Body mass index mediates the association between heavy metal exposure and overactive bladder risk: insights from a nationally representative cross-sectional study using explainable machine learning

Introduction

Overactive bladder (OAB) is a clinical disorder marked by a constellation of storage symptoms, notably urgency, which may be accompanied by elevated daytime urinary frequency, urge incontinence, and nocturia (1). As indicated by the US National Overactive Bladder Evaluation (NOBLE) program, OAB prevalence in male adults is 16%, while in females it is 16.9% (2). An epidemiological study in Europe from 2011 uncovered that approximately 36% of men and 43% of women aged over 40 exhibit OAB symptoms (3). The condition significantly impacts both the physical and mental health of affected individuals, considerably diminishing their quality of life (4). Furthermore, OAB exerts a considerable economic cost on society, with medical costs for patients in the United States exceeding 2.5 times those of individuals without OAB symptoms (5).

In recent years, environmental metal pollution has emerged as a significant factor in the etiology of OAB. Environmental contamination by chemicals, originating from both natural and human-related sources, has led to increasing concerns regarding the presence of heavy metals like lead (Pb), thallium (Tl), mercury (Hg), arsenic (As), and cadmium (Cd). These metals are of particular interest due to their widespread distribution and non-biodegradable nature (6). Exposure to heavy metals arises via multiple pathways, comprising food, water, air, and personal care products (7). The accumulation of these metals in the human body over time results in the production of reactive oxygen species (ROS), inducing oxidative stress that disrupts normal physiological functions (8). Moreover, heavy metals exposure could contribute to lower urinary tract symptoms, including urinary incontinence (9). While metals like Hg, Pb, Cd, and As have been identified as nephrotoxic and potential impairers of renal function (10), further investigation into the effects of other heavy metals on OAB remains limited. One study has suggested a complex, nonlinear relationship between blood Cd levels and OAB symptoms in middle-aged and elderly individuals in the United States (11). Future research exploring the connection between heavy metal exposure and OAB will offer valuable insights into the pathophysiology of this condition.

Advancements in machine learning (ML) technology present promising avenues for the precise diagnosis of OAB and the identification of key environmental metal exposures associated with the disorder. Traditional statistical methods often necessitate well-structured and high-quality datasets for effective disease identification and face significant challenges in modeling complex, nonlinear relationships (12). In contrast, ML algorithms require minimal data preprocessing and are adept at handling large volumes of unstructured data, offering novel solutions for disease diagnosis and risk assessment (13). By leveraging interpretable ML techniques, researchers can elucidate the intricate interactions between various variables, thereby enhancing the understanding of OAB pathogenesis. Consequently, the construction of diagnostic models based on ML to assess OAB-related environmental metal exposures holds substantial promise for advancing both research in this domain and its clinical application. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-350/rc).

Methods

Participants of the study

The National Health and Nutrition Examination Survey (NHANES) employs a variety of strategies to monitor the U.S. population via physical examinations and interviews. This study’s sample is based on NHANES data (https://www.cdc.gov/nchs/nhanes) collected over three consecutive cycles from 2005 to 2010. The exclusion criteria were as follows: (I) missing data on urine and blood heavy metal for participants; (II) lack of information confirming the status of OAB based on NHANES questionnaire data; and (III) missing values for covariates related to the participants. Ultimately, 3,201 participants were enrolled in this study with the flow chart exhibited in Figure S1. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Heavy metal concentration measurement

This study analyzes eight heavy metals in blood, including cadmium (Cd), calcium (Ca), mercury (Hg), lead (Pb), phosphorus (P), iron (Fe), sodium (Na), and potassium (K), and twelve heavy metals in urine: barium (Ba), cobalt (Co), cadmium (Cd), cesium (Cs), lead (Pb), antimony (Sb), molybdenum (Mo), thallium (Tl), tungsten (W), mercury (Hg), uranium (U), and arsenic (As) (14). Heavy metal concentrations were analyzed via inductively coupled plasma mass spectrometry with dynamic reaction cell (ICP-DRC-MS) at the National Centre for Environmental Health under rigorous quality control.

Definition of OAB

According to our previous research, all participants calculated their OAB symptom scores based on OAB scoring data. When the OAB symptom score reaches or exceeds 3, participants are diagnosed with OAB (15).

Covariates

With the aim of addressing the effects of confounding factors on the findings, this study adopted methodologies from previous research and adjusted for several covariates in the data analysis (15-18). The covariates included gender (male/female), age, race (non-Hispanic White, Mexican American, non-Hispanic Black, and other race), poverty-to-income ratio (PIR), body mass index (BMI), marital status (married or living with a partner/living alone), educational attainment (high school or less/more than high school), and lifestyle factors such as smoking and alcohol consumption. Smoking status classifications included never smoked, former smoker, and current smoker. Alcohol consumption was classified into five categories: never consumed alcohol, mild drinker, moderate drinker, heavy drinker, and former drinker. Additionally, the study considered participants’ medical history, including diabetes mellitus (DM, yes/no), hypertension (yes/no), cardiovascular disease (CVD, yes/no), and depression (yes/no). Finally, we also assessed the impact of other covariates, such as urinary creatinine, energy intake, high-density lipoprotein cholesterol (HDL-C), and total cholesterol (TC). Detailed information could be obtained in Table S1.

Metal feature selection

Initially, the metal data underwent a logarithmic transformation to enhance its normality. Pearson correlation analysis was carried out for examining the relationships among the 20 metal variables. To reduce dimensionality and mitigate the risk of overfitting, we employed feature selection techniques, including the Boruta algorithm and the least absolute shrinkage and selection operator (LASSO) regression model, to identify the metal features with the strongest relevance to OAB. Specifically, the Boruta algorithm assesses whether the importance of each feature is significantly greater than that of random features (i.e., shadow variables) by iteratively fitting a Random Forest (RF) model until all predictive features are categorized as “confirmed”, “tentative”, or “rejected” (19). The features deemed “confirmed” were utilized in subsequent analyses. In this study, the number of iterations was set to 500. For each iteration, the importance of the original variables was compared with that of their shadow variables until the importance assessment of all variables stabilized. LASSO regression applies constraints to the absolute values of the coefficients by adding an L1 regularization penalty term, effectively forcing the coefficients of unimportant variables to converge to zero for feature selection purposes (20,21). In this study, we selected the maximum lambda value (lambda.1se) within one standard error of optimal model performance to strike a balance between model complexity and explanatory power. Finally, we employed the variance inflation factor (VIF) to assess collinearity in the final model, considering a VIF value greater than 10 as indicative of multicollinearity, necessitating exclusion from model construction.

Construction of the ML models

Participants were split into a training set (N=2,241) and a test set (N=960) at a 7:3 ratio. We identified OAB resulting from metal exposure using various ML models, including logistic regression (LR), linear discriminant analysis (LDA), K-Nearest Neighbor (KNN), quadratic discriminant analysis (QDA), Naive Bayes (NB), RF, Extreme Gradient Boosting (XGBoost), Decision Tree (DT), Support Vector Machines (SVM), and Light Gradient Boosting Machine (LightGBM). To ensure the optimal performance and reliability of the models, we hyperparameter-tuned each ML model using 10-fold cross-validation. To further evaluate the model’s ability to predict OAB, we utilized the following metrics: area under the curve (AUC) of the receiver operating characteristic (ROC), average precision score (APS), accuracy, recall, positive predictive value (PPV), negative predictive value (NPV), false positive rate (FPR), false negative rate (FNR), precision, and F1 score. A flowchart outlining the methodology for the ML process is presented in Figure S2.

A variety of interpretable methods were combined with the top-performing ML predictive models to elucidate the potential correlations between heavy metal exposure and OAB. To begin with, we performed feature importance analyses with the Shapley Additive Explanations (SHAP) method and the Permutation Feature Importance (PFI) analysis to identify the variables that significantly influence OAB risk prediction. The SHAP was applied for individualized decision analysis to depict patient profiles utilized for outcome forecasting while demonstrating the capacity of ML-based predictive models to optimize individualized care strategies. PFI analysis quantifies the influence of each feature on the accuracy of the model by disrupting feature values and observing resultant shifts in model performance. In contrast to SHAP analysis, PFI analysis offers a more intuitive measurement of how each heavy metal exposure variable affects the prediction of OAB risk (22). Additionally, Partial Dependence Plot (PDP) analysis and Accumulated Local Effects (ALE) analysis are critical methods employed to interpret the model. PDP evaluates the mean variation in OAB risk prediction by gradually varying the target feature values from minimum to maximum under fixed conditions of other variables. In contrast, ALE assesses the local effect of each feature by partitioning it and measuring how changes in the feature’s value impact the output, concentrating on trends in the local area while effectively mitigating the model bias that can arise in PDP.

Statistical analysis

This study systematically compared the demographic characteristics of patients with OAB to those without OAB. Given the intricate sampling design of the NHANES survey, the research team applied multistage weight adjustments to the data to ensure statistical accuracy and the nationwide representativeness of the study findings. We described continuous variables as weighted means ± standardized error (SE) and used frequencies (weighted percentages) for categorical variables. To evaluate between-group disparities, weighted t-tests and weighted Chi-squared tests were used. We established univariate and multivariate regression models to assess the relationships of the metals with OAB risks. The metals were incorporated into the models as both continuous and categorical variables. The model specifications are as follows: Model 1 represents the unadjusted model, whereas Model 2 comprises the adjusted model, which accounts for factors including age, sex, race, education level, and marital status. Model 3 further adjusted for BMI, PIR, smoking status, drinking status, TC, HDL-C, urinary creatinine, total energy intake, depression, CVD, DM, and hypertension based on model 2. Furthermore, restricted cubic splines (RCS) were used to assess non-linear links between the metals and OAB risks. A Bayesian kernel machine regression (BKMR) model was developed to visualize the dose-response relationship between combined exposure to heavy metal mixtures and the risk of OAB. The BKMR model’s strength lies in its flexibility to capture complex exposure effects, including nonlinear correlations and interactions among the exposure components, which are common in environmental health studies (23,24). Parameter estimation was conducted using a quasi-Bayesian Monte Carlo algorithm, achieving 5,000 iterative Markov chain Monte Carlo (MCMC) samples under normal approximation. All models were adjusted for potential confounders, including demographic characteristics and clinical indicators identified in model 3. Feature selection was performed based on conditional posterior inclusion probabilities (conPIPs) to identify the heavy metal components that significantly contribute to OAB risk, while the marginal effects of individual heavy metals were assessed through stratified analysis. Additionally, mediation analyses were performed to investigate the mediating role of BMI in the association between metals and OAB risks. To perform statistical analyses, R software (version 4.4.2) was used. Statistical significance was set as P<0.05.

