Abstract Session F1: Health Policy/Advocacy/Social Justice - - PDF document

Abstract Session F1: Health Policy/Advocacy/Social Justice

Moderator: LeChauncy Woodard, MD, MPH, FACP THE ROLES OF COST AND QUALITY INFORMATION IN MEDICARE ADVANTAGE PLAN ENROLLMENT DECISIONS Rachel O. Reid1; Partha Deb4; Benjamin L. Howell2; William Shrank3. 1Brigham and Women's Hospital, Boston, MA; 2Center for Medicare & Medicaid Services, Baltimore, MD; 3Brigham and Women's Hospital, Boston, MA; 4Hunter College, New York, NY. (Tracking ID #1933518) BACKGROUND: To facilitate informed decision making in the Medicare Advantage marketplace, the Centers for Medicare & Medicaid Services publishes information about Medicare Advantage plans via the Medicare Plan Finder website, including costs, benefits, and quality ratings provided on a 1-to-5 star scale. Little is known about how beneficiaries weigh costs versus quality when making enrollment choices. In this study, we assess the variation in Medicare Advantage enrollment attributable to plan attributes and willingness to pay for quality. METHODS: We conducted a nationwide, beneficiary-level cross-sectional analysis of the 2011 Medicare Advantage and Prescription Drug (MAPD) plan choices of beneficiaries enrolling in Medicare Advantage for the first time ever in 2011 who were not eligible for the low-income subsidy. Matching beneficiaries with their choice-sets of MAPD plans by county, we used conditional logistic regression to estimate associations between plan attributes and enrollment, to assess both the proportion of explained enrollment variation attributable to plan attributes and willingness to pay for quality. The model accounted for 5-star quality ratings, costs (premiums and average estimated out-of-pocket costs), benefits (plan structure; deductibles; coinsurance; hearing, vision, dental benefits; and prescription gap coverage), and lagged county-level sponsor organization (i.e., brand) market share. Because willingness to pay for quality may vary at different rating levels, the model included both the 5-star quality rating itself and its quadratic transform. We assessed differential willingness to pay by beneficiary characteristics (age, sex, race, and urban versus rural residence) by interacting these covariates with quality and cost covariates. RESULTS: The study cohort included 847,069 first-time Medicare Advantage enrollees who selected an eligible MAPD plan in 2011. Relative to the total variation explained by the model, market share accounted for 35.3% of variation in plan choice, premiums for 25.7%, estimated out-of-pocket costs for 11.6%, and 5-star quality ratings for 13.6%. Mean cumulative willingness to pay for a plan in total annual combined premiums and out-of-pocket costs varied from $4,154.93 for a 2.5-star plan to $5,698.66 for a 5-star plan. Increases in willingness to pay diminished at higher 5-star quality ratings: $549.27 (95% CI $541.10 to $557.44) to go from a 2.5-star plan to a 3-star plan and $68.22 (95% CI $61.44 to $75.01) to go from a 4.5-star plan to a 5-star plan. Beneficiaries aged 64-65 years were more willing to pay for plans with higher quality ratings than other age groups; black and rural beneficiaries were less willing to pay for plans with higher quality ratings. CONCLUSIONS: Medicare Advantage enrollees prefer plans with higher quality ratings and lower costs; however, market share's contribution to plan choice suggests that word-of-mouth and brand recognition are also

• influential. Key subgroups' differential willingness to pay for quality and market share' influence argue for

continued efforts to advance communication of plan attributes to improve marketplace efficiency. If increased enrollment in plans with the highest quality ratings is a goal, the diminishing marginal utility for quality

• bserved supports policy interventions to make achievement of the highest ratings desirable for insurers and

enrollment in the highest-rated plans attractive and accessible to beneficiaries.

