Artificial Intelligence Drug-Resistant Tuberculosis Regimen Selection Under Sparse Local Susceptibility Surveillance: A Causal Off-Policy Evaluation of Healthcare Treatment Coverage

Drug-resistant tuberculosis regimen selection in settings with sparse local susceptibility surveillance is a causal decision problem: the regimen that maximises expected coverage under the observed prescribing policy is not necessarily the regimen that would maximise coverage under an alternative policy, because prescribing is confounded by severity, prior treatment history and programme-level access. The framework combines inverse-propensity-weighted and doubly robust estimators with a partial-identification bound for the region in which local susceptibility is unobserved, and it formalises the conditions under which off-policy estimates are point-identified rather than set-identified. A replication-based evaluation of the ordinary least squares estimator was conducted on 600 data sets of N = 500 drawn from a declared data-generating process (intercept 12.0, exposure 0.42, group -3.1, severity 1.35, error SD 4.2) to characterise the estimator behaviour on which the coverage model depends. Estimated bias was -0.03684 for the intercept and 0.00104 for the exposure coefficient; root mean squared error was 0.87737 and 0.01567 respectively; empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The contribution is a decision-theoretic account of when sparse surveillance permits defensible regimen switching and when it licenses only bounded claims about coverage.