Abstract P255: Should Models of Hospital Outcomes Include Race as a Covariate?

Abstract

Objective: There remains controversy about whether models of hospital performance should account for patient race. We used simulated data to explore the effects on hospital rankings of including or excluding race as a covariate in risk-standardized hospital outcome models, using hospitalizations for heart failure as a case study. Methods We simulated three scenarios by which patient race might affect heart failure hospital outcome: a) a treatment bias simulation in which non-white patients were 20% less likely to receive optimal treatment regardless of hospital quality performance; b) an allocative bias simulation in which non-white patients systematically received care from lower performing hospitals that lower quality care to all patients uniformly; and c) a survival bias simulation in which nonwhite patients were 10% less likely to survive than white patients regardless of hospital quality performance. We evaluated the concordance in estimated hospital rank between models that did and did not include race for a simulation of 100,000 patients hospitalized at 1,000 hospitals. We also present the extent to which each model over- or under-predicted hospital quality for hospitals that treat a high percentage of nonwhite patients. Results When allocation or treatment bias scenarios were simulated, the model results were highly consistent (kappa>0.9) regardless of whether or not patient race was included in risk-standardization models; models were most disparate for the survival bias scenario (kappa =0.689). In both the allocative bias and the treatment bias scenarios, models that include race overestimated the quality of hospital care at hospitals that treat a higher percentage of nonwhite patients (beta =91.9 and 78.9, respectively; p<0.001) while models that excluded race did not (beta=31.5; p=0.184, and 2.5; p=0.916, respectively). In the survival disparity scenario, the model that included race performed well (beta=-36.7 p=0.15), whereas the model that excluded race significantly underestimated quality at highly nonwhite hospitals (beta= -326.6; p<0.001). Conclusion The impact of including race in risk standardization models of hospital performance depends on causal pathways by which race impacts clinical outcomes.