|Home | About | Journals | Submit | Contact Us | Français|
This study examined the association between recent trends in CD4 and viral loads and cognitive test performance with the expectation that recent history could predict cognitive performance. Eighty-three human immunodeficiency virus (HIV)-infected patients with a mean CD4 count of 428 copies/ml were examined in this study (62% with undetectable plasma viral load [PVL]). We investigated the relationships between nadir CD4 cell count, 1-year trends in immunologic function/PVLs, and cognitive performance across several domains using linear regression models. Nadir CD4 cell count was predictive of current executive function (p = .004). One year clinical history for CD4 cell counts and/or PVLs were predictive of executive function, attention/working memory, and learning/memory measures (p < .05). Models that combined recent clinical history trends and nadir CD4 cell counts suggested that recent clinical trends were more important in predicting current cognitive performance for all domains except executive function. This research suggests that recent CD4 and viral load history is an important predictor of current cognitive function across several cognitive domains. If validated, clinical variables and cognitive dysfunction models may improve our understanding of the dynamic relationships between disease evolution and progression and CNS involvement.
The clinical management of human immunodeficiency virus (HIV) infection requires careful monitoring and tracking of CD4 lymphocyte cell counts and plasma viral loads (PVLs). These measures provide critical information about host immunological health and viral replicative capacity that have well-documented associations with various clinical outcomes. For example, patients with CD4 cell counts lower than 200 are more susceptible to opportunistic infections (Kaplan et al., 2000) and have higher mortality rates (Ickovics et al., 2001) and high PVL counts confer additional morbidity risks (Mellors et al., 1997). CD4 cell counts and PVL also appear to be biomarkers for various measures of cognitive performance. Many studies, especially those prior to the development of combination antiretroviral therapies (cART) regimens, have demonstrated an association between measures of disease severity, cognitive function (Brew, Pemberton, Cunningham, & Law, 1997; Childs et al., 1999; McArthur et al., 1997), and the development of dementia (Levy & Bredesen, 1998; Navia, Cho, Petito, & Price, 1986; Navia & Price, 1987) with the onset of cognitive symptoms often heralding death (Ickovics et al., 2001). In the era of cART, the direct relationship between these variables appears to be diminished (McArthur et al., 2004), though it appears that improvement in immunological function and suppressed viral replication will result in observable improvements in cognitive function (Robertson et al., 2004) as well as lessen the risk of developing frank dementia (Sacktor et al., 2001).
One common explanation for these diminished associations is that the natural course or trajectory of disease progression has been altered in the era of combination therapies (Egger et al., 1997). The alteration of disease progression has resulted in several positive consequences including increased life expectancy (Cameron et al., 1998; Hammer et al., 1997; Montaner et al., 1998) and the reduction in the severity of cognitive symptoms and/or rates of dementia (Dore et al., 1997; Ferrando et al., 1998; Robertson et al., 2004; Sacktor et al., 2006). Importantly, however, there are an increasing number of patients who clearly develop less severe cognitive symptoms in the mild to moderate range of dysfunction (McArthur et al., 2003; McCutchan et al., 2007; Sacktor et al., 2002; Ances & Ellis, 2007). The prevalence of milder forms of cognitive dysfunction range between 20% and 37% (Sacktor et al., 2002; Villa et al., 1996) and these milder forms of cognitive dysfunction are know to diminish their ability to successfully engage in many aspects of daily living (e.g., driving [Marcotte et al., 1999], cooking [Heaton et al., 2004], financial management [Heaton et al., 2004], and medication adherence [Hinkin et al., 2002, 2004]). Combined, the evidence suggests a significant change in the evolution and progression of HIV-associated CNS injury and cognitive presentation in the era of cART that is not well understood.
