Stress Hyperglycemia Ratio Predicts Death in Atrial Fibrillation Only in Patients With Normal Glucose

Critically ill patients with atrial fibrillation often show blood sugar spikes that do not reflect chronic diabetes but rather the metabolic stress of acute illness. The researchers also applied interpretable machine learning to see how strongly SHR weighed against other risk factors.

The team screened patients with atrial fibrillation in the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and stratified 4,365 individuals by HbA1c into three groups: normal glucose regulation (1,388 patients), prediabetes (1,156), and diabetes mellitus (1,821). The cohort had a mean age of 73.1 years and was 62.1% male. SHR was divided into tertiles, and 30-day all-cause mortality after intensive care unit admission served as the primary endpoint. In total, 501 patients, or 11.48%, died within 30 days.

The analysis combined traditional statistics with machine learning. Kaplan-Meier, landmark, restricted cubic spline, and Cox regression analyses assessed the association between SHR and mortality, while sixteen machine learning algorithms with SHapley Additive exPlanations (SHAP) interpretation evaluated risk factors. Elevated SHR predicted mortality across all three subgroups in unadjusted and partially adjusted models, but the picture changed after full multivariable adjustment.

In fully adjusted models, high SHR remained an independent risk factor for 30-day mortality only in patients with normal glucose regulation, with a hazard ratio of 1.88 (95% confidence interval 1.15–3.07, P = 0.011). No significant association was observed in the prediabetes subgroup (HR = 1.27, 95% CI 0.78–2.07, P = 0.338) or the diabetes subgroup (HR = 1.24, 95% CI 0.84–1.82, P = 0.275). Nonlinear associations between SHR and mortality were identified across all three phenotypes.

The machine learning models achieved good discrimination, with area under the curve values above 0.85, and SHAP analysis confirmed that SHR carried greater prognostic weight in patients with normal glucose regulation than in the other groups. The authors conclude that SHR serves as a prognostic tool for atrial fibrillation mortality, offering enhanced predictive accuracy compared with individual metrics, particularly in patients with normal glucose regulation, and that this integrated approach could inform personalized glycemic management strategies.

Subject of Research: Using the stress hyperglycemia ratio and machine learning to predict mortality in critically ill patients with atrial fibrillation across glucose metabolic states.

Article Title: Assessment of stress hyperglycemia ratio to predict mortality in patients with atrial fibrillation across different glucose metabolic states: an observational cohort study with machine learning

Article References: Sun, L., Wang, T., Yuan, J., Yuan, Q., & Ning, B. (2026). Assessment of stress hyperglycemia ratio to predict mortality in patients with atrial fibrillation across different glucose metabolic states: an observational cohort study with machine learning. Cardiovascular Diabetology. https://doi.org/10.1186/s12933-026-03303-z

Image Credits: AI Generated

DOI: 10.1186/s12933-026-03303-z

Keywords: stress hyperglycemia ratio, atrial fibrillation, mortality, MIMIC-IV, machine learning, SHAP, HbA1c, intensive care, glycemic status, cohort study, Assessment, stress