Similar neurocognitive outcomes after 48 weeks in HIV-1-infected participants randomized to continue tenofovir/emtricitabine + atazanavir/ritonavir or simplify to abacavir/lamivudine + atazanavir

Briefly, ASSURE (EPZ113734; NCT01102972) was a prospective, randomized, multicenter, open-label, and phase IV study that enrolled HIV-1-infected, ART-experienced adults (≥ 18 years of age) who had received a once-daily regimen of TDF/FTC 300 mg/200 mg (Gilead Sciences, Foster City, CA, USA) + ATV 300 mg (Bristol-Myers Squibb, Princeton, NJ, USA) boosted with ritonavir 100 mg (AbbVie, Chicago, IL, USA) for ≥6 months and had HIV-1 RNA ≤ 75 copies/mL for at least 28 days prior to the screening visit. As part of the inclusion and exclusion criteria, participants were excluded if they were HLA-B*5701-positive or had prior abacavir exposure, active CDC clinical category C disease, ongoing clinically relevant hepatitis and/or chronic hepatitis B infection (HBsAg+), or a creatinine clearance < 50 mL/min via the Cockroft-Gault method. Additional information regarding all inclusion/exclusion criteria, methods, and endpoints, including a link to the protocol, is available in the 24-week efficacy results publication (Wohl et al. 2014). All participants provided written informed consent to participate in the study, and the protocol was approved by the institutional review board for each study site.

After stratification by prior ART experience (TDF/FTC + ATV/r as the initial regimen or as the first/s switch regimen), eligible participants were randomized 2:1 to simplify their regimen to once-daily ABC/3TC 600 mg/300 mg (ViiV Healthcare, Research Triangle Park, NC, USA) plus ATV 600 mg or remain on TDF/FTC + ATV/r. The ASSURE study was powered to evaluate its primary endpoint, which was the proportion of participants with HIV-1 RNA < 50 copies/mL at week 24 by the time to loss of virologic failure (TLOVR) algorithm. Eighty-seven percent of participants in both treatment groups successfully maintained HIV-1 RNA < 50 copies/mL at week 24, demonstrating non-inferiority of simplification to ABC/3TC + ATV compared with continuation of TDF/FTC + ATV/r (Wohl et al. 2014). Results at week 48 were similar, with HIV-1 RNA < 50 copies/mL in 76% of participants taking ABC/3TC + ATV and 79% of participants taking TDF/FTC + ATV/r (Wohl et al. 2016).

Neurocognition was measured in the ASSURE study with a computerized cognitive test battery, the Cogstate Brief Battery (CBB; Cogstate; Melbourne, Australia). The CBB was selected because of its demonstrated sensitivity to HIV-related CNS impairment, (Maruff et al. 2009; Bloch et al. 2016) sensitivity to the effects of CNS active drugs generally (e.g., Grove et al. 2014; McIntrye et al. 2014), and to the effects of ART with CNS penetration (Winston et al. 2015). The CBB includes four cognitive tests which assess psychomotor function, attention, learning, and working memory which have previously been described in detail (Cysique et al. 2006). The CBB requires approximately 15 min for completion in HIV-infected adults; a single performance measure is generated for each of these four tests. Analyses of the neurocognitive data were conducted on the intent-to-treat exposed (ITT-E) population. All participants completed the entire CBB at a pre-baseline familiarization assessment, a baseline assessment and then an assessment at the week 24, week 48, and if an early withdrawal from the study occurred. Data from the familiarization visit was not used in the analyses.

