Prognosis of major bleeding based on residual variables and machine learning for critical patients with upper gastrointestinal bleeding: A multicenter study
Authors' conclusion
ML models using residuals improved the accuracy and interpretability in predicting major bleeding during ICU admission in patients with UGIB.
Study context
- Population
- Critical patients with severe upper gastrointestinal bleeding (UGIB) after first intensive care unit (ICU) admission25 machine learning (ML) models
- Participants
- 2,604 patients
- Comparison
- ML models using residual data vs models using test results directly
- Outcome
- Recurrence of major gastrointestinal bleeding
- Follow-up
- During ICU admission
Methodological detailsmachine learning, multicentre, MIMIC-IV and eICU databases, Shapley additive explanations and saliency maps
- Machine learning
- Multicentre
- MIMIC-IV and eICU databases
- Shapley additive explanations and saliency maps
Performance
As reported in the abstract
Prediction of major bleeding
| Test or model | AUC |
|---|---|
| LightGBM, pre-ICU residual data | 0.96 |
LightGBM, pre-ICU residual data
AUC0.96
Further findings
Key predictive factors
- Aspartate aminotransferase, blood urea nitrogen, albumin, ICU LOS, respiratory rate
Original publication
Prognosis of major bleeding based on residual variables and machine learning for critical patients with upper gastrointestinal bleeding: A multicenter study. J Crit Care. Published 2 October 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →