Estimating Heterogeneous Causal Effects of Maternal Healthcare on Child Nutritional Outcomes in Nepal: A Survey-Aware Analysis Using TMLE, Super Learner and BART

Abstract

Background: Maternal healthcare is an important determinant of child health, yet observational studies often rely on regression models that estimate conditional associations rather than population-level causal effects. We applied complementary causal inference methods to quantify the potential effect of adequate maternal healthcare on child stunting in Nepal. Objective: To estimate the population-average effect of adequate versus inadequate maternal healthcare on child stunting and assess whether this effect varies across socioeconomic and demographic groups. Methods: We analysed a survey-weighted sample of 2,897 children from the 2022 Nepal Demographic and Health Survey. Maternal healthcare was classified as adequate or inadequate according to the study-specific definition, and stunting was defined as a height-for-age z-score below −2. We compared survey-weighted logistic regression and inverse-probability weighting with g-computation and targeted maximum likelihood estimation (TMLE). Bayesian additive regression trees (BART) were used to examine treatment-effect heterogeneity. Positivity, unmeasured confounding, alternative exposure and outcome definitions, and missing-data strategies were assessed. Results: Stunting prevalence was 31.2% overall, 38.7% among children receiving inadequate maternal care and 23.4% among those receiving adequate care. The crude difference was −15.3 percentage points. The adjusted odds ratio was 0.58 (95% CI: 0.49–0.68) with survey-weighted logistic regression and 0.55 (95% CI: 0.46–0.65) with inverse-probability weight- ing. G-computation estimated a 13.8 percentage-point reduction in stunting, while TMLE estimated a 14.2 percentage- point reduction (95% CI: −18.5 to −9.9). BART indicated larger estimated benefits among children of more educated mothers, wealthier households, and urban residents. Positivity was acceptable, and the E-value was 2.18. Sensitivity and robustness analyses yielded consistent findings. Conclusion: Adequate maternal healthcare was consistently associated with substantially lower child stunting across complementary causal estimators. The heterogeneous effects suggest that socioeconomic and geographic conditions may shape the nutritional benefits of maternal healthcare. Because the study is observational, causal interpretation remains conditional on exchangeability, positivity, consistency, and appropriate temporal ordering.