Whole genome sequencing precision medicine strategyto shorten treatment for rifampicin-resistant tuberculosis (SMARTT): a pragmatic, randomised, single-blind phase 4 trial

Abstract

Background All patients with rifampicin-resistant tuberculosis should receive a short course of effective treatment. We aimed to evaluate the effectiveness of whole genome sequencing (WGS)-guided treatment in shortening duration of treatment for rifampicin-resistant tuberculosis. Methods We did a pragmatic, randomised, single-blind phase 4 trial in 13 hospitals and 35 clinics in South Africa. Adults with pulmonary rifampicin-resistant tuberculosis were randomly assigned (1:1, using adaptive randomisation) to standard of care (WHO all-oral 9-month, seven-drug regimen or 18-month individualised regimen) or a four-drug, 6-month WGS-guided regimen generated by a feature-based artificial intelligence (AI) model. Bacteriological effectiveness, defined as difference in time to culture conversion measured as change in mycobacterial load using a non-linear mixed-effects model, was the primary outcome. Due to operational constraints and the inability to perform culture-free WGS, we performed an analysis of clinical effectiveness in a modified intention-to-treat (mITT) population (risk difference for unfavourable treatment outcomes, 10% non-inferiority margin) and safety (serious adverse events). This trial is registered with ClinicalTrials.gov, NCT05017324. Findings 204 participants were randomly assigned between Sept 23, 2021, and Feb 9, 2023 (101 to the standard of care group, 103 to the WGS group), of which 162 culture-positive individuals were included in the mITT analysis (78 in the standard of care group, 84 in the WGS group). mITT participants were mostly male (n=120 [74%]) and living with HIV (n=100 [63%]) and had rifampicin-resistant, multidrug-resistant tuberculosis (n=142 [88%]), pre-extensively drug-resistant tuberculosis (n=15 [9%]), or extensively drug-resistant tuberculosis (n=5 [3%]). In the WGS group, 15 different individualised regimens were recommended to 81 participants, and median treatment duration was shortened by 2 & centerdot;6 months. Bacteriological response was similar in both groups with a mean mycobacterial load half-life of 0 & centerdot;30 weeks (95% CI 0 & centerdot;27 to 0 & centerdot;33) in the WGS group versus 0 & centerdot;31 weeks (95% CI 0 & centerdot;31 to 0 & centerdot;31) in the standard of care group (relative difference 0 & centerdot;97, 95% CI-0 & centerdot;86 to 1 & centerdot;07; p=0 & centerdot;26). Unfavourable treatment outcomes (bacteriological failure, loss to follow-up, or death) were less frequent in the WGS group (20 [24%] of 82 vs 32 [42%] of 77; adjusted risk difference-18 & centerdot;4 percentage points, 95% CI-35 & centerdot;6 to-1 & centerdot;2); superiority p=0 & centerdot;018), mostly due to reduced loss to followup. The frequency of serious adverse events was similar between groups (23 [27%] of 84 in the WGS group vs 26 [33%] of 78 in the standard of care group; risk difference-6% percentage points, 95% CI-8 to 20; p=0 & centerdot;41). Interpretation Using WGS and AIto select regimens with optimal drug features had similar bacteriological and superior clinical efficacy compared with standard of care and was safe. Future trials should evaluate the effectiveness of culture-free sequencing-guided treatment on relapse-free cure where the standard regimen for rifampicin-resistant, multidrug-resistant tuberculosis is bedaquiline, pretomanid, linezolid, and moxifloxacin. Copyright (c) 2026 Elsevier Ltd. All rights reserved, including those for text and data mining, AI training, and similar technologies.