Jamie Tricarico Matese, B.A. Washington, DC December 6, 2017

HOSPITAL READMISSIONS REDUCTION PROGRAM (HRRP) AND HEALTH OUTCOMES: ARE HOSPITAL READMISSIONS ASSOCIATED WITH MORTALITY RATES FOR MEDICARE PNEUMONIA PATIENTS? A Thesis Submitted to the Faculty of the Graduate School of Arts and Sciences of Georgetown University in partial fulfillment of the requirements for the degree of Master of Public Policy By Jamie Tricarico Matese, B.A. Washington, DC December 6, 2017

Copyright 2017 by Jamie Tricarico Matese All Rights Reserved ii

HOSPITAL READMISSIONS REDUCTION PROGRAM (HRRP) AND HEALTH OUTCOMES: ARE HOSPITAL READMISSIONS ASSOCIATED WITH MORTALITY RATES FOR MEDICARE PNEUMONIA PATIENTS? Jamie Tricarico Matese, B.A. Thesis Advisor: Robert W. Bednarzik, Ph.D. ABSTRACT The aging U.S. population will continue to drive Federal spending on Medicare upwards. Due to various external factors, the best approach to slow growth is for the Centers for Medicare and Medicaid Services (CMS) to work within the program. In recent years, one way that the CMS has sought to reduce costs to Medicare Part A and improve quality of care is by reducing unnecessary hospital readmissions through the Hospital Readmissions Reduction Program (HRRP). Under HRRP, CMS penalizes acute care hospitals with readmission rates that exceed the national average by reducing payments across all of their Medicare admissions. This paper examines the variation in readmission rates and mortality rates for hospitals participating in the Medicare program to determine whether hospitals with higher than expected readmissions have higher than expected mortality rates. The study focuses on hospital-level 30-day readmission rates and 30- day mortality rates for pneumonia from CMS most recent collection period beginning in 7/1/12 and ending on 6/30/15. Researchers have found evidence that hospital readmission rates and hospital mortality rates have an inverse relationship, meaning that hospitals with lower mortality rates are more likely to have higher readmission rates. However, this paper found evidence that since implementation of HRRP, hospitals with lower readmission rates are more likely to have lower mortality rates for pneumonia patients, for at least certain hospital types. iii

TABLE OF CONTENTS 1.Introduction..1 2. Background..4 2.1 Demographic Trends in the United States.....4 2.2 Medicare Hospital Insurance...........5 2.3 Hospital Readmissions....5 2.4 Medicare Hospital Readmissions Reduction Program (HRRP).....6 2.5 Criticism of HRRP.......8 3. Examining the Literature...10 3.1 Relationship Between Readmissions Rates and Mortality Rates..10 3.2 Readmissions and Mortality Rates Pneumonia Patients........12 3.3 Mortality Rates and Hospital Resources.... 12 3.4 Role of Hospital Characteristics....13 3.5 Need for New Research..... 14 4. Policy Relevance....15 5. Hypothesis...16 6. Data & Model.... 16 7. Descriptive Statistics......19 8. Regression Results...23 9. Analysis..28 10. Policy Implications......29 Appendix.......31 Bibliography...... 39 iv

LIST OF FIGURES Figure 1. Population Aged 65 and Over: 1900 to 2050.. 1 Figure 2. Definitions of Variables in the Model 18 Figure 3. Hospital Pneumonia Mortality & Readmission Rates, as Compared to National Average (July 1, 2012 June 30, 2015)...21 Figure A1. Model 1 Heteroscedasticity Plot. 31 Figure A2. Model 1 Heteroscedasticity: White s Test.32 Figure A3. Model 1 Specification: Linktest.. 32 Figure A4. Model 1 Specification: Ramsey Reset.32 Figure A5. Model 2 Heteroscedasticity: White s Test.. 33 Figure A6. Model 2 Specification: Linktest..33 Figure A7. Model 2 Specification: Ramsey Reset 34 Figure A8. Model 3 Heteroscedasticity: White s Test..35 Figure A9. Model 3 Specification: Linktest..35 Figure A10. Model 3 Specification: Ramsey Reset..36 Figure A11. Model 4 Heteroscedasticity: White s Test 37 Figure A12. Model 4 Specification: Linktest....37 Figure A13. Model Specification: Ramsey Reset.38 v

LIST OF TABLES Table 1. The First Five Years of the Hospital Readmissions Reduction Program...8 Table 2. Variation in Penalties by Hospital Characteristics, 2017...14 Table 3. Average Pneumonia Readmission and Mortality Rates (%), by Hospital Characteristics (July 1, 2012 June 30, 2015)... 20 Table 4. Readmission and Mortality Rates for U.S. Hospitals, compared to U.S. Average, 2012-2015..21 Table 5. Readmission and Mortality Rates Ranking for U.S. Hospitals, 2012-2015 (Ranked by Readmission Rate). 22 Table 6. Cross Tabulation of Readmission and Mortality Rate Ranks (July 1, 2012 June 30, 2015), using CMS Rankings 22 Table 7. Cross Tabulation of Readmission and Mortality Rate Ranks (July 1, 2012 June 30, 2015), Tertiles..23 Table 8. OLS and LPM Estimates of the Relationship between Hospital Readmission Rates and Hospital Mortality Rates..24 Table 9. OLS Estimates of the Relationship between Hospital Readmission Rates and Hospital Mortality Rates for Profit and Non-Profit Hospitals...26 Table 10. OLS Estimates of the Relationship between Hospital Readmission Rates and Hospital Mortality Rates for Teaching and Non-Teaching Hospital.....26 Table 11. OLS Estimates of the Relationship between Hospital Readmission Rates and Hospital Mortality Rates for Urban and Non-Urban Hospitals....27 Table A1. Model 1 Correlations 31 Table A2. Model 2 Correlations....33 Table A3. Model 3 Correlations.......35 Table A4. Model 4 Correlations....37 vi

