Artificial Intelligence Snakebite Severity Scoring Under Variable Antivenom Batch Potency Records: A Distributionally Robust Evaluation of Healthcare Referral Escalation

Antivenom batch potency varies between manufacturers and between lots, yet artificial-intelligence snakebite severity scores that drive referral escalation are typically trained and evaluated as if the antivenom administered were a fixed, nominal-strength product. the present analysis treats batch potency as a distributional uncertainty set and asks how a severity scorer's referral decisions degrade when the potency record attached to the antivenom is itself unreliable. The problem is formalised as a distributionally robust risk functional over potency-conditional outcome distributions, and a before-after-control-impact design is characterised for evaluating referral escalation under that uncertainty. Using a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64, the design's operating characteristics were computed across 2000 replications per cell for cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points. Power reached conventional levels only at the larger cluster counts, and the minimum detectable effect at 80% power fell as clusters increased, with 52 clusters required to detect a 5-percentage-point effect. The analysis indicates that potency-record uncertainty, not scorer accuracy alone, governs whether a referral-escalation evaluation can distinguish a safe scorer from an unsafe one, and it specifies the cluster counts at which such an evaluation becomes informative.