Related Experiment Video
Updated: Jul 15, 2026
Evaluation of Biomaterials for Bladder Augmentation using Cystometric Analyses in Various Rodent Models
Published on: August 9, 2012
Comparative Analysis of Logistic Regression and Machine Learning Algorithms for Predicting De Novo Stress Urinary
Liying He1, Yin Chen2, Qiang Ma1
1Department of Ultrasound, The First Affiliated Hospital of Wannan Medical College (Yijishan Hospital of Wannan Medical College), Wuhu, Anhui, People's Republic of China.
Background:
Whether to perform concurrent anti-incontinence surgery (AIS) during pelvic floor reconstruction (PFR) remains controversial. Traditional logistic regression (LR) models for predicting postoperative stress urinary incontinence (SUI) are limited by their inability to capture complex non-linear relationships and high-order interactions among predictors. Machine learning (ML) algorithms offer advantages in modeling such complexities, but existing ML models often lack interpretability and external validation.
Objective:
This study aims to develop and validate an interpretable ML model using multicenter data to predict de novo SUI or symptomatic worsening after PFR, comparing performance with traditional LR and seven ML algorithms.
Methods:
Clinical and pelvic floor ultrasound data from 490 women who underwent PFR between Jan 2019 and Jan 2024 were retrospectively analyzed. Univariate and multivariate logistic regression analyses identified independent predictors and constructed a conventional model. Seven ML models were developed. Model performance was assessed using area under the curve (AUC), calibration, and decision curve analysis.
Results:
A total of 590 patients were enrolled, of whom 155 (26.3%) developed de novo SUI or symptomatic worsening. The conventional logistic regression model identified seven independent predictors; while least absolute shrinkage and selection operator regression selected 11 optimal features. Among all models, the Multilayer Perceptron (MLP) demonstrated the best predictive performance, achieving an AUC of 0.922 (95% CI: 0.880-0.957) in internal validation and 0.893 (95% CI: 0.844-0.936) in external testing. DeLong test revealed significant AUC differences among models (all P < 0.05). SHapley Additive exPlanations (SHAP) analysis identified concurrent AIS, body mass index, urethral rotation angle, Valsalva retrovesical angle, and age as the most influential predictors.
Conclusion:
The MLP model provides accurate and clinically meaningful predictions of postoperative SUI risk following PFR, demonstrating strong discrimination, good calibration, and potential clinical utility. This model may support personalized surgical planning and decision-making.
