Abstract
INTRODUCTION AND HYPOTHESIS: This study aimed to investigate the relationship between the Frailty Index (FI) and stress urinary incontinence (SUI) in women and to evaluate the impact of FI levels on SUI risk and the consistency of this relationship across different population characteristics. METHODS: Data were obtained from the NHANES 2005-2018 cycles. FI was assessed as a continuous, binary, and quartile variable. The outcome was stress urinary incontinence. Logistic regression models and restricted cubic spline analysis were used to examine associations between FI and SUI. Six machine learning models, such as Light Gradient Boosting Machine (LightGBM) and eXtreme Gradient Boosting (XGBoost), were developed using recursive feature elimination and cross-validation. Model performance was evaluated using area under receiver operating characteristic curve (AUC), calibration, and decision curve analysis. SHapley Additive exPlanations (SHAP) values were used for model interpretation. All analyses were conducted using R and Python. RESULTS: A total of 19,633 participants were included, among whom 5681 reported SUI. Compared to non-SUI participants, those with SUI were older, had higher body mass index (BMI) and frailty scores, and had a higher prevalence of hysterectomy, vaginal delivery, diabetes, and hypertension. Frailty scores were significantly higher in the SUI group across all metrics-continuous, binary, and quartiles. Logistic regression analysis revealed a robust association between higher frailty levels and increased SUI risk, which remained significant after adjusting for covariates. Women categorized as frail (FI > 0.2) had a 2.23-fold higher odds of SUI in the unadjusted model, and the risk remained elevated in fully adjusted models. A restricted cubic spline analysis suggested a nonlinear association, with a steeper increase in SUI risk when frailty scores were below 14. In machine learning analyses, recursive feature elimination identified frailty score, BMI, poverty index, age, and alcohol use as the top predictors. LightGBM achieved the most stable performance across training and validation sets and was chosen for further interpretation. SHAP analysis confirmed frailty score as the most influential feature. Higher values of frailty score, age, and BMI were associated with increased SUI risk, with nonlinear patterns observed in SHAP dependence plots. CONCLUSIONS: This study demonstrates that elevated FI levels are independently associated with an increased risk of SUI in women and identifies FI as a key nonlinear predictor, underscoring its potential utility in early screening and risk assessment.