Secondary Logo

Journal Logo

Clinical Science

Addition of Etravirine Does Not Enhance the Initial Decline of HIV-1 RNA in Treatment-Experienced Patients Receiving Raltegravir

Flandre, Philippe PhD*; Marcelin, Anne-Geneviève PhD; Calvez, Vincent MD, PhD; on behalf of the ANRS AC11 Resistance Study Group

Author Information
JAIDS Journal of Acquired Immune Deficiency Syndromes: August 1, 2017 - Volume 75 - Issue 4 - p 448-454
doi: 10.1097/QAI.0000000000001435

Abstract

INTRODUCTION

Highly active antiretroviral therapy is the standard of care for patients with HIV infection and is usually the combination of 3 antiretroviral agents.1 Raltegravir (RAL) was the first integrase strand-transfer inhibitor that prevents proviral DNA-strand transfer.2,3 In randomized trials, RAL showed good tolerability and proved remarkably efficient in combination therapy at reducing viral loads both in treatment-naive and antiretroviral-experienced patients.2,4,5 Nonnucleoside reverse transcriptase inhibitors (NNRTIs) are frequently used components of combination antiretroviral therapy.1 Etravirine is an expanded-spectrum NNRTI with a potent and broad in vitro activity against HIV-1, including virus with NNRTI resistance-associated mutations.6 Its efficacy and safety in treatment-experienced patients have been demonstrated in phase III trials.7,8

Current outcomes in HIV-1 infected study are percentages of patients with plasma HIV-1 RNA level <50 copies over time, and the percentage at week 48 is widely used as primary endpoint in randomized clinical trials. RAL-based regimen had rapid antiretroviral activity in treatment-naive and treatment-experienced patients with a much higher percent of patients with HIV-1 RNA <50 copies/mL for weeks 2–16 compared with other regimen.2,5 The clinical significance of a more rapid HIV-1 RNA decline has not been clearly established, but some authors suggested that early virological response to therapy increases the likelihood of maintaining or obtaining a latter response.4,9,10 There is also legitimate concern that minority drug-resistant mutants may be selected during the initial HIV-1 RNA decay phase after antiretroviral therapy initiation, thus undermining efficacy of treatment.11 Continuous outcome such as endpoint based on the magnitude of reduction in viral load usually provides more information than a binary outcome (below or above 50 copies/mL). Reduction in plasma HIV-1 RNA is also used in trials as a complement to endpoints based on percentage of patients achieving virological suppression.12 Interpretations of magnitude of reduction, however, can be biased by limits of detection of virological assays, particularly lower limits of quantification. Survival methods, including Kaplan–Meier curves, have been proposed to take into account censoring of HIV-1 RNA reduction.12

Observational resistance studies in HIV-infected patients involve participants with treatment failure on their current regimen, who were thus given a new regimen including the drug under study for resistance. Furthermore, other drugs may be introduced in the regimen leading to a combination of more than 3 drugs, with most of the new drugs introduced for the first time. The choice of the drugs and the number of drugs in the new regimen is based on some patient's characteristics at failure including levels of viral load and previous antiretroviral exposure. Then, for a specific drug, treated and nontreated participants differ in their distribution of confounding variables. Methods using propensity scores have expanded the analytic tools available to researchers to make unbiased comparisons between treatment groups in observational studies.13 Inverse probability of treatment weighting (IPTW) using propensity score and propensity score matching have been suggested as promising methods with time-to-event outcomes.14,15 Adjusted Kaplan–Meier estimator and corresponding log-rank test have been introduced to reduce confounding effects using IPTW with time-invariant treatment.16

In this work, we investigate in treatment-experienced patients receiving a RAL-containing regimen for the first time whether the additional introduction of etravirine for the first time increases the effectiveness of such a combination. Specifically, we explore whether etravirine enhance the early and rapid decline of viral load observed previously in patients receiving a RAL-based regimen.2,4 Data from an observational resistance study are used with HIV-1 RNA reduction at week 8 as outcome. Crude and adjusted methods for confounding are applied to the data for comparisons.

