Artificial Intelligence Causal Mediation Analysis of Community Health Worker Visit Frequency Effects on Childhood Immunisation Completion Under Variable Village Health Register Attrition: A Healthcare Equity Evaluation

Village health registers in low- and middle-income settings are the operational backbone of community health worker (CHW) programmes, yet their completeness varies systematically across villages in ways that are correlated with the very outcomes the registers are used to measure. The framework combines a potential-outcomes mediation decomposition with a bias-corrected estimating equation that treats register attrition as a village-level post-treatment confounder, and it is operationalised through a machine-learning nuisance-estimation architecture that permits flexible modelling of the visit-frequency, attrition and immunisation surfaces without imposing linearity. Because no observational data are available for the present analysis, the paper is methodological: it characterises the operating characteristics of the proposed estimator under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64) using a before-after-control-impact design with 2000 replications per cell across cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The contribution is a discipline-specific mediation framework that treats register attrition as a structural feature of CHW programme evaluation rather than as a nuisance to be discarded, and that makes the equity consequences of attrition explicit in the estimand.