Risk prediction for 12-month post-prostatectomy urinary incontinence to support survivorship care: development and same-center temporal validation of a perioperative model
- 1Department of Urology, The Second Affiliated Hospital of Guizhou University of Traditional Chinese Medicine
- 2Department of Urology, Guiyang First People’s Hospital
Abstract
Background: Persistent post-prostatectomy urinary incontinence (PPI) is a survivorship problem after radical prostatectomy with continuing implications for daily function, rehabilitation, follow-up, and supportive care. Risk stratification could help target these services, but uncommon events constrain model complexity and require evaluation beyond discrimination.
Objective: To develop a perioperative model for PPI 12 months after catheter removal using LASSO-penalized logistic regression, compare it with a constrained gradient boosting machine (GBM), and examine temporal performance in a later same-center cohort.
Methods: The development cohort comprised 726 patients with 51 PPI events (7.0%) and 19 candidate predictors. LASSO tuning used the one-standard-error rule within resampling, and the strategies were compared by nested repeated stratified five-fold cross-validation. The selected LASSO model was refitted to the full development cohort and applied without coefficient updating or recalibration to 465 later patients with 38 events (8.2%). Evaluation included ROC-AUC, average precision, Brier score, calibration-in-the-large (CITL), calibration slope, threshold-dependent measures, and decision-curve analysis.
Results: Eight original predictors were retained: age, preoperative International Prostate Symptom Score, ln(1 + prostate-specific antigen), operative time, postoperative hospital stay, smoking history, hypertension, and neurovascular bundle preservation. Nested internal validation yielded a ROC-AUC of 0.913 (95% CI 0.871–0.946), average precision of 0.542 (0.404–0.675), and Brier score of 0.046 (0.035–0.058) for LASSO, versus 0.896, 0.488, and 0.048 for constrained GBM. In the temporal cohort, ROC-AUC was 0.836 (0.763–0.900), average precision 0.409 (0.277–0.564), and Brier score 0.062 (0.045–0.081). CITL was 0.445 (0.012–0.808) and the calibration slope 0.850 (0.621–1.174), indicating average underprediction. At the development-derived threshold of 0.128, sensitivity was 0.500 and specificity 0.913. Decision-curve analysis indicated potential net benefit across thresholds of 2–30%.
Conclusion: The LASSO model retained useful temporal discrimination but underestimated absolute risk, and the development-derived cutoff did not transport well. It is therefore better suited to continuous risk estimation for survivorship follow-up and rehabilitation planning than to a fixed binary triage rule. Independent geographic validation, local calibration assessment, and prospective evaluation of clinical impact are required before routine use.
Related
- Development and validation of machine learning models for preoperative prediction of perioperative outcomes in patients with thymic epithelial tumors
- Perioperative tranexamic acid and postoperative haematuria after transvesical prostatectomy: a prospective observational cohort study
- Supporting implementation of the national standards for cancer survivorship care: development of the cancer survivorship maturity model (CSMM)
- Interpretable Machine Learning Prognostic Model Integrating Inflammatory-Nutritional Biomarkers for Post-Prostatectomy Biochemical Recurrence: Development, Internal Validation, and Translational Target Exploration.
- The ''Hamburg Life After Cancer Program'': Development, Implementation and Evaluation of a Structured Survivorship Program - Study Protocol of a Hybrid Effectiveness-Implementation Study
BibTeX
@article{tang2026risk,
title = {{Risk prediction for 12-month post-prostatectomy urinary incontinence to support survivorship care: development and same-center temporal validation of a perioperative model}},
author = {Lang Tang and Ming Hao and Mengru Luo and Shizhu Li},
journal = {Frontiers in Public Health},
year = {2026},
month = oct,
doi = {10.3389/fpubh.2026.1968281},
url = {https://www.frontiersin.org/articles/10.3389/fpubh.2026.1968281/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Risk prediction for 12-month post-prostatectomy urinary incontinence to support survivorship care: development and same-center temporal validation of a perioperative model
AU - Lang Tang
AU - Ming Hao
AU - Mengru Luo
AU - Shizhu Li
PY - 2026
DA - 2026/10/21
JO - Frontiers in Public Health
DO - 10.3389/fpubh.2026.1968281
UR - https://www.frontiersin.org/articles/10.3389/fpubh.2026.1968281/full
AB - Background: Persistent post-prostatectomy urinary incontinence (PPI) is a survivorship problem after radical prostatectomy with continuing implications for daily function, rehabilitation, follow-up, and supportive care. Risk stratification could help target these services, but uncommon events constrain model complexity and require evaluation beyond discrimination. Objective: To develop a perioperative model for PPI 12 months after catheter removal using LASSO-penalized logistic regression, compare it with a constrained gradient boosting machine (GBM), and examine temporal performance in a later same-center cohort. Methods: The development cohort comprised 726 patients with 51 PPI events (7.0%) and 19 candidate predictors. LASSO tuning used the one-standard-error rule within resampling, and the strategies were compared by nested repeated stratified five-fold cross-validation. The selected LASSO model was refitted to the full development cohort and applied without coefficient updating or recalibration to 465 later patients with 38 events (8.2%). Evaluation included ROC-AUC, average precision, Brier score, calibration-in-the-large (CITL), calibration slope, threshold-dependent measures, and decision-curve analysis. Results: Eight original predictors were retained: age, preoperative International Prostate Symptom Score, ln(1 + prostate-specific antigen), operative time, postoperative hospital stay, smoking history, hypertension, and neurovascular bundle preservation. Nested internal validation yielded a ROC-AUC of 0.913 (95% CI 0.871–0.946), average precision of 0.542 (0.404–0.675), and Brier score of 0.046 (0.035–0.058) for LASSO, versus 0.896, 0.488, and 0.048 for constrained GBM. In the temporal cohort, ROC-AUC was 0.836 (0.763–0.900), average precision 0.409 (0.277–0.564), and Brier score 0.062 (0.045–0.081). CITL was 0.445 (0.012–0.808) and the calibration slope 0.850 (0.621–1.174), indicating average underprediction. At the development-derived threshold of 0.128, sensitivity was 0.500 and specificity 0.913. Decision-curve analysis indicated potential net benefit across thresholds of 2–30%. Conclusion: The LASSO model retained useful temporal discrimination but underestimated absolute risk, and the development-derived cutoff did not transport well. It is therefore better suited to continuous risk estimation for survivorship follow-up and rehabilitation planning than to a fixed binary triage rule. Independent geographic validation, local calibration assessment, and prospective evaluation of clinical impact are required before routine use.
SN - 2296-2565
ER -