Advances Brain Research and Neuroscience
ISSN: 3071-2750 | DOI: 10.64030/3071-2750
Yixiao Lu, Chengbao Yang, Fuxin Yi, Chunguang Chen, Fei Meng and Mingfei Xia
Background: The combined prognostic value of the stress hyperglycemia ratio (SHR) and glycemic variability (GV) for mortality risk stratification across different glucose metabolic states in critically ill cerebrovascular patients remains unexplored. This study aims to evaluate its predictive utility by employing machine learning to identify critical risk predictors.
Methods: This retrospective cohort study analyzed data from the MIMIC-IV database and included 2,281 adult ICU patients with cerebrovascular disease stratified by glycemic status (NGR, Pre-DM, DM). The outcomes were 28-day and 90-day all-cause mortality. Associations and predictive performance of SHR and GV were evaluated via Cox regression, Kaplan�??Meier analysis, and receiver operating characteristic (ROC) curves, with machine learning models (SHAP interpretation) applied for predictor identification.
Results: Among 2,281 patients, high levels of both SHR and GV were independently associated with increased 28-day (HR 1.53, 95% CI 1.11–2.11) and 90-day mortality (HR 1.53, 95% CI 1.15–2.03), particularly in nondiabetic subgroups. GV exhibited a nonlinear association with mortality risk. Compared with the SHR-GV model alone, the combined SHR–GV model did not significantly improve 28-day mortality prediction. For 90-day mortality in diabetic patients, the combination had a marginally greater AUC (0.584 vs. 0.557), although this difference was not statistically significant. Machine learning interpretation confirmed the SHR as the dominant predictor.
Conclusion: The SHR outperforms GV in predicting short-term mortality in critical cerebrovascular patients. Although combining both metrics does not significantly improve predictive accuracy, it enables practical risk stratification—particularly in people without diabetes—to guide personalized glucose management.