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

Area of Science:

• Cardiology
• Machine Learning
• Geriatric Medicine

Background:

• Non-ST segment elevation myocardial infarction (Non-STEMI) poses a high mortality risk for individuals aged 75 and above.
• Existing prognostic models often fail to adequately address the complexities of this elderly population.
• There is a critical need for improved predictive tools for Non-STEMI patients in this age group.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting three-year mortality in Non-STEMI patients aged 75 years and older.
• To identify key clinical factors influencing mortality in this demographic.
• To provide clinicians with a tool to aid in decision-making for personalized patient care.

Main Methods:

• A Random Forest (RF) model was developed using clinical data from 234 Non-STEMI patients (70:30 train-validation split).
• LASSO regression and cross-validation identified significant predictors: age, pulse, respiratory support, glucose, percutaneous coronary intervention (PCI), and beta-blocker use.
• SHapley Additive exPlanations (SHAP) analysis pinpointed PCI, age, and pulse as most critical for mortality prediction.

Main Results:

• The RF model achieved a high predictive performance with an Area Under the Curve (AUC) of 0.92.
• Percutaneous coronary intervention (PCI), patient age, and pulse rate (P [bpm]) were identified as the most significant predictors of three-year mortality.
• A user-friendly web-based calculator was created to integrate these predictive insights into clinical practice.

Conclusions:

• A robust Random Forest (RF) model effectively predicts three-year mortality in elderly Non-STEMI patients.
• Percutaneous coronary intervention (PCI), beta-blocker use, and management of pulse (P [bpm]) and glucose (Glu) levels are vital for improving outcomes.
• The developed web tool supports personalized decision-making, optimizing resource allocation and tailored interventions for this vulnerable population.