Switch to second-line versus continued first-line antiretroviral therapy for patients with low-level HIV-1 viremia: An open-label randomized controlled trial in Lesotho

Background Current World Health Organization (WHO) antiretroviral therapy (ART) guidelines define virologic failure as two consecutive viral load (VL) measurements ≥1,000 copies/mL, triggering empiric switch to next-line ART. This trial assessed if patients with sustained low-level HIV-1 viremia on first-line ART benefit from a switch to second-line treatment. Methods and findings This multicenter, parallel-group, open-label, superiority, randomized controlled trial enrolled patients on first-line ART containing non-nucleoside reverse transcriptase inhibitors (NNRTI) with two consecutive VLs ≥100 copies/mL, with the second VL between 100–999 copies/mL, from eight clinics in Lesotho. Consenting participants were randomly assigned (1:1), stratified by facility, demographic group, and baseline VL, to either switch to second-line ART (switch group) or continued first-line ART (control group; WHO guidelines). The primary endpoint was viral suppression (<50 copies/mL) at 36 weeks. Analyses were by intention to treat, using logistic regression models, adjusted for demographic group and baseline VL. Between August 1, 2017, and August 7, 2019, 137 individuals were screened, of whom 80 were eligible and randomly assigned to switch (n = 40) or control group (n = 40). The majority of participants were female (54 [68%]) with a median age of 42 y (interquartile range [IQR] 35–51), taking tenofovir disoproxil fumarate/lamivudine/efavirenz (49 [61%]) and on ART for a median of 5.9 y (IQR 3.3–8.6). At 36 weeks, 22/40 (55%) participants in the switch versus 10/40 (25%) in the control group achieved viral suppression (adjusted difference 29%, 95% CI 8%–50%, p = 0.009). The switch group had significantly higher probability of viral suppression across different VL thresholds (<20, <100, <200, <400, and <600 copies/mL) but not for <1,000 copies/mL. Thirty-four (85%) participants in switch group and 21 (53%) in control group experienced at least one adverse event (AE) (p = 0.002). No hospitalization or death or other serious adverse events were observed. Study limitations include a follow-up period too short to observe differences in clinical outcomes, missing values in CD4 cell counts due to national stockout of reagents during the study, and limited generalizability of findings to other than NNRTI-based first-line ART regimens. Conclusions In this study, switching to second-line ART among patients with sustained low-level HIV-1 viremia resulted in a higher proportion of participants with viral suppression. These results endorse lowering the threshold for virologic failure in future WHO guidelines. Trial registration The trial is registered at ClinicalTrials.gov, NCT03088241.


Background and rationale
The Joint United Nations Programme on HIV/AIDS (UNAIDS) launched the 90-90-90 targets for 2020 based on the result of newly-acquired scientific evidence that -irrespective of CD4 count -early antiretroviral treatment (ART) for HIV-positive individuals is beneficial to them and prevents HIV transmission. UNAIDS expects that the 90-90-90 targets will lead to a reduction in the yearly global HIV incidence from 2 million currently to 500,000 by 2020.
A crucial step to achieve the third pillar of the UNAIDS 90-90-90 targets -90% viral suppression among HIV-positive individuals on treatment -and thus ensure a successful treatment outcome is the monitoring and management of first-line ART failure.
Since 2013, the WHO recommends routine viral load (rVL) measurement as the preferred monitoring strategy in resource-limited settings and defines virological failure as confirmed VL 1000 copies/mL despite good adherence. Specifically, the guidelines recommend that in case of a VL ≥ 1000 copies/mL the patient should undergo enhanced adherence support and a second VL test 3 months later. A second VL ≥ 1000 copies/mL with confirmed good adherence would trigger the switch to a second-line regimen, whereas if the VL is < 1000 copies/mL the patient should continue unaltered first-line ART. However, the optimal threshold for defining virological failure and the need for switching ART regimen has not yet been determined. In fact, people with VL levels of less than 1000 copies/mL, but not fully suppressed (usually defined as 50-100 copies/mL), are at a increased risk for drug resistance mutations (DRM) and subsequent virological failure. A recently published study from our research consortium in Lesotho indicates similar findings, demonstrating a significant accumulation of drug resistance mutations in patients with VL levels of less than 1000 copies/mL.
The VL threshold of 1000 copies/mL recommended by the WHO and the Lesotho national guidelines for the switch to second-line ART is likely to miss a substantial number of patients on first-line ART with persisting virus replication below 1000 copies/mL with DRM. In resource-limited settings where VL monitoring is not as frequent as in high-income countries, this could have serious implications: after a VL below 1000 copies/mL the patient may not receive a follow-up VL for up to a year, and thus may continue on a failing regimen for a long period of time. In conclusion, such patients are at increased risk for DRM, accumulation of further resistance mutations, drug-resistant virus transmission, and subsequent virological failure.
We hypothesize that in patients on first-line ART with two consecutive unsuppressed VL measurements 100 copies/mL, where the second VL is between 100 and 999 copies/mL, switch to second-line ART (intervention group) will lead to a higher rate of viral resuppression (VL < 50 copies/mL) and is therefore superior compared to not switching to second-line ART according to WHO guidelines (control group, standard of care).

