Objective To investigate the factors influencing the natural progression of stress urinary incontinence (SUI) among adult women in China, construct and compare the predictive performance of SUI outcome prediction models, and provide a basis for stratified management of SUI patients.

Methods Based on the cohort data from the “Epidemiological Survey of Female Pelvic Floor Dysfunction” from 2014 to 2016, a multi-stage stratified cluster sampling method was used. Women aged ≥ 20 years, residing in the registered area for at least five years, suffering SUI at the baseline were included, excluding pregnant women and those with severe mental disorders or physical illnesses. SUI prognosis prediction models were constructed using logistic regression, random forests, and Bayesian networks, with model performance evaluated through five-fold cross-validation.

Results A total of 2 620 female SUI patients were included in the study, revealing a natural remission rate of 50.2% over four years, with an average annual remission rate of 12.6%. Multivariate analysis identified several risk factors for SUI remission: age ≥ 70 years (OR = 2.339, 95% CI: 1.358 − 4.029), overweight body mass index (BMI) 24.0 − 27.9 kg/m² (OR = 1.731, 95% CI: 1.077 − 2.782), chronic cough (> 3 weeks) (OR = 1.948, 95% CI: 1.522 − 2.494), diabetes (OR = 1.418, 95% CI: 1.129 − 1.781), and history of stroke (OR = 1.698, 95% CI: 1.135 − 2.541). Conversely, minority ethnicity (OR = 0.323, 95% CI: 0.155 − 0.675) and history of pelvic surgery (OR = 0.773, 95% CI: 0.650 − 0.919) were associated with protective effects. The logistic regression model demonstrated the highest overall performance, area under the receiver operating characteristic curve (AUC) was 0.622, while the random forest model exhibited the highest sensitivity (0.637).

Conclusion Nearly half of Chinese adult women with SUI experience spontaneous remission, significantly associated with age, BMI, and chronic cough. These findings underscore the necessity to integrate pelvic floor function screening into health management systems for high-risk populations for early identification and intervention. The validated predictive models demonstrate potential utility for clinical risk stratification and public health prioritization.