Artificial Intelligence Federated Learning for Multidrug-Resistant Tuberculosis Genotype Prediction Under Variable Laboratory Panel Heterogeneity: A Secure Aggregation Evaluation of Healthcare Surveillance Coverage

Multidrug-resistant tuberculosis (MDR-TB) surveillance depends on genotypic resistance prediction from heterogeneous laboratory panels, and the prospect of training predictive models across institutions without centralising patient data has made federated learning attractive for this task. The study is a replication-based evaluation: 600 data sets of N = 500 were drawn from a process with intercept 12.0, exposure 0.42, group -3.1, severity 1.35 and error SD 4.2, and the estimator was fitted to each. Across the replications the intercept bias was -0.00185 and the exposure bias was 0.00043; root mean squared error was 0.91155 for the intercept and 0.0164 for exposure; empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The results indicate that secure aggregation, modelled here as an exact linear operation on site-level sufficient statistics, leaves the sampling distribution of the estimator intact, and that the binding constraint on surveillance coverage is panel heterogeneity rather than the privacy mechanism. The contribution is a formal statement of the conditions under which federated genotype prediction remains statistically equivalent to pooled estimation, together with a demonstration that those conditions are met in the declared process.