Enhanced machine learning models for predicting three-year mortality in Non-STEMI patients aged 75 and above

Abstract

Abstract Background Non-ST segment elevation myocardial infarction (Non-STEMI) is a severe cardiovascular condition mainly affecting individuals aged 75 and above, who are at higher risk of mortality due to age-related vulnerabilities and other health issues. Current prognostic models are often inadequate in addressing the complexity of this population. This study aims to develop and validate a machine learning (ML) model to predict three-year mortality in Non-STEMI patients aged 75 and above, to assist clinicians in decision-making. Methods Clinical data from 234 Non-STEMI patients aged 75 and above were collected and split into a training cohort (164 patients) and a validation cohort (70 patients) using a 70:30 ratio. Six key factors—age, Pulse (P, beats per minute), respiratory support rate, Glucose (Glu) levels, percutaneous coronary intervention (PCI), and β-blocker use—were identified as significantly associated with three-year mortality through LASSO regression and ten-fold cross-validation. The Random Forest (RF) model was employed for prediction, which yielded the best performance with an area under the curve (AUC) of 0.92. SHapley Additive exPlanations (SHAP) analysis was used to determine the top contributing features influencing mortality, with PCI, age, and P(bpm) identified as the most critical factors. A web-based calculator was also developed to support clinical decision-making. Results The RF model demonstrated the best predictive performance (AUC = 0.92), significantly outperforming other models. Key features, such as PCI, age, and P(bpm), were found to be highly influential in predicting three-year mortality. The developed web-based tool offers clinicians a user-friendly platform to incorporate these findings into personalized care decisions. Conclusions This study presents a robust RF model for predicting three-year mortality in Non-STEMI patients aged 75 and above. PCI, β-blocker use, and effective management of P(bpm) and Glu levels are crucial factors for improving patient outcomes. The web-based tool enhances personalized decision-making, helping clinicians better allocate resources and provide tailored interventions for this at-risk population.

Keywords

• Non-ST segment elevation myocardial infarction
• Machine learning
• Older patients