Case control evaluation of a novel late-onset neonatal sepsis screening tool

Sepsis screening tools identify at-risk hospitalized patients and are associated with decreased mortality. The ideal sepsis screening algorithm for neonatal intensive care unit (NICU) populations remains uncertain. We previously developed a paper-based NICU sepsis screening tool (NSST) for use at a level IV referral NICU which is intermittently scored during care times. EHR integration could alter the NSST’s performance if the tool is scored continuously. To evaluate the performance of an EHR based NSST that is scored continuously using all available EHR data points. We conducted a retrospective case-control, single-center study using EHR data from a level IV NICU. We identified infants’ first episodes of blood culture positive late-onset sepsis in 2021 as cases and matched them with up to two controls. The NSST algorithm was retrospectively applied to the cases’ data from the 24 h before positive blood culture collection and the controls’ data from the corresponding days of life. We matched 31 cases to 60 controls. Baseline characteristics were similar. A positive NSST occurred in 25/31 (80.6%) cases and 13/60 (21.7%) controls. Test sensitivity was 0.81, specificity was 0.80, and the odds ratio of a positive blood culture after a positive NSST was 16.6. Heart rate was the most common contributing variable followed by respiratory status and urine output. Abnormal values for capillary refill time were only noted for cases despite routine documentation in all infants. Among cases of culture positive late-onset sepsis, alerts occurred 0 – 20 h prior to collection of blood cultures. Our case-control study suggests an EHR-based NSST algorithm may offer clinical utility for early sepsis detection in a level IV NICU setting. Findings are limited by small sample size and retrospective, single-center nature which may limit precision and generalizability. Larger, multi-center studies are needed to validate performance and further evaluate clinical workflow impact.