Artificial Intelligence Doubly Robust Estimation of Community Health Worker Triage Accuracy Under Differential Verification by Referral Distance: A Healthcare Selection-Bias Evaluation

Community health worker (CHW) triage programmes routinely verify only a subset of referrals, and verification is often more likely for patients living near a referral facility. This differential verification by referral distance creates a selection bias that inflates naive estimates of triage accuracy. Because no field data are available, the estimator's finite-sample behaviour is characterised in a replication study of 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). Across the replications the estimated bias was 0.04454 for the intercept and -0.00104 for the exposure coefficient; the root mean squared error was 0.91714 and 0.01621 respectively; and empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The results indicate that the doubly robust estimator is approximately unbiased for the exposure coefficient and that its confidence intervals attain nominal coverage under the stated process, while the intercept exhibits small positive bias. Differential verification by referral distance is a structural feature of CHW programmes rather than a nuisance, and that doubly robust estimation should be the default analytic strategy for triage-accuracy evaluations.