In order to evaluate the health and nutritional status of adults and children in the United States, the National Center for Health Statistics (NCHS) and the Centers for Disease Control and Prevention (CDC) conduct the NHANES research program. Data from the NHANES survey was accessible to the general public via the official website (https://www.cdc.gov/nchs/nhanes/about_nhanes.htm). The NHANES methods were approved by the Centers for Disease Control and Prevention’s National Center for Health Statistics, and informed permission was given by each participant. Cross-sectional data from six consecutive NHANES iterations, conducted between 2007 and 2018, were included into our study. Five iterations in a row with 59842 participants. Initially, 29814 people were removed from the sample as a whole because their OAB assessments were lacking. Next, we removed the individuals (n = 241) who were pregnant from the NHANES questionnaire data. Finally, we eliminated the individuals who lacked information on their poverty income ratio (n = 2252), sedentary time (n = 31), and education level (n = 3651). A total of 23863 individuals were kept for further analysis after these patients were eliminated ( Figure 1 ).

(1) Fasting plasma glucose levels between 6.1 and 6.9 mmol/L (2); glycohemoglobin (HbA1c) levels between 6.0% and 6.4%; and (3) self-reported physician diagnosis of prediabetes are the diagnostic criteria for borderline diabetes (prediabetes). Diabetes mellitus is defined as follows (1): plasma glucose level of 7.0 mmol/L while fasting (2); blood glucose level of 11.11 mmol/L after a 2-hour oral glucose tolerance test (3); glycohemoglobin (HbA1c) of 6.5%; and (4) self-reported medical diagnosis of diabetes (21). Serum insulin, fasting glucose, and glycohemoglobin were among the diabetes-related indicators that were derived from NHANES laboratory data.

We assessed overactive bladder symptoms in the NHANES population using the NHANES Kidney Conditions-Urology questionnaire, which covered urge urine incontinence and nocturia, based on the Overactive Bladder Symptom Score [OABSS], which was presented in Supplementary Table S2 (22). The person who had a total score of ≥3 was determined to have overactive bladder disorder.

Covariates were gathered via questionnaire, laboratory, and demographic data. Using a standardized questionnaire, we have incorporated variables such as age, sex, educational attainment, race, poverty income ratio (PIR), alcohol intake, smoking status, sedentary time, hypertension, congestive heart failure, coronary heart disease, angina, heart attack, and stroke history. The NHANES website provides descriptions of each of these variables. Three categories were used to categorize educational levels: below high school, high school, and above high school. A person was considered a smoker if they had smoked 100 cigarettes or more in their lifetime. Anyone who drank alcohol in excess of 12 times in a single year was considered an alcohol user. Weight in kilograms (kg) divided by height in meters squared (m²) yielded the BMI. Through participant interviews, the amount of time spent in sedentary activity on a normal day was determined. Through laboratory examination, the number of white blood cells, neutrophils, lymphocytes, and platelets in blood samples were determined. We also identified the diabetes controlling status from “DIQ280 - What was your last A1C level and DIQ290 What does Dr say A1C should be”. If DIQ280 higher than DIQ290, we identified diabetes controlling status was not good.

The complex sampling methodology of the NHANES meant that sample weights, clustering, and stratification were all taken into consideration in the data analysis. Using multivariable logistic regression models, we assessed the associations between diabetes, diabetes-related indicators, inflammatory markers, and overactive bladder. Model 1 was not adjusted. Model 2 was adjusted for age, sex, race, education level, poverty income ratio. Model 3 was adjusted for age, sex, race, educational level, poverty income ratio, BMI, smoking, alcohol use, sedentary time, hypertension, congestive heart failure, coronary heart failure, angina, heart attack, stroke. Restricted cubic splines were used to analyze the non-linear associations. Linear regression models were performed to analyze the relationship of diabetes related markers and inflammatory biomarkers. The association between inflammatory biomarkers and diabetes-related indicators was examined using linear regression models. For causal mediation studies, we estimated the direct effect (DE), indirect effect (IE), and total effect (TE) using the R package called “mediation” (23). There should be a connection between the mediator and the result and the exposure (24). The OAB prediction model was constructed using Python software (version 3.7). The xgboost package, adaboost package, lightgbm package, decision tree package, and MLP package execute different machine learning algorithms while modeling different machine learning methods. Subgroup analysis was conducted to analyze the prevalence differ socioeconomics including age, gender, races, education and poverty income ratio (PIR). Last, we conducted the sensitivity analysis using multiple interpolation to supplement the socioeconomic status. We have multiple interpolated the socioeconomic status including poverty income ratio (n = 2252), sedentary time (n = 31), and education level (n = 3651) and to validate the research results. All statistical analyses were performed using R software (version 4.0.1). Significance was set at p < 0.05 (two-sided).

