Chenxi Weng, Jie Su, Huijuan Wang, Qi Lu, Lixia Chen
This study aimed to characterize blood glucose trajectories during the early phase of enteral nutrition (EN) in critically ill patients and develop a predictive model for these trajectories to improve clinical management and patient outcomes. A retrospective analysis was conducted on critically ill patients who received continuous EN for ≥ 2 consecutive days, using data from the intensive care unit (ICU) of a tertiary hospital. Group-Based Trajectory Modeling (GBTM) was employed to identify distinct subgroups based on patterns of blood glucose changes. An early risk prediction model for trajectory classification was constructed using the eXtreme Gradient Boosting (XGBoost) algorithm. A total of 478 patients met the inclusion criteria. Three distinct blood glucose trajectory subgroups were identified: Mild Hyperglycemia Stable (41.84%), Moderate Hyperglycemia Peaking (36.40%), and Severe Hyperglycemia Peaking (21.76%). The XGBoost model exhibited robust discriminative ability and good calibration. Key predictors of trajectory classification included insulin use, history of diabetes, C-reactive protein, and interleukin-6. This study highlights the heterogeneity of glycemic trajectories during the early phase of EN in critically ill patients. The developed XGBoost model demonstrated satisfactory predictive performance and may serve as a valuable tool for maintaining glycemic stability and optimizing outcomes in critically ill patients receiving nutritional support in the ICU.