Combined assessment of stress hyperglycemia ratio and frailty for mortality risk prediction in critically ill patients with coronary heart disease: insights from the MIMIC-IV database

Abstract

Critically ill patients with coronary heart disease (CHD) are at high risk of early death. Traditional prognostic models may not fully capture the combined effects of acute metabolic stress and reduced physiological reserve. We hypothesized that integrating the stress hyperglycemia ratio (SHR, representing acute metabolic stress) with the laboratory-based frailty index (FI-Lab, representing baseline physiological reserve) would provide incremental prognostic value beyond traditional clinical models. This study conducted a retrospective cohort analysis based on the MIMIC-IV database, including adult ICU patients with CHD. SHR was defined as the ratio of admission glucose to estimated chronic average glucose derived from HbA1c; frailty status was assessed using the laboratory-based FI-Lab, which was constructed from routine laboratory parameters and vital signs. Patients were jointly stratified according to tertiles of SHR and FI-Lab, and multivariable Cox regression was used to examine the association between their combined status and the risk of 30-day all-cause mortality. Based on a baseline clinical model, the incremental predictive value was further assessed using time-dependent AUC, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). In addition, machine learning models were constructed as a supplementary predictive analysis, and the SHAP (SHapley Additive exPlanations) method was employed to enhance model interpretability and quantify the relative contribution of each feature to the predictive outcomes. Among 5,580 eligible patients, 496 (8.9%) died within 30 days. In the joint SHR and FI-Lab stratification, the high-SHR combined with high- FI-Lab group had the highest adjusted risk (HR = 5.86, 95% CI 3.41–10.08). Adding SHR and FI-Lab to the baseline model improved model discrimination and reclassification (AUC = 0.8166; IDI = 0.025; NRI = 0.229). Among the machine learning models, Random Forest achieved the best performance in the test set (AUC = 0.846), and SHAP analysis identified both FI-Lab and SHR as important contributing variables. The combined assessment of SHR and FI-Lab integrates information on acute stress-related glycemic burden and physiological vulnerability at admission, which may facilitate more comprehensive identification of short-term mortality risk in critically ill patients with coronary heart disease. Not applicable.