To provide a measure of global neurocognition, a z-score was derived from the four tests in the CBB for each participant at each assessment. Each time, a study participant completed the CBB; their data were uploaded to a secure database and processed to provide the four primary outcome measures. Each outcome measure was then standardized using demographic-adjusted normative data and summed to compute a z-score, which was a neuropsychological composite of the four tests that was used to define global neurocognition. For each assessment, each participant’s neurocognition was categorized as either “impaired” (defined as either a z-score ≤ − 2 or having two or more standardized individual test z-scores ≤ − 1), while higher scores (equating to better performance) were classified as “normal.” The normative dataset used as a comparator in this analysis was obtained by Cogstate from a healthy, non-HIV-infected population. The normative dataset used in this analysis was stratified by age group, and was similar in composition to the ASSURE study participants in terms of age and gender distributions and where possible, was of similar ethnicity; as previously been described (Haddow et al. 2017).

The primary exploratory analysis objective was to determine the extent to which 24 and 48 weeks of treatment with ATV + ABC/3TC versus ATV/r + TDF/FTC was associated with changes in neurocognition in the ITT-E population. Study participants must have attempted the baseline and either one or both of the week 24 and week 48 assessments to be included in analyses of neurocognition. Data were analyzed using a series of linear mixed models in which the baseline was modeled as a covariate and treatment group (ABC/3TC + ATV and TDF/FTC + ATV/r) and assessment were modeled as fixed factors. The z-scores, as well as the scores from each of the CBB tests, were modeled separately as the dependent variable. Any early withdrawal or unscheduled visit assessment done between week 12 and week 35 was reclassified into the week 24 assessment group; any assessment performed at week 36 or beyond was classified as week 48 assessments. Through week 48, results for the two treatment groups were compared using an analysis of covariance (ANCOVA) adjusted for baseline performance scores. Cohen’s d was calculated to estimate the magnitude of the differences between the two treatment groups; effect sizes of 0.2–0.4 are considered mild, 0.5–0.7 considered moderate, and ≥ 0.8 considered large (Cohen 1988). To validate the model assumption that the relationship between baseline and outcome were parallel for each treatment group, a baseline by treatment interaction was included in the model and removed only if the corresponding p value was > 0.05. A secondary analysis was also conducted to determine the extent to which 24 and 48 weeks of treatment with ATV + ABC/3TC versus ATV/r + TDF/FTC was associated with changes in neurocognition but restricted to the population of study participants with virologic suppression. The analysis population included participants with z-scores at the 24 or 48 week assessment visits and who had HIV-1 RNA < 50 copies/mL, as assessed by the time to loss of virologic response (TLOVR) algorithm, and the analysis compared the measures of cognition by treatment group and the effect sizes for treatment differences.

The enrolled participants were predominantly male and white (Table 1), although more than a third were African-American. All participants were from sites in the USA and Puerto Rico. More than 25% of participants identified as Hispanic or Latino. At baseline, nearly half of the participants had taken ≥ 1 ART regimen before starting TDF/FTC + ATV/r, and the median time on ART was approximately 3 years. The mean z-scores calculated at the baseline assessment for each of the two treatment groups were not significantly different, with a mean of − 1.155 (standard deviation 1.087; n = 180) for the ABC/3TC + ATV treatment group and a mean of − 1.1755 (standard deviation 1.136, n = 83) for the TDF/FTC + ATV/r treatment group.

At the start of the ASSURE study, 54.7% of the ART-experienced study participants were assessed as having impaired neurocognition (Table 2) and 50% or more of all participants were classified as having impaired neurocognition in the post-baseline assessments. While the overall proportions remained similar, the neurocognitive assessments for an individual participant did not always remain static. In an evaluation of the change from baseline (CFB) results at the week 24 assessment, approximately 25% of participant assessments had a change from their baseline classification status (from normal to impaired or from impaired to normal). Overall, a small proportion (13.3%) of participants went from a classification of impaired at baseline to normal at week 48, while a smaller proportion (8.7%) saw a change from normal to impaired neurocognition. Overall, at week 48, 50% of participants were classified as normal, while 50% were classified as impaired, similar to the proportions observed for the study population at baseline.