1. Introduction The U.S. population is aging. Figure 1 shows that the share of the U.S. population aged 65 and above will increase from approximately 14% (40 million) in 2010 to 21% (80 million) in 2050 (United States Census Bureau, 2016). Figure 1. Population Aged 65 and Over: 1900 to 2050. (United States Census Bureau, 2016) This trend will continue to drive Federal spending on Medicare, which provides medical insurance to Americans 65 years of age and older. Medicare currently accounts for 15% of the U.S. Federal budget (CBO, 2016) and is expected to grow at a faster rate than U.S. Gross Domestic Product (GDP) for the next two decades. 1 The Medicare Board of Trustees projected in its 2016 report to Congress that the Hospital Insurance 1 In addition to the level of enrollment, Medicare expenditures are driven by a variety of factors including complexity of medical services provided, medical inflation, and life expectancy. 1

(HI) trust fund, which finances hospitalizations for Medicare beneficiaries, will become insolvent in 2028 (CMS, 2016). It is imperative to strengthen the financial stability of Medicare while not lowering quality, so the program can effectively serve the needs of the aging U.S. population. A portion of payroll taxes paid by employees and employers is credited to the HI trust fund. The trust fund is one of the sole financing mechanisms for Medicare Part A, which covers hospitalizations premium-free for most elderly Americans. Since 1970, this trust fund has faced a projected shortfall. There is no requirement under Federal law that general revenues be used to fund Medicare Part A in the event of a shortfall, making it crucial that policymakers put Medicare on a sustainable trajectory (CMS, 2016). The Medicare HI trust fund insolvency date has been postponed many times in the past due to economic growth, adjustments to the payroll tax, and enactment of measures to slow growth in the Medicare program (CRS, 2016). While policymakers have a great interest in boosting economic growth, it is a complex task that is subject to many external factors. Adjustments to the payroll tax are unpopular among voters, difficult to get through Congress, and do little to slow the growth of the program as a percentage of GDP growth. Therefore, the best approach is for the Centers for Medicare and Medicaid Services (CMS) to work within the program to slow growth. In recent years, one way that the CMS has sought to reduce costs to Medicare Part A and improve quality of care is by reducing unnecessary hospital readmissions. Medicare beneficiaries readmissions to hospitals within 30 days of discharge are common and costly a 2009 study reported that almost 20% of Medicare beneficiaries are re-hospitalized within 30 days of discharge, costing the Medicare program $17.4 2

billion annually (Jencks, Williams & Coleman 2009). The Medicare Payment Advisory Committee (MedPAC) concluded in its analysis of 2005 Medicare claims data that approximately three-quarters of hospital readmissions within 30 days are potentially preventable (MedPAC, 2007). MedPAC subsequently recommended to Congress that Medicare reduce payments to hospitals with relatively high readmission rates for select conditions (MedPAC, 2008, p.97). President Barack Obama signed the Affordable Care Act (ACA) into law in 2010. The landmark health care reform legislation authorized the Hospital Readmissions Reduction Program (HRRP) to reduce Medicare reimbursements for certain inpatient services to hospitals with higher than expected readmission rates beginning in 2012. Hospital readmissions have decreased since the implementation of the program. The Department of Health and Human Services (HHS) estimates that 565,000 fewer readmissions occurred from April 2010 through May 2015 as a result of the program (Obama 2016). In 2016, CMS reduced payments to a total of 2,665 hospitals totaling $420 million as a result of the program. In 2017, CMS will withhold $528 million from 2,597 hospitals through HRRP (Boccuti & Casillas, 2017). While it is clear that fewer readmissions reduce costs borne by the Medicare program, health care experts do not agree that the readmissions process truly reflects health care quality. Similarly, there is a lack of consensus of what proportion of readmissions are actually preventable (van Walraven, Bennet, Jennings, Austin & Forster, 2011). There is a growing body of literature examining the relationship between hospital readmissions and health outcomes using publicly reported CMS data at the hospital and patient levels. Mortality rates are an outcome of primary interest to medical professionals 3

and the public (Gorodeski, Starling & Blackstone, 2010). Researchers have found evidence that hospital readmission rates and hospital mortality rates have an inverse relationship, meaning that hospital with lower mortality rates are more likely to have higher readmission rates (Gorodeski, Starling & Blackstone, 2010; Pandey, et al.,2016; Brotman, Hoyer, Leung, Lepley, & Deutschendorf, 2016). Hospitals that have lower mortality rates have a greater proportion of discharged patients eligible for readmission, meaning that a higher readmission rate may be a consequence of successful care (Gorodeski, Starling & Blackstone, 2010). This paper examines the variation in readmission rates and mortality rates for hospitals participating in the Medicare program in order to determine whether hospitals with higher than expected readmissions have higher than expected mortality rates. CMS maintains national, state, and hospital-level data on both hospital readmissions and mortality rates for six different conditions. The study will focus on hospital-level 30-day readmission rates and 30- day mortality rates for pneumonia from CMS most recent collection period beginning in 7/1/12 and ending in 6/30/15. This paper will study pneumonia hospitalizations specifically, because pneumonia is the leading cause of hospitalization and death among adults in the United States (Jain, S. et al. 2015). 2. Background 2.1 Demographic Trends in the United States In 2017 members of the Baby Boom generation are 53-71 years of age. This population will continue to reach retirement age and become eligible for Medicare coverage over the next several years, contributing to the growth of the program. A 4

