Prognostic value of glycemic variability for ICU and in-hospital all-cause mortality in postoperative patients with upper gastrointestinal cancer: a retrospective cohort study.
Most studies on upper gastrointestinal cancer (UGIC) only focus on static glycemic markers while neglecting glycemic variability (GV). The prognostic role of GV and its subgroup heterogeneity, especially across coronary artery disease (CAD) status, remain unclear. This study aimed to explore the independent predictive value, non-linear pattern and clinical heterogeneity of GV for Intensive Care Unit (ICU) and in-hospital mortality in UGIC postoperative patients.
This study retrospectively collected and analyzed clinical data from postoperative patients with UGIC at Quzhou People's Hospital. Survival analyses were employed. They were Kaplan-Meier curves, Cox proportional hazards models, and restricted cubic splines. The study outcomes were 56-day ICU and in-hospital mortality. Subgroup analyses were performed, and three tree-based machine-learning prediction models were implemented, including the random-forest model via the ranger package (version 0.16.0), the XGBoost gradient-boosting model via the XGBoost package (version 1.7.5), and the LightGBM model. Model performance was evaluated and we employed the area under the receiver operating characteristic curve, along with the SHapley Additive exPlanations package, to interpret feature contributions.
Elevated GV independently predicted higher ICU and in-hospital mortality, with spline curves suggesting a potential non-linear-like trend, although formal statistical tests did not confirm a statistically significant nonlinear association. GV and SOFA score were the top prognostic predictors. The GV-mortality association differed between two endpoints and presented obvious CAD-related subgroup heterogeneity. In-hospital mortality better reflected the full prognostic impact of GV, especially in mechanically ventilated and CAD-free patients.
GV is a standalone predictor for ICU and in-hospital all-cause mortality in patients with UGIC, exhibiting a non-linear-like trend and demonstrating robustness across diverse patient populations. Maintaining GV may represent a valuable clinical strategy for improving outcomes in this patient population.
This study retrospectively collected and analyzed clinical data from postoperative patients with UGIC at Quzhou People's Hospital. Survival analyses were employed. They were Kaplan-Meier curves, Cox proportional hazards models, and restricted cubic splines. The study outcomes were 56-day ICU and in-hospital mortality. Subgroup analyses were performed, and three tree-based machine-learning prediction models were implemented, including the random-forest model via the ranger package (version 0.16.0), the XGBoost gradient-boosting model via the XGBoost package (version 1.7.5), and the LightGBM model. Model performance was evaluated and we employed the area under the receiver operating characteristic curve, along with the SHapley Additive exPlanations package, to interpret feature contributions.
Elevated GV independently predicted higher ICU and in-hospital mortality, with spline curves suggesting a potential non-linear-like trend, although formal statistical tests did not confirm a statistically significant nonlinear association. GV and SOFA score were the top prognostic predictors. The GV-mortality association differed between two endpoints and presented obvious CAD-related subgroup heterogeneity. In-hospital mortality better reflected the full prognostic impact of GV, especially in mechanically ventilated and CAD-free patients.
GV is a standalone predictor for ICU and in-hospital all-cause mortality in patients with UGIC, exhibiting a non-linear-like trend and demonstrating robustness across diverse patient populations. Maintaining GV may represent a valuable clinical strategy for improving outcomes in this patient population.