Development of a machine learning prediction model for overactive bladder in female nurses: the NURS study

To develop and internally validate interpretable machine-learning models for identifying individuals with a higher probability of overactive bladder (OAB) among female nursing professionals.

A total of 13,191 female nurses participating in the Nurse Urinary Related Health Study were included and divided into training (70%) and testing (30%) cohorts. Seven distinct machine learning algorithms were implemented and evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and F1 score. The best-performing model was interpreted using SHapley Additive exPlanations (SHAP) to identify feature contributions. Recursive feature elimination based on SHAP importance scores was employed to develop a parsimonious model while maintaining optimal predictive performance.

OAB was observed in 11.2% of female nurses. The logistic regression algorithm outperformed all other models, demonstrating the highest predictive performance (validation AUC 0.751; 95% CI 0.729-0.774). Through systematic feature reduction, a streamlined 10-feature model was identified as the final predictive tool. The most influential predictors, ranked by mean SHAP value were: urine holding behavior, straining to void, sleep disorder, delayed voiding, perceived stress, fluid limitation, body mass index (BMI), anxiety, constipation, and parity. A nomogram was constructed to support practical implementation of the model.

Our study developed a clinically applicable risk assessment tool for estimating OAB probability in female nurses, with conventional logistic regression outperforming more complex machine learning algorithms. The parsimonious model provides a practical screening tool that could be integrated into occupational health programs.