INTRODUCTION AND OBJECTIVE:
Incontinence is common following radical prostatectomy (RP) for localized CaP. We sought to determine whether inclusion of data derived from preoperative multiparametric prostate MRI (mpMRI) improves the prediction of social continence, defined as ≤1 pad change per day, at 3, 6, 12, and 24 months following RP.
The Michigan Urological Surgery Improvement Collaborative (MUSIC) is a consortium of 45 diverse urology practices that maintains a prospective registry of men with CaP. Using mpMRI data from 3 large hospitals linked to MUSIC registry data, we fit random forest models to predict patient-reported social continence at 3, 6, 12, and 24 months following RP. We compared models using baseline clinical predictors (age, body mass index, grade group, PSA, EPIC-26 sexual domain score, baseline social continence, urinary incontinence score), prospectively reported MRI predictors (membranous urethra length, maximum PI-RADS v2 score, median lobe size, prostate volume), and both. We assessed model discrimination using a 10-fold cross-validated C-statistic.
We identified 1,021 patients at 3 months, 897 patients at 6 months, 861 at 12 months, and 526 at 24 months, in whom incontinence was present in 27%, 14%, 9.8%, and 6.3%, respectively. The combined models had strong discriminative performance at all time horizons, with C-statistics of 0.82, 0.85, 0.93, and 0.86 at 3, 6, 12, and 24 months (Table). Adding MRI variables did not improve the model over clinical factors alone.
Models using clinical factors alone achieved very high discriminative performance in the prediction of post-RP continence. Compared to clinical variables alone, adding prospectively reported preoperative mpMRI data did not significantly improve prediction of social continence in patients following RP.
Source of Funding:
Blue Cross and Blue Shield of Michigan