financial concern is that the ratio of workers paying payroll taxes to beneficiaries will decrease (CMS, 2016). That is, less money coming in as workers retire, and likely more money going out, as they become eligible for Medicare. 2.2 Medicare Hospital Insurance Most Americans age 65 and older are entitled to premium-free Medicare Part A coverage because they or their spouse paid Medicare payroll taxes on earnings for at least ten years. Part A is financed primarily through payroll taxes levied on current workers and their employers. That tax revenue is credited to the Hospital Insurance (HI) Trust Fund. The employer portion of the Medicare payroll tax is 1.45% and the employee portion is 4.2%. When the economy and wages grow, so does revenue to the trust fund. However, when the economy performs poorly, trust fund receipts could shrink as fewer people are working and paying payroll taxes. Medicare Part A includes hospital services, skilled nursing facility (SNF) services, hospice care, and home health visits. CMS reimburses acute-care hospitals for services they provide to Medicare beneficiaries through the Inpatient Prospective Payment System (IPPS). IPPS provides hospitals with a fixed amount per admission based on the patient s diagnosis. The rates are set each year prospectively and take effect on October 1. 2.3 Hospital Readmissions According to Medicare guidelines, a hospital readmission occurs when a patient is admitted to a hospital within 30 days of an earlier, initial hospitalization. Medicare excludes from this definition transfers to another hospital as well as readmissions considered planned. 5

Readmissions are costly and may be avoidable. Medicare spent $24 billion on 1.8 million 30-day hospital readmissions from January through November 2011 (AHRQ, 2015). The exact proportion of readmissions that may be avoidable is unknown. In 2007, MedPAC actuaries examined Medicare claims data from 2005 and determined that approximately 75% of readmissions within 30 days were preventable. (MedPAC, 2007). However, a 2011 study concluded that the proportion of hospital readmissions deemed avoidable has yet to be reliably determined, as studies examining hospital readmissions published from 1966 through 2010 had varied conclusions about avoidable readmissions, ranging from 5% to 79% (van Walraven, Bennet, Jennings, Austin & Forster, 2011, p. E397). Medicare collects data from hospitals on readmissions for six specific conditions: pneumonia, chronic obstructive pulmonary disease (COPD), heart attack, heart failure, elective hip or knee replacement, and coronary artery bypass graft (CABG). Medicare first began publicly reporting readmission rates for hospitals in 2009 for certain conditions, including pneumonia. 2.4 Medicare Hospital Readmissions Reduction Program (HRRP) In 2008 MedPAC recommended that CMS report readmission rates confidentially to providers for two years, and in the third year publicly report relative readmission rates of providers and reduce payments to hospitals with comparatively high readmission rates for certain conditions (MedPAC, 2008). Lawmakers are not required to follow MedPAC recommendations and often do not. However, this recommendation came at a politically opportune time the U.S. House of Representatives and U.S. Senate were considering health care reform, a top priority of the Obama Administration, and examining ways to 6

offset new spending in the Federal Medicaid program and private health insurance market. Section 3025 of the Affordable Care Act (P.L. 111-148) added section 1886(q) to the Social Security Act of 1935 establishing the Hospital Readmissions Reduction Program (HRRP). Specifically, the law required that CMS reduce payments to hospitals in the Inpatient Prospective Payment System (IPPS) with excess readmissions, effective for discharges beginning on October 1, 2012. Under HRRP, CMS penalizes acute care hospitals 2 with readmission rates above the national average by reducing payments across all of their Medicare admissions. For the first year of the program, fiscal year 2013, the maximum penalty a hospital could be assessed was 1% of the hospital s base inpatient payments. In 2014 the maximum penalty increased to 2%, and in 2015 it increased to 3%, where it remains. CMS uses the last three years of hospital data to calculate each hospital s readmission rate and the national average. The law allowed the Secretary of Health and Human Services (HHS) to expand the list of conditions in fiscal year 2015. In 2013 the program began with heart attack, heart failure, and pneumonia. The Secretary added hip/knee replacement and chronic obstructive pulmonary disease in 2015 and will add coronary artery bypass graft (CABG) procedures for fiscal year 2017. Table 1 (below) shows the penalties assessed by the 2 The program does not assess penalties on certain exempt hospitals that include cancer hospitals exempt from IPPS, children s hospitals, long term care facilities, psychiatric hospitals, rehabilitation hospitals, critical access hospitals, and hospitals in the state of Maryland. Maryland hospitals are exempt from IPPS as a result of a 36-year old agreement between Maryland and CMS. 7

program since it began. Indeed, the program is saving Medicare an increasing amount of money. Table 1. The First Five Years of the Hospital Readmissions Reduction Program. (Boccuti & Casillas, 2017) Year Penalties FY 2013 FY 2014 FY 2015 FY 2016 FY 2017 Applied Performance Measurement July 2008- June 2011 July 2009- June 2012 July 2010- June 2013 July 2011- June 2014 July 2012- June 2015 Period National Average Readmission Rate, for Pneumonia during Performance Measurement 18.5% 17.6% 17.3% 16.9% 16.3% Period Maximum Rate of Penalty Penalties: Percentage reduction in base payments on all Medicare inpatient admissions Average Hospital Penalty (among penalized hospitals only) Percent of hospitals penalized Percent of hospitals at max penalty CMS estimate of total penalties 1% 2% 3% 3% 3% -0.42% -0.38% -0.63% -0.61% -0.74% 64% 66% 78% 78% 79% 8% 0.6% 1.2% 1.1% 1.8% $290 million $227 million $428 million $420 million $528 million SOURCE: Adapted from Kaiser Family Foundation analysis of CMS Final Rules and Impact files for the Hospital Inpatient Prospective Payment System and Hospital Compare 2.5 Criticism of HRRP The literature suggests that reducing readmissions might have an adverse impact on mortality rates (Pandey, et al., 2016; Brotman, 2016; Gorodeski, Starling & Blackstone, 8