Results

Participants’ demographic characteristics

As presented in Table 1, this study involved a cohort of 3,201 participants with an average age of 48.85 years, and 669 of them were diagnosed with OAB. Females accounted for 54.95% of the total participants, slightly higher than the 45.05% of males. Only three variables—marital status, TC, and HDL-C—did not demonstrate noticeable distinctions between the OAB and non-OAB groups (P>0.05). Additionally, we reported the heavy metal levels from blood and urine tests of the samples, which indicated that blood cadmium, blood lead, blood calcium, blood potassium, blood iron, urinary barium, urinary cadmium, urinary lead, and urinary mercury levels significantly differed between the OAB and non-OAB populations.

Correlation analysis and independent variates selection

We examined the correlations among 20 heavy metals using Pearson correlation analysis. Figure S3 illustrates the strong correlations between these heavy metals, especially between urinary cesium and thallium (r=0.79), urinary cesium and molybdenum (r=0.60), and urinary cesium and lead (r=0.59). The Boruta algorithm identified a total of 18 ‘confirmed’ metal variables (Figure S4A), while the LASSO regression analysis identified six metal signatures strongly associated with OAB (Figure S4B,S4C). Ultimately, we selected metal variables that were jointly identified by both methods, including blood lead, blood iron, urinary barium, urinary cadmium, urinary thallium, and urinary mercury, along with 17 covariates used for ML model construction. Additionally, we assessed multicollinearity for all identified metals and their covariates with the VIF, and the results revealed that no multicollinearity issues were detected (Table S2).

Prediction model for OAB risk

Figure 1 presents the ROC curves for various ML models, including LR, LDA, QDA, NB, KNN, DT, RF, XGBoost, SVM, and LightGBM. The AUC values for these models in the test set are as follows: 0.735 (LR), 0.733 (LDA), 0.710 (QDA), 0.728 (NB), 0.675 (KNN), 0.500 (DT), 0.731 (RF), 0.736 (XGBoost), 0.653 (SVM), and 0.722 (LightGBM). Figure S5 illustrates the confusion matrices for these ten ML algorithms based on the test set. To facilitate a comprehensive evaluation, we compare their evaluation metrics, as shown in Table 2. The AUC serves as a key criterion to choose the most effective ML model. Consequently, we selected the XGBoost model, which exhibits the highest AUC value, as the basis for subsequent model interpretation. The XGBoost model achieves an APS of 0.895, an accuracy of 0.758, a precision of 0.770, and an F1 Score of 0.860.

Figure 1 ROC curves for the 10 machine learning models on training and test datasets. The x-axis represents the FPR, and the y-axis represents the TPR, both scaled from 0 to 1. The dashed diagonal line (y = x) indicates random classifier baseline. Colored curves depict model-specific ROC trajectories, where greater separation above the diagonal reflects stronger discriminatory power. AUC, area under the curve; DT, Decision Tree; FPR, false positive rate; KNN, K-Nearest Neighbor; LDA, linear discriminant analysis; LightGBM, Light Gradient Boosting Machine; LR, logistic regression; NB, Naive Bayes; QDA, quadratic discriminant analysis; RF, Random Forest; ROC, receiver operating characteristic; SVM, Support Vector Machines; TPR, true positive rate; XGBoost, Extreme Gradient Boosting.

Interpretation of the model analysis

Figure 2A depicts the ranking of feature importance in the XGBoost model, calculated by PFI analysis. The findings indicate that hypertension, urinary cadmium, age, HDL-C, and depression are the top five most significant factors influencing model performance, suggesting a strong association of these variables with OAB. Additionally, we employed the SHAP method to interpret the output of the XGBoost model. Figure 2B displays the SHAP importance plot, with features arranged in descending order of significance. As illustrated in Figure S6A, a positive SHAP value signifies a direct correlation with OAB risk, while a negative value reflects an inverse correlation. Higher SHAP values suggest a more significant impact on model predictions, with brighter colors indicating larger values. The findings imply that several factors, including age, BMI, depression, blood iron, and urinary cadmium, influence the likelihood of OAB risk. Figure S6B presents the characteristics of the 7th randomly selected subject along with their corresponding risk scores, ranked by importance. Observations in this figure indicate that BMI, age, and blood iron play a more significant role in OAB risk than other characteristics. For instance, Figure S6B demonstrates that the risk of developing OAB is reduced by a factor of 0.323 when the blood iron exposure concentration is logarithmically transformed to 1.1 µmol/L.

Figure 2 Explanation of model importance by significant variables. (A) A summary forest plot based on PFI analysis illustrates the weights associated with each variable’s contribution to overactive bladder. (B) The SHAP features importance plot. BMI, body mass index; CVD, cardiovascular disease; DM, diabetes mellitus; HDL-C, high-density lipoprotein cholesterol; PFI, Permutation Feature Importance; PIR, poverty-to-income ratio; SHAP, Shapley Additive Explanations; TC, total cholesterol.

