Benjamin P. Turner, BA. Washington, DC April 13, 2012

Transcription

1 THE DC HEALTHCARE ALLIANCE AND ACCESS TO HEALTHCARE: THE EFFECT OF INCREASED INSURANCE COVERAGE AND A DISPERSED SAFETY NET ON ACCESS TO HEALTHCARE IN THE DISTRICT OF COLUMBIA 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 in Public Policy By Benjamin P. Turner, BA Washington, DC April 13, 2012

2 Copyright 2012 by Benjamin P. Turner All Rights Reserved ii

3 The DC Healthcare Alliance and Access to Care: The effect of increased insurance coverage and a dispersed safety net on access to healthcare in the District of Columbia Benjamin P. Turner, BA Thesis Advisor: Harriet L. Komisar, Ph. D. ABSTRACT In 2001, the District of Columbia (DC) fundamentally changed its system of safety-net healthcare. The District closed its public hospital; created an insurance program for low-income residents, the DC Healthcare Alliance; shifted control of its publicly-run clinics to nonprofits; expanded Medicaid coverage; and invested in safety-net providers. DC shifted from directly providing healthcare to providing coverage and promoting care through private providers. This shift in priorities and structure offers an opportunity to examine the effects of increasing coverage and promoting primary care on access to healthcare. Using annual cross-sectional data from the Centers for Disease Control s Behavioral Risk Factor Surveillance System from 1998 to 2006, I conduct multivariate regression analysis controlling for individual demographics, socioeconomic status, and health status, as well as pre-reform trends in DC and pre- and postreform trends among a comparison group of eleven metropolitans in the region. Analysis finds that DC s reforms are associated with improved access to healthcare in three of four outcomes and improved access trends in all four outcomes. Among non-elderly adults in DC, the reforms reduced the uninsured population by 5.7 percentage points; reduced the population delaying care due to cost by 17.3 percentage points; reduced the population without a personal doctor by 6.4 percentage points; and increased the population without an annual checkup by 2 percentage points. The effectiveness of DC s reforms suggests that policymakers seeking to improve access can do so by increasing coverage and promoting safety-net providers. iii

4 TABLE OF CONTENTS Introduction... 1 Background... 3 Literature Review... 6 Access to Healthcare in Washington, DC...7 Determinants of Access to Healthcare...9 State and Local Programs to Improve Access to Healthcare...11 Conceptual Framework Data and Methods Analysis Plan...16 Results...22 Regression Results...31 Sensitivity and Stability Testing...39 Post-Regression Analysis...41 Discussion...43 Limitations...45 Policy Implications...47 Appendices...50 Appendix A. Unweighted characteristics of Washington, DC residents...50 Appendix B. Unweighted characteristics of comparison group residents...56 Appendix C. Health access outcomes of individual metropolitan residents...63 Appendix D. Weighted characteristics of Washington, DC residents...74 Bibliography...80 iv

5 LIST OF TABLES AND FIGURES Figure 1. Conceptual Model Figure 2. Observed Trends Access Outcomes, Unweighted Data Table 1. Unweighted descriptive statistics for selected years...26 Table 2. Linear probability regression model estimating effect of DC policy reform on insurance coverage and access to healthcare...31 Table 3. Linear probability regression model estimating effect of DC policy reform on access to healthcare, controlling for insurance coverage...35 Table 4. Actual and predicted percentage of DC residents with poor access to healthcare in 2006 with 95% confidence intervals...41 v

6 INTRODUCTION In 2001, then-mayor of the District of Columbia, Anthony Williams, closed DC General Hospital after years of poor quality of care, health outcomes, and financial mismanagement. In its place, he established the DC Healthcare Alliance, a program designed to provide health insurance coverage for all DC residents earning less than 200% of the federal poverty level and to promote the use of primary care through clinics and providers. With this change, the city shifted from directly providing healthcare services, through DC General Hospital and a small network of publicly run clinics, to a public-private partnership in which the city relies on nonprofit and for-profit private healthcare organizations to provide services to low-income residents insured by DC Healthcare Alliance (Offner, 2001; Sager, 2001; Ormond et al., 2010). At the time it was created, the intent of the Alliance was to decentralize health care options and provide more healthcare choices, each of which would be in competition for patients. The Alliance was built on the assumption that competition among private providers would result in higher quality services at lower cost than a non-competitive public hospital would be able to provide. Advocates for the DC Healthcare Alliance also hypothesize that payments from the Alliance to small safety-net clinics strengthened them and allowed them to increase the quantity and quality of services to uninsured and publicly insured residents (Ormond et al., 2010). During the same time period, the District invested in other programs to increase insurance coverage and to promote preventive care for low-income residents through private clinics. My analysis examines whether these reforms were effective in increasing access to healthcare and estimates the effect of the changes to the safety-net. This is relevant to current policy decisions regarding the future of the DC Healthcare Alliance. Since 2006, and with increasing 1

