AUTHOR=Yao Huilan , Yuan Shijin , Pan Hongying , Hong Sisi , Huang Chen , Zhao Linfang , Yuan Hongdi , Mei Lei , Zheng Yinghong , Liu Xiaolong , Lu Weina 

TITLE=Predicting hypoglycemia risk after gastrointestinal surgery in type 2 diabetes mellitus: a retrospective cohort study

JOURNAL=Frontiers in Endocrinology

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2025.1590780

DOI=10.3389/fendo.2025.1590780

ISSN=1664-2392

ABSTRACT=ObjectiveTo identify factors influencing hypoglycemia in patients with type 2 diabetes mellitus (T2DM) following gastrointestinal tumor surgery and construct a predictive model for assessing hypoglycemia risk.MethodsWe retrospectively collected data on 1280 patients with T2DM who underwent gastrointestinal tumor surgery and divided them into two groups—one for model building (n = 982) and another for validation (n = 298). We used multivariate logistic regression to develop a predictive model for hypoglycemia following gastrointestinal tumor surgery. The model’s predictive performance was evaluated using the area under the receiver operating characteristic (ROC) curve, and its generalization ability was evaluated using the bootstrap test and the five-fold cross-validation test.ResultsWe identified hypoglycemia following gastrointestinal tumor surgery in 124 of 982 (12.6%) T2DM patients in the developmental cohort. Finally, five predictors, including duration of diabetes, operation duration, preoperative fasting time, preoperative hypoglycemic regimen (subcutaneous insulin injection), and glucose fluctuation on the day of surgery, were integrated into the predictive model. The performance of the hypoglycemia risk prediction model for patients with T2DM undergoing gastrointestinal tumor surgery was comprehensively evaluated. The model demonstrated an area under the ROC curve (AUC) of 0.837 (95% CI: 0.792–0.882), indicating a strong discriminative ability. Internal validation via five-fold cross-validation with bootstrap resampling revealed close approximation of the calibration curve to the ideal line, refining high consistency between predicted probabilities and actual hypoglycemia occurrence. Decision curve analysis (DCA) further supported its clinical utility, indicating value in clinical decision making for hypoglycemia risk stratification and preventive intervention selection.ConclusionThe developed model exhibits high discriminative ability and good calibration. Following visualization (e.g., nomogram), it provides a clinical tool for healthcare providers to stratify hypoglycemia risk in T2DM patients undergoing gastrointestinal tumor surgery, enabling personalized perioperative glucose management and informed decision making to improve patient outcomes.