Artificial Intelligence Distributionally Robust Prediction of Snakebite Antivenom Stock-Out Risk Under Variable Cold-Chain Temperature Excursion Logging: A Healthcare Resilience Evaluation

Antivenom stock-outs in low-resource snakebite-endemic settings are driven partly by cold-chain temperature excursions that degrade product potency and force unplanned disposal, yet the logging of such excursions is sparse, irregular and heterogeneous across facilities. The framework couples a nominal stock-out probability model, estimated from complete-logging facilities, with a Wasserstein ambiguity set whose radius is calibrated to the empirical logging-completeness distribution observed across a facility tier. A tractable reformulation yields a worst-case risk bound that is computable from the nominal model and a single scalar radius. The framework is illustrated with a reanalysis of NASA POWER MERRA-2 monthly air-temperature reanalysis for Kampala, Uganda (0.35°N, 32.58°E), 2000–2023, comprising 312 monthly observations with a mean of 22.443 °C and a linear warming trend of −0.1564 °C per decade. The reanalysis supplies an ambient-temperature envelope against which cold-chain excursion thresholds can be benchmarked, and the negative trend indicates that Kampala's ambient baseline has not been warming over the observation window. The contribution is a formal, assumption-explicit route from heterogeneous excursion logging to a conservative stock-out risk bound that a district health authority can compute without facility-level imputation.