7 speed since the passage of the Patient Protection and Affordable Care Act (PPACA) in 2009, the District of Columbia has been replacing the Alliance with federally subsidized insurance coverage. Medicaid now covers families earning up to 200% of poverty in DC (Kaiser Family Foundation, 2010). In addition to replacing much of the program with Medicaid expansions, the District has added additional enrollment processes and several budgets have been proposed that would cut services covered by the Alliance (Patterson and Bell, 2011; Farmer, 2012). Current Alliance beneficiaries consist entirely of people ineligible for Medicaid despite their low income; primarily recent immigrants, having immigrated in the last five years, and undocumented immigrants. Most of this population will be exempt from the individual mandate and ineligible for subsidies to purchase insurance on the PPACA health insurance exchanges (Cox and Levitt, 2012). Since federal programs will not provide coverage to the population that the Alliance serves, reducing or eliminating the Alliance will likely result in a decline in insurance coverage. Understanding the effectiveness of the Alliance and its accompanying reforms will help inform decisions regarding the District s post-ppaca safety-net. Also, local policy-makers can examine the effectiveness of the District s reforms in extending insurance coverage and access-to-care programs beyond state and federal policies. The Congressional Budget Office (CBO) estimates that after the Affordable Care Act is fully implemented, there will still be 26 million uninsured Americans (CBO, 2012). Local policymakers can examine DC s effectiveness in extending access through local action aimed at a very similar population to those expected to be uninsured after the Affordable Care Act. This analysis examines the change in access to care among DC residents, using multivariate linear regression to control for the effects of gender, age, income, education, race, health status, 2

8 health insurance status, trends in DC before reform, and regional trends. The analysis conducts linear probability models using individual level data from 1998 to 2006 from the Behavioral Risk Factor Surveillance System (BRFSS) conducted by the Centers for Disease Control (CDC). The BRFSS produces data that are representative at the national and state level. The analysis uses data from BRFSS surveys of residents of DC as well as eleven metropolitan areas in the region. Reforms in 2001 changed the District of Columbia s system of safety-net healthcare from a centralized, directly provided government program to a dispersed network of contracted private providers. The following analysis estimates the effects of reforms on access to healthcare and will discuss the implications for other local policy makers and the future of DC s reforms after the Affordable Care Act. BACKGROUND When the DC Healthcare Alliance was first implemented in 2001, the Alliance referred to the small group of private and government organizations managing the new program. The DC Department of Health contracted Greater Southeast Community Hospital, which then contracted four other hospitals and Chartered Healthcare, a managed care organization. In the first five years of its existence, DC Healthcare Alliance beneficiaries either received primary care directly through the four Alliance hospitals or received managed care through Chartered Healthcare, which included a large network of primary care physicians. Chartered Healthcare administered fee-for-service payments to all providers serving Alliance beneficiaries (Ormond et al., 2010). After some concerns over the stability and management capacities of the Alliance s main partner, Greater Southeast Community Hospital, the DC Department of Health altered the structure of the Alliance in 2003, contracting Chartered Healthcare directly rather than through 3

9 Greater Southeast. In 2006 the DC Healthcare Alliance shifted to providing beneficiaries only managed care and contracted with United Healthcare as well as Chartered Healthcare to provide managed care (Ormond et al., 2010). To be eligible for the DC Healthcare Alliance, individuals must live in Washington DC, have a family income that is less than 200% of the federal poverty threshold and be ineligible for Medicaid or other public health insurance programs. In 2001, children in households earning less than 200% of the poverty level were covered through Medicaid, and by 2004, their families were covered as well. So the Alliance primarily provided coverage to childless adults earning less than 200% of poverty as well as recent immigrants and undocumented immigrants earning less than 200% of poverty. The goal of the Alliance was that each person covered by the program would have a medical home that is, a primary care provider who would be responsible for coordinating the individual s care efficiently and effectively (Kaiser Family Foundation, 2011). The Alliance signed-up new clients rapidly. Community health centers and hospitals that served uninsured clients had a strong incentive to sign their patients up for the new benefit to avoid providing uncompensated care. Within a year of implementation, there were 20,000 DC residents signed-up for the DC Healthcare Alliance. Six months after that, an additional 10,000 residents enrolled in the program, helping to reduce the number of non-elderly adults without insurance from 105,000 to 75,000 over a three year period (Aragón and Lillie-Blanton, 2004; Ormond et al., 2010). The DC government implemented several programs that complemented the Healthcare Alliance. Between 2001 and 2006, the District gradually expanded Medicaid eligibility for childless adults and eventually would subsidize insurance coverage for families earning less than 4

10 200% of the federal poverty level (DC Healthy Families Act, 2008). Also, the DC Primary Care Association (DCPCA) won a $2.5 million grant in 2004 promote medical homes, improving funding for safety net clinics significantly (Ormond et al., 2010). This collection of reforms and initiatives all aimed at increasing insurance coverage or the strength and capacity of non-profit clinics. Together they replaced the publicly-run hospital and clinics with a dispersed safety-net made up of private organizations reliant on the DC government for funding. It is important to note the major objections and criticisms to the reforms in DC s healthcare system. First, the public system that the reforms replaced, DC General Hospital, was a local institution; a public hospital dedicated to serving the city s poor for decades (Offner, 2001). Though DC General was widely acknowledged as poorly run and inefficient, replacing the public hospital with a private system was controversial. Critics did not trust that the new system would share the same commitment to serving the city s poor, especially as the key partner in the program was a for-profit hospital that also had a reputation for poor management. Advocates for the poor in the District believed that the reforms were cost-cutting at the expense of the city s mostly black low-income population that relied on DC General (Offner, 2001). In addition, critics of the reforms argued that they did not address the District s disparities in access to healthcare and health outcomes that split the city along racial and geographic lines. Black and Hispanic people experienced considerably poorer coverage, access to care measures, and health outcomes. There also pronounced disparities between people living in the low-income and often neglected neighborhoods east of the Anacostia River and the downtown, generally wealthier areas west of the Anacostia River (Ormond et al., 1999). These communities relied more heavily than the rest of the city on DC s public system and critics believed that the reforms 5