2010). Hospital representatives criticize CMS for not taking into account certain sociodemographic factors that may relate to readmission in its risk-adjustment. CMS calculates expected readmission rates for Medicare fee-for-service patients through a model that uses patient-level data to account for age, sex, race, as well as common clinical comorbidities 3 and smoking status. However, the model excludes sociodemographic variables that may impact access to outpatient care. The American Hospital Association (AHA) advocated against the implementation of HRRP and continues to advocate for changes to the formula to better account for sociodemographic factors (AHA Fact Sheet: Hospital Readmission Reduction Program, 2016). America s Essential Hospitals (AEH), which represents safety-net hospitals in the United States, has also criticized the program for unduly harming hospitals that treat low-income patient populations (Our View: Medicare HRRP Must Account for Social Complexities, 2015). In 2016, Congress acted to direct CMS to divide hospitals into peer groups based on similar shares of inpatients that are dual-eligibles, or qualify for both Medicare and Medicaid due to their status as low-income. It is also important to note that researchers have criticized quality measures, like readmissions rates or mortality rates, in general, due to fear of referral bias (Lilford & Pronovost, 2010; Shahian et al. 2012). Referral bias or patient selection is based on the concern that people in the poorest health may be referred to higher quality hospitals resulting in biased performance assessments and comparisons across hospitals (Doyle, 2017). However, researchers looking at whether referral bias has an effect on quality measures have not found evidence (Doyle, 2017). 3 Medical issues that include heart failure, obesity, coronary artery disease, renal disease, cerebrovascular disease, chronic obstructive pulmonary disease, cancer, and diabetes. 9

3. Examining the Literature There is a growing body of literature examining the relationship between hospital readmissions and health outcomes (Pandey, et al., 2016; Brotman, 2016; Gorodeski, Starling & Blackstone, 2010). The results of these studies are guiding CMS policy. For example, CMS is reducing payments to hospitals based on readmission rates to reduce unnecessary costs and improve quality of care. However, much of the recent evidence rejects the idea that lower readmissions are correlated with improved quality of care, as measured by improved (lower) mortality rates. 3.1 Relationship Between Readmission Rates and Mortality Rates Pandey, et.al. (2016) studied how hospitals with high excess 30-day readmissions for heart failure and hospitals with low excess 30-day readmissions for heart failure performed on the American Heart Association s Get With the Guidelines-Heart Failure measures of quality of care and outcomes, which include inpatient mortality and one year all-cause mortality rates. Pandey (2016) found that hospitals in both the high excess readmission and low excess readmission pools groups had comparable performance in the quality of care and outcome measures. They did note a trend toward higher 1-year mortality rates among centers with low 30-day risk-adjusted heart failure readmission rate, suggesting that readmission rates and mortality rates have an inverse relationship. This conclusion is consistent with Gorodeski, Starling & Blackstone (2010), which saw an inverse relationship between readmission rates and mortality rates among hospitals within the CMS Hospital Compare database using linear regression analysis. A broader study beyond just heart failure by Brotman, Hoyer, Leung, Lepley, & Deutschendorf (2016) examined the relationship between hospital-wide readmission rates 10

and six mortality measures that were first reported by CMS in the Hospital Compare database: heart failure, chronic obstructive pulmonary disorder (COPD), stroke, myocardial infarction (MI), coronary artery bypass graft (CABG), and pneumonia. Using a logistic regression model, they found that hospitals in the highest hospital-wide readmission tertile were more likely to perform in the lowest, or best, mortality tertile for heart failure, COPD, and stroke and found significant association between high hospitalwide or all-cause readmission rates 4 and low mortality for MI, CABG, and pneumonia. Their results for heart failure not only reflected that of Pandey, et.al. (2016) and Gorodeski, Starling & Blackstone (2010), but also those of Krumholz, Lin & Keenan (2013), who found that risk-standardized mortality rates and readmission rates were not associated for patients admitted with MI or pneumonia (p. 587). However, Krumholz, unlike earlier researchers, found a weak positive association, using logistic regression, between readmission rates and mortality rates for patients admitted with heart failure. There is no clear-cut relationship between readmissions and mortality but different time periods and analytical techniques were used. The Krumholz study used data collected between July 1, 2005 and June 30, 2008. Pandey (2016) used readmissions and mortality data collected from July 1, 2008 through June 30, 2011. Brotman (2016) used publically available data published by CMS from July 1, 2011 through June 20, 2014. These collection periods all include data from before implementation of HRRP and tended to focus on heart failure. The collection period beginning in 2012, after 4 Medicare also collects data from hospitals of hospital-wide or all-cause readmissions hospital stays within 30 days of initial hospitalization, regardless of whether the readmission is caused by the same condition as the initial hospitalization. 11