NEW YORK CITY GREEN CARTS: IS THE PROGRAM ALLEVIATING FOOD DESERTS? Kathleen Y. Li1,2; Ashley Fox2; Ellen K. Cromley3,4; Carol Horowitz2. 1University of California, San Francisco School of Medicine, San Francisco, CA; 2Icahn School of Medicine at Mount Sinai, New York, NY; 3University of Connecticut School of Medicine, Farmington, CT; 4Lund University, Lund, Sweden. (Tracking ID #1936558) BACKGROUND: As the proportion of overweight and obesity nears 70% in the United States, local public health departments are using their authority to implement programs to tackle this growing epidemic. Limited access to fresh fruits and vegetables in low- income neighborhoods is believed to contribute to high rates of obesity. To address the high rates of obesity and related illnesses in its low-income neighborhoods, in 2008, New York City established a fruit and vegetable street vendor program (NYC Green Carts) to promote healthier eating in neighborhoods with the lowest reported rates of fruit and vegetable consumption. Carts are free to move anywhere within designated neighborhoods, which can be large and often border wealthier neighborhoods. We aimed to study whether carts locate in areas that enable them to reach the low-income "food desert" populations they were designed to serve. METHODS: We obtained a list of Green Cart locations from the New York City Department of Health and information on census tract level demographic and food environment data from the census bureau and Esri Business Analyst Desktop. We identified "healthy" food stores, namely supermarkets, independent grocers, and fruit and vegetable specialty stores according to North American Industry Classification System codes as well as bodegas with evidence of selling fresh produce on Google Maps Street

• View. We then defined a food desert as a lack of healthy food stores within a ¼ mile. Using ArcGIS software, we mapped the existing

Green Carts and generated a list of potential Green Cart locations. Within designated Green Cart areas, the intersection closest to the geographic center of each census tract without a Green Cart was coded as a candidate site. We then analyzed census tract characteristics for actual and candidate Green Carts to determine how they differed with regards to population, income, percent below the poverty level, distance to subway stops, the number of large employers, and the number of healthy food stores at both a ¼-mile and ½-mile distance from each location. RESULTS: Our team identified 265 Green Carts and 644 candidate locations without Carts. As compared with potential Green Carts sites (see table), Green Carts were significantly more likely to be within a quarter of a mile of supermarkets, grocery and fruit & vegetable stores. Over 1/3 of candidate intersections with no Green Carts were situated in food deserts compared to fewer than 1/10 of Green Carts. Green Carts were positioned in tracts with significantly larger population sizes, tended to be closer to a subway stop, and were more likely to be within a ¼ mile of large employers, though Green Cart tracts had a lower median family income compared to tracts with candidate sites. Some (22) Green Carts were located outside the officially designated Green Cart areas, and these census tracts had significantly higher income and higher access to fruits and vegetables than those within Green Cart boundaries. CONCLUSIONS: Compared with potential Green Cart locations, census tracts with Carts tended to be located in areas with large numbers of potential customers, namely population centers and areas with more pedestrian traffic (close to subway stops and large businesses), perhaps to increase market share and profitability. However, Green Carts were rarely situated in food deserts. This suggests that many already underserved neighborhoods are still not reached by this initiative. A market-driven imperative to locate near larger numbers of potential customers may be in tension with the objective of the Green Carts program to increase access to fresh fruits and vegetables in regions with lower access to supermarkets and fresh produce. However, we also identified a number of candidate locations where foot traffic is likely to be high but where Green Carts were not located. Since people usually shop close to where they live, the Green Carts program should consider introducing added incentives for Green Cart vendors to locate in high need census tracts to ensure adequate coverage of actual food deserts.

• Table. Average Characteristics of Green Cart vs Candidate Sites

Variable Green Cart Site Candidate Site p value Tract population 5,004 3,706 <0.00001 Tract median income $37,213 $42,740 0.000284 Percent of tract below poverty level 30.1% 25.6% <0.00001 Distance to nearest subway stop (ft) 793 3,079 <0.00001 # large employers within ¼ mi 0.457 0.085 <0.00001 # supermarkets within ¼ mi 1.6 0.81 <0.00001 # other stores carrying fruits and vegetables within ¼ mi 2.09 0.84 <0.00001 # large employers within ½ mi 1.185 0.402 <0.00001 # supermarkets within ½ mi 4.72 2.86 <0.00001 # other stores carrying fruits and vegetables within ½ mi 5.4 3.1 <0.00001

EVALUATING INPATIENT HOSPITAL CHARGE VARIABILITY USING LOCAL HEALTH AND MARKET FACTORS James D. Park1; Edward Kim2; Rachel M. Werner3. 1Rutgers, Robert Wood Johnson Medical School, New Brunswick, NJ; 2The College of New Jersey, Ewing, NJ; 3Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA. (Tracking ID #1937089)

