Integrating Machine Learning for Propensity Score Matching and Causal Inference: A Causal Forest Approach to Assessing the Impact of Maternal Education on Antenatal Care Utilization

BackgroundWhile maternal education is linked to antenatal care (ANC) use, its causal effect remains uncertain. This study applies a machine learning approach, Causal Forests, to estimate the causal impact of maternal education on adequate ANC utilization in Bangladesh. MethodsWe analyzed data from 6,815 ever-married women aged 15-49 years who had a live birth within five years preceding the 2022 Bangladesh Demographic and Health Survey (BDHS). The outcome was adequate ANC utilization, defined as receiving four or more ANC visits from skilled providers. Maternal education was dichotomized as <secondary vs. [&ge;]secondary education. To estimate causal effects, we employed a machine learning-based propensity score matching approach using K-nearest neighbors (KNN), followed by treatment effect estimation using Causal Forests. We also assessed treatment heterogeneity across subgroups and conducted a Rosenbaum bounds sensitivity analysis to evaluate robustness to unmeasured confounding. ResultsWhile a strong crude association was observed between maternal education and ANC use (83.4% vs. 16.6%, p < 0.001), the estimated Average Treatment Effect (ATE) after matching was modest and statistically non-significant (ATE = 0.006, 95% CI: -0.002 to 0.016, p = 0.17). However, significant heterogeneity in Individual Treatment Effects (ITEs) was detected across subgroups, with higher effects among women aged 20-40 at first birth, urban residents, those with media exposure, and those with wealthier or more educated spouses. The sensitivity analysis indicated moderate robustness to hidden bias. ConclusionsMaternal education alone may not significantly impact ANC use once socioeconomic and contextual factors are accounted for. However, its benefits are amplified among specific subgroups, suggesting the need for integrated and context-sensitive maternal health interventions. Advanced machine learning methods can enhance causal inference and inform equity-oriented policy strategies.