Table 1 summarizes the detailed baseline characteristics of participants by OAB, and Supplementary Table S1 illustrates the participants with survey-weighted descriptive statistics. A total of 23,863 participants were included in our research, with a median age of 49.7 ± 17.6 years. Among them, 4894 participants were OAB participants, and 18969 were non-OAB participants. We found inflammatory markers, including white blood cells, neutrophil count, and platelet count, had different distributions between OAB participants and non-OAB participants. OAB participants had a higher white blood cell count and neutrophil count. Moreover, OAB participants had higher levels of diabetes-related markers, including glycohemoglobin, fasting glucose, and insulin.

Figure 2 reveals the association between diabetes mellitus and OAB. We identified diabetes mellitus as no-diabetes, borderline, and diabetes. Compared with non-diabetes participants, borderline diabetes participants’ OAB prevalence increased by 26% after adjusting for all covariates (Model 3: OR = 1.26; 95% CI: 1.03, 1.53). Diabetes mellitus participants’ OAB prevalence increased 77% compared with non-diabetes participants in the full adjusted model (Model 3: OR = 1.77; 95% CI: 1.62, 1.94).

Table 2 shows the associations of diabetes-related markers with OAB in adults using multiple logistic regression analysis. We found the log10-transformed exposures, including glycohemoglobin, fasting glucose, and insulin, had a significant positive association with OAB. As the quartiles of diabetes-related markers increased, the odds of OAB monotonically increased in three models (all p for trend <0.01). In addition, the cubic spline curves revealed non-linear associations between diabetes-related markers and overactive bladder ( Figure 3 ). Glycohemoglobin exhibited a linear association with OAB (p for nonlinearity >0.05), while associations between fasting glucose, insulin, and OAB were nonlinearity curves (p for nonlinearity < 0.05).

As shown in Table 3 , we found significant correlations between inflammatory markers (white blood cell count, neutrophil count, lymphocyte count, and platelet count) and OAB. In the multiple linear regression analysis, we found log10-transformed white blood cell count and neutrophil count were significantly associated with OAB (all p < 0.001). As the quartiles of white blood cell count and neutrophil count increased, the odds of OAB monotonically increased in three models (p for trend <0.001). Restricted cubic splines reveal the non-linear associations between inflammatory markers and OAB. Four inflammatory markers exhibited a non-linear association with OAB with a statistical difference ( Supplementary Figure S1 ).

From the multiple linear regression analysis, we found a significant correlation between the diabetes-related markers (glycohemoglobin, fasting glucose, and insulin) and inflammatory markers (white blood cells, neutrophils, lymphocytes, and platelets), which is shown in Table 4 . Diabetes-related markers all showed a positive association with inflammatory markers, with statistical differences. Among them, we found glycohemoglobin to be the most significantly correlated with inflammatory markers. Glycohemoglobin was linked to higher WBC (β = 4.67; 95% CI: 3.76, 5.58), neutrophil (β = 2.56; 95% CI: 2.11, 3.01), lymphocyte (β = 1.80; 95% CI: 1.11, 2.49), and platelet counts (β = 105.68; 95% CI: 89.89, 121.47).

Mediation analysis was used to evaluate the mediation effect between the diabetes-related markers and OAB by systemic inflammation ( Supplementary Table S3 ). We found the white blood cells significantly mediated the associations between diabetes-related markers (glycohemoglobin, fasting glucose, and insulin) and OAB, and the proportions were 7.23%, 8.08%, and 17.74%, respectively. (all p < 0.0001). Similarly, neutrophils partly mediated the correlation between glycohemoglobin, fasting glucose, and insulin with OAB at 6.58%, 9.64%, and 17.93%, respectively (all p < 0.0001) ( Figure 4 ). There were no significant mediation effects observed in the associations of diabetes-related markers with overactive bladder by lymphocyte or platelet (all p > 0.05).

As shown in Figure 5 , we used machine learning in the XGBoost model to determine the relative significance of the diabetes-related markers and inflammatory markers in OAB. We observed that glycohemoglobin was the most important indicator in determining OAB. The forest chart below shows the ROC results of OAB prediction by each machine learning model, including XGBboost, LightGBM, Decision Tree, MLP, and AdaBoost. The error lines in the chart are the ROC mean value and SD. Among all the current models, XGBboost is the best-performing training set; meanwhile, the best validation set is XGBboost as well (sorted by AUC).

As shown in Supplementary Table S7 , we conducted subgroup analysis by socioeconomics including age, gender, race, education, ratio of family income to assessed the associations between diabetes and OAB. In the age <60 group, compared with non-diabetes participants, borderline diabetes participants’ OAB prevalence increased by 146%% after adjusting for all covariates and diabetes mellitus participants’ OAB prevalence increased 67%. However, this association was not statistically different in the older population. In addition, it is likely that male, education level more than high school, non-Hispanic Black and middle income (PIR 1.0-3.0) participants had significant significance of diabetes and OAB association. We have multiple interpolated the socioeconomic status including poverty income ratio (n = 2252), sedentary time (n = 31), and education level (n = 3651) and to validate the research results. We found that multiple interpolation results had almost correlated result with delete missing values results. ( Supplementary Tables S4–S6 ). In the sensitivity analysis of model 3, we further adjusted the covariates of diabetes controlling status and they had almost same result with previous result.