Multicenter Prospective Validation of an Updated Proprietary Sepsis Prediction Model
Authors' conclusion
In this prognostic study of a new sepsis prediction model, a multicenter prospective validation performed across 4 major US health systems found improved discrimination for the early prediction of sepsis but noted high institutional variability, low positive predictive value, and high alert burden.
Study context
- Population
- Adult inpatient encounters at 4 large US health systems, August 31, 2023 to March 11, 2025, after Epic Sepsis Model v2 implementation
- Participants
- 227,091 inpatient encounters
- Comparison
- Sepsis vs No sepsis
- Sepsisn = 7,401
- No sepsisn = 219,690reference
- Outcome
- Model discrimination (AUROC) for sepsis (Sepsis-3), encounter level and prediction level (4-hour, 12-hour, hospitalization-wide horizons)
Methodological detailsmulticentre prospective validation, external validation, Data collected for a consecutive period immediately following new model implementation at each site
- Multicentre prospective validation
- External validation
- Data collected for a consecutive period immediately following new model implementation at each site
Performance
As reported in the abstract
Encounter-level AUROC across sites
Encounter-level AUROC, lowest site
0.82
95% CI 0.81–0.83
Encounter-level AUROC, highest site
0.92
95% CI 0.92–0.93
Further findings
Prediction-level AUROC, 12-hour horizon, across sites · 12-hour time horizon
- Lowest site
- 0.75 (95% CI, 0.74-0.75)
- Highest site
- 0.85 (95% CI, 0.85-0.85)
Encounter-level AUROC vs clinician recognition of sepsis, across sites
- Minor decrease in performance; lowest site
- 0.80 (95% CI, 0.79-0.81)
- Highest site
- 0.90 (95% CI, 0.89-0.90)
PPV across sites
- Lowest site
- 0.13 (95% CI, 0.13-0.14)
- Highest site
- 0.26 (95% CI, 0.25-0.27)
- High number needed to evaluate and high alert burden
Original publication
Multicenter Prospective Validation of an Updated Proprietary Sepsis Prediction Model. JAMA Netw Open. Published 2 February 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →