We report a strong and pervasive association between NNRTI resistance before treatment initiation and virologic failure for people initiating first-line ART with both DTG and EFV-based ART in the ADVANCE clinical trial. The effect was stronger among individuals in the EFV arm, but also highly significant in the DTG arms, and persisted after adjusting for self-reported and pill count-based adherence and baseline viral load. When we considered a secondary outcome, which allowed for re-suppression after an episode of VF, the effect of PDR on DTG persisted, but to a lower degree. The finding that NNRTI resistance appears to predict failure among individuals initiating DTG-based ART in LMIC was unexpected, and to our knowledge not previously reported in the literature.
Although NNRTI mutations are not known to affect susceptibility of DTG, the observed effect we identified may be due to higher replication or fitness of NNRTI mutant viruses in the context of drug pressure from integrase inhibitors.28 Although we found relatively little minority resistance and no effect of minority resistance on outcomes, existence of NNRTI resistance could be a surrogate marker of archived NRTI resistance.29 Integrase resistance mutations were not assessed in this study, but are generally believed to be rare (< 1%) in this region.30,31 Alternatively, the lack of suppression may be due to a behavioural component – pre-existing EFV mutations may be a surrogate of prior default among participants not disclosing previous ART exposure. Our multivariable logistic regression models included a measure of self-reported adherence and pill count-based adherence, both of which were highly predictive of virologic outcomes, and addition of which to our model did not meaningfully alter the effect size of PDR on virologic success. However, both self-reported and pill count-based adherence are imperfect measures, and can have a relatively low sensitivity to detect poor adherence, so residual confounding might be present.32–34 The South African public program has used EFV in first line therapy since its inception in 2004, and the programme has over 5 million on treatment. As such, the numbers who have defaulted and are re-initiating therapy are likely to be significant, and it is impossible to identify this within clinical trials using existing South African data systems. Of note, a number of studies have reported denial of ART use among individuals determined to be taking therapy based on drug level testing.35–37
Whether the mechanism of effect is due to poor adherence or virologic mechanisms, our finding that NNRTI resistance, present in 10–20% of individuals initiating DTG in the region, is associated with a reduction in efficacy of DTG-based ART is has multiple public health implications. First, ensuring adequate virologic monitoring occurs with DTG-based regimens will remain a priority. Second, treatment programs will require ongoing attention to second and third-line options, particularly if DTG failure or intolerance becomes more common than previously expected, and NNRTI-based regimens become more commonly used again. Third, integrase resistance testing, which is rarely done outside of research studies in resource-limited settings should become a consideration for referral laboratories in countries where DTG becomes the treatment of choice. Finally, our findings might signal a warning for national programs in the midst of large-scale switching from EFV-based to DTG-based ART, and support increased vigilance for the presence of treatment failure at the time of switch. Future work should explore the efficacy and feasibility of innovative means of mitigating the effect of drug resistance on treatment outcomes, such as targeted point-of-care resistance testing to identify individuals at greater risk of VF, or longer acting regimens to reduce imperfect adherence in those most susceptible to it.38–40
NAMSAL is the only other randomized controlled trial which has compared DTG versus EFV-based first-line ART in sub Saharan Africa. That study, conducted in Cameroon, compared low-dose 400 mg EFV to DTG as third agent at 48 weeks.6 DTG was non-inferior to EFV in that study, but baseline VL > 100,00 copies/ml predicted failure in both arms. NAMSAL reported a much lower prevalence of NNRTI resistance (6%) than we did (14%), which is consistent with other data in the region.41 In NAMSAL, investigators reported no impact of baseline NNRTI resistance on outcomes, although 6/16 failures on EFV had pre-existing NNRTI resistance. In that study, none of the three failures in the DTG arm at 48 weeks had baseline resistance to NNRTIs, the 6% of those that did appeared to suppress during the study. By contrast, in our study, isolated K103N in the DTG arm was associated with lower virologic success in the primary analysis, albeit at 96 weeks. As in prior studies, we identified a small number of individuals with resistance to both the NRTI and NNRTI drug classes, including K65R, M184V who we believe were unlikely to be treatment naïve and who responded poorly to first-line ART. Whilst the proportion is low, this finding is concerning from the point of view of the largescale EFV to DTG-transition in sub-Saharan Africa, during which multi-class drug resistance is likely to be more prevalent.42–46
Next generation sequencing is becoming more widely used in research studies to measure the prevalence and impact of drug resistance in LMIC, and has the added advantage of being able to detect resistant viruses at low frequencies.47–49 However, many studies have failed to demonstrate a role for these low-level mutant viruses in determining clinical outcomes.50 We also found no association between PDR and outcome when considering individuals with mutations in between 5 and 20% of viral quasispecies, which supports current practice to use major resistance mutation frequencies for determination of clinically significant drug resistance.
Our study should be generalized in light of its conduct in South Africa, and the presence of NNRTI resistance-conferring mutations as the large majority of the PDR detected. As this is the first study to show an impact of PDR on the efficacy of first-line DTG, it requires corroboration from future studies of similar cohorts. The presence of a higher prevalence of PDR in the DTG arm suggests that there might have been imbalance between groups, which is most likely due to chance, because study arm was determined by computer randomization. Nonetheless, we have low suspicion for selective dropout in the study because interest in DTG among patients and within society at the time of randomisation was minimal. Our estimates could be susceptible to unmeasured or residual confounding, particularly due to effects of adherence not captured by self-report. Notably, our estimates of the effect of PDR on virologic outcomes remained large and strongly significant after adjustment for confounders, including adherence, meaning an unmeasured confounder would have to have a strong association (OR of 2.8 or greater) with both PDR and virologic success to reduce the effect of pre-treatment drug resistance to null.27 Moreover, known predictors of treatment success, such as adherence and pre-treatment viral load, each predicted virologic success, which enhances the internal validity of our estimates. We also were unable to sequence approximately 15% of the study cohort due to unavailable specimens or failed sequencing. Despite that, our sample size remained large enough to detect relatively small changes in outcomes, and we detected no differences in characteristics between those who were and were not included in this sub-study, which reduces the risk of selection bias.
In summary, our study suggests that the presence of PDR to NNRTIs is negatively associated with outcome of both EFV- and DTG-based first-line ART in South Africa. In the context of highly prevalent PDR NNRTI resistance, our findings, if corroborated, have implications for first-line ART selection and treatment monitoring guidelines in the region. Future work should validate our findings, assess the contribution of pre-treatment integrase mutations to outcomes, elucidate the impact of prior exposure to ART on treatment outcomes, and whether treatment failure observed on DTG-based ART is associated with emergence of integrase inhibitor mutations.