ORAKLE: Optimal Risk prediction for mAke30 in patients with sepsis associated acute kidney injury (AKI) using deep LEarning
Authors' conclusion
ORAKLE is a robust deep-learning model for predicting MAKE30 in critically ill patients with AKI that utilizes evolving time series data.
Study context
- Population
- ICU patients meeting sepsis-3 criteria who developed AKI within 48 h of ICU admission (MIMIC-IV development; SiCdb and eICU-CRD external validation)MIMIC-IV 16,671
- Exposure
- ORAKLE vs XGBoost vs Cox models
- Outcome
- MAKE30: composite of death, new dialysis or persistent kidney dysfunction
- Follow-up
- 30 days of ICU admission
Methodological detailsprediction model development and external validation, deep learning (Dynamic DeepHit), three public critical care databases
- Prediction model development and external validation
- Deep learning (Dynamic DeepHit)
- Three public critical care databases
Main finding
As reported in the abstract
AUROC for MAKE30
| Group | AUROC, MIMIC-IV internal test set | AUROC, SICdb | AUROC, eICU-CRD |
|---|---|---|---|
| ORAKLE | 0.84 (95% CI: 0.83-0.86) | 0.83 (95% CI: 0.81-0.85) | 0.85 (95% CI: 0.84-0.85) |
| XGBoost | 0.81 (95% CI: 0.79-0.83) | 0.80 (95% CI: 0.78-0.83) | 0.83 (95% CI: 0.83-0.84) |
| Cox | 0.80 (95% CI: 0.78-0.82) | 0.79 (95% CI: 0.77-0.81) | 0.81 (95% CI: 0.80-0.82) |
AUROC, MIMIC-IV internal test set
ORAKLE0.84 (95% CI: 0.83-0.86)
XGBoost0.81 (95% CI: 0.79-0.83)
Cox0.80 (95% CI: 0.78-0.82)
AUROC, SICdb
ORAKLE0.83 (95% CI: 0.81-0.85)
XGBoost0.80 (95% CI: 0.78-0.83)
Cox0.79 (95% CI: 0.77-0.81)
AUROC, eICU-CRD
ORAKLE0.85 (95% CI: 0.84-0.85)
XGBoost0.83 (95% CI: 0.83-0.84)
Cox0.81 (95% CI: 0.80-0.82)
Observational data: associations, not effects.
Further findings
| Outcome | ORAKLE |
|---|---|
| Calibration (Brier score)Suggesting good calibration | |
| Brier score (internal test set, SICdb, eICU-CRD) | 0.21 |
AUPRC for MAKE30
- AUPRC significantly better for ORAKLE than XGBoost and Cox
Original publication
ORAKLE: Optimal Risk prediction for mAke30 in patients with sepsis associated AKI using deep LEarning. Crit Care. Published 26 May 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →