Establishment of predictive model for postoperative delirium in patients undergoing gastrointestinal surgery
10.3760/cma.j.cn131073-20241113-00906
- VernacularTitle:胃肠外科手术患者术后谵妄预测模型的构建
- Author:
Yichun ZHENG
1
;
Yang HAN
;
Keshi YAN
;
Jianming XIAO
;
Ju GAO
;
Yali GE
Author Information
1. 扬州大学附属苏北人民医院麻醉科，扬州 225001
- Publication Type:Journal Article
- Keywords:
Delirium;
Postoperative complications;
Forecasting;
Machine learning;
Digestive system surgical procedures
- From:
Chinese Journal of Anesthesiology
2025;45(9):1117-1123
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct a predictive model for postoperative delirium (POD) in patients undergoing gastrointestinal surgery using machine learning.Methods:This retrospective study used clinical data from patients who underwent gastrointestinal surgery at Subei People′s Hospital between September 2022 and April 2024. The entire dataset was randomly divided into the training and validation sets in an 8∶2 ratio. Multivariate logistic regression analysis was conducted to identify the factors influencing POD. Eleven machine learning models were established and compared. The performance of the models was validated using metrics, including accuracy, precision, recall, Youden′s index, F1 score, Matthews′ correlation coefficient, Kappa coefficient, log loss, and Brier score. Receiver operating characteristic and calibration curves were plotted to assess the discrimination and consistency of the model. Shapley additive explanations were used in Python for interpretative analysis of the model with the best predictive performance, and the importance of the feature parameters was ranked.Results:A total of 1, 785 patients were ultimately included, of which 833 (46.67%) experienced POD. The results of multivariate logistic regression analysis revealed that advanced age, lower preoperative serum calcium ion concentration, postoperative pulmonary infection, and higher preoperative systolic blood pressure were independent risk factors for POD in patients undergoing gastrointestinal surgery, while laparoscopic surgery was a protective factor ( P<0.05). Among the 11 machine learning models, the categorical feature gradient boosting model exhibited the best performance, with an area under the receiver operating characteristic curve of 0.82 (95% confidence interval 0.77-0.87). The ranking of feature importance indicated that age had the greatest contribution in predicting POD. Conclusions:The predictive model for POD established based on the categorical boosting algorithm has higher predictive efficacy and clinical application value in patients undergoing gastrointestinal surgery.