implementation of HRRP, is worth investigating, as is more of a focus on pneumonia, a leading cause of death and hospitalization among Medicare patients. 3.2 Readmission and Mortality Rates for Pneumonia Patients Given pneumonia s prevalence among adults in the United States and its financial impact on the Medicare program, researchers have focused studies specifically on how readmission rates and mortality rates vary for pneumonia patients. In the United States pneumonia results in approximately 1.2 million hospital admissions annually, is the second leading cause of hospitalization among patients over 65 years of age, and accounts for more than $10 billion annually in hospital expenditures (AHRQ, 2015). A 2010 cross-sectional study examined whether there were patterns of hospital and regional performance in the outcomes, as measured by mortality and readmission rates, for elderly patients with pneumonia (Lindenauer, et al. 2010). These researchers found that mortality rates, and to a lesser extent readmission rates, for patients with pneumonia vary substantially across hospitals and regions. They found that readmission rates, but not mortality rates, show strong geographic concentration (Lindenauer, et al., 2010). Although the study stops short of addressing whether there is an association between mortality and readmission rates for hospitals or regions, the results suggest that they are not associated. 3.3 Mortality Rates and Hospital Resources An early study (Ong, et al. 2009) examined readmission rates and mortality rates in California. This study looked at mortality rates and lengths of hospital stays for elderly patients with heart failure at six California teaching hospitals. They found that hospitals that used more resources had lower mortality rates (Ong, et al. 2009). This finding is 12

important because hospitals with higher readmission rates use more resources, a factor that may be important to consider when examining whether readmissions rates and mortality rates are inversely associated. 3.4 Role of Hospital Characteristics It seems logical to assume and the literature suggests that there are possible associations between several hospital characteristics and both readmission rates and mortality rates. The literature shows that several hospital characteristics that may be associated with mortality rates. These include hospital ownership, status as a teaching hospital, and urban/rural location. A 2005 study found that for-profit hospitals have lower mortality rates than non-profit hospitals (Millicent, 2005). A recent study found that teaching hospitals are associated with lower mortality rates (Burke, 2017). A 2008 study found that urban hospitals have higher mortality rates than rural hospitals (Ross, 2008). Gorodeski, Starling & Blackstone (2010) were the first to look at the relationship between readmissions and mortality rates using CMS data. They found that a higher occurrence of readmissions for heart failure was associated with a lower risk-adjusted 30- day mortality rate. They found, however, that the relationship was not readily apparent among hospitals with the lowest readmission rates. They speculated that it had to do with variations in hospital characteristics between hospitals with low readmissions versus those with high readmission rates. Subsequent studies examining mortality rates and readmission rates control for hospital characteristics to get a clearer picture of the readmissions-mortality connection. These hospital characteristics include whether a hospital is urban or rural and status as a 13

teaching hospital. The most recent CMS data show variation in penalties for hospitals, which have been grouped by characteristics. Table 2 (below) shows that large hospitals (300+ beds) and hospitals treating a higher percentage of low-income patients were more likely to be assessed a penalty under HRRP than smaller hospitals and hospitals that treat a smaller proportion of low-income patient (Boccuty & Casillas, 2017). Table 2. Variation in Penalties by Hospital Charactaristics, 2017. (Boccuti & Casillas, 2017) 3.5 Need for New Research Building upon the few empirical studies relating readmissions with mortality, this paper uses hospital-level data to estimate the association between risk-adjusted 30-day hospital readmission rates and risk-adjusted 30-day mortality rates for pneumonia. The data are collected and publicly reported by CMS. The readmission and mortality rates span the 3-year time period from July 1, 2012 through June 30, 2015, after 14

implementation of HRRP. The CMS reported rates are already adjusted for patient-level variables, including sex, age, and comorbidities. Building on models in the literature this study will control for hospital characteristics that may be associated with readmissions and mortalities. This study will use the most recent data reported by CMS, which covers hospitalizations from 2012-2015. Notably, this three-year period is the first full reporting period following implementation of the Hospital Readmissions Reduction Program (HRRP). 4. Policy Relevance Medicare spending accounted for 15% of total Federal spending in 2015, or 3.6 percent of the total U.S. GDP. According to Medicare Trustee s June 22, 2016 report, Medicare spending is scheduled to grow significantly faster than the economy, up to 5.6 percent of the GDP by 2040 and cause the HI trust fund to become insolvent in 2028 (CMS, 2016). For these reasons, policymakers are concerned about lowering Medicare program spending without affecting quality. This is a complicated task, as incentives to lower readmissions must be appropriate as not to affect quality. The HRRP program is intended to lower Medicare spending while also improving quality. While it is clear that the program results in lower spending, its relationship to quality is less clear. The literature suggests that the program s goal of lowering readmissions could adversely impact outcomes as measured by mortality rates. Since 2012, CMS has penalized hospitals for higher than expected readmissions. The number of medical conditions assessed has grown since 2012. In addition, MedPAC has since endorsed readmissions penalties for other providers that participate in Medicare 15

Part A including skilled nursing facilities (SNFs) and home health. Before the program expands, it is necessary for more research to be done to determine whether there is a relationship between readmissions and mortality rates. 5. Hypothesis In keeping with the findings detailed in this paper s literature review, the following hypothesis will be tested: 30-day readmission rates have a negative correlation with 30-day mortality rates for Medicare fee-for-service beneficiaries with pneumonia. 6. Data & Model The empirical analysis uses hospital-level 30-day mortality rates and 30-day readmission rates for all Medicare fee-for-service patients treated for pneumonia at an acute care hospital from July 1, 2012 through June 30, 2015. There are over 4,000 observations. These data are published by the Centers for Medicare and Medicaid Services (CMS). There are many demographic and health-related factors that have shown to be related to both mortality rates and readmission rates. Medicare publishes National Quality Forum (NQF) endorsed risk-standardized mortality rates (RSMRs) and riskstandardized readmission rates (RSRR) for pneumonia. The hospital-specific risk-standardized mortality rate (RSMR) is defined as the ratio of the number of "predicted" deaths to the number of "expected" deaths, multiplied by the national unadjusted mortality rate. The "denominator" is the number of deaths expected based on the nation's performance with that hospital's case-mix, which accounts 16

