Artificial Intelligence Off-Policy Inference of Community Clinic Antibiotic De-Escalation Timing Under Unmeasured Prior-Culture Ascertainment: A Healthcare Regret-Bound Stewardship Evaluation

Community clinic antibiotic stewardship must decide when to de-escalate empirical therapy, but the decision is confounded by prior-culture ascertainment that is recorded inconsistently and therefore unmeasured in routine data. The framework is methodological: it specifies identification assumptions, defines a regret functional over timing policies, and proves that the regret of any candidate policy is bounded by an ascertainment-sensitivity term plus a policy-class complexity term. The bound is then applied illustratively to a real World Bank panel of 24 African countries observed from 2000 to 2023 (N = 560), where GDP per capita (constant 2015 US$) was regressed on access to electricity and urban population; the model fit was R-squared = 0.591, adjusted R-squared = 0.5895, F = 184.764 (p = 2.82e-62), with coefficients of 1291.55405 (SE 194.35159) for access to electricity, 113.13952 (SE 7.77493) for urban population, and -104.41834 (SE 11.31314) for the linear year term. The panel is used only to calibrate the scale of the ascertainment-sensitivity term, not to estimate clinical effects. De-escalation timing policies cannot be ranked by off-policy value alone when prior-culture ascertainment is unmeasured; a regret bound that makes ascertainment sensitivity explicit is a necessary companion to any stewardship evaluation.