Logistic regression versus C5.0 decision tree in predicting female stress urinary incontinence risk using clinical and pelvic floor ultrasound data: a model development and validation study.

BACKGROUND: Stress urinary incontinence (SUI) seriously affects patients’ daily lives. Early identification of risk factors for SUI has positive social and clinical significance. This study aimed to develop and externally validate predictive models for female SUI using clinical data and pelvic floor ultrasound (PFU) parameters through multivariate logistic regression (MLR) and the C5.0 decision tree (DTC5.0). METHODS: The training cohort comprised basic data and PFU parameters from 291 patients across two hospitals. Risk factors for SUI were identified through differential analysis and incorporated into MLR and DTC5.0 models to predict SUI risk. An external validation cohort of 115 patients was obtained via telephone follow-up. Model performance was evaluated and compared in terms of the AUC, accuracy, specificity, sensitivity, Youden index, positive predictive value (PPV), and negative predictive value (NPV). RESULTS: Both models identified five key predictors: vaginal delivery, hysterectomy, bladder neck mobility, urethral funnel formation, and perineal body hypermobility. In the training cohort, the MLR model achieved an AUC of 0.873 (95% CI: 0.832–0.915), with an accuracy of 80.4% (75.8%–85.0%), a sensitivity of 90.6% (85.9%–95.5%), a specificity of 71.1% (63.9%–78.3%), a Youden index of 61.7% (53.2%–70.4%), a PPV of 74.1% (67.5%–80.7%), and an NPV of 89.3% (83.8%–94.8%) at the optimal cut-off. The DTC5.0 model yielded an AUC of 0.873 (0.832–0.913), with an accuracy of 80.4% (75.8%–85.0%), a sensitivity of 84.9% (79.0%–90.8%), a specificity of 76.3% (69.6%–83.0%), a Youden index of 61.2% (52.2%–70.2%), a PPV of 76.6% (69.9%–83.3%), and an NPV of 84.7% (78.7%–90.7%). In the validation cohort, the AUC was 0.775 (0.686–0.864) for MLR and 0.719 (0.624–0.813) for DTC5.0. CONCLUSIONS: Both MLR and the DTC5.0 show similar efficacy in predicting SUI risk. The MLR model, which incorporates PFU and clinical characteristics, may be more suitable for the clinical screening of high-risk SUI women because of its greater generalizability.