Objectives
The objective of the trial is: To challenge the WHO VL threshold for treatment failure (1000 copies/mL) for HIV positive individuals on first-line ART in a resource-limited setting.

Trial design
Full details of the trial are available in the published protocol (2). This is a multi-centre, two-arm, parallel-group, open-label randomised controlled trial, conducted in health facilities in Lesotho. Eligible persons are HIV-positive individuals on non-nucleoside reverse transcriptase inhibitor (NNRTI) based first-line ART (standard first-line ART in Lesotho) for at least 6 months with reported good adherence, attending one of the study sites, and with two consecutive unsuppressed VL 100 copies/mL with the second VL between 100 and 999 copies/mL. Full inclusion/exclusion criteria are as follows: Inclusion criteria: On first-line ART with two consecutive unsuppressed VL 100 copies/mL, with the second VL between 100 and 999 copies/mL Lives and/or works in one of the study districts and intends to seek follow-up at one of the study facilities Signed written informed consent. For children aged <16 years, a main caregiver, and for illiterate a literate witness, has to provide oral and written informed consent Exclusion criteria: On ART less than 6 months On protease-inhibitor containing ART or any other second-line ART Poor adherence (self-reported at least 1 dose missing in the last 4 weeks, respectively 2 doses of a twice-daily-regimen) Clinical WHO stage 3 or 4 at enrolment Randomisation is in a 1:1 ratio to intervention (switch to second-line ART) or control (no switch to second-line ART). Follow up is over 9 months (for primary endpoint and study completion), with extended follow up over 24 months.
The primary outcome of the trial is: Proportion of virologically suppressed (VL < 50 copies/mL) participants at 9 months The secondary outcomes of the trial are: Proportion of participants with different VL thresholds (VL <100, <200, <400, <1000 copies/mL) at 9 months Proportion of participants with viral resuppression (<50 copies/mL) at 6 months Sustained virologic failure: Proportion of participants with unsuppressed VL >50 copies/mL at 6 and 9 months Adherence at 3, 6, 9 months, assessed by self-reported missed doses Clinical outcomes at 9 months: change from baseline in body weight, haemoglobin, CD4 count, lipids (total cholesterol, HDL cholesterol, LDL cholesterol, triglycerides); proportion of participants with new clinical WHO 3 or 4 events; proportion of participants died (allcauses) Proportion of patients with any adverse events (AEs) or serious adverse events (SAEs) at 9 months Long-term follow-up endpoint: Proportion of patients that are alive, retained in care and virologically suppressed (VL < 50 copies/mL) at 24 months Exploratory outcomes are: Cost-effectiveness Prevalence of major drug resistance mutations a) on all baseline VLs and b) on all VLs that remain unsuppressed (>50 copies/ml) at 9 months for all samples for which an RT-PCR amplification is successful Effect modification of the primary outcome will be assessed by: demographic groups (children [defined as <16 years old at enrolment] vs pregnant women vs adults) VL group at enrolment (second VL 100-599 vs 600-999 copies/mL copies/mL) A subgroup analysis for the primary outcome will be done among: Individuals showing a >0.5 drop in log10 VL from the first to the second VL measurements pre-randomization.

Randomisation
Randomisation was stratified by demographic groups (children versus pregnant women versus adults); VL groups at enrolment (second VL 100-599 versus 600-999 copies/mL); and site (see Table  1; health centres combined, and Butha-Buthe and Mokhotlong Hospitals combined for logistical reasons), using randomly-varying block sizes (details on block sizes held by trial statistician). The randomisation list was generated by computer by the trial statistician, who was not involved in the recruitment or follow-up of participants. Randomisation was performed using sealed, opaque and sequentially-numbered envelopes, which were prepared by person(s) independent to the trial.
The first five randomisations in each stratum were checked by the statistician as soon as possible after randomisation to ensure that the procedures were working correctly; no errors were detected.

Sample size
Based on data from previous research which we conducted in Lesotho, we expect 9-month viral resuppression rates of 25% in the control arm and 60% in the intervention arm, assuming a LTFU-rate of 10% in each arm and assuming such participants as failure (=virologically not suppressed). Assuming a two-sided type 1 error of 5% and a power of 90%, 80 individuals (40 per group) are needed to detect a 35% difference. With this sample size we will still yield 78% power to detect a 30% difference. With an estimated 10% trial participation refusal rate, we need to screen approximately 90 participants.

Framework
This is a superiority trial.

Statistical interim analyses and stopping guidance
No interim analyses are planned.

Timing of final analysis
Outcomes will be analysed after the last participant has completed their primary outcome visit at 9 months. Long term follow-up will continue until the last patient has completed their 24 month follow-up, after which time the long term follow-up analyses will be performed. Table 2 shows the nominal visit months, the permitted ranges according to the protocol, and the ranges that will be used for analysis. One month is defined as 28 days (i.e., 4 weeks), therefore the analysis windows are defined in terms of weeks. The lower limits of the analysis windows correspond to the lower protocol-defined limit (using 28 days per month and converted to weeks). Some sites used calendar months to calculate follow-up visit dates, therefore the upper windows for analysis allow for this (plus a leeway of approximately two weeks, and ensuring that the analysis windows "touch" where possible so that all visits are assigned to a nominal visit month).

Statistical principles 4.1 Confidence intervals and p-values
Statistical tests and confidence intervals will be two-sided. All estimates will be presented with 95% confidence intervals. P-values will be presented where appropriate. No adjustments will be made for multiple testing; interpretations will be based on the strength of evidence of effect size and consistency of results for related outcomes.

Adherence and protocol deviations
Immediately following randomisation, participants randomised to the control group are expected to remain on first line ART, and participants randomised to the intervention group are expected to switch to second line ART. We will present the number of participants on second line ART after randomisation and at each scheduled follow-up visit, by randomised group (with the proportion of those who attended the given visit). We will further tabulate the ART regimens at each scheduled visit, by group.
Adherence to the ART prescribed is one of the secondary outcomes, and is described in section 6.
Any protocol deviations as reported by the trial team or detected at the time of data cleaning/analysis will be described by group.

Analysis populations
Analyses will be by intention to treat, that is including all participants as randomised.
In sensitivity analyses for the primary outcome, we will also consider a "per protocol set" which is further defined in section 6.2.

Screening data
No specific screening data were captured, aside from those used to determine eligibility for the trial (see sections 3.1 and 5.2).

Eligibility
Screening/eligibility data will be summarised in a CONSORT flowchart, showing the total number of people screened and the reasons for screening failures as per the eligibility criteria in section 3.1.

Recruitment
The CONSORT flowchart will include the numbers of participants randomised by group. Enrolment will also be presented by the stratification factors. The two viral loads that are required for determining eligibility will be summarised. A pre-specified subgroup of interest is those participants who had >0.5 log drop between the two screening VLs, therefore enrolment of such participants will also be presented.