BACKGROUND: In May 2013, the Center for Medicare and Medicaid Services released previously undisclosed information about the price of hospital care. Within a diagnosis related group, inpatient hospital charges to Medicare demonstrated a wide range of charges across US hospitals. This variation in hospital charges remains unexplained, and this study examines its relationship to the local market environment in which hospitals operate. METHODS: This descriptive study evaluated the CMS Inpatient Provider Charge data of over 3000 US hospitals using clustered, multivariate linear regression. This analytic approach examined how county health, socioeconomic, and market factors relate to hospital charges. Six common inpatient conditions - cellulitis, chronic obstructive pulmonary disease, congestive heart failure, myocardial infarction, orthopedic surgeries, and pneumonia - were chosen for the analysis. Related DRGs for each condition were grouped together, and a weighted average charge for each hospital was calculated based on the proportion of discharges per DRG. Based on a conceptual model, a common set of covariates was used to create a single regression model for each condition. The analysis used various county-level measures of health status (years of potential life lost per 100,000 population, percent of adults who report fair or poor health, percent days reported as mentally unhealthy per month, diabetes prevalence, obesity prevalence); health behavior (prevalence of smoking, no leisure time activity, or heavy alcohol drinking); clinical access and quality (prevalence of uninsured status, preventable hospital stays rate, percent of diabetics who received a hemoglobin A1c test, number of primary care physician per 100,000 population); socioeconomics status (percent of a 9th grade cohort that graduates in four years, prevalence of unemployment, of children living in poverty, and of children living in single-parent homes, median household income, violent crime rate per 100,000 population); the built environment (number of accessible recreational facilities per 100,000 population); demographics (percent of the population that is African American, Asian, Latino, or other race, and percent

• f the population over the age of 65); and the hospital market (for-profit status, proportion of hospital discharges within

the county) from every US county and hospital. All analyses controlled for other market factors such as the hospital wage index, disproportionate share hospital index, hospital cost-to-charge ratio, and the county-level Herfindahl-Hirschman

• Index. Covariates were extracted from multiple datasets including the US Census Bureau, Behavioral Risk Factor

Surveillance System, National Center for Chronic Disease Prevention and Health Promotion, CMS Impact, CMS Provider

• f Service, County Business Patterns, American Community Survey, Dartmouth Atlas of Health Care, Bureau of Labor

Statistics, Small Area Income and Poverty Estimates, Federal Bureau of Investigation Uniform Crime Reporting, National Center for Health Statistics, and the National Center for Educational Statistics. All regression analyses were clustered by county, and statistical testing was deemed significant with p-values < 0.05. RESULTS: The mean number of hospitals included across six regressions was 2603. Across all six conditions, hospital charges were associated with uninsured status and resulted in $210.28 - $623.26 higher charges for every one percent increase in the prevalence of uninsured status (p<0.01). In addition, hospital charges were associated with for-profit hospital status across all six conditions and resulted in $1230.47 - $7772.10 higher charges than not-for-profit status (p<0.05). In all conditions except myocardial infarction, hospital charges were associated with unemployment status and resulted in $295.83 - $752.53 higher charges for every one percent increase in unemployment prevalence (p<0.05). Health behavioral factors such as smoking, physical inactivity, and excessive alcohol consumption, or health status factors such as premature death, fair or poor health status, poor mental health days, or diabetes were largely unassociated with hospital

• charges. Of all health factors utilized in this study, only obesity was repeatedly associated with hospital charges in four of

the six conditions, resulting in $232.57 - $513.92 higher charges for every one percent increase in obesity prevalence (p<0.05). CONCLUSIONS: Non-health related factors, such as uninsured, unemployment, and for-profit hospital status, were consistently associated with higher hospital charges. In contrast, nearly all health factors had no association to inpatient hospital charges. This research raises concerns about a regressive pricing strategy that is indifferent to health status but charges vulnerable and uninsured populations higher prices.

STATE FIREARM LEGISLATION AND NONFATAL FIREARM INJURIES Joseph Simonetti1,2; Ali Rowhani-Rahbar3,5; Douglas Zatzick3,6; Bessie Young7,2; Frederick Rivara3,4.