11 would not only fail to address the existing disparities, but make them worse by disturbing the existing safety-net (Offner, 2001). Finally, critics point to the structure of the DC Healthcare Alliance: a managed care system where many healthcare providers would compete for Alliance beneficiaries. Critics argue that DC did not have enough healthcare providers that would be interested in Alliance beneficiaries. They predicted that with too few providers for Alliance beneficiaries, the program would not effectively promote access to care; choices would be too limited and they would not receive quality preventive care that advocates expected (Ormond, 2011). Advocates for the reform had faith that the reforms would extend access and a higher quality of medical care than DC General could and that enough clinics and private providers would seek Alliance patients that it would effectively reduce disparities while also containing healthcare spending. My hypothesis is that the reforms to DC s healthcare system improved access to healthcare, both by directly providing insurance coverage to a vulnerable population and by promoting safety-net clinics that focus on preventive care. LITERATURE REVIEW Research on insurance coverage, access to healthcare, and health outcomes in Washington, DC has found that access to healthcare improved modestly after reforms. There is a large body of research showing significant effects of insurance coverage and healthcare supply on access to care, as well as individual variables, including socioeconomic status, demographic variables, and health status. Finally, literature on state and local policy has found that coverage expansion policies can improve access to healthcare. My research contributes to each of these sets of literature. 6

12 Access to Healthcare in Washington, DC Research on health care in DC finds that access to healthcare was poor before 2001 and has not improved as much as hoped since the system of safety-net care was reformed. Both the popular and the academic consensus before 2001 was that DC s healthcare system failed to provide adequate care to low-income and vulnerable residents (Bovbjerg and Ormond, 1998; Ormond et al., 1999). There is also a consensus in the research conducted after the implementation of the Alliance that access to healthcare in DC is poor and that health outcomes are lower than the goals set by the government (DCPCA, 2005; RAND, 2008). Researchers have looked closely at access to healthcare and the persistence of health disparities in Washington, DC. Before 2001, there are several descriptive studies that show that access to care and health outcomes were worse in DC compared to the nation and unsustainable (Bovbjerg and Ormond, 1998; Ormond et al., 1999). In 1998, Randall Bovbjerg and Barbara Ormond wrote a descriptive study of the hospital sector, finding that the pre-alliance safety-net system was insufficient; 25% of DC residents were underserved by primary care physicians and almost 19% of non-elderly adults were uninsured (Ormond and Bovbjerg, 1998). After the DC Healthcare Alliance was created and DC General Hospital was closed, studies found limited access to healthcare and persistent disparities in access (Aragón and Lillie- Blanton, 2004; O Malley, 2004; Lewis and Offner, 2002; Kaiser Family Foundation, 2004; Mclure and Jerger, 2005; Lurie et al., 2008). Analysis of a 2003 survey conducted by the Kaiser Family Foundation found that access was an ongoing problem, especially when compared to other states. The survey found that over 26% of Latinos and 10% of Blacks in DC had no regular source of care or their regular source of care was the emergency room. In contrast, only 3% of 7

13 White DC residents had this lack of access (Aragón and Lillie-Blanton, 2004). Regression analysis confirmed the disparities in access to care, especially for Latinos, with Latinos 2.1 times more likely to report no medical visit in the last 12 months than whites after controlling for other factors (O Malley, 2004). These disparities remained even with regression analysis controlling for insurance status (Lurie et al., 2008). In addition to finding disparities in access to care in DC, research in 2005 through 2007 found that access to healthcare and health outcomes fell below the goals set by the DC government (DCPCA, 2005; Ormond, 2007; RAND, 2008). In 2005, nearly 300,000 DC residents lived in Medically Underserved Areas: federally-defined neighborhoods with limited access to personal health services (DCPCA, 2005). And while insurance coverage improved to the point of being higher in the District than the rest of the country (Ormond, 2007; Ormond, 2010) and a group of comparison cities (RAND, 2008), avoidable hospitilizations remained high (DCPCA, 2005) and other health outcomes were as poor as benchmark cities despite the improved insurance coverage (RAND, 2008). My thesis will contribute to the literature on access to healthcare in Washington, DC by conducting a regression analysis testing the changes in health access before and after the implementation of the Alliance. Though the current literature shows poor access to care both before the Alliance was implemented and modest improvements after, there is no analysis on whether changes in access were statistically significant or caused by DC s policies, rather than federal policy, regional changes in access to care, or demographic changes within the District. My analysis will improve our ability to assess the effects of DC s shift in policy on access to healthcare. 8

