Causal Inference in Digital Health Intervention Evaluation: Methodological Advances for Nigerian mHealth Research

The evaluation of mobile health interventions in Nigeria faces fundamental methodological challenges arising from the ethical and practical limitations of randomised controlled trials in digital health contexts, combined with the measurement complexities of multi-component mHealth programmes implemented in heterogeneous community environments. This study develops and validates novel causal inference methodologies specifically designed for evaluating digital health interventions in Nigerian community health contexts, where existing methods fail to adequately address selection bias, treatment effect heterogeneity across the country's diverse socioecological environments, and data sparsity in rural zones. An original Adaptive Propensity Score Weighting methodology is developed and validated through simulation studies incorporating data generation processes calibrated to the characteristics of three prior Nigerian mHealth evaluations in North Central, South East, and South West settings. The methodology is applied to re-evaluate the causal effect of a maternal health mobile messaging intervention in Kogi State, producing effect estimates demonstrably more robust to unmeasured confounding than those obtained through conventional regression approaches. Secondary methodological contributions include a Spatial Treatment Effect Heterogeneity Analysis Framework identifying which subpopulations and geographical zones generate the largest and most reliable intervention effects, enabling more efficient programme scaling decisions. The study advances both the mHealth evaluation literature and the broader causal inference scholarship by demonstrating the applicability of doubly robust estimators in low-resource, high-heterogeneity digital health evaluation contexts. Keywords: causal inference, mHealth evaluation, propensity score, Nigeria, digital health methodology.