Machine learning-based mortality risk prediction model for elderly diabetic patients with non-ST-segment elevation myocardial infarction using MIMIC-IV database.

Non-ST-elevation myocardial infarction (NSTEMI) in elderly diabetic patients presents unique challenges in risk assessment and prognosis prediction. This study aimed to develop and validate a machine learning-based mortality risk prediction model for this specific population using the MIMIC-IV database. We conducted a retrospective cohort study including 5,272 NSTEMI patients aged ≥ 55 years with diabetes from the MIMIC-IV database. Multiple machine learning models were developed using clinical data collected within 24 h of admission. The primary outcome was 28-day all-cause mortality. Model performance was evaluated using ROC curves, calibration plots, and decision curve analysis. SHAP analysis was employed to interpret model predictions. The XGBoost model demonstrated superior performance (AUC = 0.86) compared to other algorithms and traditional scoring systems. SHAP analysis identified PaO2, Charlson Comorbidity Index, and APSIII score as the top three prognostic factors. Lactate levels showed the broadest influence range (SHAP values - 0.5 to 1.5), while platelet count exhibited distinct bidirectional effects on prognosis. Decision curve analysis confirmed the model's superior clinical utility across all risk threshold intervals. Our machine learning-based prediction model achieved robust performance in predicting 28-day mortality risk for elderly diabetic NSTEMI patients. The model's interpretability analysis revealed complex nonlinear relationships between clinical variables and outcomes, providing valuable insights for risk assessment and clinical decision-making.