Relationships between the important metals and OAB

Figure 3 presents the PDP for six metals significantly related to OAB risk in the prediction model. These metals include blood lead, blood iron, urinary barium, urinary cadmium, urinary thallium, and urinary mercury. The PDP analysis indicated that exposure to heavy metals at low concentrations, such as blood iron, urinary barium, and urinary thallium, had a small effect on the risk of OAB. However, a decreasing trend in OAB risk was observed with increasing exposure degrees to the aforementioned heavy metals. Conversely, the expected risk of OAB displayed a positive correlation with elevated blood lead and urinary cadmium levels. Additionally, increased exposure levels of urinary mercury did not demonstrate significant associations of the risk of OAB. The ALE analysis revealed consistent findings, demonstrating that in contrast to other heavy metals, the localized accumulation of blood lead and urinary cadmium significantly contribute to the risk of OAB at high levels (Figure S7).

Figure 3 PDP analysis of risk of overactive bladder and heavy metal levels. OAB, overactive bladder; PDP, partial dependence plot.

LR analyses and RCS analyses

In unadjusted LR analyses, elevated levels of blood lead and urinary cadmium were significantly associated with an increased risk of OAB, whereas higher levels of urinary barium, blood iron, urinary thallium, and urinary mercury were linked to a reduced OAB risk (Table 3). After full adjustment for confounders (Model 3), significant associations persisted only for blood iron [odds ratio (OR) =0.72, 95% confidence interval (CI): 0.57–0.90] and urinary cadmium (OR =1.23, 95% CI: 1.06–1.42). When comparing the highest (Q4) to the lowest (Q1) exposure quartiles, blood lead (OR =1.53, 95% CI: 1.09–2.14) and urinary cadmium (OR =1.69, 95% CI: 1.19–2.40) remained positively associated with OAB risk, while blood iron showed an inverse association (OR =0.69, 95% CI: 0.52–0.92; all P for trend <0.05).

The dose-response relationships between heavy metals and OAB risk are shown in Figure 4. Significant dose-response associations were observed for urinary cadmium, blood iron, and blood lead with OAB risk (P overall <0.05). A linear relationship was identified for urinary cadmium and blood iron (P nonlinear >0.05), whereas blood lead exhibited a nonlinear dose-response pattern with OAB risk (P nonlinear <0.05).

Figure 4 RCS plots of the association between metals (ln-transformed) and OAB risk. CI, confidence interval; OAB, overactive bladder; OR, odds ratio; RCS, restricted cubic splines.

Association of the important metals with OAB by BKMR model

The BKMR analysis revealed that the combined heavy metal exposure was significantly associated with OAB risk. Based on conPIPs assessment (threshold: >0.5), urinary cadmium (conPIPs =0.986), blood iron (conPIPs =0.565), and blood lead (conPIPs =0.508) were identified as key risk components. OAB risk increased significantly when the combined exposure level of the heavy metal mixture exceeded the 55th percentile (Figure 5A). Stratified analyses demonstrated that urinary cadmium remained consistently positively associated with OAB risk after adjusting for other heavy metals at 25th, 50th, and 75th percentiles (Figure 5B). Single-component exposure-response curves further indicated that urinary cadmium and blood lead were positively associated with OAB when other metals were fixed at their median concentrations, whereas blood iron exhibited a protective effect (Figure 5C). No significant interactions between the heavy metals were observed (Figure 5D).

Figure 5 Associations between metals (ln-transformed) and OAB risk by BKMR model. (A) Joint effect of metals mixture on OAB risk. (B) Single chemical-exposure effect (95% CI) to OAB risk when other chemicals were fixed at a specific quantile (25th, 50th, 75th). (C) Univariate exposure-response functions between exposure to metals and OAB risk. (D) Bivariate exposure-response relationship between metals and OAB risk (a visualization for evaluating interactions). BKMR, Bayesian kernel machine regression; CI, confidence interval; OAB, overactive bladder.

Mediation analysis

Table 4 illustrates the mediating role of BMI in the association between blood iron, blood lead, urinary cadmium, and OAB risk. BMI mediated 14.80% of the effect for blood iron, while it accounted for −17.33% and −17.33% in the associations for blood lead and urinary cadmium with OAB, respectively.

Discussion

This study introduces a novel methodological framework integrating explainable ML with conventional epidemiological approaches to elucidate interactions between heavy metal exposure and OAB pathogenesis. Our two-phase analytical design employed ML for hypothesis generation followed by covariate-adjusted validation, overcoming inherent limitations of traditional single-pollutant models. The XGBoost algorithm demonstrated strong predictive performance (AUC =0.736), with SHAP identifying urinary cadmium and blood iron as biologically plausible determinants. Metals selected through Boruta-LASSO filtering underwent comprehensive evaluation using restricted cubic spline-based dose-response analysis, BKMR for mixture effects, and causal mediation testing. This synthesis revealed BMI’s dual role as both a partial mediator (accounting for 14.80% of iron’s protective effect) and effect modifier (amplifying cadmium/lead risks by 17.33%), offering mechanistic evidence for adiposity-mediated modulation of metal toxicity in bladder regulation. By synergizing predictive analytics with pathway characterization, this approach advances environmental epidemiology methodology for complex multi-pollutant exposure research. Nevertheless, causal inference is precluded by the cross-sectional study design.

