Additive Bayesian network analysis of social determinants of health and adverse pregnancy outcomes
Ummayhany Bharmal, Tetsuya KawakitaAbstract
Introduction
Adverse pregnancy outcomes (APOs) disproportionately affect non‐Hispanic Black individuals compared with non‐Hispanic White individuals. Race and ethnicity are social constructs that reflect structural inequities, and social determinants of health (SDoH) are also associated with APOs. Prior studies have largely evaluated race/ethnicity and individual SDoH factors using conventional regression approaches, which estimate adjusted associations with APO but do not fully capture how these factors are interrelated or whether their associations are direct or indirect. We therefore used additive Bayesian network (ABN) analysis to characterize the complex interdependencies among race, SDoH, clinical factors, and APO.
Methods
This was a secondary analysis of the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers‐to‐Be (nuMoM2b). Individuals with missing data, those who delivered before 23 weeks of gestation, or those identifying to any other racial group other than Non‐Hispanic Blacks or Non‐Hispanic Whites were excluded because the analysis was restricted to these groups to examine disparities. The primary outcome was a composite of APOs, including preterm births (PTB) <37 weeks, hypertensive disorders of pregnancy (HDP), small for gestational age (SGA) (<fifth percentile), stillbirth, and gestational diabetes mellitus (GDM). We examined direct and indirect associations between APO and SDoH as well as clinical factors, including body mass index (BMI), pregestational diabetes mellitus (PDM), and chronic hypertension (CHTN), using additive Bayesian network analysis. A final Directed Acyclic Graph (DAG) was derived from 1000 resamples created using nonparametric bootstrapping. Results are reported as odds ratios (ORs) along with their 95% credible intervals (CrIs) for categorical variables and beta coefficients (β) with 95% CrI for continuous variables.
Results
Among 5708 participants included in the analysis, 1296 (22.7%) experienced an APO. In the final network, only PDM (OR, 3.55; 95% CrI, 2.25–5.67) and higher BMI (OR, 1.36; 95% CrI, 1.28–1.44) were directly associated with APO. Non‐Hispanic Black race was the most upstream node and was not directly associated with APO; instead, it was associated with public insurance, experienced discrimination, higher BMI, lower employment, and lower emotional and social support, through which it was indirectly related to APO.
Conclusion
In this network analysis, social and psychosocial conditions were highly interconnected and related to APO primarily through cardiometabolic pathways, whereas only PDM and higher BMI showed direct associations with APO, and non‐Hispanic Black race acted as an upstream structural determinant whose association with APO was mediated. These findings suggest that disparities in APO may arise through interconnected social and clinical pathways and support multilevel interventions that address both structural and social conditions and proximal cardiometabolic risk factors.