Venue: The Fuqua School of Business, Duke University, 1 Towerview Drive, Durham, NC 27708-0120
Presentation
Understanding Geographic Disparities in Ill-Defined Stroke Diagnoses Among Fee-For-Service Medicare Beneficiaries
Background: Stroke is the third leading cause of death and a leading cause of long term disability, costing about $62.7 billion in 2007 (American Heart Association, 2007). Two stroke types exist: ischemic (about 66% of cases, occurs when an artery that supplies blood and oxygen to the brain is blocked) and hemorrhagic (about 10% of cases, occurs when an artery in the brain bursts). Ill-defined stroke (about 20% of cases) occurs when no stroke type is assigned upon discharge from the hospital, which is problematic because thrombolytic drugs are recommended for ischemic but contraindicated for hemorrhagic stroke.
Purpose of this study: Our study focuses on documenting the geographic disparities in ill-defined stroke diagnoses over time and examining the contributions of hospital and community factors, including both economic and social resources, to the observed variation in proportions of ill-defined stroke across time and space.
Methods: We obtained hospital claims data for fee-for-service (FFS) Medicare beneficiaries aged 65 years or older for each year during 1996?2004 from the Centers for Medicare & Medicaid Services (CMS) Medicare Provider Analysis and Review (MEDPAR) file. Data were collected for beneficiaries from all 50 states. Ill-defined stroke was defined as principal diagnosis with ICD-9 CM codes 436-437. We employ spatial regression in a small-area analysis of ill-defined stroke proportions by county, and estimate three equations for three time periods (1996-1998, 1999-2001, and 2002-2004). We use a seemingly unrelated regressions setup to improve efficiency and allow statistical assessment of changes in coefficient parameter estimates over time (Mobley, 2003). Our goal is to explain small-area variation with contextual and demographic factors, and then identify geographic locations with simultaneously high levels of the three predictors as ?worst case scenarios?. Finally, we examine residuals for spatial clustering in the larger positive values, to identify geographic locations where the model does the worst job in explaining the ill-defined stroke proportions.
Preliminary Results: In preliminary descriptive analysis using the LISA test (Anselin 1995) for 1996-1998, we found statistically significant clusters of counties with high proportions of ill-defined stroke concentrated in Oklahoma, North and South Dakota, Minnesota, and Utah (Map 1). Nationally, proportions of ill-defined stroke were higher for women and African Americans, but the observed small-area patterns were consistent across all gender and age groups within areas, suggesting that unexplored place-specific factors are perhaps quite important. We also found that statistically significant county clusters in 1996-1998 persisted in 1999-2001 (Map 2), suggesting persistent problems in specific geographic areas.
Conclusions: The findings will be a useful starting point for designing programs and polices at the local and state levels that are geared towards reducing the proportions of stroke events generally categorized as ?ill defined? and thereby improving treatment for stroke patients.