Variable Cognitive Function Response to HAART: HIV suppression by HAART preserves cognitive function in advanced, immune-reconstituted AIDS patients
AIDS:Volume 21(9)31 May 2007p 1109-1117
McCutchan, J Allena; Wu, Julia Wb; Robertson, Kevinc; Koletar, Susan Ld; Ellis, Ronald Ja; Cohn, Susane; Taylor, Michaela; Woods, Stevena; Heaton, Roberta; Currier, Judithf; Williams, Paige Lb
From the aUniversity of California, San Diego, California, USA
bHarvard School of Public Health, Boston, Massachusetts, USA
cUniversity of North Carolina, Chapel Hill, North Carolina, USA
dOhio State University, Columbus, Ohio, USA
eUniversity of Rochester School of Medicine, Rochester, New York, USA
fUniversity of California, Los Angeles, California, USA.
".....In conclusion, advanced AIDS patients who have responded to HAART for prolonged periods usually have stable or improving cognitive function, but remain more likely than the general population to be cognitively impaired. During prolonged HAART, cognitive test scores continued to improve, probably reflecting a combination of practice effects and, because improvement is associated with the suppression of HIV replication, continuing neurological recovery.... Non-impairment (not diagnosed with NPI) across visits was marginally associated with both the improved suppression of plasma HIV-RNA levels from above to below 500 copies/ml over the previous 16 weeks (P = 0.054) and current HIV-RNA suppression (< 500 copies/ml) at the time of neuropsychological testing (P = 0.06)..."
Introduction: HIV can damage neurons leading to cognitive impairment. Epidemiological observations suggest that neuropsychological impairment might progress despite successful HAART therapy, but available prevalence estimates are based on populations that were selected for impairment.
Methods: Of 433 advanced AIDS patients with documented immune reconstitution (CD4 lymphocyte counts < 50 before and > 100 cells/μl after HAART), 286 had brief assessments of cognition (Trailmaking A/B and Digit Symbol Tests) at least once, no confounding neurological conditions, and available neuropsychological norms with comprehensive demographic corrections. At entry, most were immune reconstituted on HAART (median CD4 cell count 230 cells/μl) and HIV was suppressed (65% < 500; only 14% > 20 000 RNA copies/ml).
Over one quarter (27%) of participants exhibited impairment at their initial neuropsychological assessment, a rate nearly twice that expected in a normal (HIV-uninfected) reference population (14%).
These impaired participants did not differ from the unimpaired group with respect to age, sex, education, race, CD4 lymphocyte counts, or HIV-RNA levels.
Improved performance on neuropsychological tests was documented over a 2-year period 3-5 years after initiating HAART. This improvement was marginally associated with the continued or improving control of plasma HIV-RNA levels, but not with concurrent levels of immune recovery (CD4 lymphocyte counts).
Conclusion: Most advanced AIDS patients responding to HAART for prolonged periods have stable or improving cognition, but remain more likely to be impaired than the general population. During HAART, improving test performance probably reflects both practice effects and continuing neurological recovery after more than 3 years of HAART.
HIV enters the central nervous system (CNS) during the initial phases of infection and later damages neurons without infecting them, leading to cognitive dysfunction that ranges in severity from subclinical neuropsychological impairment (NPI) to clinically obvious HIV-associated dementia [1-3]. Combinations of three or more antiretroviral drugs (HAART) prolong survival and reduce the incidence of HIV-associated dementia . Less severe, but often functionally significant, cognitive abnormalities can be detected reliably only by formal neuropsychological testing, and may persist in a substantial minority of patients on HAART .
Although antiretroviral treatment improves HIV-induced cognitive impairment in many patients [5-10], some drugs appear to reach only very low levels in the CNS [11-13], potentially providing effective systemic, but not CNS, treatment. Widely used dideoxy-nucleoside reverse transcriptase inhibitors (e.g. stavudine and didanosine) are toxic to peripheral nerve mitochondrial function and may be toxic to brain mitochondria as indicated by magnetic resonance imaging spectroscopy [14,15]. Therefore, either failure of therapy in the CNS compartment, CNS neurotoxicity of HAART, or both could contribute to CNS dysfunction. Accordingly, we evaluated cognition in advanced AIDS patients who had experienced prolonged recovery of immune function on HAART.
