Abstract 200: Is Risk-standardized In-hospital Stroke Mortality an Adequate Proxy for Risk-standardized 30-day Stroke Mortality Data? Findings From the GWTG-Stroke Program

Abstract

Background: Hospital profiling is typically undertaken using risk standardized 30-day mortality but obtaining this data is complicated as it requires tracking patients post discharge. We undertook this analysis to determine if risk standardized in-hospital mortality could serve as an adequate proxy for risk standardized 30-day mortality data for the purposes of identifying outlier hospitals with high or low ischemic stroke mortality. Methods: Acute ischemic stroke cases entered into Get With The Guidelines (GWTG)-Stroke between 2003-2013 were linked to fee-for-service Medicare files. Hospitals with fewer than 25 cases were excluded. Risk-standardized mortality rates (RSMR) for both in-hospital and 30-day mortality were generated using a previously developed risk model, and the proportion of hospitals classified as statistical outliers (based on random effects models) were compared between the two. To assess the impact of stroke severity we conducted a sensitivity analysis limiting cases to those with NIHSS data. Results: A total of 535,332 ischemic stroke patients from 1494 GWTG-Stroke hospitals were included; mean age was 80 years, 59% were female, 19% were non-white, NIHSS (mean = 7.7) was recorded in 58%. Average in-hospital and 30-day mortality was 5.7% and 14.5%, respectively. At the hospital level, mean in-hospital RSMR and 30-day RSMR was 6.0% (SD 1.9) and 14.6% (SD 2.0), respectively, but the correlation between the two was only modest (r = 0.53). Overall agreement between the designation of outlier hospitals based on in-hospital and 30-day RSMRs was 78% (Table), but after correcting for chance agreement, concordance was quite modest (Kappa = 0.29). Findings based on models that included NIHSS were similar. Conclusions: When used to identify outlier hospitals with high or low ischemic stroke mortality the agreement between in-hospital mortality and 30-day mortality was only modest. Using in-hospital mortality data proved to be a poor proxy for 30-day mortality.