Causal Inference for Artemisinin Resistance Forecasting Under Variable Plasmodium falciparum K13 Propeller Surveillance Coverage: A Target-Trial Emulation of Artificial Intelligence Treatment Failure Prediction

Artemisinin-based combination therapies remain the frontline treatment for uncomplicated Plasmodium falciparum malaria, yet the emergence and spread of K13 propeller mutations threaten their efficacy across the Greater Mekong Subregion and increasingly in sub-Saharan Africa. Artificial intelligence models that forecast treatment failure from molecular surveillance data are typically trained on convenience samples in which K13 propeller genotyping coverage varies systematically across sites and time periods, creating a form of selection bias that standard predictive metrics do not detect. The framework is evaluated through a design-based power analysis of a before-after-control-impact estimator under a declared variance structure with an intraclass correlation of 0.64, between-cluster standard deviation of 6.0, and residual standard deviation of 4.5. Across 2000 replications per design cell, 54 clusters are required to detect a 5 percentage-point effect at 80% power. The analysis clarifies how surveillance coverage heterogeneity biases AI forecasts and specifies the design conditions under which target-trial emulation yields valid causal estimates of treatment failure.