Withdrawal/follow-up
The CONSORT flowchart will summarise follow-up for each of the scheduled visits, by randomised group. Reasons will be given for participants who did not complete follow-up as expected. We will also present time from randomisation to last follow-up (categorised as <12, 12-<24, 24-<36, 26-<96, ≥96 weeks).

Baseline patient characteristics
Baseline characteristics will be summarised by randomised group, using medians and interquartile ranges for continuous variables and numbers and percentages for categorical variables. Shell tables showing the variables and categories are included in section 8. Assessment will be made for baseline imbalances between the randomisation groups by visual inspection only, by the trial team before looking at outcome data. In sensitivity analyses, we will further adjust the outcome analyses for any covariates so identified (erring on the side of inclusivity). There will be no formal testing of baseline characteristics across randomised groups (3,4).

Analysis
The screening visit and randomisation are expected to happen on the same date. Time will be measured from randomisation.
The "baseline" viral load result is defined as that closest to randomisation, with a window of -14 weeks to 0 days relative to the date of randomisation. Most participants are expected have a baseline VL result up to a maximum of 4-8 weeks before randomisation, but we allow up to 14 weeks for the few patients whose next regular visit was scheduled up to three months later, and it was not possible to get the patient to return sooner. This was anticipated and allowed for in the protocol. For all other laboratory results, the "baseline" result is defined as that closest to randomisation, with a window of -14 days to +14 days relative to the date of randomisation. We will present data summarising the timing of baseline viral load and other laboratory results relative to the date of randomisation. The same window of -14 to +14 days will be used for defining baseline clinical measurements such as weight and blood pressure.
There will be no independent programming of the primary outcome, rather we rely on the experience of the team for the accuracy of the data, analyses and interpretations; and the results will be assessed as a whole for consistency.

Outcome definitions
The primary and secondary outcomes based on VL categories are as defined in section 3.1. For participants who did not have VL measured within the required window (including due to LTFU or death, or had VL measured but had changed regimen line due to clinical failure [see below]), VL are considered as missing.
Change in regimen line due to clinical failure is defined as a change to a new drug class, e.g. from NNRTI to protease inhibitor, due to clinical failure defined by a new WHO 3 or 4 event. Therefore substitutions for side-effects, e.g. changing from tenofovir to zidovudine because of renal impairment, are not considered change in regimen line. Further, changes to regimen line due to other reasons, for example due to stock outs, are not included in this definition. There are anticipated to be few participants changing regimen line due to clinical failure or stock outs.
The secondary outcome of adherence will be assessed by self-reported doses missed. We define participants reporting having missed no doses in the last 4 weeks as having good adherence.
As per the protocol, AE and SAE are graded by the study staff according to CTCAE (version 4, published 28 May 2009) (6).
In addition to the AEs reported, we will assess the laboratory measurements of ALT, AST, haemoglobin and creatinine (after conversion to estimated glomerular filtration rate (eGFR)) according to CTCAE grades 3 and 4 ( Table 2). eGFR will be calculated using the Cockcroft-Gault equation (7): where constant = 1.23 for men and 1.04 for women, and assessed as in Table 2. ALT (IU/L) "Alanine aminotransferase increased" Major drug resistance mutations will be determined according to the Stanford University Resistance Database (8).

