Sexual function recovery trajectories and predictive modeling in men undergoing rectal cancer surgery

Sexual dysfunction is a common and distressing sequela of rectal cancer surgery. However, the longitudinal patterns of its recovery and reliable early predictive factors are not well established, which hinders proactive management strategies. This prospective observational cohort study enrolled 154 male patients undergoing curative-intent surgery for rectal cancer. Sexual function was assessed using the International Index of Erectile Function-5 (IIEF-5) questionnaire preoperatively (T1) and at 3, 6, 9, and 12 months postoperatively (T2–T5). Latent profile analysis (LPA) identified distinct recovery trajectories. Feature selection algorithms (Lasso, Boruta, and Recursive Feature Elimination) were used to identify key predictors of sexual dysfunction (IIEF-5 ≤ 11) at T2, the time point with the highest dysfunction rate. Ten machine learning models were subsequently developed and evaluated. LPA revealed three distinct recovery trajectories: “Severe Impact with Slow Recovery” ( n = 53, 34.4%), “Major Impact with Fast Recovery” ( n = 40, 26.0%), and “Mild Impact with Fast Recovery” ( n = 61, 39.6%). The incidence of sexual dysfunction was highest at T2 (83/154, 53.9%), declining thereafter. All three feature selection algorithms consistently identified age, preoperative chemoradiation, tumor location, surgical approach, and surgical procedure as key predictors. Among the ten machine learning models, XGBoost demonstrated favorable performance in predicting T2 sexual dysfunction, achieving an area under the curve (AUC) of 0.866 (95% CI 0.802–0.931). SHapley Additive exPlanations (SHAP) analysis identified age as the most influential predictor. Sexual function in men after rectal cancer surgery follows distinct recovery trajectories. The XGBoost model, based on five readily available clinical predictors, is an internally validated tool for identifying patients at high risk for early severe sexual dysfunction. However, external validation in independent cohorts is required before clinical application.