MATERIAL AND METHODS

Study Population

In 2012, the Agence Nationale de Recherche sur le SIDA (ANRS) conducted a resistance study in patients receiving a RAL-containing regimen for the first time.17 All patients included in the study experienced virological failure and were treatment-experienced but had not previously received RAL or another integrase strand-transfer inhibitor. In each center, sociodemographic, clinical, and virological data were collected for all patients treated with the studied drug and an optimized background regimen. The study investigated the potential impact of natural polymorphisms in the integrase gene on the virological response. The main analysis, however, did not suggest that these baseline integrase polymorphic mutations were associated with the virological response.17 In addition to RAL, etravirine, and darunavir, a protease inhibitor (PI), were 2 other “recent” drugs that were introduced for the first time in many patients. Patients receiving a PI, including darunavir, and a NNRTI, except etravirine, for the first time, were excluded of the present analysis. Then, in this study, we analyzed the 349 patients receiving either RAL for the first time or RAL and etravirine for the first time.

Statistical Methods

The outcome was the HIV-1 RNA reduction from baseline to week 8 (±2 weeks). The crude method is simply to define all HIV-1 RNA levels at week 8 below the limit of quantification to be equal to the limit of quantification. The 2 limits of quantification used by the virology laboratories at this time were 40 and 50 copies/mL. Usually HIV-1 RNA levels are transformed using the log10 scale. For example, a patient having HIV-1 RNA measurements of 2000 copies/mL at baseline and 200 copies/mL at week 8 has an observed reduction of 1 log10. In the crude method, a magnitude of reduction can be calculated for all patients, and standard statistical procedures can be used to compare treatment groups. For instance, a standard analysis with a linear model and a Wilcoxon nonparametric test were performed to compare treatment groups.

Assuming that the baseline HIV-1 RNA measurement is within the range of quantification, a censored reduction is observed when the measurement at week 8 is below the limit of quantification. An equivalent structure of data exists between a censored reduction in HIV-1 RNA and censored time-to-event.12 Thus, Kaplan–Meier method is used to estimate the cumulative frequency distribution of the magnitude of HIV-1 RNA reduction. Such method has already been used with reduction of viral load and other biomarkers.18,19 Further analyses based on Kaplan–Meier estimates can also be made such as log-rank test to compare treatment groups with respect to viral load reduction. It was demonstrated that the crude method provides a biased estimate of the median reduction and that censored method greatly improves such estimates.12 An apparent difficulty is that a reduction in HIV-1 RNA level may be negative (ie, when HIV-1 RNA levels rise from baseline to week 8), although time-to-event data remain positive. This can be easily circumvented by adding any single number (+2 in our data) to each HIV-1 RNA reduction, so that all observations are positive and then subtracting this number from all observations subsequent to the analysis.

Considering that HIV-1 RNA levels are at least approximately normally distributed after transformation on the log10 scale, parametric analyses may be used.12 Such parametric analyses assumed that the cumulative frequency distribution follow the form of a normal distribution. Comparison of residuals obtained with the Kaplan–Meier method to those obtained with a parametric model provides a check of the normality assumption. Such parametric analyses provide estimated mean adjusted for censoring.

An implicit assumption of the Kaplan–Meier and other survival analysis is the so-called noninformative censoring assumption. In our setting, noninformative censoring corresponds to the assumption that the magnitude of reduction is not related to the baseline HIV-1 RNA level.12 If this is not satisfied, further analyses on treatment comparisons should be adjusted by baseline HIV-1 RNA levels. We investigated the dependence of censoring and baseline HIV-1 RNA level by an analysis of covariance introduced for censored data.12 Both parametric and nonparametric approaches have been used to investigate the noninformative censoring assumption. Baseline HIV-1 RNA levels, however, is one of the main confounding variables considered in the propensity score approach described below.

Propensity score methods are increasingly being applied to reduce or minimize the effects of confounding when estimating the effects of treatments when using nonrandomized studies.13,15 The propensity score is the probability of receiving a given treatment, conditional on observed baseline variables. It is a balancing score: conditional on the propensity score, the distribution of measured observed baseline covariates is expected to be balanced in the 2 treatment groups. Among the different propensity score methods, IPTW and propensity score matching are recommended for survival outcomes.14 Propensity scores are unknown and usually estimated using a parametric model. Stabilized weights were used as IPTW using logistic regression models. For the ith subject, the stabilized weight is where is the marginal probability of receiving RAL and etravirine and is the probability of receiving RAL and etravirine conditional on the baseline covariates X. Covariates X were selected among the following variables: baseline viral load, baseline CD4 cell count, age, sex, subtype (B or non-B subtype), number of previous drugs used, and the ANRS global resistance score for drugs in the optimized background regimen (drugs other than RAL and etravirine). The global score is calculated as follows: For each antiretroviral drug, we determined a resistance score of 0 or 1 depending on whether the virus was classified as susceptible/intermediate resistant or resistant according to the ANRS algorithm, respectively. The global resistance score is the sum of all these scores. The score represents, in addition to the number of drugs previously received, the treatment failure resistance history of the patient.

Stabilized weights were estimated in following recent recommendations.20 In particular, we investigated different specifications for continuous covariates including linear terms (continuous), 4-knot restricted cubic splines, and categories using the median value of the corresponding variable. Criteria to retain the best subset of variables with best specifications were based on mean, SD, and extreme values of estimated weights and balance diagnostics including computation of standardized difference in the weighted sample.20,21 Adjusted Kaplan–Meier estimates using IPTW provide confounder-adjusted “survival” curves and log-rank statistics.16 All statistical analyses were performed with SAS version 9.4. In particular, the Proc Lifetest with the weight option was used to compute adjusted Kaplan–Meier curves.

RESULTS

Of the 349 patients included, 191 (55%) patients started RAL and etravirine (RAL/etravirine group) for the first time, and 158 patients started only RAL (RAL group) for the first time. Baseline characteristics are summarized in Table 1. Patients in the RAL/etravirine group had higher levels of baseline HIV-1 RNA (median: 4.6 versus 3.9 log10 copies/mm3) and a lower number of CD4 cell count (123 versus 242 cells/mm3) than patients in the RAL group. Furthermore, patients in the RAL/etravirine group had previously received more antiretroviral drugs than patients in the RAL group (median, 13 versus 11), leading also to a higher number of resistant drugs according to the ANRS algorithm (median, 13 versus 9). There was no difference in the number of nucleoside reverse transcriptase inhibitors and PIs in the optimized background regimen between the 2 treatment groups. Resistance-associated mutations to etravirine are displayed in Table 2 (IAS-USA list),22 and prevalence of integrase polymorphic mutations is shown in the Supplemental Digital Content, Figure 1, https://links.lww.com/QAI/B50. Overall, 178 (51%) patients had a plasma HIV-1 RNA level below the limit of quantification (50 or 40 copies/mL) at week 8. There was a lower percentage of patients below the limit of quantification in the RAL/etravirine group than in the RAL group (48% versus 55%), which may be partially explained by the lower level of baseline HIV-1 RNA in the latter group.

TABLE 1.
TABLE 1.:
Characteristics of the 349 Patients Participating in This Study
TABLE 2.
TABLE 2.:
Prevalence of Resistance-Associated Mutations to Etravirine (IAS-USA Drug Mutations List January 2017)
FIGURE 1.
FIGURE 1.:
Kaplan–Meier estimates and corresponding log-rank test of the distribution of HIV-1 RNA reductions between baseline and week 8.

Using the crude method (in which all values below the limit of quantification were replaced by the value of 40 or 50 copies/mL), estimated mean and median reductions were 2.08 and 2.35 in the RAL/etravirine group and 1.82 and 1.75 in the RAL group, respectively (Table 3). The means difference was statistically significant, P = 0.025, and a Wilcoxon rank test between the 2 groups was also significant (P = 0.018). The use of censored methods provided not only larger estimated mean values and medians but also much smaller differences between treatment groups. Estimated median reductions were 2.77 and 2.93 in the RAL/etravirine and RAL groups, respectively. The Kaplan–Meier estimates of the distribution of HIV-1 RNA reduction lead to a nonsignificant difference between the 2 groups (Fig. 1, P = 0.92). Using parametric analysis adjusted for censoring, estimated mean reduction were 2.78 and 2.67 in the RAL/etravirine and RAL groups, respectively (Table 3, P = 0.58).

TABLE 3.
TABLE 3.:
HIV-1 RNA Reductions in log10 copies per milliliter Differences and Significance of Differences Between Treatment Groups

To assess the extent and effect of informative censoring, treatment comparisons were repeated, adjusting for the effect of baseline HIV-1 RNA using an analysis of covariance. The baseline HIV-1 RNA effect was significant using both nonparametric (P = 0.02) and parametric analyses (P = 0.04); patients with higher baseline levels tend to have larger reductions. Such results implied that treatment comparisons are affected by informative censoring resulting from a dependence on baseline HIV-1 RNA level. Although differences between treatment groups were slightly altered after adjustment for baseline level, conclusions based on P value remain unchanged (P = 0.96 and P = 0.53, Table 3).

