Evaluation of a real time machine learning sepsis risk algorithm for Emergency Department waiting rooms (SAFE-WAIT)

ObjectivesTo evaluate and compare the real-time Sepsis risk Artificial intelligence algorithm For Emergency department WAITing room (SAFE-WAIT) model with the standard SIRS-based sepsis alert in recognizing and managing sepsis. Study DesignA retrospective analysis of the AI algorithm predicting sepsis risk using an ongoing emergency department sepsis archive. Setting and ParticipantsAdults presenting to a metropolitan emergency department in Western Sydney between July 2022 and June 2024 who received either a SAFE-WAIT risk category and/or the standard sepsis alert. Main Outcome MeasuresThe primary outcomes: recognition of the development of sepsis and septic shock. Secondary outcomes: the time to physician review and initial antibiotic administration. ResultsAmong 108,401 patients analysed, 104,904 received a SAFE-WAIT risk category. Of these, 5,149 (4.9%) had a confirmed sepsis diagnosis, and 312 (3%) developed septic shock. SAFE-WAIT categorized 270 (86.5%) of septic shock patients as high risk at triage, with event onset at 92 minutes (Inter Quartile Range: 17.25-233.00). Adjusted {beta}-coefficients showed significantly faster antibiotic administration in moderate and high-risk SAFE-WAIT groups (Moderate: -58.81 minutes, 95% Confidence Interval (CI): -72.26 to -35.80, p < ; High: -116.99 minutes, 95% CI: -135.34 to -98.65, p < 0.001). High-risk patients had a slightly shorter time to first physician review ({beta} = -3.5 minutes, 95% CI: -5.41 to -1.58, p < 0.001). These effects of SAFE WAIT grouping on the time to antibiotics and clinician review were both modified by triage category (p for interaction < 0.001). ConclusionsSAFE-WAIT effectively predicts sepsis-related adverse events at triage, positively impacting sepsis management. These findings underscore the potential role of AI-augmented clinical practice.