1University of Washington, Seattle, WA; 2School of Public Health, University of Washington, Seattle, WA; 3University of Washington, Seattle, WA; 4University of Washington, Seattle, WA; 5School of Public Health,

University of Washington, Seattle, WA; 6University of Washington, Seattle, WA; 7Northwest Center of Excellence, VA Puget Sound, Seattle, WA. (Tracking ID #1938977) BACKGROUND: Stricter state firearm legislation has been shown to be associated with lower rates of fatal firearm injuries. However, less is known regarding nonfatal firearm injuries (NFI), which are more common than fatal injuries, and differ in several ways, including age of the injured, intent, and type of firearm involved. The aim of this study was to determine whether stricter state firearm legislation is also associated with lower rates of NFI. METHODS: We performed a cross-sectional, ecological analysis using available retrospective data from 17 states collected through the 2010 Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases, which include information on more than 95% of all discharges from community hospitals. We defined each case of hospitalized NFI as a patient discharged alive with an External Cause of Injury Code (E-code) for firearm injury (Assault [E965.0-.4], intentional self-inflicted [E955.0-.4], unintentional [E922.0-.3, .8, .9], or undetermined [E985.0-.4]). The primary predictor of interest was the Brady Score, a measure of state firearm legislative strength based on 2009 scorecards created by the Brady Campaign to Prevent Gun Violence. Scores are based on laws related to firearm trafficking, background checks, child safety, and restrictions on military- style assault weapons and firearms in public places. The Brady Score ranges from 0 to 28, with higher scores indicating more stringent laws, which we categorized into quartiles. The primary outcome was total hospitalized

• NFI. Secondary outcomes were assault-related, self-inflicted and unintentional hospitalized NFI. We calculated

age-adjusted total hospitalized NFI rates for each participating state and used Poisson regression with clustered robust sandwich standard error estimates to determine if states in the highest quartile of legislative strength had fewer total, assault-related, self-inflicted and unintentional hospitalized NFI than states in the lowest quartile. Using data from the 2010 U.S. Census, we adjusted each model for the following state characteristics: age distribution, gender composition, racial/ethnic composition, population density, percentage with a college degree, percentage living below the federal poverty line, and unemployment rate. RESULTS: The median Brady Score in 17 HCUP-participating states was 7 (interquartile range 4-16). Rates of total hospitalized NFI ranged from 1.7 (Hawaii) to 12.0 (Maryland) per 100,000 individuals. In unadjusted analyses, compared to states in the lowest quartile of legislative strength, those in the highest quartile of legislative strength had lower rates of self-inflicted (incident rate ratio [IRR] 0.33; 95% confidence interval [CI] 0.27-0.41) and unintentional (IRR 0.53; 95% CI 0.43-0.66) hospitalized NFI. In multivariable models, states in the highest quartile of legislative strength had lower rates of total (IRR 0.42; 95% CI 0.34-0.50), assault-related (IRR 0.44; 95% CI 0.35-0.57), self-inflicted (IRR 0.26; 95% CI 0.22-0.31), and unintentional (IRR 0.27; 95% CI 0.22-0.32) hospitalized NFI. CONCLUSIONS: Findings of this cross-sectional, ecological study indicate that stricter state firearm legislation is associated with lower rates of all types of hospitalized nonfatal firearm injuries.

PAYMENT REFORM IN MASSACHUSETTS: EFFECT OF GLOBAL PAYMENT ON HEALTH CARE SPENDING AND QUALITY 4 YEARS INTO THE ALTERNATIVE QUALITY CONTRACT Zirui Song1,2; Dana G. Safran3,4; Bruce E. Landon1,5; Sherri Rose1; Matthew Day3; Michael E. Chernew1,2.

1Harvard Medical School, Boston, MA; 2National Bureau of Economic Research, Cambridge, MA; 3Blue Cross

Blue Shield of Massachusetts, Boston, MA; 4Tufts University School of Medicine, Boston, MA; 5Beth Israel Deaconess Medical Center, Boston, MA. (Tracking ID #1939467) BACKGROUND: Slowing the growth of health care spending while improving quality of care remains a national priority. Global payment within accountable care organizations has received increasing attention as a potential way to control spending while improving quality. In 2009, Massachusetts began moving away from fee-for-service with the Blue Cross Blue Shield of Massachusetts Alternative Quality Contract (AQC). The AQC pays provider organizations a global budget that covers the entire continuum of care, while rewarding up to 10 percent of the budget in additional quality bonuses. In the ensuing years, membership in the AQC has

• grown. We evaluated the effect of the AQC on spending and quality after 4 years.

