Artificial Intelligence Inference of Community Clinic Snakebite Antivenom Dosing Under Unmeasured Species-Identification Delay: A Healthcare Distributionally Robust Mortality Comparison

Community clinics in low-resource settings must decide how many vials of polyvalent antivenom to administer before the biting species has been identified, and species-identification delay is unmeasured in routine records. The framework is formalised as a minimax expected-mortality problem over dosing rules, and a finite-sample bound on the out-of-sample mortality gap is derived under a smoothness condition on the mortality surface. A comparative evaluation is then conducted against three benchmarks: fixed-dose, species-conditional, and empirical-risk-minimising rules. The comparison is analytical and replication-based, not a field trial; no patient data were collected. Under the reference distribution, the empirical-risk-minimising rule attains the lowest expected mortality, but its advantage over the robust rule shrinks as the ambiguity radius grows, and the robust rule dominates once the radius exceeds a threshold that depends on the curvature of the mortality surface and the delay distribution. The analysis yields a dosing rule that is implementable from vial-count and weight data alone, together with an explicit statement of the conditions under which species-conditional dosing is unsafe. The contribution is a formal bridge between distributionally robust optimisation and bedside antivenom logistics, and a calibrated warning that inference under unmeasured delay can reverse the ranking of dosing rules.