Nurse-Led Enhanced Recovery and Explainable Machine Learning Prediction of Rehabilitation Adherence after Robot-Assisted Radical Prostatectomy

This single-center retrospective cohort study evaluated rehabilitation adherence and early continence recovery within a standardized nurse-led enhanced recovery care pathway after robot-assisted radical prostatectomy (RARP) and developed an interpretable model for poor 3-month adherence. The analysis included 168 men aged ≥50 years who underwent RARP from June to December 2023 and completed follow-up through June 2024. The pathway combined preoperative education, pelvic floor muscle training (PFMT), catheter and urinary symptom management, psychological support, discharge planning, home rehabilitation recording, and structured follow-up. Poor adherence was defined as a 3-month composite score <80/100. Logistic regression, least absolute shrinkage and selection operator regression, random forest, support vector machine, and extreme gradient boosting (XGBoost) were evaluated using repeated stratified fivefold cross-validation; Shapley additive explanations (SHAP) were used to interpret the final model. At 3 months, 112 patients (66.7%) had adequate adherence and 56 (33.3%) had poor adherence. At 6 months, the adequate-adherence group had lower ICIQ-UI SF scores (3.4 ± 2.5 versus 7.2 ± 3.4), higher EPIC-26 urinary function scores (83.1 ± 9.5 versus 67.5 ± 12.2), and a higher no-pad rate (78% versus 35%). Poor adherence was also associated with lower baseline PFMT self-efficacy, higher anxiety, and less frequent 1-month diary completion (all P < 0.001). XGBoost achieved the best internal performance (AUC, 0.89; accuracy, 0.84; sensitivity, 0.79; specificity, 0.87; F1-score, 0.77; Brier score, 0.12). Adequate rehabilitation adherence was associated with faster early continence recovery and better prostate cancer-specific quality of life. The interpretable landmark model identified modifiable nursing-related factors that may support risk-stratified follow-up, but external validation and prospective impact evaluation are required before clinical implementation.