Analysis methods
Continuous variables will be inspected using histograms: 1) to assess for outliers which may be queried for accuracy, and 2) to assess whether appropriate transformations are required for modelling.
Outcomes will be summarised using means and standard errors for continuous variables and numbers and percentages for categorical variables, by randomised group. Percentages will be reported to zero decimal places, unless <0.5% when they will be given to one decimal place.
We will present numbers and proportions of participants with VL below different thresholds (<50, <100, <200, <400, <1000 copies/m) at 9 months (and <50 copies/ml at 6 months), with those missing VL results within the required analysis windows as defined in section 3.7 (including due to LTFU, death or regimen line change due to clinical failure (defined in section 6.1)) counted as "failures". We will provide details about the reasons for missing VLs (i.e., LTFU for that visit, died, regimen line change due to clinical failure).
For the analysis of the primary outcome, we will use a logistic regression model adjusted for the randomisation stratification factors (demographic groups (children vs pregnant women vs nonpregnant adults), VL at enrolment (second VL 100-599 vs 600-999 copies/mL)), but not centres since we have few participants at a number of centres. LTFU, death and regimen line change due to clinical failure prior to determination of the primary endpoint will be considered as failures. Results will be reported as odds ratios with 95% confidence intervals. Further, we will estimate the risk difference, with 95% confidence intervals estimated using the delta method (9).
In a sensitivity analysis for the primary outcome, we will analyse the per protocol set, considering only participants that finished the 9 months according to the protocol (=alive, retained in care, no change in regimen line other than that indicated by the randomisation (regardless of the reason), VL measurement available at 9 months). We will summarise the data (numbers and proportions of participants included in this analysis, and meeting the primary outcome), and fit logistic regression models as for the main primary outcome analysis.
In a further sensitivity analysis for the primary outcome, we will repeat the analyses using only VLs measured within the protocol-defined windows (summarising the data, and fitting logistic regression models as above).
For assessment of effect modification of the primary outcome by demographic groups and VL at enrolment (see section 3.1), analyses will be performed as for the main primary outcome analysis, and including in the models an interaction between the potential effect modifiers and randomised group, with recognition that power may be low. If there are small numbers of participants within certain subgroups (for example, children and/or pregnant women), then such an approach may not be feasible and instead we would restrict the model to the subgroups of sufficient size.
For assessment of the subgroup analysis (those with >0.5 drop in log10 VL), analyses will be performed as for the main primary outcome analysis, with the dataset restricted to individuals showing a >0.5 drop in log10 VL from the first to the second VL measurements pre-randomization.
For the secondary outcomes defined by different VL thresholds or time points, we will follow the same approach as for the main primary outcome analysis.
For the outcome of sustained virologic failure, we will also follow the same approach as for the main primary outcome analysis. We will summarise the number and proportion of participants with both 6 and 9 month VL results, and whether <50 copies/ml at one or both. For participants missing either the 6 or 9 month VL (but not both), if the non-missing VL is <50 copies/ml then the participant will be considered NOT to have sustained virologic failure. In line with the main primary analyses, participants missing both 6 and 9 month VL results will be considered as failures (i.e., has sustained virologic failure). Similarly, participants missing either the 6 or 9 month VL and with the non-missing VL ≥50 copies/ml, then the participant will be considered as a failure. As for the main primary outcome analyses, "missing VL" here means no VL measured within the required analysis window (including due to LTFU or death, or regimen line change due to clinical failure).
For the adherence outcome, we will report the number and proportion of participants with good adherence (see definition in section 6.1) at each of the scheduled visits, by randomised group. To compare between groups, we will use a logistic regression model for each of these timepoints separately, adjusted for the same variables as the primary outcome analysis. The analysis of this outcome will be restricted to participants who attended and answered the adherence question at the respective follow-up visits.
We will also assess overall adherence over the whole trial, reporting the number and proportion of participants with good adherence at every one of the 3, 6 and 9 month visits, versus poor adherence at ≥1 of these three visits. For participants without adherence data at all three visits, if the participant had poor adherence reported for at least one of the visits then he/she will be included as having overall poor adherence; otherwise he/she will be excluded from the analysis. To compare between groups, we will fit a logistic regression model, adjusted for the same variables as the primary outcome analysis.
We will also consider a repeated measures analysis, using a mixed effects logistic regression model with the outcome of good adherence and each participant contributing up to three measurements (one from each of the three visits at 3, 6 and 9 months). Visit will be included in the model as a fixed effect, along with the same variables as the primary outcome analysis. Participant will be included as a random effect. Assessment will be made of the quadrature fit, with generalised estimating equations considered as an alternative if necessary.
For the clinical outcomes of changes at 9 months versus baseline in weight, haemoglobin, CD4 count, and lipids (total cholesterol, HDL cholesterol, LDL cholesterol, triglycerides), we will present the mean values and mean changes with standard errors, by group. We will compare between groups using linear regression, adjusted for the same variables as in the primary outcome analysis plus the baseline value of the respective outcome. The analysis of these outcomes will initially be restricted to participants who have the respective outcome data at 9 months. If there is a lot of missing outcome data, then we will also consider using multiple imputation for these outcomes, using baseline variables and measurements from the 3 and 6 month visits as appropriate (10). In addition, we will perform similar analyses for BMI, systolic and diastolic blood pressure, and total cholesterol/HDL cholesterol ratio. Of note, BMI and systolic and diastolic blood pressure will only be assessed among adults.
We will present the number and proportion of participants who had any new WHO 3 or 4 events, or died, by 9 months. We will describe the events. If data allow (i.e., enough events), we will compare between groups using Cox proportional hazards regression (with outcome of time to any new WHO 3 or 4 events, or death), adjusted for the stratification factors. This analysis will include follow-up time to the last visit for each participant, with those not observed to have any such events censored at the minimum of the last visit or 130 weeks after randomisation.
We will present the number and proportion of participants who had any grade 3 or 4 AEs by 9 months, and summarise the overall numbers of AEs, grading, relationship to treatment, treatment changes, outcomes, event names. We will compare the number of participants with any grade 3 or 4 AEs between groups using logistic regression, adjusted for the stratification factors. This analysis will be done firstly among all participants (with the limitation that there may be unreported adverse events among participants who did not attended the final visit) and secondly restricted to participants who completed follow-up at 9 months. We will present and analyse SAEs similarly.
For the analyses of outcomes such as WHO events, deaths, AEs and SAEs which are described above as "by 9 months", we will include any such events occurring up to the upper limit of the 9 month analysis window, namely 130 weeks (see section 3.7). If we are notified of any events occurring after this time, we will describe such events but they would not be included in the analyses.
We will present the number and proportion of participants with major drug resistance mutations at baseline (overall, and separately by randomised group; also reporting if any samples did not undergo successful RT-PCR amplification). We will present similar results for participants with unsuppressed viral load (>50 copies/ml) at 9 months, by randomised group.

Missing data
Where applicable, percentages will be of non-missing values, with the number (%) of missing values given if data are not complete. As detailed in section 6.2, main analyses of the primary outcomeand the virological secondary outcomes -will include all participants as randomised with missing data counted as failures. For the remaining secondary outcomes, section 6.2 details how missing data will be handled in each case.

Additional analyses
The long term follow up endpoint at 24 months will be assessed once all participants have passed that time point. All participants will be included as randomised, with those LTFU, died or changed regimen counted as failures like for the primary outcome analysis. Of note, dolutegravir (an integrase strand transfer inhibitor, INSTI) is being rolled out in Lesotho in 2020, and it is likely that a substantial proportion of our trial participants will be transitioned to dolutegravir before their 24 month follow up, potentially rendering the planned long term follow up analysis of limited clinical interest.
A separate analysis plan will be developed for the cost-effectiveness analyses (an exploratory outcome).

Harms
Safety data are included as secondary endpoints (see section 6.2).
For the laboratory parameters indicated in Table 3, the CTCAE categorisations will be applied as indicated in the table. We will present the number and proportion of participants with a measurement at each scheduled visit, by randomised group; and the numbers and proportions of participants with gradings 3 and 4 for each scheduled visit, by randomised group, of those with a measurement at the given visit. Of note, laboratory parameters graded 1-4 according to CTCAE would not have been reported as adverse events if they were determined by the clinical team to be not clinically significant. Laboratory abnormalities which occurred before study enrolment would be captured by the comorbidity question on the baseline form but would not otherwise be captured as an adverse event.

Shell tables
Presented below are the shell tables for the baseline characteristics, to illustrate the variables included. Further tables will be developed to present the analyses described above. [2] Female: at least 2 drinks every day or regularly 4 or more drinks on one occasion; Male: at least 3 drinks every day or regularly 5 or more drinks on one occasion.   (7).