A total of 287 participants were enrolled into Cohort A between 2013 and 2015, representing 53% of the overall A5288 study population [5]. As described in the Methods, we excluded nine participants, giving 278 participants in the study population for this report. Table 1 describes the characteristics at study entry of these 278 participants: 56% were female and median age was 40 years. At study entry, median time on ART was 7.8 years, median time on a PI-based regimen was 3.2 years, and median CD4 count was 169 cells/mm3. Median HIV-1 RNA was 20,444 copies/mL and 52 (19%) had HIV-1 RNA < 1000 copies/mL at study entry despite having confirmed HIV-1 RNA ≥ 1000 copies/mL during screening.

The PI continued was LPV/r for 173 participants (62%) and ATV/ritonavir (ATV/r) for 103 (37%); for two other participants it was ATV (because of intolerance to ritonavir) or saquinavir (SQV). All participants took either 3TC or FTC. The second NRTI was TDF for 256 participants (92%), ZDV for 20 participants (7%), and ABC for one participant (< 1%); the remaining participant (< 1%) took both TDF and ZDV (plus FTC). Of the 278 participants, 104 (37%) had NRTI resistance based on the genotype obtained at screening including 89 (32%) with resistance to one NRTI in the regimen taken during follow-up (or to two NRTIs for the participant who took both TDF and ZDV plus FTC). Thus 189 participants (68%) had virus susceptible to both the two NRTIs and the PI taken. Among those with resistance to the NRTI component on the regimen taken, 80 participants (29% of the study population of 278 participants) had high-level resistance to 3TC/FTC (due to the M184V mutation), two (< 1%) had intermediate-level resistance to 3TC/FTC (one of the two took TDF and ZDV plus FTC also had intermediate resistance to TDF), and seven (3%) had low-level resistance or potential low-level resistance to the other NRTI (which was to TDF for all seven) (Additional file 1:  Table S1).

Retention was very high: during median follow-up of 72 weeks (quartiles: 60, 96), only 6 participants (2%) were lost to follow-up without first experiencing virologic failure. Only 12 (4%) of participants had a change in their ART regimen prior to virologic failure or prior to their last HIV-1 RNA measurement if they did not experience virologic failure. Six changed due to adverse events, four due to noncompliance or participant decision to stop ART, one due to a drug supply issue, and one due to a potential drug interaction with tuberculosis treatment.

Of the 278 participants, 139 (50%) experienced confirmed virologic failure at or after week 24 of follow-up. Figure 1 shows the cumulative proportion experiencing virologic failure by time. Most (103 of 139) of the failures occurred at week 24: the estimated proportion failing at that time was 39% (95% confidence interval [CI] 33%, 45%). The remaining 36 participants who experienced virologic failure achieved suppression below 1000 copies/mL but then experienced virologic failure after week 24. The estimated cumulative proportion failing increased with time, from 39% at week 24 to 48%, 53% and 60% at weeks 48, 72 and 96. Conversely, an estimated 40% achieved suppression below 1000 copies/mL through to 96 weeks (95% CI 33%, 49%).

Estimated Cumulative proportion of participants with virologic failure (confirmed HIV-1 RNA ≥ 1000 Copies/mL at or after 24 weeks) over time, in the overall study population and by selected characteristics of participants at study entry

Figure 1 also shows the cumulative proportion experiencing virologic failure over time by sex; age, HIV-1 RNA, CD4 count, duration of prior ART at study entry; NRTI resistance at screening; and randomized adherence support intervention. There was significant variation in time to virologic failure among the categories of each of these variables except for the randomized comparison of CPI + SOC versus SOC adherence support. Based on visual inspection of Fig. 1, categories of some variables were combined for further analysis where the distribution of time of virologic failure appeared very similar: for age, the revised categorization used was < 30 versus ≥ 30 years; for HIV-1 RNA, < 10,000 versus ≥ 10,000 copies/mL; for CD4 count, < 200 versus ≥ 200 cells/mm3; and for duration of ART, < 10 versus ≥ 10 years. Using these revised categories, Table 2 shows descriptive statistics for the proportion observed with virologic failure during follow-up, as well as estimated hazards ratios from univariable and multivariable proportional hazards models. In the multivariable analysis, the strongest and significant predictors of virologic failure were higher HIV-1 RNA at study entry (adjusted hazard ratio [aHR]: 2.20 for ≥ 10,000 versus < 10,000 copies/mL), lower age at study entry (aHR: 1.96 for < 30 versus ≥ 30 years), NRTI resistance at screening (aHR: 1.74 for those with versus without resistance), lower CD4 count at study entry (aHR: 1.73 for < 200 versus ≥ 200 cells/mm3), and shorter duration of ART (aHR 1.62 for < 10 versus ≥ 10 years). Female participants had somewhat higher risk of virologic failure than male participants (aHR: 1.30) though this was not statistically significant (p = 0.16). Although the estimated hazard ratio for CPI + SOC versus SOC was in the direction of a small benefit of CPI (aHR: 0.80), it was not statistically significant (p = 0.45). We further evaluated whether there was effect modification between each possible pair of variables included in the multivariable model and found no significant evidence (p > 0.05 for all interaction terms). We also found no significant evidence that the PI taken (LPV or ATV) was predictive when added to the model or that any of the associations were modified by which PI was taken. Similarly, with the caveat of limited power, there was no significant evidence that country was predictive of virologic failure when added to the multivariable model or that any of the associations varied among countries. Sensitivity analyses using thresholds of ≥ 50 copies/mL and ≥ 200 copies/mL instead of ≥ 1000 copies/mL to define virologic failure showed the same five variables as the strongest predictors though the hazard ratios for both HIV-1 RNA and CD4 count at study entry showed weaker associations as the threshold used decreased from 1000 to 50 copies/mL (Additional file 1: Tables S2 and S3).

