Key result
RandomForest model outperforms GRACE score for post-AMI mortality prediction with an AUC of 0.81.
Why the study?
Current risk assessment models for acute myocardial infarction may not reflect changes in risk factors in current medical practice.
Does a RandomForest machine learning model improve prediction of one-year post-discharge cardiac mortality in AMI survivors compared to the GRACE score?
Cohort (n=14,345)
Yes
Does a RandomForest machine learning model improve prediction of one-year post-discharge cardiac mortality in AMI survivors compared to the GRACE score?
Effect estimate: AUC 0.81 (95% CI 0.75-0.86)
A machine learning model incorporating environmental and angiographic data outperforms the GRACE score in predicting 1-year cardiac mortality after AMI discharge.
ML model may aid post-AMI mortality prediction; leaves open prospective validation before clinical use.
Background Acute myocardial infarction (AMI) is a leading cause of cardiac mortality, necessitating accurate risk prediction models; however, current risk assessment models may not reflect the changes in risk factors in current medical practice. This study aimed to develop and validate a machine learning-based risk prediction model for one-year post-discharge cardiac mortality among patients with AMI who survived hospitalisation, utilising a multidimensional dataset that integrates environmental determinants and angiographic findings. Methods Data were sourced from a multicentre prospective cohort study (NCT03297164) along with the supplementary AMI cohort from the coordinating medical centre, covering the period from January 2017 to July 2022. Patients from the coordinating centre constituted the discovery cohort ( n =11,752), while patients from the participating centres served as the external validation cohort (n=2,593). RandomForest, XGBoost, logistic regression, CatBoost, decision trees, and LightGBM algorithms were applied, and the feature selection was performed using SelectFromModel. The model with the strongest overall performance in the internal testing cohort underwent external validation. Results The RandomForest model was selected, achieving areas under the curve (AUCs) of 0.86 (95% CI, 0.82–0.89) and 0.81 (95% CI, 0.75–0.86) in the internal testing and external validation cohorts, respectively. Frequently selected predictors included age, body mass index, temperature, PM2.5 concentration, Gensini score, weighted diffuse lesion value, and ejection fraction. RandomForest outperformed the global registry of acute coronary events (GRACE) score in both cohorts, with AUCs of 0.80 (95% CI, 0.75–0.84) and 0.77 (95% CI, 0.71–0.83), respectively. Compared with the GRACE-based stratification, the risk stratification based on the RandomForest algorithm also resulted in greater discrimination in both cohorts (AUC: 0.83 vs. 0.64; 0.77 vs. 0.68). Conclusion This study presents a discharge-oriented model for one-year post-discharge cardiac mortality risk stratification in AMI survivors. Developed and validated using multidimensional data, the model demonstrates superior predictive performance compared with the GRACE score.
No takes yet. Share an insight, caveat, or question.
Hou et al. (2026) conducted a cohort in Acute myocardial infarction (n=14,345). RandomForest machine learning risk prediction model vs. GRACE score was evaluated on One-year post-discharge cardiac mortality (AUC 0.81, 95% CI 0.75-0.86). A RandomForest machine learning model predicted one-year post-discharge cardiac mortality in AMI survivors with an AUC of 0.81 (95% CI 0.75-0.86), outperforming the GRACE score (AUC 0.77).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: