Artificial Intelligence Drug Interaction Alerting Under Variable Pharmacy Dispensing System Coding Granularity: A Graph-Neural Calibration Evaluation of Healthcare Prescribing Safety

Drug–drug interaction (DDI) alerting systems deployed in African hospital pharmacies must operate across dispensing records whose coding granularity varies from coarse Anatomical Therapeutic Chemical (ATC) class codes to fully specified ingredient-level identifiers. the present analysis treats that variation not as a data-cleaning nuisance but as a calibration problem for graph-neural alert classifiers, and asks whether a single alerting model can remain safe when the same drug is represented at different levels of the ATC hierarchy. A formal framework is developed in which the dispensing record is a graph whose nodes carry a granularity label, and in which alert calibration is defined as the agreement between predicted interaction risk and a reference severity distribution conditional on that label. A graph-neural architecture with granularity-conditioned calibration heads is specified, and its operating characteristics are characterised analytically under a declared variance structure. The design analysis uses a before–after control–impact (BACI) estimator with an intraclass correlation of 0.64, between-cluster standard deviation of 6.0 and residual standard deviation of 4.5; power was estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. Fifty-four clusters are required for 80% power at a 5-percentage-point effect. The contribution is a calibration-aware formulation of DDI alerting that makes coding granularity an explicit modelling quantity rather than an unexamined source of alert fatigue.