for patients age, past medical history, and comorbidities 5 (National Quality Measures Clearinghouse PN RSMR, 2015). Similarly, the hospital-specific risk-standardized readmission rate (RSRR) is defined as the ratio of the number of "predicted" readmissions to the number of "expected" readmissions, multiplied by the national unadjusted readmission rate. The "denominator" is the number of readmissions expected based on the nation's performance with that hospital's case-mix (National Quality Measures Clearinghouse PN RSRR, 2015). Therefore, the 30-day mortality rates and 30- day readmission rates used in this analysis are already controlling for patient-level characteristics. There are many hospital characteristics that have shown to be related to mortality rates, including hospital ownership, whether a hospital is located in an urban area, and whether the hospital is a teaching hospital. Therefore, these factors are controlled for in the empirical analysis. Data for these controls were taken from the Centers for Medicare and Medicaid Services (CMS) Provider of Service (POS) files. This paper will utilize ordinary least squares (OLS) and linear probability model (LPM) regression to examine the relationship between readmissions rates and mortality rates. 5 Comorbidities are the simultaneous presence of two or more chronic diseases or conditions in one patient. 17

Definition Variable Name Expected Justification Data Source Sign Dependent Variable Y 1 Continuous variable indicating hospitals 30- day mortality rate for pneumonia patients MORT_SCORE N/A Centers for Medicare and Medicaid Services (CMS) Hospital Y 2 Dummy variable indicating hospital s 30-day mortality rate for pneumonia patients is worse than national average Independent Variable X 1a X 1b Continuous variable indicating hospital s 30-day readmission rate for pneumonia patients Dummy variable indicating hospital s 30-day readmission rate for pneumonia patients is worse than national average Control Variable X 2 X 3 X 4 Dummy variable indicating hospital is for profit Dummy variable indicating hospital is a teaching hospital Dummy variable indicating hospital is urban Figure 2. Definitions of Variables in the Model. Compare MORT_WORSE N/A Centers for Medicare and Medicaid Services (CMS) Hospital Compare READM_SCORE - Pandey (2016), Brotman (2016), Gorodeski (2010) READM_WORSE - Pandey (2016), Brotman (2016), Gorodeski (2010) Centers for Medicare and Medicaid Services (CMS) Hospital Compare Centers for Medicare and Medicaid Services (CMS) Hospital Compare PROFIT - Millicent (2005) Centers for Medicare and Medicaid Services (CMS) Provider of Service (POS) Files TEACHING - Burke (2017) Centers for Medicare and Medicaid Services (CMS) Provider of Service (POS) Files URBAN + Ross (2008) Centers for Medicare and Medicaid Services (CMS) Provider of Service (POS) Files 18

The analysis in this paper is based on the following regression models: Y = β 0 + β 1 X 1 + β 2 X 2 + β 3 X 3 + β 4 X 4 + E Where: Y = MORT_WORSE or MORT_SCORE X 1= READM_WORSE or READM_SCORE X 2 = PROFIT X 3 = TEACHING X 4 = URBAN E = Unexplained variance, error term β 0, β 1, β 2, β 3, β 4 = Coefficients of respective independent variables; partial slope coefficients. 7. Descriptive Statistics The mean readmission rate for the 4,812 acute care hospitals for which CMS reports data is 17.1%. The mean mortality rate is 16.4%. In Table 3 the rates are broken down by hospital characteristics of interest to the model. There appear to be very little differences in mortality rates and readmission rates when hospitals are broken down by characteristics. 19

Table 3. Average Pneumonia Readmission and Mortality Rates (%), by Hospital Characteristics (July 1, 2012 June 30, 2015). Readmission Rate Mortality Rate All Hospitals (100%) 17.1 16.4 For-Profit (17.9%) 17.4 16.4 Non-Profit (82.1%) 17.1 16.5 Teaching (26.5%) 17.3 16.5 Non-Teaching (74.5%) 17.0 16.1 Urban (60.4%) 16.9 16.2 Rural (39.6%) 17.2 16.7 Source: CMS, 4,812 Hospitals CMS also reports its hospital-level readmission and mortality rates for pneumonia as whether they are no different than, worse than, or better than the national average. The distribution is displayed in Figure 2. 6 There is not much deviation in readmission or mortality rates for pneumonia, as over three- fourths of all hospitals cluster around the mean. 6 For 12.2% of hospitals, mortality rates for pneumonia are not available or the number of cases is too small to report. For 15.14% of hospitals, readmission rates for pneumonia are not available or the number of cases is too small to report 20

79.8% 75.1% Mortality Rate Readmission Rate 4.7% 1.4% 5.1% 3.6% 15.1% 12.2% Better No Different Worse N/A Figure 3. Hospital Pneumonia Mortality & Readmission Rates, as Compared to National Average (July 1, 2012- June 30, 2015). Source: CMS Hospital Compare, for 4,812 Acute Care Hospitals Table 4 shows the readmission rates and mortality rates for hospitals grouped into the CMS-reported categories of better than the national average, no different than the national average, and worse than the national average. There appears to not be much of a relationship between readmission rates and mortality rates; as the readmission rate changes, the mortality rate does not. Table 4. Readmission and Mortality Rates for U.S. Hospitals, compared to U.S. Average, 2012-2015. Readmission Rate Mortality Rate Total (100%) 17.1% 16.4% Better than Nation (1.9%) 14.1% 16.1% No difference (93.7%) 17.0% 16.4% Worse than Nation (4.3%) 20.5% 16.2% Source: CMS Hospital Compare, for 4,047 Hospitals Table 5 shows the readmission rates and mortality rates for hospitals grouped by readmission rate into nearly equal-sized tertiles representing high, medium, and low readmission rates. There again appears to be not much of a relationship between 21