14 Determinants of Access to Healthcare Many researchers have examined the determinants of healthcare access using multivariate regression analysis. My research builds on these studies and their findings on how insurance coverage, healthcare supply, demographics, socioeconomic factors, and health status affect access to healthcare. Studies into access to healthcare have found one of the strongest relationships to be that between insurance coverage and access to health care (Hafner-Eaton, 1993; Long, Coughlin, and King, 2005; Weissman et al., 2008). Regression analysis of the National Health Interview Survey found that those with insurance coverage were percentage points more likely to have a doctor s visit, depending on their health status (Hafner-Eaton, 1993). Regression model using the National Survey of America s Families found that having insurance coverage improved access to care measures by 5-28 percentage points (Long, Coughlin, and King, 2005). A more recent study using the Current Population Survey found that income-based gaps in access to health care were smaller in states with higher levels of Medicaid eligibility, suggesting that eligibility for Medicaid insurance improves access to healthcare (Weissman et al., 2008). These three papers, along with many others in the literature around insurance coverage and access to healthcare, conclude that insurance coverage is associated with increased access to healthcare. Literature on the availability of clinics and hospitals finds that these factors are associated with significant improvements in an individuals access to healthcare (Mathematica Policy Research, 1999; Shi and Stevens, 2007; Cunningham and Hadley, 2004; Falik et al., 2001). Uninsured clinic-users are about half as likely as uninsured non-clinic users to delay care, fail to receive the medical care they felt they needed, or fail to fill prescriptions that they had received 9

15 (Mathematica Policy Research, 1999). Regression analysis of the Community Health Center User Survey and the National Health Interview Survey finds that health center users are 14 percentage points more likely to have had a generalist physician visit, percentage points more likely to have access to healthcare. The research notes that insured clinic-users had better outcomes than uninsured clinic-users (Shi and Stevens, 2007). Regression analysis of the Community Tracking Survey and the Uniform Data System shows that proximity to health centers, measured by the total grant revenues of all community health centers within five miles of the respondent, is associated with significant increase in access to healthcare for low income people. The analysis found that the effects of community health center proximity were greater for the insured than the uninsured (Cunningham and Hadley, 2004). The literature on healthcare supply concurs that being a community health center user or being in close proximity to a community health center is associated with increased access to healthcare. The research acknowledges that community health centers alone are not enough to meet the needs of the uninsured, insurance coverage is necessary to promote access to specialty care and improve long term access and health outcomes. All of the studies mentioned control for socioeconomic variables, demographic variables, and individual health status. Often they note that some of these variables were strongly associated with access to care, most often income, health status, and race/ethnicity. Some researchers split their analyses into groups based on one or more of these factors (Weissman et al., 2008; Hafner- Eaton, 1993). Education, employment, and family size are also linked with insurance status and access outcomes. 10

16 My research will contribute to the existing literature on access to healthcare by analyzing a program that focused on both increasing health insurance coverage and promoting the supply and use of safety-net clinics. State and Local Programs to Improve Access to Healthcare Finally, literature on state and local policy changes is highly relevant to this analysis. States, counties, and cities can increase access to healthcare by expanding Medicaid coverage, subsidizing private insurance, creating public health insurance programs, mandating the purchase of health insurance, public education and health promotion programs, and by supporting safetynet coverage (Berk and Schur, 1998; Finkelstein et al., 2011; Leininger et al. 2011; Weissman et al., 2008; Pande et al., 2011). An analysis of state administrative data in Wisconsin found that Medicaid expansion resulted in higher enrollment of low-income residents in insurance, establishing that Medicaid improves insurance coverage (Leininger et al., 2011). Regression analysis of data from the National Health Interview Study found that persons with Medicaid were two to four times as likely to have a regular source of care compared to the uninsured (Berk and Schur, 1998). In Oregon, the beneficial effects of Medicaid were confirmed, with a lottery selecting a group of uninsured lowincome adults to be given Medicaid creating a randomized control experiment. Analysis found the treatment group having higher health care utilization, lower out-of-pocket medical expenditures, and better self-reported health (Finkelstein et al., 2011). As mentioned earlier, analysis of the Current Population Survey found that states with higher Medicaid eligibility had smaller income-based access gaps (Weissman et al., 2008). 11

17 Analysis of Massachusetts 2006 healthcare reform using BRFSS data found the reform associated with higher probabilities of being insured, having a personal doctor and a lower probability of foregoing care due to cost, compared to the expected levels of insurance coverage and access if there had been no reform. The effects of the reform on access to care were much larger for groups defined as disadvantaged (ethnic minorities, low-income, low-educated, or selfemployed) than for the rest of the state (Pande et al., 2011). The existing literature consistently shows that states can increase access to care by expanding insurance coverage through direct expansions or mandates. My analysis will add to this literature by analyzing the combination of reforms in DC: expanded insurance coverage, privatization of indigent care, and increased financial support of safety-net clinics. I build on the existing research to show the effects of the District s reforms on access to healthcare. CONCEPTUAL FRAMEWORK This analysis is built on the conceptual framework that access to healthcare is affected by insurance coverage and the availability and strength of local healthcare providers. The analysis controls for other variables that could affect access to healthcare, including socio-economic factors, demographics, and individual health status. The primary hypothesis is that reforms to the DC safety-net healthcare system increased access to healthcare by increasing insurance coverage and promoting the network of safety net clinics. The framework is illustrated in Figure 1 below. 12

18 The first assumption of this analysis is that access to health insurance reduces barriers to health care. The literature strongly supports this assumption (Cunningham and Hadley, 2004; Hafner-Eaton, 1993; Long et al., 2005; Pande et al., 2011; Politzer et al., 2001; Shi and Gregory, 2007; Weissman et al., 2008). Many insurance programs, including Medicaid and the DC Healthcare Alliance, have provider networks that coordinate care and promote the use of preventive care. Health insurance makes it more likely that a person would be able to afford health care services, would seek preventive care, and would have a medical home. I hypothesize that the reforms to DC s healthcare system increased insurance coverage, mainly through the DC 13

