Hemoglobin glycation index may contextualize the association between time-varying stress hyperglycemia ratio and mortality in critically ill patients with heart failure: a multicenter retrospective cohort study

BACKGROUND: Stress hyperglycemia is associated with adverse outcomes in critically ill patients, yet the optimal metric and the influence of chronic glycemic background remain unclear. The stress hyperglycemia ratio (SHR) adjusts acute glucose for chronic glycemic status, but whether the prognostic association of time-varying SHR differs across hemoglobin glycation index (HGI)-defined glycation phenotypes in critically ill heart failure patients remains unclear. OBJECTIVES: To evaluate the association between time-varying SHR and mortality across HGI tertiles in critically ill heart failure patients. METHODS: This multicenter retrospective cohort study included 2,488 adult heart failure patients from MIMIC-III-CareVue (n = 307), MIMIC-IV v3.1 (n = 1,654), and NWICU v0.1.0 (n = 527). HGI was calculated as the residual from regressing HbA1c on admission glucose within each database, and patients were stratified into tertiles. SHR was modeled as a time-varying exposure using extended Cox proportional hazards models with progressive covariate adjustment. Dose-response relationships were examined using restricted cubic splines, and incremental predictive value was assessed via time-dependent ROC analysis. RESULTS: In fully adjusted models (hazard ratios per 0.1-unit increase in SHR), time-varying SHR was significantly associated with 28-day mortality in MIMIC-III (HR 1.13, 95% CI 1.06-1.21) and MIMIC-IV (HR 1.10, 95% CI 1.07-1.13; both P < 0.001), and with in-hospital mortality across all three cohorts (all P ≤ 0.002). Twenty-eight-day mortality was not available in NWICU. HGI-stratified analyses revealed a numerical gradient in MIMIC-IV (T1: HR 1.13, 95% CI 1.08-1.17; T3: HR 1.06, 95% CI 1.01-1.12), though SHR × HGI interactions were not significant in individual databases. Restricted cubic spline analyses demonstrated J-shaped dose-response curves in larger analyses, with significant nonlinearity in MIMIC-IV and the pooled in-hospital mortality cohort. Adding mean SHR to clinical models modestly improved discrimination (ΔAUC = 0.029-0.032, P ≤ 0.019). Subgroup analyses identified significant effect heterogeneity by age, sex, and Charlson index (all P-interaction < 0.05). Eight sensitivity analyses corroborated the robustness of primary findings. CONCLUSIONS: Time-varying SHR was independently associated with mortality in critically ill patients with heart failure, with a numerical tendency toward stronger associations among patients with low HGI phenotype. These hypothesis-generating findings suggest that dynamic SHR assessment combined with HGI-informed interpretation may potentially aid individualized glycemic risk assessment in this population.