The logistic model was used to estimate stabilized weights (IPTW). The final stabilized weights involved the following 4 variables: baseline HIV-1 RNA, baseline CD4, the number of previous drugs received, and the ANRS global resistance score for drugs other than RAL and etravirine. The “best” weights were obtained with the 4 variables included in the model as categories using median values of each variable. The mean of estimated stabilized weights was 0.997 (SD, 0.45); the 1/minimum and maximum estimated weights were 1.72 and 2.49, respectively. Absolute standardized difference in the weighted sample were 2.2, 1.2, 1.8, and 0.9% for baseline HIV-1 RNA, baseline CD4, the number of previous drugs received, and the ANRS global resistance score, respectively. Other diagnostics including the comparison of interactions and higher-order moments in the 2 treatment groups could not be computed because none of the 4 variables included in the propensity score model was used as continuous variables. Adjusted Kaplan–Meier survival curves (Kaplan–Meier curves in the weighted sample) for the 2 treatment groups are reported in Figure 2. The 2 curves are somewhat similar to those displayed in Figure 1 and provided a nonsignificant difference between the 2 treatment groups (P = 0.59). Then, after adjusting for both censoring due to the limit of quantification and imbalanced baseline characteristics in patients between the 2 treatment groups, there was no statistical difference in the HIV-1 RNA reduction at week 8 between the RAL/etravirine and RAL groups.

FIGURE 2.
FIGURE 2.:
Adjusted Kaplan–Meier estimates using IPTW and corresponding log-rank test of the distribution of HIV-1 RNA reductions between baseline and week 8.

DISCUSSION

The crude method defining all HIV-1 RNA levels below the limit of quantification to be equal to the limit of quantification is straightforward but is known as providing biased estimates.12 Such a bias concerned not only estimates of the magnitude of HIV-1 RNA reduction but also difference estimates between treatment groups. The method, however, is still widely used in the publication and communication of randomized and nonrandomized studies. In our data, applying the crude method showed an apparent larger mean and median HIV-1 RNA reductions in patients receiving RAL and etravirine compared with patients receiving RAL (P < 0.05). A first analysis with the Kaplan–Meier method, that takes into account the censoring of HIV-1 RNA measurements by the limit of quantification, showed a nonsignificant difference in HIV-1 RNA reduction between these 2 treatment groups (P = 0.92). Although a covariance analysis demonstrated that the assumption of noninformative censoring was violated, implying that further analyses should be adjusted by the baseline HIV-1 RNA levels, such analyses lead also to nonsignificant differences between treatment groups. Such findings based on survival methods, however, may be suspected to bias because of the lack of randomization in our study. Survival analyses based on propensity score approach, taking into account confounding factors, confirmed that there was no significant difference in HIV-1 RNA reduction between patients receiving RAL plus etravirine and patients receiving RAL.

There is a clinical and virological interest of a more rapid HIV-1 RNA decline, although this question is still under debate.4,9,10 Percentages of patients below a limit of quantification in weeks 2–16 provide only a partial picture of a rapid decline in viral load. For example, in patients with viral suppression (HIV-1 RNA <50 copies/mL), it has been shown that a higher percentage of patients receiving a nevirapine-containing regimen had a residual viremia below 1 copy/mL compared with patients receiving an efavirenz-containing regimen.23 Some authors argued that patients achieving an early virological response to therapy and who continued therapy had a strong likelihood of maintaining or improving those responses for up to 96 weeks.10 Not only is initial suppression of plasma viremia critical for achieving desired treatment responses, but durability of that response is essential because it indicates tolerability and the lack of emergence of drug-resistant virus.4 It has been shown that drug-resistant viruses can be selected and be replicated even in the first weeks of suppressive ART; thus, potency of antiretroviral therapy during the initial treatment period is of particular interest.24 In particular, selection of preexisting minority variants of drug-resistant HIV-1 can lead to virological failure in patients who receive antiretroviral therapy with low genetic resistance barriers.25 Therefore, we think that there is still an interest in this topic, although in our study, antiretroviral-experienced patients receiving etravirine plus RAL did not exhibit a significantly higher reduction in HIV-1 RNA at week 8 than patients receiving RAL.