Previous studies provided limited evidence regarding the risk of OAB and heavy metals (25,26). A recent epidemiological investigation that included middle-aged and elderly individuals in the U.S. identified a nonlinear correlation between blood cadmium and OAB (11). The connection between mixed metal exposure and stress urinary incontinence (SUI) is well-established, with research indicating that exposure to mixtures of heavy metals like lead, cadmium, and mercury may elevate the SUI risk in women (27,28). Furthermore, as blood cadmium levels increase, women have a higher probability of developing SUI, with risk peaking at 4 µg/liter (29). However, direct studies linking heavy metals and OAB remain scarce. In this research, we conducted the first comprehensive analysis of the effects of various heavy metal exposures on OAB risk and found that blood iron emerged as a promising predictor of OAB risk. Our study provides initial evidence revealing a negative relationship between blood iron levels and OAB risk, potentially related to the role of oxidative stress in OAB pathogenesis. Oxidative stress is regarded as a significant causative factor in bladder dysfunction (30). Iron, an essential trace element, contributes to enhancing antioxidant capacity when present at moderate levels (31). Disruptions in iron metabolism, particularly when levels are excessively high or low, may impair bladder smooth muscle function, influencing the risk of OAB (32). Therefore, we speculated that maintaining optimal blood iron levels may mitigate oxidative stress and, consequently, reduce OAB risk (33,34). Meanwhile, alterations in cortical networks or emotional-affective state modulation can significantly influence bladder sensation, representing a key underlying mechanism in the onset and persistence of OAB (35). For example, occupational stress has been positively linked to OAB, with toileting behaviors acting as a mediator (36). Additionally, an analysis of NHANES 2011–2018 data revealed a significant association between depression and OAB, with age serving as an effect modifier (37). Notably, iron deficiency has been implicated in depression across multiple studies, and a systematic review and meta-analysis demonstrated the clinical efficacy of iron supplementation in alleviating postpartum depression (38). Given these connections, the hypothalamic-pituitary-adrenal axis and subsequent immune dysregulation may constitute a critical pathway linking iron status, emotional-affective states, and OAB (39). However, it must be emphasized that the existing literature on this relationship remains limited, and the precise molecular mechanisms demand further investigation. Notably, confounding factors, such as nutritional status, must be addressed in future research, -and longitudinal cohort or intervention studies are warranted to confirm the possible protective effects of iron supplementation on OAB.

The results of the PDP and ALE analyses indicate that urinary cadmium exposure may serve as a potential risk factor of OAB, particularly in people with a history of prolonged high exposure to cadmium. Cadmium is a heavy metal commonly found in both natural and industrial environments, and the long-term health hazards associated with its exposure should not be underestimated. With a biological half-life of 25 to 30 years, the effects of cadmium on the body can be both prolonged and cumulative. Research has demonstrated a connection between cadmium neurotoxicity and bladder dysfunction, potentially linked to neurological damage (40). A study by Ghoochani et al. proposed a mechanism for impaired bladder control, suggesting that diminished inhibition of the voiding reflex by the central nervous system may serve as the pathophysiological basis for OAB (41). Our findings further support this hypothesis, indicating that elevated urinary cadmium concentrations are significantly related to an elevated risk of OAB. Cadmium may promote the pathogenesis of OAB by influencing bladder smooth muscle function through the activation of biological pathways associated with inflammatory responses. Additionally, studies have shown that the toxic effects of cadmium can impair bladder function by inhibiting muscarinic receptor activity and reducing the expression of related neuroreceptors (42,43). In addition, we also elucidated the association between blood lead with the risk of OAB. Lead, a well-known heavy metal, is extensively documented as a cause of damage to the nervous system and renal function. While direct evidence linking lead to OAB is insufficient, its detrimental effects on the kidneys and nervous system may indirectly elevate the risk of developing OAB (44,45).

ML effectively identifies critical factors and their corresponding effect patterns that influence outcomes without requiring human intervention or continuous improvement. Using ROC curve analysis and the associated AUC values, we identified the XGBoost model as the optimal choice for predicting OAB. XGBoost utilizes a gradient boosted DT framework to enhance its predictive power by combining multiple weak classifiers. The model captures non-linear relationships and intricate interactions among features, which is vital for analyzing the intricate link between heavy metal exposure and OAB risk (18). During training, XGBoost assigns weights to features based on their significance, effectively eliminating irrelevant or redundant attributes. In contrast, other models (e.g., NB regression and LR), although they also incorporate feature selection, tend to perform poorly in handling high-dimensional features or intricate data interactions (46). Moreover, in our dataset, which contains a limited number of OAB cases, XGBoost enhances predictive performance through modifying category weights or employing a sampling strategy. PFI analysis pinpoints the critical heavy metal exposures, guiding policymakers to allocate resources for priority surveillance and intervention. The influence of age on OAB is pronounced (47); studies reveal an age-related increase in OAB prevalence, and SHAP analysis further validates the significant positive contribution of advanced age to predicting OAB risk.

