Dynamic blood glucose trajectory is 28-day all-cause mortality in critically ill patients with heart failure: A survival prediction analysis

Background Heart failure (HF) is a global public health issue. Blood glucose influences HF progression, but research on its dynamic changes and mortglity remains limited. Objective This study aimed to investigate the association of baseline admission blood glucose and its dynamic changes with all-cause mortality in patients with HF. Methods This retrospective cohort study used data from MIMIC-IV and a tertiary hospital. We identified distinct glycemic trajectories using group-based trajectory modeling (GBTM) and assessed their association with 28-day all-cause mortality via Cox proportional hazards models and restricted cubic spline analysis (RCS). Machine learning models were applied to evaluate the predictive value of glycemia. Results A total of 6142 patients from the MIMIC-IV database and 391 local patients were included. Results showed that five dynamic blood glucose trajectory groups were identified. Among them, the moderate-to-high and high-level groups exhibited elevated mortality risk [HR (95% CI): 1.546 (1.231–1.942); 1.572 (1.140–2.168)]. Incorporating blood glucose trajectories into predictive models moderately improved the predictive performance for outcomes, and the Deepsurv model demonstrated better predictive accuracy and stability. Conclusion Although current retrospective design limits causal inference and warrants prospective validation, the present study provides evidence that dynamic blood glucose fluctuations demonstrate superior prognostic value compared to baseline measurements for predicting mortality in severe HF patients. Furthermore, machine learning based on glucose trajectory modeling enables early identification of high-risk individuals.