Pediatric anesthesia & critical care
Machine Learning for Predicting Critical Events Among Hospitalized Children
Authors' conclusion
This retrospective cohort study describes the development of a novel hospitalwide model for continuously predicting the risk of critical events through the entirety of a child's stay.
Study context
- Population
- Pediatric (age <18 years) admissions to inpatient and ICUs at 3 tertiary care academic hospitals
- Participants
- 135,621 participants
- Exposure
- XGB model vs 2 existing ward-focused models
- Outcome
- Critical events: invasive mechanical ventilation, vasoactive medications, or death within 12 hours of an observation
- Follow-up
- 12 hours
Methodological detailsprognostic model development, external hold-out test cohort, 3 tertiary care academic hospitals
- Prognostic model development
- External hold-out test cohort
- 3 tertiary care academic hospitals
Main finding
As reported in the abstract
Discrimination (C statistic)
| Group | Value | Number needed to alert (at 80% sensitivity) | C statistic |
|---|---|---|---|
| XGB | C statistic 0.86 | 6 | |
| Ward-focused models | 9 | 0.82 |
Value
XGBC statistic 0.86
Ward-focused models— not stated in the abstract
Number needed to alert (at 80% sensitivity)
XGB6
Ward-focused models9
C statistic
XGB— not stated in the abstract
Ward-focused models0.82
- XGB vs Ward-focused models: P < 0.001 — vs model with C statistic 0.82
- XGB vs Ward-focused models: P < 0.001 — vs model with C statistic 0.70
Observational data: associations, not effects.
Further findings
Deep learning models
- The deep learning models did not exhibit improved performance
XGB vs unit-specific models
- XGB performed better or equivalent to models trained for a specific hospital unit
Original publication
Machine Learning for Predicting Critical Events Among Hospitalized Children. JAMA Netw Open. Published 1 May 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →