Artificial Intelligence Counterfactual Allocation of Emergency Obstetric Blood Units Under Variable Donor Deferral Log Completeness: A Doubly Robust Evaluation of Healthcare Transfusion Equity

Emergency obstetric haemorrhage remains a leading cause of maternal mortality, and the allocation of scarce blood units across obstetric and non-obstetric demand is a consequential decision problem. Counterfactual allocation rules learned from observational transfusion data promise to improve equity, but their validity depends on the completeness of donor deferral logs, which determine whether the donor pool is correctly characterised. The estimator combines an outcome regression with an inverse-propensity weighting scheme in which the propensity model is augmented by a log-completeness indicator. A before-after-control-impact design is used to evaluate the estimator's power to detect allocation effects of 3, 5 and 8 percentage points across 8, 16, 24 and 32 clusters, with 2000 replications per design cell. Under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, 54 clusters are required to detect a 5-percentage-point effect at 80% power. The analysis clarifies the conditions under which doubly robust allocation rules remain informative when deferral logs are incomplete, and it identifies the cluster counts at which equity-relevant effects become detectable. The contribution is a formal characterisation of the bias-variance trade-off induced by log incompleteness and a design benchmark for future evaluations of counterfactual transfusion allocation.