Geographic Clusters in Sepsis Mortality and the Role of Targeted Regionalization: A Simulation Study

Introduction: Sepsis outcomes are variable between hospitals, and some have proposed using this variability to improve aggregate sepsis outcomes. The objective of this study was to develop geospatial sepsis clusters using Medicare data, identify sources of variation between geospatial clusters, and test the hypothesis that redistributing sepsis patients from low-performing hospitals to higher-performing hospitals within a cluster will improve sepsis outcomes.


Methods: We conducted a cohort simulation study using Medicare administrative claims data from 2013 to 2015. We developed geospatial clusters based on the similarity index, then we calculated risk-standardized mortality for hospitals and for clusters. Finally, we used a simulation to model the effect of reallocating sepsis patients to higher-performing hospitals within the same cluster using 4 different methods, and we compared the effect with local performance improvement.


Results: We included 1,125,308 patients from 2,668 hospitals, and they were grouped into 222 regional clusters. High-performing clusters were located largely in the Midwest, and they tended to be in less urban regions with fewer teaching hospitals. In our simulation, the most impactful regionalization strategy was reassigning cases from the worst hospital in a cluster to the best hospital in the cluster, which was predicted to prevent 1,705 deaths (95% CI 1,702-1,708) per year in the U.S. with 5% reassignment (reducing aggregate mortality by 0.09%, 95% CI 0.091–0.093%). This finding was primarily related to the observation that hospitals in a cluster had similar risk-standardized mortality, suggesting that the impact of finding suitable alternative facilities within a cluster was limited. This aggregate benefit was lower than the 5,702 deaths predicted from reducing absolute mortality by 1% in hospitals in the lower half of the performance distribution.


Conclusions: Geospatial clusters provide insight into regional approaches to system-based acute care. In a simulation study, targeted sepsis regionalization appears less effective than local performance improvement in reducing preventable sepsis deaths.