Artificial Intelligence Inference of Community Clinic Hepatitis B Birth-Dose Timing Under Unmeasured Cold-Chain Temperature Logger Loss: A Healthcare Instrumental Variable Safety Evaluation

Hepatitis B birth-dose vaccination within 24 hours of delivery is a cornerstone of perinatal transmission prevention, yet in many low-resource community clinics the cold-chain temperature loggers that would verify vaccine viability are lost, leaving the true thermal exposure of doses unmeasured. The framework treats logger loss as a missingness mechanism that is informative about clinic-level cold-chain reliability, and proposes a clinic-level instrument constructed from the interaction of seasonal ambient temperature and documented refrigerator maintenance intervals. Under explicit assumptions of exclusion and monotonicity, the instrumental variable estimand is shown to bound the causal effect of delayed birth-dose administration on seroprotection failure. The analysis is methodological and theoretical: no clinic-level outcome data are analysed, and the temperature series is used only to characterise the ambient exposure distribution that conditions the instrument. The contribution is a transparent, assumption-explicit identification strategy that separates the safety question from the missing-data problem, and a set of derived bounds that can be reported alongside conventional birth-dose timeliness indicators. The framework is intended for adoption in settings where logger loss is routine and where the alternative is either silent omission of affected doses or unverified assumption of viability.