19 Healthcare Alliance program and Medicaid expansions, and that the increased insurance coverage improved access to healthcare. Presence, proximity, and strength of community health centers also increase an individual's access to healthcare (Cunningham and Hadley, 2004; Politzer et al., 2001; Shi and Gregory, 2007; Falik et al., 2001). Federally Qualified Health Centers (FQHC s) are required to provide primary medical care services regardless of an individual's ability to pay. Most community health centers, even those without FQHC status, have a similar mission to provide care to uninsured or low-income individuals. Individuals living in communities that have many wellfunded FQHC's are more likely to have access to healthcare, even if they are uninsured. I hypothesize that the reforms to DC s healthcare system supported safety-net clinics, through direct grants and the insurance coverage limiting the amount of uncompensated care, and that the strengthened safety-net clinics increased overall access to care. Socioeconomic factors such as income, education, and employment are strongly correlated both with access to healthcare and with access to health insurance (Cunningham and Hadley, 2004; Hafner-Eaton, 1993; Long et al., 2005; Pande et al., 2011; Politzer et al., 2001; Shi and Gregory, 2007; Weissman et al., 2008). A person with a high income is more likely to be able to purchase health services when needed, and is more likely to be able to purchase health insurance. Similarly, employed people are far more likely than unemployed people to be insured and more likely to have income that would allow them to purchase health services whether or not they were insured. Finally, a person with more years of education may have greater health literacy, making them more likely to be insured and to seek medical care, and is more likely to have the kind of job that would provide insurance coverage. 14

20 In addition, an individual's health status has a strong impact on their access to healthcare (Hafner-Eaton, 1993). An individual with a chronic condition is more likely to seek regular medical care, simply because they need to. Similarly, they are more likely to be insured simply because they need health insurance more than a healthy person does. This relationship could push in the opposite direction, with people who have low access to healthcare or health insurance being more likely to have poor health status. The need for insurance coverage and medical services is associated with changes in access to health care and insurance coverage. Finally, as mentioned in the introduction, there are a number of trends or policy changes that could affect access to healthcare, either directly or indirectly. Changes in Medicaid reimbursement rates could affect access to care for all Medicaid beneficiaries. Similarly, changes in the self-purchased insurance market, changes in employers policies and prices for employersponsored insurance, or changes in healthcare costs will affect access to care. These trends or policy shifts could increase or decrease access to care and could change year-by-year. My primary hypothesis that the shift in the District of Columbia s safety-net model increased access to healthcare is built on this framework. I hypothesize that the District increased access to healthcare by promoting insurance coverage and supporting safety-net clinics, controlling for the effects of socioeconomic factors, demographic factors, individual health status, and regional trends. DATA AND METHODS Analysis relies on the Behavioral Risk Factor Surveillance System (BRFSS) from 1999 through The BRFSS is a project of the Centers for Disease Control (CDC) designed to measure behavioral risk factors, preventive health practices, and health care access. It is a 15

21 publicly available, annual cross-sectional telephone survey reaching over 350,000 people annually with large samples in each state, including DC. The survey contacts households using random-digit-dialing and stratified weighting sampling. The sample size in DC averages around 1,500 non-incarcerated, non-elderly adults annually from 1998 to Analysis will be limited to respondents from age 18-64, as policy changes are not expected to have a significant effect on the Medicare-eligible population. The analysis compares the changes in outcomes in Washington, DC to changes in a group of eleven regional metropolitan areas from 1998 to These metros include: Baltimore, MD; Hagerstown, MD; Newark, NJ; Philadelphia, PA; Pittsburgh, PA; Providence, RI; Richmond, VA; Scranton, PA; Trenton, NJ, Virginia Beach/Norfolk, VA; and Wilmington, DE. These metropolitan areas were chosen due to proximity to DC and availability of data for each year of analysis. Sample sizes varied by year; the number of DC residents surveyed increased from 1,170 in 1998 to 3,094 in 2006 and the number of residents of control groups increasing form 8,055 in 1998 to 19,903 in The total sample used in the analysis is 139,869 individuals. Analysis Plan This paper relies on an interrupted time-series analysis of individual data, using linear probability regression models to estimate the change in measures of access to healthcare before and after the implementation of the DC Healthcare Alliance. The implementation of the DC Healthcare Alliance is treated as a quasi-experiment with all DC residents answering surveys after 2001 as having been exposed to the treatment and all DC residents surveyed 2001 and before as unexposed to the treatment. 16

22 The methodology of the analysis is based on a 2011 study that used BRFSS data to estimate the effects of Massachusetts healthcare reform on access to care (Pande et al., 2011). Using residents of metropolitan areas in the region as a comparison group, the analysis estimates the difference between the changes in access trends among DC residents after reform from the change in access trends among the comparison group. This effectively controls both for prereform trends in access to care in the District of Columbia, and regional post-reform trends which could include exogenous shocks to health access. Analysis includes four outcome variables, the first of which is insurance status. Insurance status was defined using the BRFSS survey question: Do you have any kind of health care coverage, including health insurance, prepaid plans such as HMOs, or government plans such as Medicare? Analysis treats respondents that answer no as uninsured, omits those who say that do not know their insurance status or refuse to answer, and treats all others as insured. Next, to examine changes in healthcare access, the primary outcome for this paper, the analysis relies on three questions: (1) Do you have one person you think of as your personal doctor or health care provider? For those who respond that they do not, the survey clarifies whether they have multiple personal doctors. If they have one or more personal doctor, this analysis treats them as having access to a personal doctor. Those who do not know or refuse to answer are omitted. (2) Was there a time in the past 12 months when you needed to see a doctor but could not because of cost? All respondents who answered one year or less are considered to have access to preventive care. Those who do not know or refuse to answer are omitted. 17

