AI based prediction of post prostatectomy urinary incontinence and its impact on quality of life: development and validation study

Urinary incontinence can affect up to 53% of patients after radical prostatectomy, yet prediction tools rely on binary outcomes, missing heterogeneous recovery patterns. We developed machine-learning models that for the first time predict incontinence presence, severity, and quality-of-life impact at 3 and 12 months post-surgery. XGBoost models were trained on 21 perioperative features from 2,586 patients (2018–2021) and validated on 728 patients (2022). For patients incontinent at 3 months (n = 962), 12-month predictions achieved strong performance (AUC 0.82–0.86), with early quality-of-life impact (SHAP 1.4) outperforming symptom severity (SHAP 1.0) as a predictor. Three-month predictions using only baseline features showed moderate performance (AUC 0.52–0.59). Patients with identical total ICIQ-UI scores demonstrated divergent recovery trajectories when symptom severity and quality-of-life components were analyzed separately, with some achieving continence while others remained severely affected. Decomposing outcomes into multidimensional components rather than using aggregate scores enables personalized prediction of continence recovery, facilitating targeted counseling and rehabilitation strategies.