Virological Outcomes of Second-line Protease Inhibitor–Based Treatment for Human Immunodeficiency Virus Type 1 in a High-Prevalence Rural South African Setting: A Competing-Risks Prospective Cohort Analysis

Summary A community based, second-line treatment cohort of 102 subjects was followed monthly in community clinics with a total of 178 patient-years of follow-up; second-line virological failure was frequent in this setting.

demographic factors such as education level, employment, and data on household members have not been studied together. Furthermore, the potentially critical association between concomitant tuberculosis (TB) [17] and second-line bPI failure is inconclusive.
Previously published cohort studies have used a Kaplan-Meier survival analysis and standard Cox regression models that can result in inflated estimates due to competing events. Competing risk regression analysis overcomes this limitation by accounting for events, such as death, that preclude the subject from experiencing the study outcome. This is particularly important in sub-Saharan Africa where mortality remains significant following ART initiation due to advanced disease stage as well as loss to follow-up [19]. Here we have applied competing risk methods to comprehensively and prospectively investigate risk factors for VF of bPI-based second-line ART in a rural population of South Africa. In this setting ART was readily available and follow-up optimum according to local standards, including real-time viral load monitoring.

Study Setting and Study Design
This cohort study was nested within the French National Agency for AIDS and Viral Hepatitis Research (ANRS) 12249 Treatment as Prevention (TasP) trial (ClinicalTrials.gov identifier: NCT01509508), an ongoing cluster-randomized trial evaluating the impact of immediate vs deferred ART initiation (according to South African guidelines) on HIV incidence [20] (see Supplementary Materials). The trial started recruiting in March 2012 in Hlabisa subdistrict, Umkhanyakude district, Northern KwaZulu-Natal, where the antenatal prevalence of HIV is 44%, one of the highest in South Africa [21,22]. This rural setting is also one of the poorest districts in South Africa with very high unemployment rates.
This analysis is based on the treatment outcomes documented prospectively in the subgroup of patients on second-line bPIbased treatment within the TasP trial. Participants from all clusters irrespective of trial arm were included in this cohort. Participants were aged >15 years, resident in Hlabisa subdistrict, and included from the date of initiation of second-line treatment till the earliest of the date last seen in clinic, death, or loss to follow-up. No participants initiating bPI ART after May 2015 were included to allow at least 6 months' follow-up for the primary outcome to occur. The participants' outcomes were recorded until November 2015.

Outcome and Prespecified Explanatory Variables
The primary outcome was VF defined as a VL >1000 copies/mL on at least 1 occasion ≥6 months after commencing second-line treatment. Any death occurring during the trial period was recorded and follow-up time was censored at the date of death.
Loss to follow-up was defined as ≥3 consecutive missed clinic appointments. Follow-up time was censored at the last clinic visit. For all the other participants who remained in the trial with virological suppression (VL < 1000 copies/mL), follow-up was censored as the latest of either the date of the last laboratory test or last clinic visit.

Clinic Procedures and Laboratory Methods
At presentation to the trial clinics, all consenting HIV-infected participants were asked to complete the study questionnaires and underwent clinical examination by trial nurses. CD4 count was measured in TasP clinics using a commercial point-ofcare CD4 test (Alere PIMA device tool, Alere Inc, Waltham, Massachusetts). VL was measured using The Abbott m2000 RealTime System with HIV type 1 (HIV-1) VL determination from human plasma of HIV-1-infected individuals in the range of 40-10 000 000 copies/mL (Abbott Molecular Inc, Des Plaines, Illinois). The viral load assay was CE (Conformité Européene) marked and performed at the Africa Centre laboratory; the laboratory participated in Quality Control for Molecular Diagnostics (QCMD) for VL quality assurance testing. Both CD4 and viral load were measured at baseline, and at months 3 and 6 after ART initiation and every 6 months thereafter. Full-genome deep sequencing was performed after 2 consecutive VL measurements >1000 copies/mL at least 6 months apart following second-line VF, as per adapted protocol from Gall et al [23]. In brief, 4 overlapping fragments, spanning approximately 9 kb of the HIV genome, were amplified and purified as per adapted protocol from Gall et al [23]. The library preparation was done on equimolar pooled amplicons, using the Nextera XT Library preparation kit, followed by sequencing on the Illumina MiSeq platform. Next-generation sequencing (NGS) data were analyzed on Geneious software and a threshold of 2% was used for minority variants detection, with a minimum coverage of 1000 reads. In the case of a single VL >1000 copies/mL, Sanger sequencing was done following Manasa et al's protocol [24]. The external quality assurance for Sanger sequencing was with QCMD. Drug resistance mutations were identified according to the Stanford HIV Drug Resistance Database (http://hivdb.stanford.edu/) [25]. We reported mutations where they were detected above 2% frequency. Safety monitoring blood samples were also taken as per protocol [20].

Statistical Analysis
Participant characteristics were reported using frequency and percentage for categorical variables and median and interquartile range (IQR) for continuous variables. The incidence of second-line failure per 100 person-years was estimated with its 95% confidence interval (CI).
We estimated the cumulative incidence function of VF on second-line treatment, taking into account death and loss to follow-up as competing risks. Competing risk regression was used to estimate the subdistribution hazard ratio (SHR) of the associations between VF and participant clinical and demographic factors, accounting for the competing risks of death and loss to follow-up, according to the model of Fine and Gray [26].
In the final multivariable model, mutually adjusted estimates of the SHRs were determined by including those factors with evidence of an association in the univariable analysis and a P value of <.1. Although age and sex were not significantly associated with VF in the univariable analysis, they were kept in the final model as they were a priori specified confounders. Analysis was done using Stata software version 13.

Ethical Committee Approval
Ethics approval was granted by the Biomedical Research Ethics Committee of the University of KwaZulu-Natal (BFC 104/11) and the Medicines Control Council of South Africa. The study was also authorized by the KwaZulu-Natal Department of Health in South Africa. Written informed consent was obtained from all participants.

RESULTS
One hundred one participants were included in this analysis. Sixteen (15.8%) individuals were already on second-line treatment at enrollment into the trial for a median of 2.7 years (IQR, 1.1-3.9 years). Three of these participants had VF at the baseline clinic visit. Seven (6.9%) participants had not initiated ART at enrollment into the trial. The remaining 78 (77.2%) were on first-line NNRTI-based ART for a median duration of 4.9 years (IQR, 3.2-6.7 years) at the time of enrollment; 41 (52.6%) of these had VF at the baseline clinic visit in the TasP trial, necessitating a switch to bPI along with 2 nucleoside reverse transcriptase inhibitors (NRTIs). Median duration on bPI for the 85 patients was 0.6 years (IQR, 0.3-0.9 years).
The majority of participants were female (65.4%) and the median age at initiation of second-line treatment was 37.4 years (IQR, 31.6-45.3 years). There was a high level of unemployment (91.9%) in this cohort of participants residing in a rural setting (Table 1). Thirteen individuals (12.9%) were diagnosed with TB during the study period.

Virological Failure and Associated Risk Factors
The 101 participants contributed 178.7 person-years of follow-up to the analysis. The overall incidence of VF on second-line ART was 12.9 per 100 person-years. At administrative censoring, 76 participants were alive and in care, 1 had died, and 1 was lost to follow-up before any VF was documented, and 23 participants had VF at least 6 months after initiating bPI second-line ART (Supplementary Figure 1). Following intensification of adherence counseling, 13 of the 23 participants (56.5%), including the 3 patients with bPI VF at the baseline clinic visit,  resuppressed with a VL <1000 copies/mL after a median of 8.0 months (IQR, 2.8-16.8 months) (Supplementary Figure 2). Eight of these 13 participants subsequently rebounded with a VL >1000 copies/mL. The prevalence of VF in patients alive and on bPI ART at the time of administrative censoring was 17.8% (18/101).
Thirteen participants were diagnosed with TB and received antituberculosis treatment during the study observation period; 4 of them were treated within 6 months of VF; 1 participant, within 6 months of censoring, did not experience VF; the remaining 8 participants were diagnosed and treated for TB at time points distant from their study exit and did not exhibit VF after initiation of antituberculosis treatment.

Drug Resistance
Genotypes were available at first-line NNRTI ART failure in 9 participants and at bPI ART failure in 6 participants of the 23 with VF failure of second-line ART ( Table 3). The reasons for missing genotypes were as follows: 4 patients receiving care in Department of Health clinics and therefore no sample was available; resuppression in 10 patients and therefore no viremic confirmatory sample available for genotyping; and no confirmatory sample in 2. Three of 5 patients exposed to tenofovir in their first-line regimen developed high-level tenofovir resistance with the K65R mutation, and 2 had accessory tenofovir mutations A62V, V75I, or F77L in the reverse transcriptase gene. In one of these individuals, K65R was detected only by NGS (at 12% frequency; Table 3). All first-line failures had major NNRTI resistance and 5 of 9 had high-level lamivudine/emtricitabine resistance (M184V/I). An additional drug resistance-associated variant (at <20% frequency) conferring NNRTI resistance was detected in patient 6 (K103N at 12%). NGS did not detect minority variants in the protease gene in any of the 9 patients with second-line failure.
At second-line bPI ART failure, only 1 of 8 (12.5%) participants had acquired major PI mutations: M46I, I54V, L76V, and V82A. This individual received rifampicin containing TB treatment started at the same time as double-dose bPI. DISCUSSION We determined the incidence rate for VF on second-line ART and associated risk factors in rural KwaZulu-Natal within the TasP trial. The incidence rate of VF was 12.9 per 100 person-years (95% CI, 8.6-19.4), and prevalence was 17.8% at the end of the observation period. A meta-analysis of studies conducted in resource-limited settings reported a pooled prevalence of VF of 23.1% (range, 11.4%-39.9%) after 12 months of treatment with bPI ART [6], although our prevalence estimate was based on a follow-up period of <1 year in those initiated on second-line ART within TasP. More recent randomized trials reported a lower prevalence of VF, between 14% and 19% at 48 and 96 weeks of treatment [13][14][15]27]. However, the lower prevalence within trials may not be generalizable to the "real world" because trial participants are closely monitored.
There are considerable differences between studies in the definition of VF (from VL >50 to >1000 copies/mL), and guidelines are not clear on a definition of second-line failure that should trigger a switch of treatment regimen. We therefore chose a pragmatic definition of a single VL >1000 copies/ mL, particularly as the follow-up time was relatively short for patients in TasP who switched to second-line ART within the trial. We found that many patients with VF on bPI ART resuppressed to VL <1000 copies/mL following a period of intense adherence counseling. This phenomenon has been reported in patients treated with first-line ART in both South Africa [28] and other parts of sub-Saharan Africa [29], and demonstrates that the efficacy of second-line treatment could be optimized if regular VL monitoring is widely available, adding to the impetus for development of point-of-care VL testing in tandem with effective adherence counseling.
The association between TB treatment and VF of NNRTIbased first-line ART has previously been reported [17]. Here we found evidence of an association between TB treatment and VF of second-line treatment. Our study design did not allow us to infer causality between TB treatment and VF, as some patients were never virologically suppressed even before the rifampicin treatment. TB disease itself could be a marker of virological and clinical failure; indeed, the 2010 World Health Organization (WHO) clinical criteria for treatment failure include a new diagnosis of TB [30]. In patients failing during the rifampicin treatment, the mechanism leading to VF may be due to drug interactions between PIs and rifampicin. Rifampicin, a potent cytochrome P450 3A4 inducer, significantly reduces the serum levels of PIs, and concurrent use can lead to VF [31,32]. To compensate for this phenomenon, double-dosed bPI has been proposed-from twice-daily 400 mg/100 mg to 800 mg/200 mg of lopinavir/ritonavir, as is in this clinical setting [33]. However, the success of such a strategy may be limited by intolerance to the higher doses of PIs [34][35][36]. Nonadherence to treatment may also contribute to failure as the inherent polypharmacy required to treat both conditions may be a challenge for patients.
Poor adherence has been shown to be associated with VF in other studies [6,8,10,17,18]; however, we found no association between adherence measured by pill count and VF. This might in part be due to the fact that adherence was high in this trial context, although announced pill count is susceptible to pill dumping by participants and therefore not be a particularly good marker for adherence: indeed, 88% of participants on first-line NNRTI ART had an overall adherence of ≥95% at 6 months [37]. We found multiple major NRTI and NNRTI mutations at first-line failure in the participants who had genotype testing (9 participants), consistent with data from resource-limited settings [2,38]. Though numbers were small, NGS increased detection of significant tenofovir resistance (K65R mutation) by around 50%, consistent with previous NGS studies in this setting [39]. We found acquired major protease mutations in only 1 (11%) failing bPI ART, consistent with other data from South Africa [8,18,[40][41][42]. NGS did not increase the detection rate of  mutations in protease. The high genetic barrier of bPI to resistance development could be a reason for this. It is worth noting that standard genotype tests based on pol sequencing ignore the influence of mutations in other genes such as gag [43][44][45][46][47] and env [48] on PI resistance. Notably, the individual with major protease resistance had received double-boosted PI treatment, and coupled with previous reports of multiple major protease resistance mutations in children treated with double-dose PI [49,50], further work regarding this approach is warranted. The main methodological strength of this study is the application of regression methods, which account for the presence of competing risks to estimate the rate of VF on second-line treatment and the association between covariates of interest and VF. Other cohort studies reporting outcomes on second-line treatment and factors associated with VF have used standard Kaplan-Meier survival analysis and Cox regression models, which can lead to biased or inflated estimates of association. As this analysis was done on prospectively collected data within a trial context with preset procedures rather than routine data, we also minimized the common problem of missing data and information bias.
There are some limitations of the study. First, although the sample size is small, the prevalence of VF on bPI we found is in line with other published studies. The inclusion of whole-genome sequencing data, albeit from 9 patients, is unique to this cohort. Second, the study was nested in a randomized community trial and as such the patients may not be truly reflective of the general HIV population, given the intense monitoring and provision of counseling and adherence interventions that may impact their attitude toward health and promote more successful treatment outcomes. We did not measure drug levels to assess pharmacokinetic drug-drug interactions or adherence before a clinical visit. Finally, 16 (15.8%) patients were included who were already on second-line ART at enrollment, for a median of 2.7 years, representing a potential bias.
There was also heterogeneity in the patient population. We included all participants from the parent study irrespective of study arm, though study arm had no effect on outcome, largely due to poor linkage to care in this setting (Dabis et al, 21st International AIDS Conference 2016, Durban, abstract FRAC0105LB). We included newly diagnosed as well as ARTexperienced patients, as both present for first-line therapy under "real-world" programmatic conditions, sometimes with evidence of drug resistance to thymidine analogues arising from prior ART [51] or with evidence of transmitted drug resistance [52].
In conclusion, this study found that second-line PI-based VF was common in this population accessing ART in rural South Africa under trial circumstances but recruited at the population level. Further research is needed to understand the mechanisms behind VF of bPI ART in TB-coinfected patients. Novel approaches to optimize second-line ART in resource-limited settings are still urgently needed as this population is likely to grow rapidly, owing to the WHO 2015 guidelines to test and treat all people living with HIV [53].

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes
Author contributions. D. C., R. K. G., F. D., D. P., and C. I. designed the study; R. G., D. C., N. O., C. I., F. D., A. C., and D. P. collected and analyzed the data; D. C. and R. G. wrote the first draft; A. D. and T. de O. performed experiments and analyzed data.
Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the International Initiative for Impact Evaluation (3ie) or the Bill & Melinda Gates Foundation. The funders had no role in the design, analysis, and interpretation of the study or the decision to submit for publication.
Financial support. The French National Agency for AIDS and Viral Hepatitis Research (ANRS) is the sponsor and co-funder of the trial. Research discussed in this publication has been co-funded by 3ie