This lack of understanding has prompted many to examine relationships between immune function, viral replication, and cognitive performance. For example, several researchers have examined the predictive value of historical markers of disease severity. Nadir CD4 cell count, which is a patient's lowest observed CD4 lymphocyte cell count, has been shown to predict the diagnosis of HIV-associated dementia even after adjusting for confounding factors such as educational attainment, current CD4 counts, or self-reported duration of illness (Valcour et al., 2006). A 50-cell difference in nadir CD4 count has also been linked with increased risk (OR = 1.08; 95% CI = 1.03–1.13) for global cognitive impairment (Robertson et al., 2007). Though it is clear from these findings that a patient's degree of immunosupression is important, there remain many unanswered questions regarding the importance of clinical history and its influence on development of cognitive dysfunction.
The purpose of this study was 3-fold. Our first objective was to characterize the relationship between nadir CD4 cell count and cognitive domain performance. This was an attempt to replicate recent findings in the literature (Robertson et al., 2007; Valcour et al., 2006) that suggest a relationship between severely immunosuppressed HIV-infected patients and poor cognitive/functional outcomes. Our second objective was to examine the relationship between 1-year history of CD4/PVL and cognitive domain performance. Examining the recent trends and the variability in the trend trajectories could improve the ability of clinicians and researchers to find associations between CD4 counts, PVL, and cognitive performance not observed in traditional cross-sectional designs or when using a single historical time point (e.g., nadir CD4). The delineation of accurate statistical models could also improve our understanding of the temporal relationships between markers of disease severity and the evolution and/or progression of CNS injury as manifested by cognitive dysfunction. Finally, we investigate the combined effects of 1-year CD4/PVL history and nadir CD4 cell count on cognitive performance to determine the statistical interactions between these measures. In addition, we expand the current literature base by examining the relationship between these measures of disease severity and five cognitive domains separately, rather than focusing on global cognitive performance or dementia rates. We hypothesized that motor, attention, and executive function measures would be associated with worse clinical 12-month history as these are consistently cited as areas of dysfunction in HIV-infected patients experiencing cognitive dysfunction (Heaton et al., 1995).
Patients (N = 83) were recruited for participation as part of two NIH-funded grants (one examining cognitive factors influencing treatment adherence and the other examining the relationship between cognitive factors and neuroimaging variables) that utilized the same cognitive battery and had identical inclusion/exclusion criteria. Participants were HIV-positive, English-speaking adults between the ages of 18 and 65 (actual age range 23–61). In order to limit additional confounding effects due to neurologic and psychiatric disorders, participants were excluded if they had a history of head injury with loss of consciousness >5 min, diagnosed learning disability, neurologic disease (e.g., seizures, multiple sclerosis [MS]), a major psychiatric diagnosis (e.g., schizophrenia, bipolar disorder), active use or abuse of alcohol and/or drugs 6 months prior to enrollment, or an active HIV-associated opportunistic infection. The local institutional review board approved all procedures and consent forms. Additional demographic data are shown in Table 1.
All participants underwent a battery of self-administered computerized neuropsychological tests (IntegNeuroTM, Brain Resource Company, Melbourne, Australia) measuring verbal intelligence and five cognitive domains: speed of processing and motor function, attention, executive function, language fluency, and learning and memory. Two to five tests were used to evaluate each cognitive domain (Table 2). The statistical reliability and validity of this battery has been established (Paul et al., 2005; Silverstien et al., 2007) and results within HIV-infected patients have been reported elsewhere (Tate et al., 2010). The individual subtests for this battery were administered using standardized protocols and scoring. The individual raw test performance scores for each HIV-infected patient were then normalized using age, gender, and education-corrected normative data available from the Brain Resource Company (http://www.brainresource.com/uploads/).
To provide additional demographic background information regarding our particular sample, drug and alcohol use/abuse histories were assessed using the Kreek–McHugh–Schluger–Kellogg scale (KMSK; Kellogg et al., 2003). This self-report questionnaire quantifies the frequency, duration, and amount of alcohol and drug use separately for the period in their life when use was heaviest and results for this instrument are reported in Table 1.
The Miriam Hospital Immunology Center Database (ICDB) and patient chart review were used to retrospectively capture all CD4 cell counts, PVL measurements, and each patient's nadir CD4 cell count. CD4 cell counts and PVL were obtained and quantified using standard procedures during each routine clinic visit. For the purposes of this study, we examined all the CD4 cell counts and PVL measurements during the 13 months prior to and 1 month following cognitive testing.
Nadir CD4 cell count was obtained by reviewing each patient's CD4 history since entry into HIV care at the immunology clinic. Since patients had CD4 counts measured unevenly, the 1-year history of CD4 count was summarized using individual-specific random effects from a longitudinal random effects regression model fit to all CD4 measures collected and the number of CD4 tests performed during the 13 months prior to and 1 month following cognitive testing. The model assumed individual-specific linear time trends in CD4 over the measurement period; this assumption was checked using standard model fit techniques. The estimated individual-specific intercept and slope were used to estimate each participant's CD4 count 1 year prior to the date of cognitive testing and the change in CD4 (cells/day) during the year prior to testing (Bycott & Taylor, 1998; Laird, Donnelly, & Ware, 1992). The extra months prior to and following the study period were included to improve precision of the estimated subject-specific random effects.
PVL was categorized as detectable or undetectable to alleviate complications associated with its highly skewed distribution and lower limit of detection (75 copies/ml). PVL 1 year prior to cognitive testing was taken to be the measurement closest to 1 year, but not more than 13 months, prior to cognitive testing. Similarly, PVL at testing was taken to be the value of the PVL test closest to the cognitive testing date, but not more than 1 month after cognitive testing. Each subject's change in PVL during the year was coded as −1 if the PVL worsened (undetectable 1 year prior but detectable at cognitive testing), 0 if status remained the same, and 1 if it improved (detectable 1 year prior to testing but undetectable at testing).
Eighteen cognitive tests were administered to each patient; however, some participants had invalid results for one or more tests (Table 2). Multiple imputation was used to handle the missing scores. The 18 test scores were assumed to have a multivariate normal distribution with unrestricted variance–covariance matrix. Parameters were estimated using an EM algorithm, and missing values were imputed from the fitted model. Ten complete data sets were generated via multiple imputation as described by Honaker, King, and Blackwell (2009).
For each completed data set, patient-specific domain scores are a weighted average of the test scores, with weights computed using a one-factor factor analysis. Factor analysis was used for “speed of processing,” “attention,” and “executive functioning,” each of which had three or more cognitive tests. The domains “language and fluency” and “learning and memory” have only two test scores each. Hence, the scores for these two domains were estimated by averaging the two within each domain, since factor analysis cannot be use with fewer than three variables.
Three regression models were fit to each domain score. The first included only nadir CD4 cell count as a predictor. The second included (a) CD4 cell count 1 year prior to cognitive testing, (b) 1-year change in CD4 cell count during the year prior to cognitive testing, (c) the indicator of detectable PVL 1 year prior to cognitive testing, and (d) the change in PVL status at cognitive testing compared with 1 year prior. In the third model, nadir CD4 cell count was added to the second model. All models were adjusted for patient age, gender, race/ethnicity, and the number of PVL measures taken during the year. The number of tests was included, since it may be related to unmeasured patient characteristics, such as recent treatment failure, which may also explain cognitive performance. PVL and CD4 are usually tested during the same blood draw, and the number of CD4 and PVL tests was very similar (Pearson's correlation = .9). The effects of the CD4 variables and change in PVL were assumed to be linear, and the effects of age and the number of PVL measures were modeled using non-parametric smooth functions. We examined residual plots to assess model fit and examined correlation between pairs of independent variables to evaluate possible multicollinearities.
Domain scores were modeled by fitting the regression models described above to each of the ten completed data sets; coefficient estimates and standard errors were computed using standard methods for multiple imputation (Little & Rubin, 2002). This analysis was performed using R (R Development Core Team, 2009). p-values of <.05 were considered statistically significant.
