Source record
External validation of a widely deployed sepsis model
The source is Wong and colleagues, JAMA Internal Medicine, 21 June 2021, an external validation of the Epic Sepsis Model at Michigan Medicine. Across 38,455 hospitalizations the hospitalization-level AUC was 0.63 (95% CI 0.62–0.64). At a score of 6 or higher the model alerted on 18% of hospitalizations, did not identify 67% of sepsis cases, and found only 7% of sepsis patients who had not already received timely antibiotics. The paper is a caution about local validation of a widely deployed proprietary score.
- Source date
- Jun 21, 2021 · Year 2021
- Evidence label
- Published
- Source type
- Paper
Citation
Wong et al., JAMA Internal Medicine
Wong et al. External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients. JAMA Internal Medicine, 21 June 2021. https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2781307
Key public facts
- The study covers 38,455 adult hospitalizations at Michigan Medicine. The score was recalculated through the stay.
- Sepsis was defined from surveillance criteria and the record, not from the model’s own label.
- The hospitalization-level AUC was 0.63 (95% CI 0.62–0.64), below the discrimination the developer had reported.
- At a score of 6 or higher the model alerted on 18% of hospitalizations and did not identify 67% of sepsis cases.
- It found only 7% of sepsis patients who had not already received timely antibiotics.
Directory entry for education and market scanning. Not medical advice and not an endorsement.