Several assumptions were required in the analyses of this work. A covariance analysis showed that the noninformative censoring assumption required for the Kaplan–Meier method was violated. In our context, this implied that the value at which a patient's HIV-1 RNA reduction becomes censored was determined by the patient's baseline HIV-1 RNA level.12 However, taking into account baseline HIV-1 RNA levels in both parametric and nonparametric survival analyses provided similar conclusions without such an adjustment. In addition, the baseline HIV-1 RNA variable is included in the propensity score model, and then both methods based on it are adjusted for baseline HIV-1 RNA levels. The use of a parametric version of the Kaplan–Meier method to estimate mean of HIV-1 RNA reduction assumes that the HIV-1 RNA level on the log10 scale is at least approximately normally distributed. Comparing both residuals obtained from the Kaplan–Meier method and from a parametric model confirmed that assumption (data not shown).

Causal inference using the propensity score requires 4 assumptions: consistency, unmeasured confounders, positivity, and no misspecification of the propensity score model.20 Consistency means that a subject's potential outcome under the treatment actually received is equal to the subject's observed outcome. Our work implicitly assumes that consistency holds, which is reasonable assumption when estimating the effect of medical treatments. No unmeasured confounding is a critical assumption in propensity score analyses. For the assumption of no unmeasured confounding to hold, we have to measure enough joint predictors of exposure and outcome such that, within the levels of these predictors, associations between exposure and outcome that are due to their common causes will disappear.20 It is known that variables that are unrelated or weakly related to the treatment exposure but related to the outcome should be included in the propensity score model.26 We believed that the 4 variables (baseline HIV-1 RNA, baseline CD4, antiretroviral treatment history, and the ANRS global resistance score) retained in the final propensity score model are the most important predictive factors of our week-8 outcome. Positivity is the condition that all subjects have a nonzero probability of receiving each treatment. Cole and Hernan recommend that the assumption is accepted when the mean of stabilized weight is close to one and when there are no extreme values.20 Such conditions are satisfied in our study with a mean of 0.997 and a 1/minimum and maximum estimated of 1.72 and 2.49. It was also noted that estimated weights with the mean far from one or very extreme values are indicative of misspecification of the propensity score model.20 We investigated different specifications of our model to choose the “best” specification according to Cole and Hernan,20 although we did not use other regression models or machine learning.

This study has several limitations. First, because this was an observational cohort study, selection bias may have been an issue. However, this was partly addressed by adjusting for potential baseline confounders. Second, optimized background regimens were heterogeneous, and no information of compliance on regimens was recorded. Third, sample sizes in the 2 treatment groups were relatively small.

In conclusion, we did not demonstrate that antiretroviral-experienced patients receiving RAL and etravirine for the first time had a higher HIV-1 RNA reduction at week 8 compared with patients receiving RAL for the first time. We aimed at comparing crude method, censored methods, and adjusted censored methods, both to estimate and compare HIV-1 RNA reduction between treatment groups. Our findings confirmed that crude methods provided biased estimate of viral load reduction. The different steps of our analyses may be used for many other biological variables censored with some limits of quantification.