All participants had at least two CD4 count measures during the period of 13 months prior to and 1 month following cognitive testing (median four tests, range 2–14). The mixed-effects model estimated mean CD4 count 1 year prior to testing for a participant with four CD4 tests was 394 cells/ml (standard error = 57), the estimated mean 1-year change in CD4 counts prior to testing was positive at 55 cells/year (standard error = 23, p = .01), and for each additional CD4 test performed, the mean CD4 count was lower by 24 cells (standard error = 10, p = .01). CD4 count and change in CD4 count were not found to be correlated (Pearson's ρ = −.02).
Forty-nine participants (59%) had detectable PVL 1 year prior to testing. Of these, 24 (49%) had undetectable PVL at the time of testing. Of the 34 participants with undetectable PVL 1 year prior to testing, 3 (9%) had detectable PVL at testing. The remaining 56 participants had no change in their PVL status (25 remained detectable and 31 remained undetectable) during the year. The participants had a median of four PVL tests during the year (range 1–16). In the regression analysis, the two participants with only one PVL test were coded 0, assuming that they had no change in PVL during the year. These two participants had undetectable PVL during their 36 and 7 available PVL test results spanning the 3 years prior to and 2 years following cognitive testing and lending support to this assumption.
The sample included 37 African American (45%), 10 Hispanic (12%), 34 Caucasian (41%), and 2 patients of other ethnicity (2%; defined as Asian, American Indian, Alaska native, or multiple races). Median age was 45 years old (range 23–61) at the time of testing. Most patients were on stable cART for more than 1 year (82%), with 66% having undetectable viral loads (≤75 copies/ml) at the latest time point. Additional patient demographic data are contained in Table 1.
Group mean standardized z-scores (SD) for each test are presented in Table 2. Performance on the cognitive test measures indicated that average performance for this cohort generally falls below that of healthy normative data, but generally within normal limits (between −1 and 1). Tests for which HIV-infected patients were on average impaired (>1 SD below the mean) relative to normal limits included: spot the word (mean = −1.74, ±1.59); tapping dominant hand (mean = −1.49, ±1.95), reaction time (mean = −2.09, ±1.60), delayed verbal memory (mean = −1.42, ±1.18), immediate visual memory (mean = −1.92, ±1.45), switching of attention task 2 (mean = −1.14, ±2.14), time to complete mazes (mean = −2.57, ±1.44), and total number of errors for mazes (mean = −2.27, ±1.81). These represent deficits in fine motor dexterity, response speed, visual memory, and executive function.
Our analysis found a significant relationship between nadir CD4 cell counts and executive function. Higher nadir CD4 cell count was associated with higher executive function scores (0.21 units per 100-unit increase in nadir CD4 cell count; 95% CI: 0.07–0.35). The remaining cognitive domains were not significantly associated with nadir CD4 cell counts (Table 3).
Scores from three cognitive domains were found to be associated with changes in CD4 count and/or PVL: executive functioning, attention, and learning and memory (Table 4). Executive functioning scores were positively associated with CD4 count 1 year prior to cognitive testing and with improvement in PVL during the year prior to testing. Average executive function scores were 0.17 units greater (95% CI: 0.07–0.28) for each 100-unit difference in CD4 cell count. Those whose PVL status improved during the year had, on average, executive functioning scores 0.70 units higher (95% CI: 0.22–1.19). Higher attention domain scores were associated with higher CD4 count 1 year prior to testing, such that the mean difference was 0.14 (95% CI: 0.04–0.23) per 100-cell difference in CD4. Learning and memory cognitive scores were higher among those exhibiting improvement in PVL status during the year prior to testing (mean difference 0.75, 95% CI: 0.04–1.47). Language and Fluency and Learning and Memory domain scores were not found to be significantly associated with the CD4 or PVL measures.
