Genome-guided prioritization of rifabutin MIC testing among rifampicin high-MIC Mycobacterium tuberculosis isolates: a retrospective development and internal evaluation study

Background Quantitative rifabutin testing can resolve rifamycin cross-resistance patterns, but testing capacity is limited. We evaluated whether whole-genome mutation profiles could concentrate rifabutin low-MIC isolates into a small, high-yield confirmatory-testing worklist among rifampicin high-MIC Mycobacterium tuberculosis isolates. Methods This retrospective development and internal evaluation study used paired whole-genome and broth-microdilution data from the CRyPTIC consortium. The research endpoint was rifabutin MIC ≤ 0.12 mg/L among isolates with rifampicin MIC > 0.5 mg/L. Subject-grouped nested cross-validation included fold-specific model selection and fold-local ranking at testing budgets of 5%, 10%, 15%, and 20%. A compact rpoB logistic procedure was compared with broader learned models and rule-based genomic comparators. Results Among 3,466 eligible isolates from 3,397 subjects, 210 (6.1%) met the rifabutin low-MIC endpoint. Testing the highest-ranked 10% selected 348 isolates and recovered an expected 151.0 of 210 low-MIC events (71.9%; 95% subject-cluster confidence interval, 65.2–77.4%), increasing worklist yield to 43.4%. A cross-validated carrier ≥ 10 rpoB logistic procedure recovered 71.5% (64.9–76.7%) and had an area under the precision–recall curve of 0.425, compared with 0.438 for the nested procedure. Collection-site and phylogenetic-lineage holdouts showed heterogeneous performance. The compact procedure was separately refitted on the full cohort as a frozen 37-feature candidate. Conclusions Genome-guided ranking substantially enriched a limited rifabutin MIC testing worklist among rifampicin high-MIC isolates. The compact rpoB procedure provides a transparent candidate for independent prospective evaluation in specialist clinical microbiology laboratories while retaining phenotypic MIC testing as the confirmatory measurement.