The prevalence of NPI was 27% in advanced AIDS patients who had experienced prolonged immunological reconstitution on HAART. This rate of NPI is nearly twice that expected in normal individuals, which we calculated from the distribution of performance on the three tests in a normal population (14%, P < 0.001). The prevalence of NPI in the HAART era has previously been estimated only in populations that were selected for a high risk of cognitive impairment by admission criteria to a research cohort  or clinical referral for neuropsychological evaluation . As anticipated, the prevalence of impairment found in these latter cohorts (65-75%) was higher than in our participants who were selected without a bias towards impairment. A striking increase in the period prevalence of HIV-associated 'dementia' was reported in a less selected clinic population from the Johns Hopkins University HIV Clinic between 1994 and 2001 (approximately 6.6-10.1 cases/100 person-years), but the diagnostic methods used were not explained .
Several kinds of studies have suggested that HAART may not be as effective in treating or reversing HIV infection in the brain as in other tissues. Two epidemiological studies have documented a dramatically decreased incidence of cognitive impairment in the early HAART era, indicating that HAART treats the CNS in most patients . The decrease was, however, less than for other complications of AIDS, suggesting a failure of CNS treatment in some patients. For example, during a marked decline in the incidence of AIDS in HIV-infected Australian individuals at the onset of the HAART era (1992-1997), dementia increased from 4.4 to 6.5% as a proportion of all initial AIDS-defining illnesses. In contrast, the proportion of HIV-associated opportunistic infections of the CNS remained stable (8.1-8.2%) . In the Multicenter AIDS Cohort Study , fold decreases in the incidence of CNS opportunistic conditions such as cryptococcal meningitis (X 3.3), toxoplasmal encephalitis (X 2.5), or primary CNS lymphomas (X 7) exceeded those of dementia (only twofold from 21 to 10.5% per year) during the early HAART era.
Multiple autopsy studies of the neuropathology of HIV have concluded that HIV-specific effects on the brain remain frequent in the era of HAART [29,30]. Several studies have suggested changing patterns of neuropathology since the introduction of HAART [29,31-33]. These findings may indicate the failure of HAART, but many patients fail or discontinue HAART before death, providing time for HIV to replicate in and damage the CNS terminally.
Performance on cognitive tests improved over 2 years in our participants who had been on HAART for an average of nearly 3 years before their initial testing. Improvement appeared to be associated with concurrently sustained or recently achieved complete plasma HIV suppression, but not with the overall degree of immunological restoration on HAART, i.e. increases in CD4 cell count. Neurological recovery may thus continue for prolonged periods if HIV replication is suppressed. This finding extends reports of an early improvement in neuropsychological functioning in impaired patients who have a suppression of HIV replication in the cerebrospinal fluid on HAART for 3-6 months . This improvement over short periods has not been complete or seen in all patients. For example, of 16 impaired patients starting HAART, seven remained impaired after 3 years of HAART .
The continuing neurological recovery over 5 years on HAART parallels the continuing recovery of CD4 lymphocyte counts in this same population over the period of this study . Given this pattern and the increased prevalence of dementia at lower CD4 cell counts , the lack of correlation of cognitive improvement with rising CD4 cell counts is surprising and suggests that the suppression of HIV replication, but not immune recovery, promotes cognitive recovery.
Improvements in test scores resulting from practice rather than from increased cognitive function may confound longitudinal neuropsychological assessments . Practice effects may have contributed to the improved test scores that we observed, but are unlikely to explain them completely. The method we used to remove the effects of this confounder, reliable cognitive improvement plus practice, has been validated in HIV-infected individuals .
After adjusting for potential practice effects, a significant improvement in TMA scores that correlates with the suppression of HIV replication suggests that this improvement was at least partly caused by antiretroviral treatment. The marginal significance of the improvement in TMB scores and a trend towards worsening performance in the DSY score over time after adjusting for practice effect might reflect an over-adjustment for practice effects in this sample. Woods et al.  found that TMA scores were almost free of practice effects, whereas TMB and DSY scores were affected much more by practice in their HIV-uninfected cohort.
Neuropsychological functioning could potentially have improved for reasons other than HAART or practice effects. First, the more impaired subjects may have withdrawn from the study at a higher rate than the unimpaired subjects. Second, abnormal scores may have regressed toward the mean, i.e. NPI subjects were more likely than others to improve after a poor performance. Third, NPI subjects could have received specific interventions to help improve their response.
These three possibilities seem unlikely to account for our observations. NPI prevalences at each visit in subjects who had measurements at all three assessments were similar to those in the whole group. Differential loss was thus unlikely to be the only source of improvement. Regression to the mean probably accounted for some proportion of the reduction in the prevalence of NPI over time, but is difficult to assess because of the practice effect. Although we cannot directly evaluate whether subjects received interventions to improve their neurocognitive functioning, none of them was clinically demented, milder forms of cognitive dysfunction are nearly universally under-recognized, and no treatment for HIV-associated NPI other than HAART has been developed.
