Key result
Extreme gradient boosting shows no mortality prediction benefit over logistic regression in STEMI survivors.
Why the study?
Machine learning may enhance the predictive performance of traditional risk models after MI, prompting the development and validation of an ML-based model to predict 5-year all-cause mortality after hospital discharge in STEMI patients.
Does a machine learning-based model improve the prediction of 5-year all-cause mortality compared with logistic regression in hospital survivors of STEMI who underwent coronary angiography?
Cohort (n=4,754)
Yes
Does a machine learning-based model improve the prediction of 5-year all-cause mortality compared with logistic regression in hospital survivors of STEMI who underwent coronary angiography?
Machine learning models (specifically extreme gradient boosting) did not substantially improve 5-year mortality discrimination compared with traditional logistic regression in STEMI survivors, though ML demonstrated improved calibration in low-risk patients.
Answers are grounded in this paper and cite their sources.
ML models may enhance post-STEMI mortality prediction in registries; leaves open prospective validation before clinical use.
Background Machine learning (ML) may enhance the predictive performance of traditional risk models after myocardial infarction (MI). This study aimed to develop and validate an ML-based model to predict 5-year all-cause mortality after hospital discharge in patients with ST-elevation myocardial infarction (STEMI). Methods Data were obtained from the FAST-MI 2010 and 2015 multicenter, prospective registries. A total of 4,754 STEMI patients who underwent coronary angiography and survived to hospital discharge were included and randomly split into a training cohort (3,334 patients) and a validation cohort (1,420 patients). The primary endpoint was 5-year all-cause mortality. Among 34 candidate variables, the Boruta algorithm identified the most relevant predictors, including age, hemoglobin, creatinine, left ventricular ejection fraction, Killip class, and histories of stroke, heart failure, peripheral artery disease, cancer, and atrial fibrillation. Three ML classifiers—extreme gradient boosting (XGB), random forest, and support vector machine—were compared with logistic regression (LR). Results During 5-year follow-up, 555 deaths (11.7%) occurred. XGB showed the best performance among ML methods and was selected as the final model. In the validation cohort, the ML model achieved an AUROC of 0.81 [95% CI 0.77–0.85], which was comparable to that of LR (0.80 [95% CI 0.77–0.84]; DeLong’s test p=0.79). The corresponding PRAUC values were 0.45 [95% CI 0.37–0.53] and 0.42 [95% CI 0.34–0.49], respectively. The ML model demonstrated improved calibration, particularly in low-risk patients. Conclusions Among hospital survivors of STEMI who underwent coronary angiography, ML did not substantially improve 5-year mortality prediction compared with LR based on routinely available clinical variables.
No takes yet. Share an insight, caveat, or question.
Singh et al. (2026) conducted a cohort in ST-elevation myocardial infarction (STEMI) (n=4,754). Machine learning mortality prediction model (extreme gradient boosting [XGB]) vs. Logistic regression (LR) was evaluated on 5-year all-cause mortality prediction performance (AUROC) (AUROC 0.81, 95% CI 0.77-0.85, p=0.79). Extreme gradient boosting machine learning achieved comparable 5-year mortality prediction to logistic regression in STEMI survivors.
Synapse has enriched 10 related papers. Consider them for comparative context: