Combined assessment of stress hyperglycemia ratio and hemoglobin glycation index to predict mortality after coronary revascularisation: a machine learning study

Background Stress hyperglycemia ratio (SHR) and hemoglobin glycation index (HGI) reflect acute glucose elevation and chronic-to-acute glycemic mismatch, which correlate with adverse outcomes in cardiovascular diseases. However, the prognostic significance of SHR or HGI evaluation for mortality in critically ill patients following coronary revascularization remains unclear. Purpose This study examines the associations of SHR, HGI, and their synergistic effects with mortality in this specific population, incorporating machine learning (ML) to identify critical risk factors influencing mortality. Methods Patients undergoing coronary revascularization were screened from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. The primary endpoint was 30-day mortality, with 365-day mortality as the secondary outcome. SHR and HGI levels were categorized into quartiles. Associations with mortality were analyzed using Kaplan-Meier (KM) curves, Cox proportional hazards models, restricted cubic splines (RCS), receiver operating characteristic (ROC) curves, and subgroup analyses. Five ML algorithms were employed for mortality risk prediction, with Shapley Additive exPlanations (SHAP) applied to identify and interpret critical predictors. Results A total of 4518 patients were included, with a median age of 69.41 years, and 74.79% were female. Overall, 139 (3.1%) and 375 (8.3%) patients died within 30 and 365 days of ICU admission, respectively. In the Cox regression analysis, the combined "High SHR + Low HGI" group demonstrated the poorest survival (log-rank P < 0.001). RCS revealed a consistent U-shaped association for HGI across all time points, while SHR exhibited a J-shaped relationship for long-term mortality. For 30-day mortality prediction, the combined SHR-HGI assessment demonstrated superior predictive performance versus HGI alone [AUC 0.750 (0.701–0.799) vs. 0.633 (0.576–0.689), P < 0.001]. Consistent significant advantages of the combined model were observed for 180-day and 365-day mortality predictions. Furthermore, among the ML models, the Random Forest (RF) algorithm achieved the highest predictive capability (AUC = 0.848). SHAP interpretability analysis confirmed both SHR and HGI as critical top-ranking contributors to mortality risk. Conclusions Both SHR and HGI serve as novel, independent prognostic indicators for critically ill patients following coronary revascularization, exhibiting distinct non-linear risk patterns. This integrated approach, supported by interpretable ML, could inform personalized glycemic management strategies, potentially improving clinical outcomes in the intensive care setting.