23 (3) About how long has it been since you last visited a doctor for a routine checkup? All respondents who answered one year or less are considered to have access to preventive care. Those who do not know or refuse to answer are omitted. The key assumption of this model is that if there had been no reform, access trends for DC residents would have followed the trends of the control group, controlling for the pre-reform trends in the District. To capture these trends, the model relies on seven constructed variables based on the respondent s DC residency and the year the respondent was surveyed: DC is a binary variable indicating whether or not the respondent lives in Washington, DC. The coefficient for this variable indicates the effect of being a DC resident rather than a member of the control group on the probability of having poor access to healthcare. Year is a continuous variable set 0 to 8 depending on whether the respondent took the survey in 1998 through 2006, and estimates a baseline (pre-reform) trend. DCxYear is an interaction multiplying the first two. This variable allows there to be a difference between the baseline trend for DC residents and the baseline trend for the control group of other regional metropolitan areas. Shift is a binary variable set to zero for respondents surveyed before 2002 and 1 for respondents surveyed 2002 and on. This variable estimates a shift in access to care immediately after PostTrend is a semi-continuous variable set to zero for respondents surveyed before 2003 and 1 to 4 for respondents surveyed in 2003 to This variable estimates the difference between the pre-2002 trend and the post-2002 trend in access to healthcare. 18

24 DCxShift is a binary variable constructed from an interaction between DC and Shift. It is set to zero for anyone surveyed in 2001 and before and 1 for District residents surveyed after It estimates the difference between the DC s and the control group s shift in access to healthcare immediately after reforms in DC. DCxPostTrend is a semi-continuous variable constructed from an interaction between DC and PostTrend. It is set to zero for non-dc residents and for DC residents surveyed before 2003 and 1 to 4 for District residents surveyed from 2003 to It estimates the difference between DC s and the comparison group s change in trend after reforms in DC. The final two variables are the predictors of interest. The coefficient on DCxShift can be interpreted as the immediate and lasting policy effect of the Alliance. The coefficient on DCxPostTrend can be interpreted as the effect of the Alliance on access trends. By including two predictors of interest, the model will be able to show a change in the trends associated with the reforms to DC s healthcare system, DCxPostTrend, while differentiating that trend-change from any one-time change in access associated with the reforms. This generates a clearer estimate of the effects of reform. The analysis includes control variables for demographic characteristics, socioeconomic variables, health status, and insurance status. The demographic variables included in the analysis are age, gender, race, marital status, and the number of children. The analysis uses eight age categories: 18-24; 25-29; 30-34; 35-39; 40-44; 45-49; 50-54; 55-59; and The analysis includes five variables to indicate whether the respondent describes him or herself as non-hispanic white, non-hispanic black, Hispanic, Asian, 19

25 or other/mixed race. The respondents are classified status as married if the respondent is married or with a permanent partner. Respondents are classified as unmarried if he or she has never been married, is widowed or divorced. The respondents are classified status as married if the respondent is married or with a permanent partner. Respondents are classified as unmarried if he or she has never been married, is widowed or divorced. Next, the analysis includes socioeconomic variables: employment status, education and income. Analysis divides employment status into three categories, the individual is either employed, a student, or other (unemployed, unable to work, retired, or a homemaker).analysis divides education into four categories: not a high-school graduate, high school graduate without college, high school graduate with some college, or college graduate. Annual household income is collected as seven categories in the BRFSS survey: less than $10,000; $10,001-$15,000; $15,001 to $20,000; $20,001-to $25,000; $25,001-$35,000; $35,001-$50,000; $50,001-$75,000; and more than $75,000. The analysis includes five controls for individual health status. The first is a simple general health status measure, based on individuals response to a question asking them to rate their health status from poor to excellent. My analysis simplifies the response to good, if the respondent answered good, very good, or excellent, and not good if they responded fair or poor. The second measures health status based on whether respondents answered that their activity was limited by poor health in the last month. The third measure is whether the respondent reports having diabetes. The fourth is whether the respondent is a current smoker, based on two questions: whether the respondent has smoked 100 cigarettes in his or her life and whether the respondent currently smokes cigarettes. Finally, the analysis controls for whether the 20

26 respondent s reported height and weight classifies them as obese on the Body Mass Index (a score of 30 or greater is categorized as obese). The BRFSS collects reported height and weight and calculates a score for each respondent that answers. Respondents that refused to answer questions used for the control variables were set to the median answer for each question. The exception was income, where rather than assign a fairly large number of people to an income category, a new group was created for the respondents who refused to give their annual income. Complicating the analysis and interpretation, the BRFSS did not ask the outcome questions needed in every year for three of the four outcome variables. The only outcome for which there is data in every year is insurance coverage. The outcome variable, Medical Cost as Barrier, based on whether the respondent has had to delay seeking medical care due to cost in the last year, is missing data in 2001 and 2002, which does not significantly change interpretation of the regression. However, the outcome variable, No Personal Doctor, based on whether the respondent has one or more particular doctor they can call when needed, is missing before This changes interpretation of analysis quite significantly. Without more than one year of data before reform, the model cannot control for pre-reform trends, so the Year and DCxYear variables are omitted from regressions. Instead of a difference in trend-change, the DCxPostTrend variable only shows the difference in post-reform access trends between DC and the comparison group. And the DCxShift variable is less compelling as the pre-reform period it relies on for comparison consists of only one year,

