Accuracy of real-time sepsis-3 surveillance across ten inpatient facilities - a prospective cohort study.

Objective

To evaluate portability and diagnostic performance of an automated sepsis surveillance algorithm applying objective Sepsis-3 clinical criteria across an entire adult population.

Methods

In this prospective cohort study, a Sepsis-3 algorithm developed in one region in Sweden, was implemented across ten inpatient facilities in another region, covering all adult admissions. Classifications were compared against blinded expert chart review (reference standard) in a random sample. Diagnostic accuracy, incidence, bloodstream infection (BSI), and 30-day mortality were assessed.

Results

Between July 2022 and December 2024, 82,112 admissions (49,574 adults) were included. The algorithm identified 5,101 sepsis events (59.6% occurred in men). Most were community-onset (CO; 84.0%), whereas hospital-onset (HO) sepsis carried higher 30-day mortality (17.6% vs 12.1%). In validation (700 patients; 1087 admissions), sensitivity was 90.5% (95% CI 80.4-96.4) and specificity 98.4% (95% CI 97.5-99.1). Sepsis incidence was 850/100,000 population/year (95% CI 662-1,084). BSI did not increase mortality in CO sepsis, whereas HO sepsis was associated with increased hazards irrespective of BSI (with BSI HR 2.36, 95% CI 1.64-3.40; without BSI HR 2.20, 95% CI 1.83-2.63).

Conclusion

Objective automated Sepsis-3 surveillance is scalable and transferable, representing a critical next step in infection surveillance to improve sepsis care quality.