readmission rates and mortality rates; as the readmission rate changes, the mortality rate does not. Table 5. Readmission and Mortality Rates Ranking for U.S. Hospitals, 2012-2015 (Ranked by Readmission Rate). Readmission Rate Mortality Rate Total (100%) 17.1% 16.4% High (30.8%) 17.8% 16.5% Medium (35.8%) 17.0% 16.4% Low (33.4%) 15.7% 16.3% Source: CMS Hospital Compare, for 4,047 Hospitals Table 6 is a cross-tabulation of readmission and mortality rate ranks, using CMS rankings. The table shows that hospitals with readmission rates that are better than the national average, as reported by CMS, are equally as likely to have better than the national average mortality rates as worse than the national average mortality rates. Table 6. Cross Tabulation of Readmission and Mortality Rate Ranks (July 1, 2012 June 30, 2015), using CMS Rankings. Mortality Rate Rank Better than the Readmission National Rate Rank Average Total 247 6.1 Better than the National Average No different than the National Average 9 11.4 No Different than the National Average 3,539 87.5 65 82.3 206 5.4 3,352 88.4 Worse than the 32 122 National Average 18.1 68.9 Source: CMS Hospital Compare, for 4,047 Hospitals Worse than the National Average 261 6.5 5 6.3 233 6.2 23 13.0 Key: Frequency Row % Total 4,047 100.0 79 100.0 3,791 100.0 177 100.0 Table 7 shows the cross-tabulation of readmission rate and mortality rate ranks after hospitals have been sorted into tertiles. The table shows that hospitals with low 22

readmission rates are slightly more likely (35.3%) to have low mortality rate rank than a high mortality rate rank (30.8%). Hospitals with high readmission rates are slightly more likely to have high mortality rate (36.8%) than low mortality rate (33.6%). Table 7. Cross Tabulation of Readmission and Mortality Rate Ranks (July 1, 2012 June 30, 2015), Tertiles. Readmission Mortality Rate Rank Rate Rank Low Medium High Total Total 1,365 33.7 1,337 33.0 1,345 33.2 4,047 100.0 Low (Bottom Tertile) 478 35.3 459 33.9 416 30.8 1,353 100.0 Medium 466 506 468 1,440 (Middle Tertile) 32.4 35.1 32.5 100.0 High 421 372 461 1,254 (Top Tertile) 33.6 29.7 36.8 100.0 Source: CMS Hospital Compare, for 4,047 Hospitals An overall conclusion from the descriptive data alone is that there is not a very clear relationship between readmission and mortality rates. Key: Frequency Row % 8. Regressions Results The models use hospital-level 30-day mortality rates and 30-day readmission rates for all Medicare fee-for-service patients treated at an acute care hospital for pneumonia from July 1, 2012 to June 30, 2015. These data are available in three forms raw scores, risk-standardized scores that control for patient characteristics, and risk-standardized scores reported as whether they are below, equal to, or above the national average. The four models developed here use the latter two forms of the data. The four models control for three hospital-specific characteristics: hospital ownership, teaching status, and whether the hospital is urban. 23

Table 8. OLS and LPM Estimates of the Relationship between Hospital Readmission Rates and Hospital Mortality Rates. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The first regression (Model 1) has as the dependent variable the hospital-level 30-day mortality rate for Medicare patients with pneumonia and as the independent variable the hospital-level 30-day excess readmission rate for Medicare patients with pneumonia. Readmission rates are a significant predictor of mortality rates in this model, at the 95% confidence level. That is, lower readmissions go hand-in-hand with lower mortality rates. All models are hederoskedastic; robust standard errors are used. The low R- squared and additional tests indicate that all models have an omitted variable problem. 7 The second regression (Model 2) has as the dependent variable the hospital level 30- day mortality rate for Medicare patients with pneumonia and as the independent variable, a dummy variable indicating the hospitals 30-day readmission rate for pneumonia 7 Model specification issues were not improved when variables were logged or when the model controlled for interactions between the independent variable and control variables. See appendix for details of diagnostic tests. 24

patients is worse than the national average. Readmissions rates are not a significant predictor of mortality rates in this model. The third regression (Model 3) has as the dependent variable a dummy variable indicating the hospital s 30-day mortality rate for pneumonia patients is worse than the national average and as the independent variable, a dummy variable indicating the hospital s 30-day readmission rate for pneumonia patients is worse than the national average. Like in the first model, the independent variable is significant at the 95% confidence level and illustrates a positive relationship between readmission and mortality. The fourth regression (Model 4) has as the dependent variable a dummy variable indicating the hospital s 30-day mortality rate for pneumonia patients is worse than the national average and as the independent variable, the hospital-level 30-day excess readmission rate for Medicare patients with pneumonia. The independent variable is significant at the 99% confidence level. Thus, three of the four models indicate that readmissions and mortality move in tandem. The relationship between mortality rates and readmission rates was also examined when the hospital sample was divided into subsamples of for-profit/ non-profit, teaching/non-teaching, and urban/non-urban. 25

Table 9. OLS Estimates of the Relationship between Hospital Readmission Rates and Hospital Mortality Rates for Profit and Non-Profit Hospitals. (FOR-PROFIT) (NON-PROFIT) VARIABLES MORT_SCORE MORT_SCORE READM_SCORE 0.203*** 0.0255 (0.0624) (0.0244) TEACHING -0.427* -0.208** (0.225) (0.0805) URBAN -0.853*** -0.444*** (0.210) (0.073) Constant 13.61*** 16.26*** (1.101) (0.698) Observations 652 3,395 R-squared 0.049 0.0167 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 10. OLS Estimates of the Relationship between Hospital Readmission Rates and Hospital Mortality Rates for Teaching and Non-Teaching Hospitals. (TEACHING) (NON TEACHING) VARIABLES MORT_SCORE MORT_SCORE READM_SCORE 0.0511 0.0655** (0.0402) (0.0280) URBAN -0.8477*** -0.4312*** (0.165) (0.0765) PROFIT -0.0852 0.1411 (0.188) (0.1020) Constant 15.83*** 15.57*** (0.698) (0.476) Observations 1,102 2,945 R-squared 0.024 0.0123 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 26