METHODS: We used claims and quality data in a difference-in-differences design to assess the association between the AQC and changes in spending and quality. For the 2009 AQC cohort, the pre-intervention period was 2006-2008, while the post-intervention period was 2009-2012. In addition to basic unadjusted analysis, we used propensity-weighted linear regression models at the enrollee level adjusting for demographics, health status, secular trends, and plan benefit design. We conducted a number of sensitivity analyses, consistent with

• ur prior work. We decomposed the AQC effect by year, as well as by type of care, site of care, health status,

risk contracting experience, and price versus utilization effects. RESULTS: The 2009 AQC cohort comprised about 500,000 unique enrollees across the study period whose primary care physicians belonged to organizations that entered the AQC in 2009. Building on prior work demonstrating AQC-associated savings of 1.9% in year-1 and 3.3% in year-2, preliminary analyses show that the AQC continued to slow medical spending in years 3 and 4, on average exceeding 5% in these years. Medical savings were larger than the sum of budget surpluses and quality payments. Accounting for shared savings and quality bonuses, total health care spending growth after 4 years was lower than the targeted state economic growth of 3.6%. Savings were larger among organizations without prior risk contracting experience. Changes in both prices (through referral pattern changes) and volume helped explain the results, which were evident in imaging, tests, and procedures. Achievement of quality in chronic care management improved from 79% in 2007 to 86% in 2012; adult preventive care from 74% to 82%; and pediatric care from 79% to 85%. Analogous results for subsequent cohorts were broadly consistent. CONCLUSIONS: After 4 years, the AQC was associated with a slowing of spending growth and improved

• quality. Combining global budgets with pay-for-performance for accountable care organizations may be a

successful model for slowing spending and improving quality. Global payment may be able to encourage greater price competition without threatening quality. The AQC could be informative for Medicare in both its Shared Savings and Pioneer ACO programs, as well as for other commercial insurers across the country engaged in payment reform.

ARE PATIENT-CENTERED MEDICAL HOME PRACTICES ASSOCIATED WITH BETTER ACCESS TO PRIMARY CARE SERVICES? Jaya Aysola1; Karin V. Rhodes2; Daniel Polsky1. 1University of Pennsylvania, Philadelphia, PA; 2University of Pennsylvania, Philadelphia, PA. (Tracking ID #1940136) BACKGROUND: Patient-centered medical homes (PCMH) have gained prominence as models designed to promote high quality cost effective primary care. An integral component of the PCMH model is providing enhanced access to primary care. However, the potential effect of PCMHs on access to primary care services has not been examined. Our study evaluated whether PCMH practices in ten US states were associated with better access to care, measured using simulated patient (audit) methodology. METHODS: We conducted a cross-sectional analysis of publically available data on practices recognized as National Committee Quality Assurance (NCQA) patient-centered medical homes (PCMH) before June 2013 and data collected through a simulated patient (audit) study. Our primary predictor was whether or not a practice was a NCQA recognized PCMH. Our primary outcome measure was the ability of the caller (simulated patient) to schedule a new appointment. Trained field staff contacted primary care offices in 10 US states between November 2012 and March 2013 posing as individuals seeking a new patient appointment with varying insurance status (private, Medicaid, or uninsured). All eligible primary care offices were identified using a proprietary database of existing physician clinics (SK&A physician file). Callers contacted a stratified random sample of primary care practices treating non-elderly adults in each state, for a total of 10572 requests for appointments at 7266 practices. We merged detailed practice characteristics (number of sites, number of providers, average patient volume, number of exam rooms) from SK&A physician file, and geographic socioeconomic and health care resource information (county-level percent persons below poverty level, percent minority, percent non-elderly uninsured, number of office based primary care providers, and Health Professional Shortage Area designation) from the Area Resource File, into our audit data. We performed multivariable logistic regression model, adjusting for the caller's insurance status and race/ethnicity as well as the practice and geographic characteristics listed above, to estimate the association between PCMH status and

• ur outcome variable. We then fitted a multivariable logistic regression model, including interactions of practice

PCMH status and caller's insurance status, adjusting for the same covariates above. RESULTS: Out of 10572 requests to practices for appointments, 646 (6.1%) were to NCQA recognized

• PCMHs. Significant predictors of scheduling new appointments included the caller's insurance status, number
• f practice exam rooms, percent minority and percent persons in poverty where the practice was located, and

the practice's NCQA PCMH recognition status. There were no significant interactions between practice PCMH status and caller's insurance status. Adjusted analysis revealed that calls placed to PCMH practices were significantly more likely to result in scheduled new appointments then calls placed to non-PCMH practices. (Adjusted OR 1.3; 95% CI 1.0, 1.5; p=0.02). CONCLUSIONS: Our findings suggest that practices recognized as patient centered medical homes, after robust adjustment for individual caller, practice, and geographic variables, were associated with better access to new primary care appointments for non-elderly adults, those most likely to gain insurance under the Affordable Care Act.