Abstract Rationale Thresholds for white blood cell (WBC) count (4 - 12 ×109/L) and temperature (36 - 38 °C), established by expert consensus in the early 1990s, determine both clinical suspicion and electronic health record (EHR) based alerts for sepsis. However, the adequacy of these thresholds has never been systematically evaluated in large, real-world, modern cohorts. We therefore derived empirically optimized WBC and temperature cutoffs for sepsis detection across multiple U.S healthcare systems using a standardized, reproducible analytic framework. Methods We conducted a retrospective multicenter cohort study of adult hospitalizations at five healthcare systems (27 hospitals) in the Common Longitudinal ICU data Format (CLIF) consortium. Eligible hospitalizations included adults admitted to inpatient wards or intensive care units with at least one documented WBC and temperature within 24 hours of admission. Each site applied a standardized analytic pipeline to local CLIF registries to identify hospitalizations meeting Adult Sepsis Event (ASE) criteria (defined by obtaining blood cultures and administering antibiotics, plus end-organ dysfunction) using local EHR data. We delineated revised thresholds using three data-driven approaches: (1) Youden Index - maximizing sensitivity and specificity (2) Stepwise Odds Ratio - where odds ratio of meeting ASE criteria ≥ 3 and (3) ASE Prevalence by Interval - thresholds above and below which ASE prevalence ≥ 10%. Results Across 472,006 adult hospitalizations, 27,841 (5.9%) met ASE criteria (site range 2.9-8.1%). Empirically derived thresholds differed from conventional “normal” reference ranges and are shown for one of the CLIF institutions in Figure 1. The data-driven derived thresholds were as follows: For WBC (x109/L) the site ranges of lower bounds by the three approaches were 3.1 - 3.3, 2.4 - 3.9, 1.1 - 3.1, and the site ranges of upper bounds were 11.2 - 12.4, 8.6 - 13.5, 13.4 - 15.8, respectively. For temperature (°C) the site ranges of lower bounds by the three approaches were 35.9 - 36.4, 35.6 - 36.2, 35.5 - 36.3, and the site ranges of upper bounds were 37.0 - 37.2, 37.0 - 37.2, 37.2 - 38.3, respectively. Although numeric values varied by site, directional trends were largely concordant. Conclusion Across 27 hospitals in five healthcare systems, revised WBC bounds and upper temperature bounds were both consistently below non-data-driven, traditional definitions. Reliance on long-standing “normal” cutoffs may delay recognition of early infection, while data-driven recalibration of sepsis screening parameters could enhance timely identification and intervention in sepsis. This abstract is funded by: None
Hunt et al. (Fri,) studied this question.