Table 11. OLS Estimates of the effects of Hospital Readmission Rates on Hospital Mortality Rates for Urban and Non-Urban Hospitals. (URBAN) (NON-URBAN) VARIABLES MORT_SCORE MORT_SCORE READM_SCORE 0.0715* 0.0333 (0.0283) (0.0394) PROFIT -0.0515 0.4225** (0.109) (0.1605) TEACHING -0.360*** 0.9419 (0.0898) (0.1499) Constant 15.08*** 16.09*** (.486) (0.668) Observations 2,400 1,647 R-squared 0.0085 0.0053 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 The results show that the positive association between mortality rates and readmission rates disappears when the Model 1 is run for non-profit hospitals only, teaching hospitals only, and non-urban hospitals only. In other words, there is no statistically significant association between mortality rates and readmission rates for them. However, importantly for policymakers to know, there is a statistically significant positive association between mortality rates and readmissions rates for for-profit hospitals, non-teaching hospitals, and urban hospitals 8. 8 The sample was divided into subsamples for Model 2, Model 3, and Model 4. This statistically significant positive association between mortality rates and readmissions rates for subsamples non-teaching and urban hospitals was found in Model 3 and Model 4. This statistically significant positive association between mortality rates and readmissions rates for subsample for-profit hospitals was found in model 4. Detailed results available upon request. 27

9. Analysis The regression results seem to contradict the expected hypothesis that hospital-level mortality rates and readmission rates are inversely associated. However, the results show a positive significant association between readmission rates and mortality rates, when controlling for a hospital s status as a for-profit, teaching hospital, and whether the hospital is urban or rural. In other words, hospitals with lower readmission rates are more likely to have lower mortality rates for pneumonia patients. However, these findings do not offer entirely convincing evidence that the hypothesis is incorrect. There is reason to believe that the models and primary independent variable are incomplete in describing the relationship between mortality rates and readmission rates. Other studies in the literature controlled for geographic location (Pandey et. al. 2016, Lindenhauer, et. al. 2010) and number of beds (Pandey et. al 2016, Wang et. al. 2016) and whether a hospital s Medicaid caseload was greater than one standard deviation above the mean (Krumholz, Lin & Eenan 2013). These data were not available for this paper. The absence of these variables may be responsible for the model s specification issue. Additionally, the research contributing to this paper involves hospital admissions for pneumonia patients from July 1, 2012 to June 30, 2015. This specific population and time period has not been studied in previous research. The time period is especially important because the Hospital Readmissions Reduction Program (HRRP) and its impact on hospital payments began in 2012. The collection period analyzed in this paper is 28

entirely after the implementation of HRRP. The existing literature involves earlier collection periods. Most of the literature from collection periods prior to the implementation of HRRP showed that readmissions rates and mortality rates are inversely associated. This study, using data for after the program was implemented and during which readmission rates for pneumonia patients decreased nationwide, suggests that for at least in certain hospitals and settings, readmissions rates and mortality rates are positively associated. In conclusion, the findings show that the relationship between mortality rates and readmission rates is very complex and warrants further study. 10. Policy Implications Although this paper did not find evidence supporting the hypothesis of a negative relationship between mortality and readmissions, it contributes to the growing literature suggesting that the relationship is not clear. Therefore, this section will briefly address the policy implications of expanding Medicare s Hospital Readmissions Reduction Program (HRRP) without a clear picture of the relationship between mortality rates and readmission rates. The goal of HRRP is to lower Medicare spending while also improving quality, or at least to not lower quality. While it is clear that HRRP has resulted in lower spending, the program s relationship to quality is less clear. Existing literature suggests that the program s goal of lowering readmissions could adversely impact outcomes as measured by mortality rates. However, this study shows that at least for certain hospital types, HRRP may be helpful in controlling costs without jeopardizing quality. 29

Since HRRP began in 2012, the number of conditions for which hospitals can be penalized has grown. In addition, MedPAC has since endorsed readmissions penalties for other providers that participate in Medicare Part A including skilled nursing facilities (SNFs) and home health. Before the program expands, more research should be done to determine whether there is a definitive relationship between readmissions and mortality rates. Based on the evidence here, they appear to move in tandem. Thus, lowering readmissions could lower mortality. However, research into the differences in the relationship between readmissions rates and mortality rates before and after the implementation of HRRP is needed. In closing, this paper supports the conclusion of Pandey et al. 2016, that future prospective studies are needed to determine how readmission penalties levied by CMS have affected quality of care and outcomes in hospitals over time. 30

Appendix: Model Information Model 1: Diagnostic Notes Table A1. Model 1 Correlations. MORT_SCORE READM_SCORE PROFIT TEACHING URBAN MORT_ 1.0000 SCORE READM_ 0.0254 1.0000 SCORE PROFIT 0.0090 0.0868 1.0000 TEACHI -0.0798 0.0964-0.0446 1.0000 NG URBAN -0.1265 0.0990 0.1168 0.2729 1.0000 The variable correlation table does not indicate multicollinearity. Residuals -10-5 0 5 10 15.5 16 16.5 17 17.5 Fitted values Figure A1. Model 1 Heteroscedasticity Plot. The plot of the standardized residuals does not indicate that the model is heteroscedastic. 31