大腸癌に対する腹腔鏡下前方切除術後の高齢者における永続的回腸瘻の機械学習ベース予測

Conversion from temporary stoma to a permanent stoma (PS) following loop ileostomy (LI) significantly impairs patients' quality of life and prognosis. However, few predictive tools are currently available in clinical practice. This study aims to develop and validate a machine learning (ML) model to predict the risk of PS following laparoscopic anterior resection (LAR) in older adults with rectal cancer.

Clinical data were retrospectively collected from 956 older adults (age ≥ 60 years) who underwent LAR with LI between June 2015 and December 2024 at two campuses of the First Hospital of China Medical University. After random stratification into training and test sets (7:3 ratio), variables were selected using the Least Absolute Shrinkage and Selection Operator (LASSO) regression and multivariate logistic regression. Seven ML algorithms-logistic regression (LR), support vector machine (SVM), decision tree (DT), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LGBM), and neural network (NNet)-were employed to construct predictive models. Hyperparameter optimization was performed using five-fold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), the precision-recall area under the curve (PRAUC), confusion matrix metrics, calibration plots, and decision curve analysis (DCA). Model interpretability was enhanced using SHapley Additive Explanation (SHAP) values, and a nomogram was developed to facilitate clinical utility.

The incidence of PS in the cohort was 22.4% (214/956). The LR model demonstrated superior predictive performance, achieving an AUROC of 0.822 (95% CI: 0.763-0.881) and a PRAUC of 0.587 (95% CI: 0.464-0.705), along with optimal balanced accuracy (0.752), recall (0.719), and F1 score (0.582). SHAP analysis identified age, diabetes, and postoperative adjuvant therapy as the most influential predictors of PS.

The LR-based model exhibited robust discriminative ability and clinical applicability for predicting PS in older patients following LAR. This tool facilitates the early identification of high-risk individuals, thereby enabling personalized interventions to mitigate adverse outcomes.