Introduction: White blood cell count (WBC) and temperature thresholds used in the SIRS criteria (WBC 12 ×109/L; temperature 38°C) have long been flagged as abnormal in electronic health records (EHRs), often triggering clinical actions such as ordering blood cultures or antibiotics. These thresholds were originally derived from expert consensus in the early 1990s and continue to play a prominent role in EHR identification of potential sepsis cases. We hypothesized that revising WBC and temperature thresholds using real-world data may improve their ability to detect sepsis. Methods: We performed a retrospective cohort study of adult ED presentations to Emory Healthcare hospitals (2018–2019). Patients with at least one WBC and temperature recorded in the first 24 hours of admission were included. Sepsis cases were identified as patients meeting Adult Sepsis Event (ASE) criteria during their hospitalization. We evaluated the original thresholds for WBC and temperature as stand-alone variables using sensitivity, specificity, and AUC for capturing sepsis cases. We then evaluated alternative thresholds using an automated approach to identify revised thresholds that optimized performance of sepsis identification. Results: Among 120,234 ED presentations, 6,000 (4.99%) met ASE criteria. Original WBC thresholds had 0.59 sensitivity and 0.69 specificity (AUC 0.64), while revised WBC (11.5), had sensitivity increased to 0.65 but specificity to 0.59, slightly reducing AUC (0.62). Original temperature thresholds had 0.53 sensitivity and 0.84 specificity (AUC 0.68) while revised thresholds, (37.5°C), improved sensitivity to 0.58 with a minor loss in specificity to 0.82 and improved AUC to 0.70. Combining revised WBC and temperature thresholds improved AUC significantly from 0.69 to 0.71 (p = < 0.0001.) Conclusions: Traditional WBC and temperature thresholds, based on SIRS criteria and flagged as abnormal in EHRs, offer limited ability for sepsis detection. Modest gains in sensitivity and AUC can be achieved through revised thresholds, particularly with temperature. Updating EHR alert thresholds may optimize early sepsis detection and minimize unnecessary interventions.
Hunt et al. (Sun,) studied this question.