Operational Integration and Temporal Validation of a Continuously Deployed ICU Prediction Model
Authors' conclusion
A continuously deployed, EMR-integrated ICU prediction system achieved strong temporal discrimination and generally good calibration.
Study context
- Population
- All ICU admissions, tertiary mixed medical-surgical ICU in Japan, August 2017 to March 2025; excluded age <16 years or ICU stay <4 hours
- Participants
- 11,176 vs 1,127
- Comparison
- Development vs Temporal validation
- Developmentn = 11,176
- Temporal validationn = 1,127
- Outcome
- Discrimination and calibration of hourly EMR-integrated BEST-AI predictions for 6 ICU outcomes (in-hospital mortality, ICU mortality, ICU discharge ≤ 72 hr, intubation ≤ 72 hr, extubation ≤ 72 hr, tracheostomy at ICU discharge)
Methodological detailssingle-centre, hybrid study with stepwise clinical deployment, forward-in-time temporal validation
- Single-centre
- Hybrid study with stepwise clinical deployment
- Forward-in-time temporal validation
Performance
As reported in the abstract
Temporal validation, 6 prediction tasks
| Test or model | AUC (ROC), range | Area under precision-recall curve, range |
|---|---|---|
| Temporal validation | 0.856 to 0.960 | 0.302 to 0.786 |
Temporal validation
AUC (ROC), range0.856 to 0.960
Area under precision-recall curve, range0.302 to 0.786
Further findings
24-hour landmark sensitivity analysis (one prediction per patient at 24 hr after ICU admission)
- Discrimination and calibration preserved relative to main analysis
Workflow integration
- System maintained with automated hourly updates and electronic medical record (EMR)-embedded patient- and unit-level visualizations, without prescriptive alerts
Original publication
Operational Integration and Temporal Validation of a Continuously Deployed ICU Prediction Model. Crit Care Med. Published 3 June 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →