Objective To construct a model to predict the risk of postoperative fatigue syndrome (POFS) in patients undergoing anorectal surgery under intravenous general anesthesia.

Methods The patients who underwent anorectal surgery under intravenous anesthesia at Fuyang Hospital of Traditional Chinese Medicine from December 2023 to February 2025 were selected as the research subjects. The clinical data of 230 patients were collected, and randomly divided into the training set (n = 160) and test set (n = 70) at a ratio of 7:3. Multivariate logistic regression analysis was used for variable selection. The POFS prediction model was developed by using seven machine learning methods (decision tree, multi-layer perceptron, K-nearest neighbor algorithm, optical gradient lifter, random forest, support vector machine and extreme gradient lifter). The performance of the machine learning model was evaluated using the area under the receiver operating characteristic curve, precision rate-recall curve, 5-fold cross-validation and decision curve analysis. Global and local interpretations of machine learning models were provided through SHAP value appending.

Results The incidence of POFS in 230 patients undergoing anorectal surgery was 27.0% (62/230). The age, preoperative Pittsburgh Sleep Quality Index score, preoperative hospital anxiety score, propofol dosage, intraoperative hypotension and postoperative digital pain score were the independent risk factors of the occurrence of POFS in patients undergoing anorectal surgery (P ＜ 0.05). Among the seven machine models, the support vector machine model had the best predictive performance, with an AUROC of 0.889 (95%CI: 0.805–0.972) and PRROC of 0.799 (95%CI: 0.621–0.917), showing the best performance. The SHAP value histogram showed that the importance of the feature variables of the SVM model from high to low was: hypotension, propofol dosage, Pittsburgh Sleep Quality Index score, age, hospital anxiety score and digital pain score. The SHAP values of the six characteristic variables in the SHAP scatter plot showed a "two-end separation" phenomenon in predicting the risk of POFS. Based on the SHAP map SVM model, the risk of POFS in a patient undergoing anorectal surgery was as high as 0.972, and the risk of non POFS in a patient undergoing anorectal surgery was 0.229.

Conclusions The support vector machine model based on age, PSQI score, HADS-A score, dosage of propofol, intraoperative hypotension and NRS score as characteristic variables has the best performance in predicting the risk of POFS in anorectal surgery patients under intravenous anesthesia. The SVM model explained based on SHAP value can better predict the risk of POFS in patients.