27 Additionally, the outcome variable, No Annual Checkup, based on whether the respondent has had an annual preventive checkup in the last year, is missing in 2001, 2002, 2003, and Since there are only two years of post-reform data, the change-in-trend variable, DCxPostTrend, is less convincing. The analysis relies on linear probability models rather than logistic regression for ease of interpretation. The magnitudes and significance levels of all coefficients are very similar regardless of whether linear probability models are used rather than logistic regression models. RESULTS Observed trends in the data collected from the BRFSS show varied patterns. Figure 2 suggests that DC s trends in health outcomes changed after 2001 in a way that the trends in the comparison group did not. The observed data show that the percentage of people without health insurance declined more than three percentage points in DC between 2001 and 2006, compared to an increase of about two percentage points in the comparison group. However, between 1998 and 2001 DC residents experienced a decline in uninsurance twice the magnitude of the comparison group. The percent uninsurance for the control group increased after reform while the pre-reform decline that DC residents experienced accelerated. Figure 2 also shows DC and the comparison group following significantly different trends in the percentage that have delayed care due to cost. Before reform, the percent that delayed care in DC increased by almost two percentage points each year while dropping slightly in the comparison group. After reform, these trends reversed, with the percent of DC residents delaying care declining while the comparison group increased, though with a sharp increase between 2003 and 2004 before declining. 22

28 The third outcome, the percent without a personal doctor, only includes one year of prereform data so it does not present pre-reform trend-lines. Both DC and the comparison group experienced a decline in the percent without a personal doctor after reform. The decline among 23

29 DC resident is greater than the comparison group s decline. However, without pre-reform trends, it is impossible to tell when this improvement relative to the comparison group started. The final outcome variable, the percent without a preventive checkup in more than a year, also is missing several years of data. The fact that there are only two years of post-reform data makes trend analysis difficult. In 1998, the largest age-group in the population surveyed in DC was the year old group, with 29% and about two-thirds of DC residents were under the age of 45. By 2006, the largest age-group in DC was the year-old group, with only 52% of residents under the age of 45. In 1998, 60.4% of DC residents were female, decreasing to 60.3% in The racial composition of the District changed considerably between 1998 and 2006: in 1998, 32.7% of residents were non-hispanic White, 60.7% were non-hispanic Black, 4.4% were Hispanic, and 2.2% were in other racial categories. In 2006, 50.5% of DC residents were non-hispanic White, 39.8% were non-hispanic Black, 4.3% were Hispanic, and 5.4% were in other racial categories. In 1998, 70.7% of DC residents had no children, 23.2% had one or two children, 4.9% had three or four children, and 1.7% had more than 4 children. In 2006, 69.3% of DC residents had no children, 24.6% had one or two children, 5.4% had three or four children, and 0.7% had more than four children. In 1998, 85% of DC residents were unmarried, decreasing to 58.8% in The largest income group in 1998 was the $25,000-$35,000 group, with 21.9% of DC residents. More than half, 51.7%, of DC residents earned less than $35,000, 11.8% earned $75,000 or more and 4.1% did not know or refused to report their income. In 2006, the largest income-group in DC was the more-than $75,000 group, with 42.8%. Only 23.6% of DC residents earned less than $35,000 and 10.5% did not know or refused to report their income. In 1998, 24

30 78.2% of DC residents were unemployed or students, decreasing to 70.5% in The largest education group in 1998 was the college-degree or more group with 43.9% of DC residents. An additional 21.5% had some college education, 25.8% had a highschool diploma without college education, and 8.7% had not completed highschool. In 2006, 62.2% of DC residents had completed college, 15.8% had some college education, 17% had a highschool degree without college, and 5% had not completed highschool. In 1998, 91.2% of DC residents described themselves as in good, very good, or excellent health, declining slightly to 90.4% of residents in Residents were not asked about whether their physical health limited them until 2000, where 13.3% of DC residents described themselves as limited, increasing to 15.7% by Among DC residents, 4.4% reported that they were diabetic in 1998 and 6.8% reported being diabetic in Based on DC residents reported height and weight, 22% were obese in 1998 and 26% in Finally, 16.7% of DC residents were smokers in 1998, declining to 9.9% in Table 1, below, presents four years of observations 1998, 2000, 2003, and 2006 describing the characteristics of the study population in each of these years for DC and the comparison group. All years of observations for DC residents are available in Appendix A and for comparison group residents in Appendix B. Appendix C shows observations for outcomes for each member of the comparison group. Table 1 presents unweighted data for both DC and the control group. The weighting practices that the BRFSS uses for metropolitan areas changed in 2002, complicating the use of weighted data. Rather than use weighted data for the treatment group and unweighted for the comparison group, Table 1 presents both groups unweighted. Weighted observed data for DC residents is available in Appendix D. 25

