Related Experiment Video
Updated: Nov 1, 2025
05:37
An R-Based Landscape Validation of a Competing Risk Model
Published on: September 16, 2022
2.3K
Predicting 30-day mortality after ST elevation myocardial infarction: Machine learning- based random forest and its
Amir Hadanny1, Roni Shouval2, Jianhua Wu3
1The Mina and Everard Goodman Faculty of Life Sciences, Bar-Ilan University, Ramat-Gan, Israel; Sackler School of Medicine, Tel-Aviv University, Tel-Aviv, Israel.
Journal of Cardiology
|June 22, 2021
Summary
Machine learning models accurately predict 30-day mortality after ST-segment elevation myocardial infarction (STEMI), outperforming the GRACE score. This validated random forest model offers improved risk stratification for STEMI patients upon admission.
Area of Science:
- Cardiology
- Medical Informatics
- Machine Learning in Healthcare
Background:
- Numerous prognostic models exist for predicting mortality post-ST-elevation myocardial infarction (STEMI).
- Previous research indicated machine learning (ML) models surpass traditional risk scores for 30-day mortality prediction in STEMI.
- This study aimed to refine and externally validate an ML-based random forest model for 30-day mortality post-STEMI.
Purpose of the Study:
- To redevelop a machine learning (ML) random forest prediction model for 30-day mortality post-ST-elevation myocardial infarction (STEMI).
- To externally validate the developed ML model on a large, independent cohort.
- To compare the performance of the ML model against the established Global Registry of Acute Cardiac Events (GRACE) score.
Main Methods:
- Retrospective, supervised learning data mining study utilizing the Acute Coronary Syndrome Israeli Survey (ACSIS) registry and the Myocardial Ischemia National Audit Project (MINAP) for external validation.
- Development of two random forest models (full and simple) on the ACSIS cohort, assessing feature importance.
- External validation on the MINAP cohort, comparing discrimination (Area Under the Curve - AUC) and calibration of ML models against the GRACE score.
Main Results:
- Random forest models demonstrated superior discrimination compared to the GRACE score in the MINAP cohort (AUC 0.804/0.787 vs. 0.764).
- Initial calibration was suboptimal for all models in the MINAP data.
- Platt scaling improved the calibration of the random forest models, but not the GRACE score.
Conclusions:
- The externally validated random forest model provides a more accurate tool for early risk stratification of 30-day mortality in STEMI patients.
- The ML-based model outperforms the widely used GRACE score.
- This represents a significant advancement as the first externally validated ML model for STEMI mortality prediction.