When nadir CD4 cell count was added to the CD4/PVL 1-year history models, generally there were no significant changes noted in the effect sizes and/or significance across the cognitive domains (data not shown). For example, the association between CD4 count 1 year prior to testing and attention remained significant with a similar effect size (0.18 per 100-cell difference) and nadir CD4 remained insignificant with a smaller, negative effect size (−0.08 per 100 cell difference). Neither nadir CD4 nor CD4 count 1 year prior to testing was significant for the other domains. The only exception was for executive function where the effect size and significance for the predicted CD4 cell count was attenuated (0.17 down to 0.11, p = .14). Because nadir CD4 and CD4 count 1 year prior to testing were positively correlated (ρ = .7) and even though both nadir CD4 and CD4 1 year prior to testing were individually associated with executive function, their high correlation precludes estimating their joint and conditional effects. For the executive functioning and attention domains, the mean AICs of model 2 (220.4 for executive function and 207.7 for attention) were smaller compared with the mean AICs of model 2 with nadir CD4 substituted for CD4 count 1 year prior to testing (221.9 for executive function and 214.3 for attention). Therefore, we conclude as a single predictor, CD4 count 1 year prior to testing is more strongly associated with these domains.
This manuscript investigated the relationship between nadir CD4 cell count, CD4 and PVL counts 1 year prior to cognitive testing, the change in CD4 and PVL during the year prior to cognitive testing, and current cognitive performance. We used a multivariate modeling approach to determine whether these clinical variables would be useful in predicting current cognitive performance. Nadir CD4 cell counts have been used successfully to predict global cognitive function and rates of dementia illustrating a relationship between a patient's clinical history and CNS injury. Results from the current study suggest that even in the era of cART lower nadir CD4 cell counts are predictive of worse executive function. Language, attention, learning and memory, and processing speed were not found to be significantly related to nadir CD4 cell counts. To our knowledge, this is the first attempt to examine the association between nadir CD4 cell counts and specific cognitive domains. The specificity of this finding for executive function suggests that there may be special significance of the severity of immunosuppression and executive function.
We also examined 1-year clinical history as a predictor of current cognitive function. This allows incorporation of within-subject variability of clinical measures (CD4 cell counts and PVL) which is not possible in traditional cross-sectional designs. We found that CD4 cell count is associated with performance in two cognitive domains, namely executive function and attention. The trend in CD4 cell counts (change over the previous year) was not found to be significantly associated with any cognitive domains after controlling for predicted CD4 cell count. In addition, compared with those whose PVL status stayed the same or worsened, patients whose PVL status improved demonstrated better cognitive performance for executive function and learning/memory domains.
For these cognitive domains, this suggests, in general, that recent clinical history is more important than nadir CD4 cell counts. In addition, the addition of nadir CD4 to these models did not improve the fit to scores in the attention, language/fluency, speed of processing, or motor function domains. Although nadir CD4 cell counts and the predicted CD4 cell count were significantly associated with executive function when examined separately, the significance and effect sizes were attenuated and became insignificant when both variables were included in the same model. As a result, we could not compare the joint effects of nadir CD4 and CD4 count 1 year prior to testing on executive function due to multicollinearity. Regardless, these findings emphasize the need to examine common measures of immune function for trends that might provide additional insight into the development and progression of cognitive dysfunction in HIV-infected patient cohorts.
Further, elucidation of these relationships in other HIV-infected patient cohorts may prove to be particularly important for several reasons. First, these findings provide additional insight into the evolution of cognitive dysfunction where a recent history of improving immunological function heralds improved cognitive performance, especially for attention, executive function, and learning/memory. This is similar to the results of the Bornstein and colleagues (1991) pre-cART era study that described improving cognitive benefits (reaction time and memory) in HIV-infected patients who experienced better improvement in CD4 cell count percentage change. It is clear that additional studies will be needed to understand the exact nature of these temporal relationships, as the current study does not include any prospective cognitive data. This fact precludes us from making definitive statements regarding cause and effect, but it does illustrate the value of examining such relationships. Second, as we know, deficits in these domains can negatively affect participation in activities of daily living, including medication management (Hinkin et al., 2004). Understanding these relationships will be increasingly important as we continue to manage the long-term chronic aspects of this disease. Third, further elucidation of these temporal relationships will not only provide us with relevant biomarkers of CNS injury, but it will also elucidate the necessary changes or trajectories in these markers that are sufficient to cause CNS injury or repair. In many ways, these findings also emphasize the need to broaden our search for additional biomarkers that may better explain CNS injury and/or provide clinical markers for evaluating treatment efficacy. For example, there have been studies examining other biomarkers of immune activation and neurologic complications like monocyte chemoattractant protein, tumor necrosis factor, and cell surface markers. These studies have found significant associations between these markers of immune activation and neurologic complications (Kusdra, McGuire, & Pulliam, 2002; McArthur, Brew, & Nath, 2005) though it is not clear what effect they may have on cognitive performance. Though modeling recent CD4 and PVL history is significantly predictive of attention/working memory and executive dysfunction in this study, studies of additional biomarkers are required.
