Virological failure on first-line antiretroviral therapy; associated factors and a pragmatic approach for switching to second line therapy–evidence from a prospective cohort study in rural South-Western Uganda, 2004-2011

Introduction We investigated factors affecting Virological failure (VF) on first line Antiretroviral Therapy (ART) and evaluated a pragmatic approach to switching to second line ART. Methods Between 2004 and 2011, we assessed adults taking ART. After 6 months or more on ART, participants with VL >1000 copies/ml or two successive VL > 400 copies/ml (Conventional VF) received intensified adherence counselling and continued on first-line ART for 6 more months, after which participants who still had VL > 1000 copies/ml (Pragmatic VF) were switched to second line ART. VF rates were calculated and predictors of failure were found by fitting logistic regression and Cox proportional hazards models. Results The 316 participants accrued 1036 person years at risk (pyar), 84 (26.6%) had conventional VF (rate 8.6 per 100 pyar) of whom 28 (33.3%) had pragmatic VF (rate 2.7 per 100 pyar). Independent predictors of conventional VF were; alcohol consumption, (adjusted Hazard Ratio; aHR = 1.71, 95% CI 1.05-2.79, P = 0.03) and ART adherence: per 10% decrease in proportion of adherent visits, (aHR = 1.83, 95% CI 1.50-2.23; P < 0.001). Using reference age group < 30 years, among conventional failures, the adjusted odds ratio (aOR) of pragmatic failure for age group 30-39 years were 0.12, 95% CI 0.03-0.57, P = 0.02 and for age group > 40 years were 0.14, 95%CI 0.03-0.71, P = 0.02. Alcohol consumers had a threefold odds of pragmatic failure than non-alcohol consumers (aOR = 3.14, 95%CI 0.95-10.34, P = 0.06). Conclusion A pragmatic VF approach is essential to guide switching to second line ART. Patient tailored ART adherence counselling among young patients and alcohol users is recommended.


Introduction
Antiretroviral therapy (ART) has changed the natural history of HIV infection [1]. By the end of 2015 about 36.7 million individuals were living with HIV globally and 25.8 million of these were living in sub-Saharan Africa (SSA) [2,3]. By June 2016, ART coverage had increased to 46% globally and 54% in SSA [2]. Uganda's HIV prevalence by 2014 was estimated at 7.4% and out of the estimated 1.5 million people living with HIV-AIDS (PLWHA)by the end of 2014, 0.75 million people were on ART and about 10% of these had HIV viral load of above 1000 copies/ml [4]. ART initiation reduces HIV replication in peripheral blood [5][6][7][8] suppresses plasma HIV viral loads (VL) to unquantifiable levels within 4-6 months [9,10], reduces morbidity and mortality, with resultant improvement in survival [1,5]. If adequate viral suppression is not achieved, therapy is considered to be failing and may require switching to a second line ART regimen [11,12]  Many other factors are independently associated with failure to achieve optimal HIV viral suppression [19]. Knowledge of such predictors will enable clinicians forecast ART outcomes and design interventions to prevent virological failure and meet the UNAIDS 90-90-90 treatment targets. Diagnosis of virological failure based on a single plasma VL measurement of > 1000 copies/ml or two successive VL measurements above 400 copies/ml, at any time after 6 months on ART, (conventional VF), has led to unnecessary switching from first to second line ART regimen which is costly and almost currently the last available treatment option in most SSA countries [22]. If patients with detectable VL after 6 months on ART receive further intensified ART adherence counselling and continue on first-line ART for another 6 months, only those whose VL remains (Roche, Roche molecular systems, NJ, USA) assay was used (lower detection limit of 400 copies per ml).

Data management and statistical analysis
Software and participant follow up: Data was managed using an Ms Access database and was analysed in Stata version 13.0.
Participants' characteristics at ART initiation between January 2004 and March 2011 were summarised by cross tabulation. The study had outcomes based on 2 virological failure definitions (conventional and pragmatic) at different time points. A participant's follow-up started at ART initiation and ended at the earliest time of diagnosis of virological failure, death, loss to follow up, or date last seen (whichever was first). Loss to follow up was defined as being late for 3 months or more after the next scheduled quarterly visit and still known to be alive. Participants were censored at the end of the study.
Ethics and consent: this study was approved by the Science and Ethics Committee of the Uganda Virus Research Institute and the Uganda National Council for Science and Technology. Participants gave signed or thumb-printed written informed consent to participate in the study and confidentiality procedures were adhered to throughout the study Outcome: The main outcome was virological failure. Conventional virological failure was defined as a single VL of > 1000 copies/ml or two successive VL > 400 copies/ml taken 6 months apart from a participant on ART for 6 months or more. Participants with conventional virological failure received intensified adherence counselling and continued on first line ART for 6 more months after which those with persistent VL > 1000 copies/ml were regarded to be true or Pragmatic failures and all were switched to second line ART regimen.

Exposures:
The main exposure was ART adherence which was measured monthly at each ART refill visit using both self-report and pill count. Participants reported the number of days they missed taking their doses over the previous four days and since the last ART refill visit. Adherence was categorised as: good (> 95%) if no doses were skipped over the last 4 days or since the last refill visit and poor (<95%) if any doses were missed over the same period.
ART adherence data for each participant over the entire follow up period or up to the visit of virological failure was collapsed to create an overall adherence summary measure. The proportion of visits with good adherence was presented for each participant as categorical for crude analysis. In addition, the continuous variable (proportion of visits with good adherence) was rescaled so that 1 unit represented a 10% decrease in the proportion of visits with good adherence (e.g. from 100% to 90%) and this was used in the regression model. Secondary exposures included characteristics measured at ART initiation including gender, age, marital status, number of partners, occupation, alcohol consumption, CD4 counts, HIV viral loads, weight and body mass index (BMI).
Conceptual framework: Our analysis was based on a conceptual framework which assumed; ART adherence was independently associated with virological failure. Other confounders included; socio-demographic (age, gender, marital status, occupation and level of education) behavioural (alcohol consumption and number of sexual partners) and biological (body mass index, baseline CD4 counts and baseline VL) risk factors. The association with ART adherence was considered distal for apriori confounders age and gender and proximal for alcohol consumption. Biological factors such as, high baseline VL was associated with low baseline CD4 counts, poor health, low BMI and poor ART adherence. We also assumed that, a high baseline VL provides a potential for ART drug  (Table 1). We censored 22 (6%) individuals who initiated ART, 12 (4.1%) died before a diagnosis of virological failure (4 males and 8 females) while 10 (3.2%) were lost to follow up (4 males and 6 females). Twelve (4.1%) participants died after conventional virological failure and so were censored in the analysis of pragmatic failure. Participants who died were older than those who did not die (P = 0.05), were more immunosuppressed (P = 0.01) and more wasted (P = 0.01).
Participants accrued 1035 person years at risk (pyar) and the median follow up time was 2.97 years (IQR), range 0.23 to 7.11 years. Five years after ART initiation, about 30% of the participants had developed conventional failure ( Figure 2) and 10% of the participants had developed pragmatic virological failure ( Figure 3).

Effect of adherence on conventional virological failure:
Overall, 84 (26.6%) participants developed conventional virological failure, a rate of 8.6 per 100 pyar. There were no significant differences in failure rates by gender, age and alcohol consumption.
The median proportion of visits on which good adherence was reported was 97.0% (IQR (92-100). Conventional failure rates were significantly higher among participants who reported adherence less than 90% compared with participants who reported good adherence of above 95% of the visits (crude HR 1.38 for linear association with 10% drop in the proportion of visits with good adherence, 95% CI 1.23-1.55, P<0.001) ( Table 2).

Effect of adherence on pragmatic virological failure:
After the intensified ART adherence counselling, 28 (33.3%) of the 84 participants with conventional failure still had persistent VL > 1000 copies/ml, a pragmatic failure rate of 2.7 per 100 pyar. The pragmatic failure rate decreased significantly with increasing age (P = 0.04), but increased with decrease in proportion of visits on which good adherence was reported (P < 0.001) ( Table 3). The unadjusted linear hazard ratio indicated that pragmatic failure rate increased by 45.0% for every 10% decrease in the proportion of visits on which good adherence was reported (HR 1.45, 95%CI 1.26-1.71; P < 0.001). After adjusting for age, gender, alcohol consumption, baseline CD4 counts and baseline VL, the adjusted linear hazard ratio for the failure rate per 10% decrease in the proportion of visits on which good adherence was reported increased to 2.12 (95%CI 1.61-2.79, P < 0.001).

