The Unpredictability of Social Continence After Radical Prostatectomy
INTRODUCTION
Patients with prostate cancer (CaP) undergoing radical prostatectomy (RP) commonly develop urinary incontinence requiring the temporary use of pads. We sought to determine whether it is possible to individualize a patient's likelihood of experiencing short- and long-term "social continence" after RP--one pad change per day or fewer--to guide pre-operative counseling.
METHODS
The Michigan Urological Surgery Improvement Collaborative (MUSIC) is a consortium of 44 diverse urology practices that maintains a prospective registry of men with CaP. We identified men with patient-reported outcomes (PROs) for urinary incontinence at 3, 6, and 12 months after RP. We developed random forest models to predict incontinence at each of these time points based on information available prior to surgery, including age, race, baseline urinary function and QOL, body mass index, diabetes, prostate-specific antigen, biopsy Gleason score, clinical T-stage, number of positive and total cores on biopsy, prostate gland volume by ultrasound, and nerve-sparing surgery. We divided the practices into a training and test set. We assessed internal validity using 5-fold cross-validated (CV) area-under-the-curve (AUC) on the training practices and external validity by applying the model to the test set.
RESULTS
We identified men with PROs available after RP: 1,435 at 3 months, 1,068 at 6 months, and 705 at 12 months. We stratified the practices to allot 2/3rd of patients to the training set and 1/3rd to the test set. The models achieved modest discrimination in 5-fold cross-validation with poor external validity at 3 and 6 months, with the best test set discrimination at 12 months (AUC 0.71) (Figure 1). Incorporating the surgeon into the model led to a large improvement in 5-fold CV AUC to 0.70 at 12 months and smaller improvements at 3 (AUC 0.69) and 6 (AUC 0.66) months.
CONCLUSION
Social continence after RP appears to be difficult to predict based on demographic and tumor characteristics alone, and incorporating the surgeon into the model leads to the largest improvement in performance at 12 months. Understanding whether the unpredictability of social continence--particularly in the short term--is the result of key information lacking from the model or an imprecise outcome will be important in guiding how patients should be counseled pre-operatively.
Funding: Blue Cross and Blue Shield of Michigan and grant K12 DK111011 from the National Institute of Diabetes and Digestive and Kidney Disease.