These findings should be interpreted cautiously for several reasons. First, this is a retrospective archival study. It could be argued that the data used were not designed to specifically address the research questions in this study. Certainly, there are several potentially confounding factors not directly addressed given the archival nature of the disease severity measures including uncontrolled demographic features like potential changes in drug/alcohol abuse histories. It should be noted that a significant number of our sample met the addiction criteria for drug/alcohol (>50%; Table 1) sometime in their past as assessed by the KMSK. It is clear from the literature that drug/alcohol use in HIV-infected patients is a potential confound and changes in their use histories during the course of the study could affect their neuropsychological performance as well. However, inclusion criteria for both studies required that each participant to not be using alcohol or drugs 6 months prior to enrollment that afforded some control of this potential confound. Second, the normative data that were used to calculate standard scores were based on an international database of healthy subjects. The use of this specific normative data may be considered a weakness, as there may be important demographic features that vary from that of the HIV-infected patients. Importantly, however, the data did allow us to produce standardized scores that were corrected for age, education, and gender.
Regardless, these findings extend the current literature in several ways. First, to our knowledge, this is the first paper to examine the relationship between recent history of CD4 and PVL and cognitive performance using random effects statistical modeling in the cART era. As already discussed, in the pre-cART era, Bornstein and colleagues (1991) demonstrated a significant relationship between rates of CD4 change (using two time points) and measures of reaction time and memory. Though their method (calculating the percent change between two time points) is a simpler approach to this question, it is unable to account for individual variance in each patient's disease severity measures. Additionally, random effects modeling has proven useful in examining several clinical outcomes for progressive neurological conditions, such as MS and Alzheimer's disease, and if validated further in larger prospective HIV-infected cohorts could reasonably result in the development of mathematical models useful in predicting cognitive dysfunction. Second, use of recent history of key clinical variables may have additional utility over the use of other historical markers of disease severity such as nadir CD4. Though nadir CD4 cell count has been used with a significant amount of success when predicting cognitive outcome (Robertson et al., 2004; Tozzi et al., 2005; Valcour et al., 2006), there are still many questions left unanswered by using a single historical time point. We know that clinical variables may fluctuate and we also know that a certain percentage of patients experience a pattern of fluctuating cognitive performance over the course of the disease. Thus, examining the relationship between measures from a single time point may or may not yield any significant findings especially if there are temporal lags in the relationship between variables. Together, these observations require new approaches to examining the variability of these data to answer important questions regarding clinical/cognitive relationships. Third, we examine the effects of these variables on specific tests of cognitive function rather than global measures of cognitive function or rates of dementia. By using a broader based battery of cognitive tests, we were able to assess multiple domains. This may be increasingly important in the era of cART and in the context of better viral suppression where the cognitive profile (or phenotype) may be evolving (Brew, 2004).
In summary, these findings suggest that the patient's recent clinical history is an important predictor of current attention/working memory and executive dysfunction. If validated, modeling of recent clinical variables may dramatically improve our understanding of the evolution and progression of HIV-associated CNS injury providing researchers and clinicians with the much needed objective markers of CNS involvement among HIV-infected cohorts.
This manuscript was supported in part by the following grants: K23-MH073416 (D.F.T.), K23-MH065857 (R.H.P.), and P30-AG013846 (D.F.T.).