OR06

Bedaquiline is a core drug for rifampicin-resistant-tuberculosis (RR-TB) treatment. Accurate prediction of bedaquiline-resistant phenotype from genomic data remains  challenging. We used a Bayesian approach to combine expert opinion and phenotype-genotype data on 756 isolates and estimate the posterior median probability of bedaquiline resistance (PBR) and 95% credible intervals (CrIs) for variants in bedaquiline candidate resistance genes. To assess how the novel PBR concept may be used in clinical care, we performed a discrete choice experiment with 45 experienced physicians. The median PBR was highest for missense mutations in atpE (60.8%, 95% CrI 44.0-76.1) and nonsense mutations in Rv0678 gene (55.1%, 95% CrI 27.3-80.7) and lowest (≤3%) for synonymous mutations in atpE and Rv0678 and missense mutations in Rv0679and pepQ.  PBR played the most important role in determining the physician’s treatment decision, followed by patient’s response to treatment at 1 month and the strain’s resistance profile. The influence of the PBR on physicians’ decisions depended on the 95% CrI width and history of exposure to bedaquiline. Physicians were most likely to stop bedaquiline when the PBR was >75%, continue bedaquiline when the probability was <50%, and strengthen the bedaquiline-containing regimen when PBR was between 50% and 75%. Probability of resistance (and 95% CrIs) is a novel promising approach to communicate genomic information when accurate binary classification is not yet possible. Developing a clinical decision support tool for PBR, together with ensuring access to next-generation sequencing has the potential to improve the treatment outcomes of RR-TB.