The results for each of the individual tests and the z-scores were obtained for study participants with evaluable data and tabulated by treatment group (Table 3). There were no significant differences in the baseline-adjusted performance between the participants taking ABC/3TC + ATV and those taking TDF/FTC + ATV/r for any individual neurocognitive test or for the composite z-score. Consideration of the effect sizes indicated the magnitude for these group differences was uniformly small (i.e., d < 0.2).

As not all participants remained virologically suppressed through 48 weeks in either treatment group, a secondary analysis assessed whether there was any differential impact by treatment group on neurocognition, as assessed by the z-scores, for those participants who maintained virologic suppression (HIV-RNA < 50 copies/mL) through 24 or 48 weeks. By treatment group, for the comparison of ATV + ABC/3TC to ATV/r + TDF/FTC, there were 143 and 73 participants, respectively at week 24 who had remained virologically suppressed and had neurocognitive test results at baseline and week 24 and 142 and 70 virologically suppressed participants, respectively at week 48 with z-scores at baseline and week 48. The adjusted mean difference and 95% upper and lower CI for the comparison at weeks 24 and 48 were − 0.195 (− 0.435; 0.045) and 0.097 (− 0.332; 0.137), respectively. Cohen’s d effect size at weeks 24 and 48 were − 0.230 and − 0.120, and the p values were 0.111 and 0.414, respectively. Similar results were observed for analyses performed for each of the individual tests within the test battery and for all measures of neurocognition, the effect sizes for treatment differences remained small in magnitude and none were statistically significant.

Analysis of the demographic and medical factors associated with a classification of impaired neurocognitive performance at the baseline assessment was performed to explore whether specific demographic or illness-related variables could moderate the effect of treatment. For the detection test, while the regression analyses identified high-sensitivity C-reactive protein as a covariate of interest, it no longer had any significance on the relationship between treatment and neurocognition after the backwards regression was conducted. For the identification test, the covariates psychiatric disorder not otherwise specified and CD4 nadir lymphocyte cell counts were found to be significant at the alpha level of 0.1 and remained in the model after baseline, treatment and visit were added. After adjusting for psychiatric disorder not otherwise specified and CD4 nadir lymphocyte count, the effect of treatment on neurocognition was similar (p value = 0.234 vs. p value = 0.178). For the one card-learning test, psychiatric disorder not otherwise specified, type II diabetes, stratification, and depression were significant after the univariate regression, however none of these covariates remained significant after the backwards regression process. For the one back test, depression was significant in the univariate analysis and remained in the model when treatment, baseline and visit were controlled. However, after adjusting for the effect of depression history on neurocognitive performance, the effect of treatment on neurocognition was similar (p value = 0.340 vs. p value = 0.571). None of the covariates were significant in the univariate analysis for the z-score (all p values > 0.1).

When performance on the neurocognitive tests at baseline was considered categorically as being impaired or normal (Table 4), differences were found for four factors: lower CD4 lymphocyte count at nadir, higher scores on the Framingham risk factor and higher values for interleukin 6, and having a history of psychiatric disorder not otherwise specified. For the continuous variables (Framingham risk, CD4 nadir lymphocyte count, and interleukin 6), the magnitude of differences in covariate score between impaired and normal neurocognition was expressed using Cohen’s d. For the categorical variable history of psychiatric disorder, the effect was expressed by computing the odds of impaired neurocognition with a history of psychiatric disorder compared to no medical condition. The magnitudes of differences in covariates for Framingham risk, CD4 nadir lymphocyte count, and interleukin 6 covariates were small (d < 0.2). However, a history of psychiatric disorder increased the odds of current classification of impaired neurocognition tenfold.

Taken together, these analyses indicate that after adjusting for baseline and relevant demographic and medical covariates there were no differences in performance between the ATV + ABC/3TC and ATV/r + TDF/FTC treatment arms for any of the Cogstate outcomes. While these risk factors were identified to be associated with neurocognitive performance over all assessments (baseline, week 24 or 48), none of these risk factors were found to moderate the effect of treatment on neurocognition.