To our knowledge, this represents the first comprehensive investigation of BMI’s differential mediation in heavy metal-OAB associations, elucidating novel metabolic-environmental pathway interactions. Blood iron analysis revealed BMI mediated 14.80% of its protective effect against OAB (total effect β=−6.57e−02, P<0.001), consistent with iron’s role in mitochondrial energy metabolism. As an essential electron transport chain cofactor, adequate iron levels may attenuate OAB risk through obesity mitigation via enhanced lipid oxidation and suppressed adipogenesis (48,49). This mechanism substantiates clinical correlations between obesity indices and OAB at the molecular level. Notably, blood lead and urinary cadmium exposure exhibited paradoxical mediation: while directly increasing OAB risk (β=2.87e−02, P<0.05), their BMI-mediated indirect effects reduced risk (β=−4.24e−03, P<0.001), resulting in negative mediation proportions (−17.33%). This counterintuitive finding suggests adiposity may mitigate metal toxicity through dual mechanisms: (I) adipose sequestration of lipophilic toxicants reducing bioavailability (50,51), and (II) obesity-associated metabolic compensation counteracting oxidative stress (52,53). While these findings provide mechanistic insights, the limited existing literature on triadic obesity-metal-OAB interactions necessitates cautious interpretation of these novel pathway relationships.

This study advances environmental epidemiology research through synergistic integration of explainable ML with conventional epidemiological methods to decode multifactorial relationships between heavy metal mixtures, obesity, and OAB risk. To our knowledge, this represents the first application of ensemble ML algorithms combined with interpretability frameworks (SHAP, PFI) in OAB research, enabling both high predictive accuracy (AUC: 0.736) and clinically actionable insights into nonlinear exposure-response patterns. Methodological robustness was ensured through multi-technique validation combining LR, BKMR, and RCS analysis, which consistently confirmed prioritized metal-OAB associations. Our mediation analysis revealed BMI’s dual role as mediator-suppressor (14.80% indirect effect for iron; −17.3% for lead/cadmium), providing mechanistic insights into metabolic pathways linking environmental exposures to OAB pathogenesis. These findings establish urinary cadmium and blood iron as clinically relevant biomarkers while demonstrating the feasibility of ML-driven risk stratification for preventive urology interventions.

There are also some limitations in this study. Firstly, missing metal exposure data for some NHANES participants necessitated data censorship, potentially impacting the findings’ generalizability. Future research should ensure complete medical records and detailed environmental exposure data for validation and model improvement. Secondly, the cross-sectional feature of NHANES may obscure causal relationships between metal exposure and OAB. Upcoming studies should focus on large prospective cohorts with regular follow-ups and utilize methods like Mendelian randomization to determine causality (54). Additionally, long-term cohort data will refine predictive models. Lastly, while data were collected from various U.S. regions, this remains a single-center study lacking external validation.

Specifically, future models should be validated across multiple centers using extensive public or private databases that represent diverse populations, including those from Europe and Asia, to enhance the models’ predictive validity. Moreover, the future study should focus more on the mediation role of BMI through longitudinal design and the mechanisms of regarding the role of iron in bladder smooth muscle, as well as the relationship between anemia or ischemia with OAB.

Conclusions

In conclusion, our study uniquely integrates multiple ML techniques with interpretable models to fill critical knowledge gaps in the relationship between environmental metal exposures and OAB. It establishes a scientific foundation for future medicine and preventive strategies. Subsequent research can build upon this groundwork to investigate the specific mechanisms through which various metal exposures influence OAB risk, as well as how this knowledge can be leveraged to enhance individual health outcomes.

Acknowledgments

We thank all participants and investigators who contributed to the NHANES, from which our study data were derived.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-350/rc

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-350/prf

Funding: This study was supported by the “Talent 555 Project” of Jiangxi Province (No. 700238003).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-350/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Krhut J, Kobberø H, Kanaan R, et al. The mechanism of action of neuromodulation in the treatment of overactive bladder. Nat Rev Urol 2025;22:414-26. [Crossref] [PubMed]
2. Wei B, Zhao Y, Lin P, et al. The association between overactive bladder and systemic immunity-inflammation index: a cross-sectional study of NHANES 2005 to 2018. Sci Rep 2024;14:12579. [Crossref] [PubMed]
3. Mikami K, Kaga K, Ichikawa T. Analysis of Overactive Bladder Symptom Score Improvement in Lower Urinary Tract Symptom Patients During Behavioral Therapy While Using the Smartphone Application "USAPO". Cureus 2025;17:e82406. [Crossref] [PubMed]
4. Miyajima S, Omaru T, Ishii T, et al. Real-World Evidence for Risk Factors of Bruises and Fractures from Falls in Patients with Overactive Bladder: A Medical Record Analysis. Int J Clin Pract 2023;2023:3701823. [Crossref] [PubMed]
5. Durden E, Walker D, Gray S, et al. The economic burden of overactive bladder (OAB) and its effects on the costs associated with other chronic, age-related comorbidities in the United States. Neurourol Urodyn 2018;37:1641-9. [Crossref] [PubMed]
6. Qin G, Niu Z, Yu J, et al. Soil heavy metal pollution and food safety in China: Effects, sources and removing technology. Chemosphere 2021;267:129205. [Crossref] [PubMed]
7. Witkowska D, Słowik J, Chilicka K. Heavy Metals and Human Health: Possible Exposure Pathways and the Competition for Protein Binding Sites. Molecules 2021;26:6060. [Crossref] [PubMed]
8. Rehman K, Fatima F, Waheed I, et al. Prevalence of exposure of heavy metals and their impact on health consequences. J Cell Biochem 2018;119:157-84. [Crossref] [PubMed]
9. Matsumoto T, Hatakeyama S, Imai A, et al. Relationship between oxidative stress and lower urinary tract symptoms: results from a community health survey in Japan. BJU Int 2019;123:877-84. [Crossref] [PubMed]
10. Orr SE, Bridges CC. Chronic Kidney Disease and Exposure to Nephrotoxic Metals. Int J Mol Sci 2017;18:1039. [Crossref] [PubMed]
11. Gao F, Lu Y, Cheng Q, et al. Blood cadmium levels and overactive bladder in middle-aged and older adults in the United States: Insights from NHANES 2007-2020 data. Environ Pollut 2024;363:125148. [Crossref] [PubMed]
12. Alber M, Buganza Tepole A, Cannon WR, et al. Integrating machine learning and multiscale modeling-perspectives, challenges, and opportunities in the biological, biomedical, and behavioral sciences. NPJ Digit Med 2019;2:115. [Crossref] [PubMed]
13. Peng GCY, Alber M, Tepole AB, et al. Multiscale modeling meets machine learning: What can we learn? Arch Comput Methods Eng 2021;28:1017-37. [Crossref] [PubMed]
14. Liu J, Li X, Zhu P. Effects of Various Heavy Metal Exposures on Insulin Resistance in Non-diabetic Populations: Interpretability Analysis from Machine Learning Modeling Perspective. Biol Trace Elem Res 2024;202:5438-52. [Crossref] [PubMed]
15. Zhang Y, Song J, Li B, et al. Association between body roundness index and overactive bladder: results from the NHANES 2005-2018. Lipids Health Dis 2024;23:184. [Crossref] [PubMed]
16. Yu Y, Meng W, Kuang H, et al. Association of urinary exposure to multiple metal(loid)s with kidney function from a national cross-sectional study. Sci Total Environ 2023;882:163100. [Crossref] [PubMed]
17. Shang Y, Chen J, Tai Y. Association between weight-adjusted waist index and overactive bladder syndrome among adult women in the United States: a cross-sectional study. BMC Womens Health 2024;24:488. [Crossref] [PubMed]
18. Li X, Zhao Y, Zhang D, et al. Development of an interpretable machine learning model associated with heavy metals' exposure to identify coronary heart disease among US adults via SHAP: Findings of the US NHANES from 2003 to 2018. Chemosphere 2023;311:137039. [Crossref] [PubMed]
19. Saleem J, Zakar R, Butt MS, et al. Application of the Boruta algorithm to assess the multidimensional determinants of malnutrition among children under five years living in southern Punjab, Pakistan. BMC Public Health 2024;24:167. [Crossref] [PubMed]
20. Bainter SA, McCauley TG, Fahmy MM, et al. Comparing Bayesian Variable Selection to Lasso Approaches for Applications in Psychology. Psychometrika 2023;88:1032-55. [Crossref] [PubMed]
21. Tan MY, Zhang YJ, Zhu SX, et al. The prognostic significance of stress hyperglycemia ratio in evaluating all-cause and cardiovascular mortality risk among individuals across stages 0-3 of cardiovascular-kidney-metabolic syndrome: evidence from two cohort studies. Cardiovasc Diabetol 2025;24:137. [Crossref] [PubMed]
22. Yu Q, Ji W, Prihodko L, et al. Study becomes insight: Ecological learning from machine learning. Methods Ecol Evol 2021;12:2117-28. [Crossref] [PubMed]
23. Hu Y, Kong Y, Tian X, et al. Association between Heavy metals and triglyceride-glucose-related index: a mediation analysis of inflammation indicators. Lipids Health Dis 2025;24:46. [Crossref] [PubMed]
24. Cheng Y, Su J, Wang X, et al. Associations between brominated flame retardants exposure and non-alcoholic fatty liver disease: Mediation analysis in the NHANES. Ecotoxicol Environ Saf 2025;290:117762. [Crossref] [PubMed]
25. Zhang K, Tai Y, Gong Y, et al. The association between urinary cadmium exposure levels and overactive bladder syndrome in the U.S. adults from NHANES database. Sci Rep 2025;15:12870. [Crossref] [PubMed]
26. Bian H, Zhang Y, Liu K. Association between magnesium depletion score and overactive bladder among U.S. Adults using data from NHANES 2005-2018. Sci Rep 2025;15:32193. [Crossref] [PubMed]
27. Yao X, Jiang M, Dong Y, et al. Association between exposure to multiple metals and stress urinary incontinence in women: a mixture approach. Environ Geochem Health 2024;46:149. [Crossref] [PubMed]
28. Fu M, Zhu Z, Xiang Y, et al. Associations of Blood and Urinary Heavy Metals with Stress Urinary Incontinence Risk Among Adults in NHANES, 2003-2018. Biol Trace Elem Res 2025;203:1327-41. [Crossref] [PubMed]
29. Ni J, Li Z, Lu Y, et al. Relationship between exposure to cadmium, lead, and mercury and the occurrence of urinary incontinence in women. Environ Sci Pollut Res Int 2022;29:68410-21. [Crossref] [PubMed]
30. Wu YH, Chueh KS, Chuang SM, et al. Bladder Hyperactivity Induced by Oxidative Stress and Bladder Ischemia: A Review of Treatment Strategies with Antioxidants. Int J Mol Sci 2021;22:6014. [Crossref] [PubMed]
31. Galaris D, Barbouti A, Pantopoulos K. Iron homeostasis and oxidative stress: An intimate relationship. Biochim Biophys Acta Mol Cell Res 2019;1866:118535. [Crossref] [PubMed]
32. Halon-Golabek M, Borkowska A, Herman-Antosiewicz A, et al. Iron Metabolism of the Skeletal Muscle and Neurodegeneration. Front Neurosci 2019;13:165. [Crossref] [PubMed]
33. Jomova K, Alomar SY, Nepovimova E, et al. Heavy metals: toxicity and human health effects. Arch Toxicol 2025;99:153-209. [Crossref] [PubMed]
34. Khalaf EM, Taherian M, Almalki SG, et al. Relationship between exposure to heavy metals on the increased health risk and carcinogenicity of urinary tract (kidney and bladder). Rev Environ Health 2024;39:539-49. [Crossref] [PubMed]
35. Grundy L, Caldwell A, Brierley SM. Mechanisms Underlying Overactive Bladder and Interstitial Cystitis/Painful Bladder Syndrome. Front Neurosci 2018;12:931. [Crossref] [PubMed]
36. Xu D, Zhu S, Li H, et al. Relationships among occupational stress, toileting behaviors, and overactive bladder in nurses: A multiple mediator model. J Adv Nurs 2019;75:1263-71. [Crossref] [PubMed]
37. Li T, Di X, Li Y, et al. The Association between Depression and Overactive Bladder: A Cross-Sectional Study of NHANES 2011-2018. Int Urogynecol J 2025;36:373-80. [Crossref] [PubMed]
38. Tian Y, Zheng Z, Ma C. The effectiveness of iron supplementation for postpartum depression: A protocol for systematic review and meta-analysis. Medicine (Baltimore) 2020;99:e23603. [Crossref] [PubMed]
39. Reid BM. Early life stress and iron metabolism in developmental psychoneuroimmunology. Brain Behav Immun Health 2024;40:100824. [Crossref] [PubMed]
40. Bhardwaj JK, Siwach A, Sachdeva D, et al. Revisiting cadmium-induced toxicity in the male reproductive system: an update. Arch Toxicol 2024;98:3619-39. [Crossref] [PubMed]
41. Ghoochani M, Rastkari N, Yunesian M, et al. What do we know about exposure of Iranians to cadmium? Findings from a systematic review. Environ Sci Pollut Res Int 2018;25:1-11. [Crossref] [PubMed]
42. Chen L, Shen Q, Liu Y, et al. Homeostasis and metabolism of iron and other metal ions in neurodegenerative diseases. Signal Transduct Target Ther 2025;10:31. [Crossref] [PubMed]
43. Moyano P, de Frias M, Lobo M, et al. Cadmium induced ROS alters M1 and M3 receptors, leading to SN56 cholinergic neuronal loss, through AChE variants disruption. Toxicology 2018;394:54-62. [Crossref] [PubMed]
44. Su Q, Zhang W, Li D, et al. Association between blood lead levels and serum creatinine: a cross-sectional study. Int Urol Nephrol 2025;57:973-80. [Crossref] [PubMed]
45. Lian CY, Chu BX, Xia WH, et al. Persistent activation of Nrf2 in a p62-dependent non-canonical manner aggravates lead-induced kidney injury by promoting apoptosis and inhibiting autophagy. J Adv Res 2023;46:87-100. [Crossref] [PubMed]
46. Wang SY, Ravindranath R, Stein JD, et al. Prediction Models for Glaucoma in a Multicenter Electronic Health Records Consortium: The Sight Outcomes Research Collaborative. Ophthalmol Sci 2024;4:100445. [Crossref] [PubMed]
47. Carpenter L, Campain NJ. Overactive bladder: not just a normal part of getting older. Br J Nurs 2022;31:S16-22. [Crossref] [PubMed]
48. Winter WE, Harris NS. Iron Biology - An Overview for Laboratorians. Ann Clin Lab Sci 2023;53:681-95. [PubMed]
49. Alshwaiyat NM, Ahmad A, Wan Hassan WMR, et al. Association between obesity and iron deficiency Exp Ther Med 2021;22:1268. (Review). [Crossref] [PubMed]
50. Aaseth J, Javorac D, Djordjevic AB, et al. The Role of Persistent Organic Pollutants in Obesity: A Review of Laboratory and Epidemiological Studies. Toxics 2022;10:65. [Crossref] [PubMed]
51. Jandacek R, Liu M, Tso P. Interactions of Body Weight Loss with Lipophilic Toxin Storage J Nutr 2024;154:801-3. Commentary. [Crossref] [PubMed]
52. Lolescu BM, Furdui-Lința AV, Ilie CA, et al. Adipose tissue as target of environmental toxicants: focus on mitochondrial dysfunction and oxidative inflammation in metabolic dysfunction-associated steatotic liver disease. Mol Cell Biochem 2025;480:2863-79. [Crossref] [PubMed]
53. Tinkov AA, Aschner M, Ke T, et al. Adipotropic effects of heavy metals and their potential role in obesity. Fac Rev 2021;10:32. [Crossref] [PubMed]
54. Zhang Y, Gui J, Song J, et al. Unraveling the relationship between metabolic syndrome and epigenetic aging: evidence from NHANES 1999-2002 and Mendelian randomization study. J Gerontol A Biol Sci Med Sci 2025;80:glaf134. [Crossref] [PubMed]