OAB

OAB predicts anticholinergic treatment outcomes for patients with overactive bladder using machine learning applied to retrospective clinical data.

Key Features:

• Machine Learning Model: The core predictive model is a random forest trained to predict treatment failure among patients with OAB treated with anticholinergic drugs.
• Retrospective Dataset: The model was trained on a retrospective cohort of 559 female patients from January 2010 to December 2017.
• Cross-Validation: Model development used k-fold stratified cross-validation across patient strata defined by age and number of previously failed medications.
• Stratification by Age and Medication History: Patients were categorized into age groups (<40 years, 40–60 years, >60 years) and by prior anticholinergic treatment failures (0 or ≥1).
• External Validation: External validation was performed on a prospective dataset of 82 women collected from January 2018 to December 2018 at a different clinical site.
• Performance Metrics: Reported overall accuracy 80.3% (95% CI 79.1–81.3) with AUC 0.77 (95% CI 0.74–0.79), sensitivity 80.4% (95% CI 66.5–89.7%), and specificity 77.4% (95% CI 58.6–89.7%).
• Age-Specific Performance: Age-stratified performance included AUC 0.84 for women younger than 40 years and AUC 0.71 for women older than 60 years with prior medication failure.
• Data-Driven Approach: The model leverages retrospective clinical data and patient stratification to generate individualized predictions of anticholinergic treatment outcomes.

Scientific Applications:

• Treatment Outcome Prediction: Predicts the likelihood of anticholinergic treatment success or failure during a standard three-month trial period for patients with OAB.
• Treatment Optimization: Provides patient-level predictions to inform selection of anticholinergic therapy and reduce exposure to ineffective treatments.

Methodology:

Model construction used machine learning on a retrospective dataset with a random forest algorithm selected as the predictive model, refined by k-fold stratified cross-validation within age and medication-history strata, and externally validated on a prospective dataset.