Integrated Assessment of Stress Hyperglycemia and Glycemic Variability for Mortality Prediction in Non-St-Segment Elevation Myocardial Infarction and St-Segment Elevation Myocardial Infarction: A Glycometabolic Phenotype-Stratified Analysis

Background: Precise risk stratification in myocardial infarction (MI) remains challenging, particularly when integrating acute dysglycemia with chronic metabolic status. Prior evaluations of stress hyperglycemia ratio (SHR) and glycemic variability (GluCV) have seldom accounted for underlying glucose metabolic phenotypes, potentially obscuring their prognostic utility in distinct MI subtypes. Methods: In this retrospective cohort study, we identified 3628 non-ST-segment elevation myocardial infarction (NSTEMI) and 825 ST-segment elevation myocardial infarction (STEMI) patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Participants were stratified by glycemic status: normal glucose regulation (NGR), pre-diabetes (Pre-DM), and diabetes mellitus (DM). We assessed the individual and combined prognostic value of SHR and GluCV for one-year all-cause mortality. Machine learning (ML) models were subsequently developed to optimize risk prediction. Results: The combination of SHR and GluCV effectively stratified mortality risk across all glycemic phenotypes in both NSTEMI and STEMI patients. Notably, within NSTEMI cohorts, the subgroup characterized by both high GluCV and low SHR exhibited the poorest survival. In comparative analyses, GluCV demonstrated superior prognostic performance over SHR for NSTEMI outcomes. Among five ML algorithms tested, the Random Forest model achieved the highest predictive accuracy [NGR (area under the curve (AUC): 0.885, 95% confidence interval (CI): 0.882-0.888), Pre-DM (AUC: 0.854, 95% CI: 0.85-0.857), DM (AUC: 0.87, 95% CI: 0.868-0.871)], with GluCV consistently identified as a top-ranking feature. Conclusions: By concurrently evaluating SHR and GluCV within a glycemic stratification framework, this study refines mortality prediction in MI patients. Our findings advocate for a shift from uniform glycemic targets to phenotype-specific management strategies. The developed ML tool offers a practical approach for individualized risk assessment in clinical practice.