Identifying risk factors and predicting stroke using Bayesian networks: Evidence from NHANES 2011–2020

Background Stroke is a leading cause of morbidity and mortality worldwide, representing a major cerebrovascular disorder. Early identification of stroke-related risk factors is essential for implementing effective prevention and management strategies. This study aimed to develop an interpretable Bayesian network (BN)-based predictive model to identify key risk factors associated with stroke and to elucidate their complex interdependencies. Methods This study analyzed cross-sectional data derived from the National Health and Nutrition Examination Survey (NHANES) spanning the period 2011–2020. Feature selection was performed using univariate and multivariate logistic regression analyses. The BN structure was constructed using the hybrid HPC algorithm (H2PC), with conditional probability distributions estimated via maximum likelihood estimation. Both qualitative and quantitative analyses were conducted to examine node probabilities and elucidate dependencies between stroke and associated risk factors. Model performance was primarily assessed using the area under the receiver operating characteristic curve (AUROC) and compared against established machine learning algorithms. Results The final analytical sample comprised 20,535 individuals. Bayesian network analysis identified five variables with direct dependency relationships to stroke occurrence: age, sleep disorders, alcohol consumption, coronary heart disease, and diabetes. The BN model demonstrated superior predictive performance with an AUROC of 0.803 (95% CI: 0.773–0.833), significantly outperforming other machine learning approaches. Conclusions The developed BN model provides an intuitive visualization of the probabilistic interdependencies among stroke risk factors while achieving competitive predictive accuracy. These findings demonstrate its exploratory value in unmasking complex risk pathways and suggest its potential to inform future stroke risk assessment and prevention strategies upon further longitudinal validation.