1097B: A NOVEL HOSPITAL CAPABILITY MEASURE OUTPERFORMS CASE VOLUME IN EXPLAINING SEPSIS MORTALITY

Abstract

Introduction: Regionalization of sepsis care could improve sepsis outcomes. There are no validated measures of sepsis capability to guide the identification of hospitals that can best serve sepsis patients. We previously empirically derived Capability-Based (CB) scores based on principal component analysis of several hospital characteristics. We compared the performance of CB scores and case volume as system predictors of mortality among adults with sepsis. Methods: We used the 2018 State Inpatient Databases to identify 90051 adult sepsis encounters at 157 non-federal New York hospitals (derivation cohort), and 130,249 sepsis encounters at 220 hospitals in Florida and Massachusetts (validation cohort). We evaluated the correlation of sepsis volume and each CB score with hospital mortality among non-transferred patients, and with outward sepsis transfer proportions. We fitted linear, nested, predictive models to compare the system predictive abilities of CB scores and sepsis volume in relation to hospital mortality. Results: In the derivation cohort, 83963 (93.2%) patients were non-transferred, of which 20230 (24.1%) died. The mean (range) score was 0 (-3 - +5) with higher scores denoting more capable hospitals. Higher scores were weakly and inversely correlated (spearman’s coefficient [r]: - 0.28) with outward sepsis transfer proportions. Higher scores had weak but better positive correlation with hospital mortality (r: 0.33), than sepsis volume (r: 0.24). The capability scores explained markedly more variation in sepsis mortality (R2 = 0.24, P <.001) than sepsis volume (0.09, P = .01). Predictive mortality models including the capability scores outperformed sepsis volume model (Derivation AUC: 0.53 vs. 0.50; Validation: 0.54 vs. 0.50). A clinical prediction model based on patient characteristics (AUC: 0.76) was not improved by adding sepsis volume or the CB scores (AUC: 0.76). Conclusions: Capability-based hospital scores account for three times more variation in sepsis mortality than sepsis volume and outperform sepsis volume as a system predictor of mortality. With further refinement and validation, these scores may find utility for improving system-based approaches to sepsis care.