Causal analysis of traditional and environmental risk factors for long-term development of type 2 diabetes using a conditional survival Bayesian network: evidence from the Korean Genome and Epidemiology Study

Over the past decade, traditional demographic, lifestyle, and metabolic factors, along with air pollutants, have increasingly been recognized as key contributors to type 2 diabetes (T2D). However, the comprehensive causal structure among these factors and their individual and interacting interventional effects have seldom been characterized in long-term population studies. Using 11-year follow-up data from 2,102 adults without T2D in the Ansan cohort of the Korean Genome and Epidemiology Study (KoGES), we investigated causal pathways among demographic, lifestyle, metabolic factors, and multiple ambient air pollutants leading to long-term T2D incidence. We employed a Conditional Survival Bayesian Network (CSBN), which integrates survival analysis with Bayesian network modeling to accommodate censored and incomplete data, to visualize the causal structure among risk factors, and to estimate both individual and joint (interaction) interventional effects. The CSBN depicted a holistic causal structure showing how multiple risk factors jointly shape T2D development over the 11-year follow-up and helped distinguish putative direct/indirect pathways from associations likely reflecting confounding. Interventional analysis quantified each factor’s causal contribution to the 11-year T2D incidence. Obesity produced the largest individual effect: setting BMI to the obese category approximately doubled 11-year T2D risk compared with normal weight. High alanine aminotransferase (ALT) and older age increased risk by about 40–50%, while family history of T2D, dyslipidemia, overweight, $$\text {PM}_{2.5}$$, and gaseous pollutants had intermediate effects. Furthermore, the CSBN uncovered synergistic interactions mainly among metabolic factors. In particular, ALT with family history, dyslipidemia, or obesity displayed strong additive interactions. By contrast, air pollutants were found to influence T2D independently rather than through interactions with other risk factors. These findings underscore the importance of integrated public health strategies targeting multiple risk factors to effectively curb T2D incidence. The CSBN’s capability to explicitly model complex causal interactions highlights the necessity for advanced epidemiological analyses to inform targeted preventive measures and efficient resource allocation.