Artificial Intelligence Inference of Community Clinic Tuberculosis Contact-Tracing Prioritisation Under Unmeasured Household Stigma Reporting: A Healthcare Fairness-Guarantee Evaluation

Contact-tracing prioritisation algorithms for tuberculosis (TB) in community clinics must allocate scarce screening resources across households whose stigma-reporting behaviour is systematically unmeasured. The dataset comprises 109 country-survey observations from 35 African countries between 1990 and 2023, with a mean of 9.018 and a standard deviation of 12.67. The paper proves that when household stigma reporting is missing not at random, standard prioritisation rules violate a fairness guarantee defined as equalised tracing yield across reporting strata. A bound is derived on the fairness gap as a function of the unmeasured reporting bias, and a robust prioritisation rule is proposed that satisfies the guarantee under stated assumptions. The framework is evaluated analytically rather than empirically; no original data were collected. The contribution is a formal characterisation of when AI-driven TB contact-tracing prioritisation can be certified fair under unmeasured stigma reporting, and a decision rule that clinics can use to test whether their prioritisation is certifiable.