Interpreting Xpert MTB/RIF Ultra in a High-Risk Tuberculosis Referral Cohort in Taiwan: Treatment-Associated Culture Discordance and Rifampicin-Indeterminate Results

Background/Objectives: Interpretation of positive and indeterminate rifampicin (RIF) calls from Xpert MTB/RIF Ultra (Ultra) in previously treated, high-risk populations is poorly characterised. Methods: We retrospectively analysed all 2271 Ultra records (2024) at Taiwan’s National Reference Mycobacteriology Laboratory, a high-risk referral cohort, against referring-laboratory culture and phenotypic drug-susceptibility testing (pDST), with line-probe assay (LPA) as a reflex test and treatment history defined by referral indication. Results: Among 1662 culture-paired records, sensitivity was 91.7% and specificity 75.1%. Specificity was 92.7% in screening or risk-based referrals but 56.4% in previously or currently treated patients, who accounted for 84.8% of culture-negative Ultra-positive results. Positive predictive value fell from 89.3% (HIGH) to 36.2% (Trace). All 77 Trace results were RIF indeterminate (77 of 78 indeterminate calls). In 362 of 382 pDST-paired records with determinate calls, RIF sensitivity and specificity were 94.3% and 96.8%. Eight of 10 pDST-discordant resistance calls, none sequenced, were LPA-resistant. Reflex LPA was interpretable in 98.9% of resistance-detected records but in 19.2% of indeterminate records. Conclusions: Low specificity was concentrated among previously or currently treated patients, and indeterminate calls reflected Trace detection. LPA supported most pDST-discordant resistance calls, whose mutational basis remains unconfirmed. Stewardship should reflect the limited yield of reflex LPA in paucibacillary specimens.