Day Potentially Avoidable Readmissions Due to Adverse Drug Events. J Patient Saf March 2017. doi:10.1097/PTS.0000000000000346 49. Deschepper M, Vermeir P, Vogelaers D, Devulder J, Eeckloo K. Is pain at disc harge a risk factor for unplanned hospital readmission? Acta Clin Belg 2017;72(2):95 102. doi:10.1080/17843286.2017.1293311 50. Agency for Healthcare Reasearch and Quality. Hospital Guide to Reducing Medicaid Readmissions. August 2014:82. 51. Healthcare C ost and Utilization Project (HCUP). HCUP Nationwide Readmissions Database (NRD) Rockville, MD: Agency for Healthcare Research and Quality; 2013. https://www.hcup us.ahrq.gov/nrdoverview.jsp. 52. About AHRQ. /cpi/about/index.html. Published November 2017. Accessed June 11, 2018. 53. Fingar K, Washington R. Trends in Hospital Readmissions for Four High Volume Conditions, 2009 2013 #196. https://www.hcup us.ahrq.gov/reports/statbriefs/sb196 Readmissions Trends High Volume Conditions.jsp. Published November 20 15. Accessed October 19, 2017. 54. A Step by Step Approach to Using SAS for Factor Analysis and Structural Equation Modeling SAS Institute; 2013. 55. Caraceni P, Tufoni M, Bonavita ME. Clinical use of albumin. Blood Transfus 2013;1 1(Suppl 4):s18 s25. doi:10.2450/2013.005s 56. Frank G, Walton Ziegler O. Albumin (Blood) Health Encyclopedia University of Rochester Medical Center. University of Rochester Medical Center. https://www.urmc.rochester.edu/encyclopedia/content.aspx?conten ttypeid=167&con tentid=albumin_blood. Accessed June 3, 2018. 57. Fisher SR, Kuo Y F, Sharma G, et al. Mobility After Hospital Discharge as a Marker for 30 Day Readmission. J Gerontol A Biol Sci Med Sci 2013;68(7):805 810. doi:10.1093/gerona/gls252 58. Office of Disease Prevention and Health Promotion. ADE Action Plan Adverse Drug Events health.gov. https://health.gov/hcq/ade action plan.asp. Accessed June 2, 2018.
118 59. TM http://childrenshealthwatch.org/p ublic policy/hunger vital sign/. Accessed June 12, 2018. 60. Manning J, Zhao S, Anshutz M, Carman C, Philipson K. PERFORMANCE OF THE MASSACHUSETTS HEALTH CARE SYSTEM SERIES: A FOCUS ON PROVIDER QUALITY. http://www.chiamass.gov/assets/Uploads/A Focus on Pro vider Quality Jan 2015.pdf. Accessed June 5, 2018. 61. Akirov A, Masri Iraqi H, Atamna A, Shimon I. Low Albumin Levels Are Associated with Mortality Risk in Hospitalized Patients. Am J Med 2017;130(12):1465.e11 1465.e19. doi:10.1016/j.amjmed.2017.07.020 6 2. Tool 3H: Morse Fall Scale for Identifying Fall Risk Factors. /professionals/systems/hospital/fallpxtoolkit/fallpxtk tool3h.html. Published January 31, 2013. Accessed June 1, 2018. 63. Billings J, Zeitel L, Lukomnik J, Carey TS, Blank AE, Newman L. Impac t of socioeconomic status on hospital use in New York City. Health Aff (Millwood) 1993;12(1):162 173. doi:10.1377/hlthaff.12.1.162 64. Havelka M, Lucanin JD, Lucanin D. Biopsychosocial model -the integrated approach to health and disease. Coll Antropol 2 009;33(1):303 310. 65. https://www.uptodate.com/contents/hospital discharge and readmission#H9. Accessed June 1, 2018. 66. Lachman ME, Teshale S, Agrigoroaei S. Midlife as a Pi votal Period in the Life Course: Balancing Growth and Decline at the Crossroads of Youth and Old Age. Int J Behav Dev 2015;39(1):20 31. doi:10.1177/0165025414533223 67. Office of Disease Prevention and Health Promotion. Older Adults | Healthy People 2020. https://www.healthypeople.gov/2020/topics objectives/topic/older adults. Accessed May 3, 2018. 68. Hainer V, Aldhoon Hainerov I. Obesity Paradox Does Exist. Diabetes Care 2013;36(Suppl 2):S276 S281. doi:10.2337/dcS13 2023 69. Khan SS, Ning H, Wilkins JT et al. Association of Body Mass Index With Lifetime Risk of Cardiovascular Disease and Compression of Morbidity. JAMA Cardiol 2018;3(4):280 287. doi:10.1001/jamacardio.2018.0022 70. Ohlson LO, Larsson B, Svrdsudd K, et al. The influence of body fat dis tribution on the incidence of diabetes mellitus. 13.5 years of follow up of the participants in the study of men born in 1913. Diabetes 1985;34(10):1055 1058.
119 71. Larsson B, Svrdsudd K, Welin L, Wilhelmsen L, Bjrntorp P, Tibblin G. Abdominal adipose tis sue distribution, obesity, and risk of cardiovascular disease and death: 13 year follow up of participants in the study of men born in 1913. Br Med J (Clin Res Ed) 1984;288(6428):1401 1404. 72. Rose G, Khaw K T, Marmot M. icine Oxford University Press; 2008.
120 BIOGRAPHICAL SKETCH Denny Fe G. Agana recei ved her Bachelor of Science in biology and a m inor in Spanish from the University of Florida (UF) in Gainesville, FL in 2007. A year after graduation, she was accepted to the Master of Public Heath (MPH) with a concentration in e pidemiology at UF in Gainesville, FL in 2012. She graduated with her MPH in 2014. Soon after, she worked as a r esearch coordinator at the H. James Free Center for Primary Care Education and Innovation (Free Center) at UF. In 2014, the Free Center and the Department of Community Health and Family Medicine agreed to fully fund her Doctor of Philosophy (PhD). During the PhD, her primary mentor was Catherine W. Striley, PhD, MSW, MPE and her co mentor was Peter J. Carek, MD, MS. Denny Fe received her PhD in Epidemiology from UF in the Summer of 2018. public health. Particularly, Denny Fe is interested in social determinants of health related to health service utilization and health outcomes in the primary care setting Fur ther, she will continue on the path of a primary care researcher as a T 32 fellow for a primary care research fellowship at Baylor College of Medicine in Houston, TX.