Overall, therefore, participants of age < 30 years, participants with uncontrolled HIV disease at study entry as measured by HIV-1 RNA ≥ 10,000 copies/mL or CD4 count < 200 cells/mm3, participants with NRTI resistance at screening, and participants on ART for < 10 years had significantly higher risk of virologic failure during follow-up. As a simple illustration of the variation in risk of virologic failure in the study population, we categorized each participant according to the number of these risk factors they had, giving a score between 0 and 5. Figure 2 shows the estimated cumulative proportion of participants experiencing virologic failure at weeks 24 and 96 by this score (as only six participants had all five risk factors and so a score of 5, participants with scores of 4 and 5 are grouped together). There was a strong trend in cumulative proportion with virologic failure at week 24 by number of risk factors varying from 0% for a score of 0 to 16%, 29%, 45% and 75% for scores of 1, 2, 3 and 4/5; and at week 96, varying from 8% for a score of 0 to 31%, 40%, 73% and 100% for scores of 1, 2, 3 and 4/5.

Estimated Cumulative proportion of participants with virologic failure (confirmed HIV-1 RNA ≥ 1000 Copies/mL At or after 24 weeks) at Weeks 24 and 96 by Number of Risk Factors for Virologic Failure a Participant Had. Bars Show Breakdown of Those With Versus Without New NRTI and/or PI Mutations. Risk Factors Were Age at Study Entry < 30 Years, HIV-1 RNA at Study Entry ≥ 10,000 Copies/mL, CD4 Count at Study Entry < 2000 Cells/mm3, Any NRTI Resistance Identified at Study Screening, and Duration of Prior ART at Study Entry < 10 Years

We also evaluated whether self-reported adherence at weeks 12 and 24 was predictive of virologic failure. Among the 180 participants reporting taking ≥ 90% of their doses in the month prior to study visits at both weeks 12 and 24, 88 (49%) experienced virologic failure. In comparison, among the 98 participants reporting taking < 90% of their doses at either or both of weeks 12 and 24, 51 (52%) experienced virologic failure. The difference was not statistically significant in univariable analysis (p = 0.15) or when added to the multivariable model (p = 0.17).

Of the 139 participants who experienced virologic failure, 134 had a resistance genotype available at virologic failure, including 36 who developed new NRTI or PI resistance mutations compared with the genotype obtained during screening for study entry (Additional file 1: Table S4). Thirty-one participants (11%) had new NRTI resistance mutations and seven (3%) had new PI resistance mutations which were not present at screening (including two who had both new NRTI and new PI resistance mutations). Twenty-six of the 36 participants developed new resistance mutations to drugs in the regimen being taken. This included high-level resistance to 3TC/FTC (17 participants), to the other NRTI being taken (two participants, one of whom also developed high-level resistance to 3TC/FTC), and to the PI being taken (one participant). Assuming that the five participants without a genotype at virologic failure had no new NRTI or PI resistance mutations, 13% of the 278 participants experienced virologic failure with new NRTI or PI resistance mutations. The same proportion of participants (13%) with versus without NRTI resistance at screening developed new resistance mutations.

Figure 2 also shows the breakdown of participants experiencing virologic failure by number of risk factors for failure between those with new NRTI or PI resistance mutations (shaded part of bar) and those without new NRTI or PI resistance mutations (unshaded part of bar). There was a significant association over time of cumulative proportion experiencing virologic failure without new mutations and risk score (p < 0.001). Despite the small number of participants experiencing virologic failure with new resistance mutations, there was also a significant association over time of cumulative proportion experiencing virologic failure with new resistance mutations and risk score (p = 0.039).