This study has five important limitations. First, because participants were selected for sustained immune recovery that requires good adherence to their drugs, they may have been more cognitively intact at entry than less-selected HIV-infected individuals on ART. Second, the lack of normative data for other ethnicities limits our study to Caucasian and African-Americans. Third, attrition of those with cognitive dysfunction (i.e. survival bias) may partly explain the observed improvements in the neuropsychological measurements. Fourth, the predominance of PI-based antiretroviral regimes prevented us from comparing the effects of various classes of drugs. Finally, assessments of potential confounders such as mood, drug abuse, and head trauma were unavailable.
In conclusion, advanced AIDS patients who have responded to HAART for prolonged periods usually have stable or improving cognitive function, but remain more likely than the general population to be cognitively impaired. During prolonged HAART, cognitive test scores continued to improve, probably reflecting a combination of practice effects and, because improvement is associated with the suppression of HIV replication, continuing neurological recovery.
This observational substudy originated in AIDS Clinical Trials Group (ACTG) 362, a randomized, double blind, placebo-controlled trial designed to evaluate the need for and the efficacy of azithromycin prophylaxis for Mycobacterium avium complex after immune reconstitution on HAART. At entry to ACTG 362, participants had a history of CD4 cell counts of less than 50 cells/μl, which had risen on HAART to over 100 cells/μl on two occasions at least 4 weeks apart. The primary analysis of ACTG 362 found that M. avium complex prophylaxis was unnecessary after immune reconstitution .
In October 1999, ACTG 362 was converted to an observational cohort study of advanced AIDS patients who had responded to HAART. Participants were assessed every 16 weeks with medical histories and examinations, plasma HIV-RNA levels, and blood CD4 lymphocyte counts. They received antiretroviral therapy (ART) according to community standards from their primary physicians, and reported changes in and adherence to ART at each visit. At week 16 of the observational phase and every third visit (48 ± 6 weeks) thereafter, three short tests of cognition, which comprise this substudy, were administered.
Of the 433 subjects who enrolled in the observational phase of ACTG 362, 401 participants (93%) completed at least one of three neuropsychological assessments, and 370 of these had no exclusionary CNS conditions such as a history of meningitis, encephalitis, tumors, or stroke. Of these 370, 286 qualified for this substudy because they completed at least one assessment and self-identified as black or white, racial groups for which appropriate age, education, and race-specific normative data for the three neuropsychological tests are available.
The screening neuropsychological assessments consisted of three tests: (i) the Digit Symbol from the Wechlser Adult Intelligence Scale version III (DSY), the number of symbols corresponding to a list of numerals (0-9) that are recorded within 90 s using a key that matches symbols to numerals; (ii) Trailmaking A (TMA), the time taken to connect by a line a sequence of numbers from 1 to 25 scattered over a page (1-2-3 ... 25); (iii) Trailmaking B (TMB), the time taken to connect by a line a sequence of alternating numbers and letters (1-A-2-B-3-C ... L-13).
This screening battery assesses domains that are often impaired by HIV infection: speed of information processing, mental flexibility, and working memory. The raw scores from these three neuropsychological assessments were combined to create two summary measures: a categorical diagnosis of NPI and a continuous, normalized score for neuropsychological function.
Raw scores on each of the tests were converted to standardized t-scores, which in the normal population are normally distributed and have a mean of 50 and standard deviation (SD) of 10. These t-scores are corrected for relevant demographic factors (age, education, sex and ethnicity) using large, normative databases [17,18].
Impairment points were assigned for each of the three tests as follows: 2 for two or more SD below their specific norms; 1 for one to two SD below norms; and 0 if above or within one SD of norms. Total impairment points were summed for the three tests. NPI was defined as two impairment points or greater. At a minimum NPI reflects either mild impairment on more than one test or moderate impairment on one test [17,19,20].
This categorical approach to neuropsychological assessment (NPI) emphasizes below-expected performance and ignores better-than-expected results, and has been validated in another ACTG study . We also considered an alternative definition for NPI based on global deficit scores, as outlined in Carey et al. . Because the results were nearly identical (overall kappa for agreement of NPI and global deficit score diagnoses of impairment was 0.83), we present results based on the definition of NPI above.