31 26 Table 1. Unweighted descriptive statistics for selected years Residents of Washington, DC Residents of Comparison Group STUDY POPULATION: Total Population 1,170 1,455 1,663 3,094 8,056 8,980 12,037 19,903 DEPENDENT VARIABLES: Insurance Status Sample for Insurance Status 1,167 1,452 1,659 3,091 8,042 8,964 12,005 19,864 Uninsured 13.1% 12.3% 8.1% 7.4% 11.3% 10.3% 11.5% 11.2% Insured 86.9% 87.7% 91.9% 92.6% 88.7% 89.7% 88.5% 88.8% Has had to delay care in last 12 months Sample for Delayed Care 1,170 1,455 1,663 3,094 8,056 8,980 12,037 19,903 Has had to delay seeking care 8.1% 12.3% 9.8% 8.9% 8.8% 8.7% 10.9% 11.5% Has not delayed care 91.9% 87.7% 90.2% 91.1% 91.2% 91.3% 89.1% 88.5% No particular doctor when sick Sample for Personal Doctor 0 1,564 a 1,657 3, ,193 a 12,021 19,865 No doctor N/A 25.2% 23.0% 18.3% N/A 14.5% 14.5% 12.0% Has doctor N/A 74.6% 77.0% 81.7% N/A 85.5% 85.5% 88.0% Has had routine medical checkup in last year Sample for Annual Checkup 1,170 1, ,091 8,053 8, ,894 No visit 19.3% 19.4% N/A 25.9% 24.2% 23.1% N/A 28.2% At least one visit 80.7% 80.6% N/A 74.1% 75.8% 76.9% N/A 71.8% a Since the question, Do you have a particular doctor? was not asked in 2000, data from 2001 is included for this variable. All percentages listed for other variables in this column are proportions of the 2000 Total Population sample at the top of each page.

32 27 Table 1. Unweighted descriptive statistics for selected years (continued) Residents of Washington, DC Residents of Comparison Group DEMOGRAPHIC VARIABLES: Age % 12.9% 9.0% 6.4% 10.6% 9.9% 8.0% 5.8% % 29.6% 27.1% 21.6% 24.6% 22.2% 19.1% 15.4% % 23.9% 24.2% 24.6% 29.2% 29.4% 27.6% 24.6% % 21.1% 22.3% 24.4% 21.3% 23.7% 25.7% 28.5% % 12.5% 17.4% 23.0% 14.3% 14.8% 19.6% 25.7% Gender Male 39.6% 40.2% 42.7% 39.7% 41.3% 39.8% 39.7% 37.8% Female 60.4% 59.8% 57.3% 60.3% 58.7% 60.2% 60.3% 62.2% Race/Ethnicity White, non-hispanic 32.7% 34.8% 51.4% 50.5% 78.9% 78.0% 75.4% 78.1% Black, non-hispanic 60.7% 51.3% 38.1% 39.8% 13.8% 12.3% 12.9% 12.5% Hispanic 4.4% 8.1% 4.6% 4.3% 4.5% 6.8% 6.5% 5.3% Other 2.2% 5.8% 5.8% 5.4% 2.8% 3.0% 5.2% 4.1% Number of children in household None 70.7% 69.0% 71.8% 69.3% 55.4% 52.6% 54.1% 57.5% 1 to % 23.6% 23.3% 24.6% 35.7% 37.5% 36.7% 33.7% 3 to 4 4.9% 6.0% 4.4% 5.4% 8.3% 8.9% 8.5% 8.0% 5 or more 1.2% 1.4% 0.5% 0.7% 0.7% 1.0% 0.6% 0.8%

33 28 Table 1. Unweighted descriptive statistics for selected years (continued) Residents of Washington, DC Residents of Comparison Group Marital Status Married 15.0% 29.0% 36.6% 41.2% 54.5% 56.1% 57.5% 59.2% Not married 85.0% 71.0% 63.4% 58.8% 45.5% 43.9% 42.5% 40.8% SOCIOECONOMIC VARIABLES Annual Income Less than $10, % 5.2% 4.0% 4.5% 3.6% 3.3% 3.6% 4.5% $10,000-$15, % 4.0% 3.2% 3.3% 3.2% 3.1% 2.9% 3.7% $15,000-$20, % 5.8% 4.9% 3.8% 5.5% 4.8% 4.6% 5.0% $20,000-$25, % 8.5% 6.3% 4.4% 7.6% 6.9% 6.1% 5.9% $25,000-$35, % 15.0% 10.0% 7.6% 12.9% 11.5% 10.2% 9.1% $35,000-$50, % 15.1% 13.3% 10.3% 17.3% 17.7% 15.4% 13.4% $50,000-$75, % 15.8% 14.8% 12.7% 16.5% 18.1% 17.3% 16.6% $75,000 or more 11.8% 20.3% 34.8% 42.8% 15.1% 22.6% 28.1% 30.3% Unknown 4.1% 10.3% 8.7% 10.5% 18.4% 12.0% 11.9% 11.6% Employment status Employed 72.1% 69.9% 66.9% 66.1% 71.6% 70.0% 64.9% 63.1% Student 6.1% 6.5% 6.2% 4.4% 3.4% 3.0% 3.5% 2.5% Not working 24.6% 22.8% 22.2% 24.2% 21.9% 22.7% 26.1% 28.4%