Probe-level performance of Xpert MTB/RIF versus phenotypic culture for rifampicin resistance detection and a triage algorithm based on 3,360 clinical specimens in China

Abstract Background The Xpert MTB/RIF (Xpert) assay is a cornerstone for the rapid detection of rifampicin resistance. However, discordant results between Xpert and phenotypic drug susceptibility testing (pDST) present diagnostic challenges, particularly in regions with genetically diverse Mycobacterium tuberculosis populations. This study aimed to evaluate the performance of Xpert at the probe level to resolve such discrepancies and to inform an optimized diagnostic workflow. Methods Using a retrospective cohort of 3,360 clinical specimens from a high-burden setting in China, we compared Xpert results with pDST as the reference standard. We then performed a stratified analysis based on specific probe mutation patterns to determine their predictive value for phenotypic resistance. Results Overall, Xpert demonstrated substantial agreement with pDST ( κ = 0.722, 95% CI: 0.685–0.759), exhibiting a sensitivity of 80.4% and a specificity of 95.8%. However, concordance was highly dependent on the specific probe mutation. Mutations in Probe E were strongly predictive of phenotypic resistance (Positive Predictive Value: PPV: 91.4%). In contrast, The mutations detected by Probe ΔCt and Probe A exhibited a low PPV of only 28.0% and 14.3%, respectively. Based on these findings, we propose a diagnostic triage strategy: results indicating Probe E mutations can reliably guide immediate treatment adjustments, whereas isolated ProbeΔCt resistance, or Probe A mutations, warrant confirmatory pDST before initiating second-line regimens. Conclusion The clinical interpretation of Xpert results requires a nuanced understanding of probe-specific mutation patterns. Our probe-stratified analysis and the resulting triage algorithm provide a practical framework for optimizing the management of rifampicin-resistant tuberculosis (RR-TB). This approach effectively balances the need for diagnostic speed with therapeutic accuracy in programmatic settings.