REFERENCES

1. Gunthard HF, Saag MS, Benson CA, et al. Antiretroviral drugs for treatment and prevention of HIV infection in adults: 2016 recommendations of the International Antiviral Society-USA Panel. JAMA. 2016;316:191–210.
2. Lennox JL, DeJesus E, Lazzarin A, et al. Safety and efficacy of raltegravir-based versus efavirenz-based combination therapy in treatment-naive patients with HIV-1 infection: a multicentre, double-blind randomised controlled trial. Lancet. 2009;374:796–806.
3. Summa V, Petrocchi A, Bonelli F, et al. Discovery of raltegravir, a potent, selective orally bioavailable HIV-integrase inhibitor for the treatment of HIV-AIDS infection. J Med Chem. 2008;51:5843–5855.
4. Markowitz M, Nguyen BY, Gotuzzo E, et al. Rapid and durable antiretroviral effect of the HIV-1 Integrase inhibitor raltegravir as part of combination therapy in treatment-naive patients with HIV-1 infection: results of a 48-week controlled study. J Acquir Immune Defic Syndr. 2007;46:125–133.
5. Steigbigel RT, Cooper DA, Kumar PN, et al. Raltegravir with optimized background therapy for resistant HIV-1 infection. N Engl J Med. 2008;359:339–354.
6. Andries K, Azijn H, Thielemans T, et al. TMC125, a novel next-generation nonnucleoside reverse transcriptase inhibitor active against nonnucleoside reverse transcriptase inhibitor-resistant human immunodeficiency virus type 1. Antimicrob Agents Chemother. 2004;48:4680–4686.
7. Lazzarin A, Campbell T, Clotet B, et al. Efficacy and safety of TMC125 (etravirine) in treatment-experienced HIV-1-infected patients in DUET-2: 24-week results from a randomised, double-blind, placebo-controlled trial. Lancet. 2007;370:39–48.
8. Madruga JV, Cahn P, Grinsztejn B, et al. Efficacy and safety of TMC125 (etravirine) in treatment-experienced HIV-1-infected patients in DUET-1: 24-week results from a randomised, double-blind, placebo-controlled trial. Lancet. 2007;370:29–38.
9. Murray JM, Emery S, Kelleher AD, et al. Antiretroviral therapy with the integrase inhibitor raltegravir alters decay kinetics of HIV, significantly reducing the second phase. AIDS. 2007;21:2315–2321.
10. Raffi F, Katlama C, Saag M, et al. Week-12 response to therapy as a predictor of week 24, 48, and 96 outcome in patients receiving the HIV fusion inhibitor enfuvirtide in the T-20 versus Optimized Regimen Only (TORO) trials. Clin Infect Dis. 2006;42:870–877.
11. Noguera-Julian M, Casadella M, Pou C, et al. Stable HIV-1 integrase diversity during initial HIV-1 RNA decay suggests complete blockade of plasma HIV-1 replication by effective raltegravir-containing salvage therapy. Virol J. 2013;10:350.
12. Marschner IC, Betensky RA, DeGruttola V, et al. Clinical trials using HIV-1 RNA-based primary endpoints: statistical analysis and potential biases. J Acquir Immune Defic Syndr Hum Retrovirol. 1999;20:220–227.
13. Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70:41–55.
14. Austin PC. The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments. Stat Med. 2014;33:1242–1258.
15. Austin PC. The performance of different propensity score methods for estimating marginal hazard ratios. Stat Med. 2013;32:2837–2849.
16. Xie J, Liu C. Adjusted Kaplan-Meier estimator and log-rank test with inverse probability of treatment weighting for survival data. Stat Med. 2005;24:3089–3110.
17. Marcelin AG, Delaugerre C, Beaudoux C, et al. A cohort study of treatment-experienced HIV-1-infected patients treated with raltegravir: factors associated with virological response and mutations selected at failure. Int J Antimicrob Agents. 2013;42:42–47.
18. Ceccherini-Silberstein F, Cozzi-Lepri A, Ruiz L, et al. Impact of HIV-1 reverse transcriptase polymorphism F214L on virological response to thymidine analogue-based regimens in antiretroviral therapy (ART)-naive and ART-experienced patients. J Infect Dis. 2007;196:1180–1190.
19. Cozzi-Lepri A, French MA, Baxter J, et al. Resumption of HIV replication is associated with monocyte/macrophage derived cytokine and chemokine changes: results from a large international clinical trial. AIDS. 2011;25:1207–1217.
20. Cole SR, Hernan MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol. 2008;168:656–664.
21. Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med. 2015;34:3661–3679.
22. Wensing AM, Calvez V, Gunthard HF, et al. 2017 update of the drug resistance mutations in HIV-1. Top Antivir Med. 2017;24:132–133.
23. Haim-Boukobza S, Morand-Joubert L, Flandre P, et al. Higher efficacy of nevirapine than efavirenz to achieve HIV-1 plasma viral load below 1 copy/mL. AIDS. 2011;25:341–344.
24. Metzner KJ, Allers K, Rauch P, et al. Rapid selection of drug-resistant HIV-1 during the first months of suppressive ART in treatment-naive patients. AIDS. 2007;21:703–711.
25. Metzner KJ, Rauch P, von Wyl V, et al. Efficient suppression of minority drug-resistant HIV type 1 (HIV-1) variants present at primary HIV-1 infection by ritonavir-boosted protease inhibitor-containing antiretroviral therapy. J Infect Dis. 2010;201:1063–1071.
26. Brookhart MA, Schneeweiss S, Rothman KJ, et al. Variable selection for propensity score models. Am J Epidemiol. 2006;163:1149–1156.
Keywords:

antiretroviral therapy; propensity score; Kaplan–Meier estimates; HIV-1 RNA reduction; limits of quantification

Supplemental Digital Content

Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.