Machine learning-based mortality risk prediction model for elderly diabetic patients with non-ST-segment elevation...

Area of Science:

• Cardiology
• Artificial Intelligence
• Geriatrics

Background:

• Non-ST-elevation myocardial infarction (NSTEMI) in elderly diabetic patients poses significant challenges for risk stratification and prognosis.
• Existing risk assessment tools may not fully capture the complexity of this specific high-risk population.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based model for predicting 28-day all-cause mortality in elderly diabetic NSTEMI patients.
• To compare the performance of ML models against traditional scoring systems.

Main Methods:

• A retrospective cohort study utilizing the MIMIC-IV database, including 5,272 NSTEMI patients aged ≥55 years with diabetes.
• Development and evaluation of multiple ML models using clinical data within 24 hours of admission.
• Performance assessment via ROC curves, calibration plots, decision curve analysis, and SHAP for interpretability.

Main Results:

• The XGBoost ML model achieved superior predictive performance (AUC = 0.86) compared to other algorithms and conventional scores.
• SHAP analysis identified key prognostic factors including PaO2, Charlson Comorbidity Index, and APSIII score.
• Lactate levels demonstrated a wide influence, and platelet count showed bidirectional effects, highlighting complex nonlinear relationships.

Conclusions:

• The developed ML model provides robust and clinically useful prediction of 28-day mortality for elderly diabetic NSTEMI patients.
• The model's interpretability offers valuable insights into the multifactorial nature of mortality risk in this population.
• This AI-driven approach enhances risk assessment and clinical decision-making for improved patient outcomes.