To calculate the expected rates of NPI in the normal population using this three-test battery, we simulated 10 000 triplets of test scores each assigned values based on a normal distribution and observed interscore Pearson's correlations of approximately 0.5 between scores on the three tests. We then calculated the expected rates of impairment as the proportion of the 10 000 triplets that yielded impairment points of 2 or greater. The significance of the difference between the observed and expected rates of impairment at each examination was tested with a one-sample binomial test.
Neuropsychological function (NPZ-3 norm-adjusted score) TOP
Continuous, normalized measurements of neuropsychological function were derived by averaging the three demographically corrected t-scores described above to obtain the summary neuropsychological z-score from three tests (NPZ-3). The advantage of the categorical diagnosis of NPI is its relevance to clinical diagnosis, whereas NPZ-3 provides a continuous measure of functioning with greater power to detect associations with factors such as ART and plasma HIV levels (viral load).
Adherence to antiretroviral medications was assessed at entry into ACTG 362 and at the initial substudy visit based on a question about the 'Last time you skipped medications?' Answers at six levels were possible: (i) never skip; (ii) more than 3 months ago; (iii) 1-3 months ago; (iv) 2-4 weeks ago; (v) less than 2 weeks ago; (vi) less than 1 week ago. In the generalized estimating equation (GEE) models two dichotomized adherence outcomes were explored as covariates: (i) never skip (33%) versus ever skip (67%); (ii) skip within past 1-2 weeks (28%) versus not skip within past 1-2 weeks (72%).
Data submitted to the ACTG data management center from October 1999 to August 2003 are included in this analysis. Differences in patient characteristics between participants meeting the criteria of NPI and those who did not were examined using Fisher's exact, chi-square, and Wilcoxon rank sum tests, as appropriate. The Wilcoxon signed rank test was used to assess the significance of within-subject changes over time in NPZ-3 norm-adjusted scores. Generalized estimating equation (GEE) repeated-measures models, were used to examine the correlates of NPI and NPZ-3 scores in all 286 subjects over 2 years as well as to evaluate changes in prevalence of impairment over time. These GEE-based models take into account the correlation between the multiple measurements collected on each participant over the time period. Mixed effect models were also used to confirm the GEE model results for the continuous NPZ-3 norm-adjusted scores.
The Reliable Change Index plus practice effects (RCI+P) model was applied for an assessment of the neuropsychological change adjusting for possible practice effects. This model predicts a given individual's follow-up test score by adding the mean practice effect of a reference group to that individual's baseline test score . Normative standards for this RCI+P battery approach were derived from healthy controls, and it was validated on a neuromedically stable HIV-infected sample . For each test of the three tests (TMA, TMB, and DSY), changes (Δ) in raw scores from the initial visit were computed. z-Scores were then generated using the following formula: (Δ - mean practice effect)/SD(Δ). The TMA and TMB tests were multiplied by (-1) so that positive z-scores indicate improvement. The z-scores for the three tests were summed to create the practice-adjusted summary z-score.
To evaluate whether changes from baseline were significant after adjusting for the practice effect, individual z-scores and summary z-scores were compared against 0 using one-sample t-tests. Although such z-scores should be normally distributed, results were confirmed using a non-parametric Wilcoxon signed rank test. The 90% RCI+P prediction intervals, derived from the empirical z-score distribution of the healthy controls, were used to define improvement.
Demographics, immune recovery, and level of HIV suppression
Participants were predominantly male (90%), white (80%), mature (median age 40 years), well educated (median 14 years of schooling) and had remained on HAART since their initial visit (median exposure 161 weeks; Table 1). At the initial neuropsychological assessment, the median CD4 cell count was 326 and 73% had achieved HIV suppression (< 500 copies/ml). Of the 286 subjects, less than 1% reported current intravenous drug use at entry to ACTG 362 and 12% reported previous use. At the time of the initial neuropsychological visit, HAART included a protease inhibitor (PI) in 245 (86%), non-nucleoside reverse transcriptase inhibitor and no PI in 21 (7%), and neither a PI nor a non-nucleoside reverse transcriptase inhibitor in one. The remaining 19 subjects (7%) were on non-HAART regimens or not receiving ART.
Compared with the original enrollees in ACTG 362 who did not participate in or were not assessable, participants in this substudy were more likely to be men (90 versus 84%) and white (81 versus 48%), but were similar in age, proportion with a history of intravenous drug use, duration of HAART, lowest or baseline CD4 lymphocyte counts, and plasma HIV-RNA levels.
