Overactive bladder affects tens of millions of women yet remains chronically under-screened, partly because conventional risk tools ignore the reproductive biography that shapes pelvic floor function over a lifetime. A machine learning framework applied to a large nationally representative dataset now offers a more nuanced picture of who is at elevated risk — and which factors matter most.
Drawing on nearly 7,900 women from four cycles of the National Health and Nutrition Examination Survey (2011–2018), investigators used LASSO regression to winnow an initial pool of variables down to 14 clinically meaningful predictors, then built and compared multiple classification algorithms. SHAP (Shapley Additive Explanations) analysis — a method that quantifies each variable's marginal contribution to individual predictions — identified age, body mass index, the ratio of family income to the poverty threshold, age at menarche, and number of vaginal deliveries as the five strongest independent signals. Dose-response curves derived from restricted cubic splines revealed non-linear relationships between several of these factors and OAB probability, meaning risk does not climb in simple straight lines.
The inclusion of reproductive milestones alongside socioeconomic and anthropometric variables is the study's most notable methodological contribution. Prior clinical scoring systems have leaned heavily on age and BMI while treating parity as a binary yes/no; here, the number of vaginal deliveries carries graded weight. That said, several limitations temper enthusiasm. The NHANES design is cross-sectional, so causality cannot be established, and self-reported urinary symptoms introduce recall bias. The model has not been externally validated in a prospective cohort or in populations outside the United States. OAB diagnosis in this dataset also relies on symptom questionnaires rather than urodynamic confirmation. This work is best characterized as incremental but directionally useful — it demonstrates that machine learning can surface overlooked reproductive predictors at population scale, a template that prospective screening tools could build upon.