Independent predictors of conventional virological failure:
After adjusting for age and gender, the only independent predictors of conventional virological failure were adherence and alcohol consumption. For each drop of 10% in the proportion of visits at which a participant achieved good adherence, the rate of failure increased by 83% (aHR 1.83, 95%CI 1.50-2.23, P < 0.001).

Independent predictors of pragmatic failure among
individuals with conventional failure: We adjusted for apriori confounders; age, gender and alcohol consumption to determine predictors of pragmatic virological failure among participants with conventional virological failure. There was an 88% reduction in the odds of pragmatic failure from age group < 30 years to age group 30-39 years (adjusted odds ratio (aOR) 0.12, 95% CI 0.03-0.57, P = 0.01) and an 86% reduction in the odds of pragmatic failure from age group <30 years to age group 40 years and above (aOR = 0.14, 95%CI (0.03-0.71) P = 0.02). There was borderline evidence of a threefold increase in the odds of pragmatic failure among participants who reported consuming alcohol compared to those who reported not consuming alcohol (aOR = 3.14, 95% CI 0.95-10.34), P = 0.06). However, adherence was not significantly associated with pragmatic failure among participants with conventional failure (P = 0.44) ( Table 5). Death, migration and missed appointments were well recorded and loss to follow up was low at 3.2%, which minimised the likelihood of selection bias. The few participants with missing baseline data were comparable to those without missing data. Regular viral loads measurement reduced the likelihood of misclassification of outcome.

Discussion
Participants were recruited from a general population based cohort minimising selection bias and enhancing generalizability of findings.
The study had a robust data collection mechanism, with trained staff and laboratory equipment were subjected to regular external validity checks and quality assurance to minimise errors in the measurement of viral load and CD4 count.
Limitations: Our estimates were subjected to random error by having few participants aged below 30 years and above 50 years. This undermined exploration of the age effect like to find the actual age below or above which failure is more or less likely. Few participants had pragmatic virological failure which might have led to imprecise estimates and low power to detect associations between risk factors and outcome. Participants who died differed significantly from those who did not die and were likely to be older, have lower BMI and were more immunosuppressed. Treating deaths as censored observations might have biased the estimates of virological failure especially if those who died were more likely to have developed virological failure. The measured self-reported alcohol consumption was prone to social desirability bias, and the quantity, type and frequency of alcohol consumption was not recorded. The likely recall bias arising from self-reported ART adherence was minimised by asking participants if they skipped any doses in the last 4 days, however, the analysis of adherence as a summary variable ignored the temporal variation of adherence. We assessed a limited number of potential confounders so virological failure estimates might be subjected to residual confounding. We did not establish mortality or test ART drug resistance mutations within the 6 months participants with virological failure were subjected to strict adherence counselling, which would have been important in evaluating the safety of the pragmatic failure approach.

Conclusion
Our findings reinforce the need to sustain viral load testing as a key tool in monitoring treatment progress and to use the safe and cheap pragmatic approach when switching participants from first to second line ART in resource limited settings to preserve second line ART regimen. To achieve the 90-90-90 treatment target by 2020 in SSA, the strategy of rolling out viral load testing should be supplemented with innovative adherence enforcement to achieve sustained viral suppression among patients on ART. Maintaining adequate stocks of ART and patient tailored counselling especially for the young and alcohol users will prevent poor adherence and virological failure.
National governments should increase funding for recruiting or training staff in ART adherence counselling. Qualitative studies are however needed to assess the factors responsible for virological failure among the alcohol consumers and young people as this will help design strategies to address adherence issues and virological failure in these groups.

What is known about this topic
 WHO guidelines currently recommend viral load testing as the gold standard for monitoring and confirming treatment response;  WHO also recommends Intensified adherence counselling following an initial high viral load (> 1000 copies/mL) before conducting a second viral load test;  Adherence is a known independent predictor of virological failure.

Acknowledgments
We acknowledge the contribution of RCC study participants, RCC study team and staff of the Clinical Diagnostic Laboratory Services (CDLS).