Prevalence and correlates of neuropsychological assessment
At the initial visit, the prevalence of NPI was 27%, which was nearly twice the calculated rate expected from the distributions of performance of a normal (HIV-uninfected) reference population on these three tests (14%, P < 0.001). The 78 participants with NPI at their initial assessments did not differ from the 200 unimpaired participants with respect to age, sex, education, race, duration of exposure to HAART, CD4 lymphocyte counts, or HIV-RNA levels at their initial visit (See Table 1).
The initial prevalences of NPI (27%) decreased to 16 and 14% at 48 and 96 weeks, respectively (P < 0.001 for each based on a GEE model, Table 2). The difference between the prevalences at weeks 48 and 96 was not significant (P = 0.36). Among the 210 patients who were tested at all three visits, the prevalence of NPI was similar at each assessment (24, 16, and 14% at weeks 0, 48, and 96), suggesting that improvement was not solely attributable to the differential loss of the more impaired participants (Table 2).
Changes in neuropsychological status at 48 and 96 weeks
The changes in NPI status from 0 to 48 weeks and from 48 to 96 weeks are summarized in Table 3. Of 57 participants with NPI at baseline and follow-up at 48 weeks, only 27 (47%) had a repeated NPI diagnosis. An additional 11 out of 177 without NPI at baseline progressed to NPI when re-examined at 48 weeks. From 48 to 96 weeks, 21 out of 33 (64%) retained the NPI diagnosis and eight out of 177 became newly impaired. Only 21 participants remained impaired at all three assessments, 19 became impaired at one follow-up visit, and 42 participants who were impaired at baseline or week 48 were not impaired at the next assessment. Substantial shifts among the NPI categories in both directions thus occurred during the study.
Assessments of neuropsychological change adjusted for practice effect
The RCI+P method was applied to the raw scores of individual tests over time. A one sample t-test showed that, after adjustment for practice effects, the mean TMA score still improved significantly from weeks 0 to 48 (P = 0.03) and from weeks 0 to 96 (P < 0.001). After adjustment for practice effect, the mean TMB scores did not change significantly from weeks 0 to 48 (P = 0.99), but marginally improved from weeks 0 to 96 (P = 0.07). After adjustment for practice effect, the mean DSY score declined (i.e. failed to show expected practice effects) from weeks 0 to 48 (P = 0.001) and from weeks 0 to 96 (P = 0.01). The practice adjusted summary z-score based on the three individual tests' z-score did not change significantly over time. After adjusting for practice effects and using the 90% RCI+P prediction intervals, 13% (30/230) had a real improvement in TMA scores and 9% (21/230) in TMB scores. Slightly less than the expected 5% (4% or nine out of 230) had an improvement in the DSY scores from weeks 0 to 96.
Changes in NPZ-3 scores
Consistent with the decreasing prevalence of NPI, average norm-adjusted NPZ-3 scores improved over the 2 years of observation. The scores improved significantly from 0 to both 48 (P < 0.001) and 96 weeks (P < 0.001; Fig. 1). A GEE model that included only time estimated an improvement in the NPZ-3 score of 2.0 points every 48 weeks (P < 0.001).
Correlates of changing neuropsychological functioning as measured categorically (neuropsychological impairment) and continuously (NPZ-3)
Multivariate GEE models of both norm-adjusted NPZ-3 scores and NPI diagnosis that controlled for demographic characteristics and time on study were used to explore the relationship of neuropsychological functioning to plasma HIV suppression. As expected, most of the demographic variables were not significantly associated with neuropsychological function because NPZ-3 scores had been adjusted for these characteristics. Age was associated with worse neuropsychological function with marginal significance (P = 0.06) in univariate, but not multivariate, GEE models. Education level higher than high school was also found to be marginally associated with worse neuropsychological function (P = 0.07) in the univariate GEE model. This finding was confirmed in the multivariate GEE models for NPZ-3 scores, possibly reflecting an over-adjustment for the effects of more education in that sample.
Non-impairment (not diagnosed with NPI) across visits was marginally associated with both the improved suppression of plasma HIV-RNA levels from above to below 500 copies/ml over the previous 16 weeks (P = 0.054) and current HIV-RNA suppression (< 500 copies/ml) at the time of neuropsychological testing (P = 0.06; Table 4). In contrast, continuous NPZ-3 scores were not significantly associated with plasma HIV suppression.
No measure of previous immune function or immune restoration (lowest lifetime CD4 cell counts or CD4 cell counts measured contemporaneously or at 16 weeks before the neuropsychological testing) correlated with changes in NPZ-3 scores. Adherence at entry to the original ACTG 362 study and to this substudy was dichotomized at several levels and was found to be unassociated with changing neuropsychological function.