Participant characteristics

Adult PLWH (301 in number) enrolled in the CNHC CTRSC cohort met inclusion criteria and completed at least one neurocognitive assessment. Of the 301 participants, 132 underwent multiple neurocognitive assessments and comprised the longitudinal sample for the study’s primary analysis. There were no significant differences in any demographic or HIV-related characteristics between participants with one versus multiple visits (Supplemental Table 1). However, there was a trend towards a between-group difference in latest viral load (p = 0.054), with a higher percentage of detectable viral load among people without follow-up visits. Of the 132 patients with at least two visits, 50 had at least three visits, and 14 had four visits, for a total of 328 visits. The average follow-up time was 4.9 years (SD = 2.1, range = 0.4–8.5) across all follow-up visits and 5.2 years (SD = 2.3) at last visit.

Participants ranged in age from 25–68 (M = 51.8), with a high school level of education on average (M = 11.8 years; Table 1). The sample was predominantly Black (92%), with somewhat more men (59%) than women. The participants were well treated, with 98% currently on cART, 91% with undetectable viral load, and clinically normal current CD4 levels (M = 686, SD = 302). Thirty eight percent of the participants had a CD4 nadir below 200. The average time since CD4 nadir at baseline was 6.5 years (SD = 3.8, range = 0.0–17.9) and the average time since CD4 nadir at the last visit was 12.4 years (SD = 6.1), with a maximum of 26.7 years. Individuals with CD4 nadir < 200 were less likely to be Black (p = 0.039) and had more years of education (p = 0.027), lower current CD4 counts (p < 0.001), and a longer time since their nadir (p < 0.001) than the CD4 nadir > 200 group.

Table 1 Demographics, HIV-related, and cognitive characteristics of the longitudinal sampleDefining cognitive domains and baseline cognitive status

In a PCA of baseline cognitive data from all participants with complete data in all 9 NC measures (n = 263), three components had eigenvalues greater than 1 and explained 43%, 15%, and 13% of the variance in cognitive test scores. Subsequent components each explained < 7.4% of the variance, and the scree plot revealed minimal added benefit after the third component. The three-factor solution had at least two items per factor, rotated factor loadings were strong (> 0.6), and each factor represented a distinct meaningful construct. The speed and executive function (S/EF) factor was comprised of Trail Making Test Parts A and B and Grooved Pegboard Dominant and Non-dominant Hands. Visuospatial memory (VM) was comprised of BVMT-R Immediate Recall, Delayed Recall, and Recognition. Verbal Fluency (VF) was comprised of Letter Fluency and Category Fluency (Supplemental Table 2). The rotated factors accounted for 29%, 24%, and 18% of variance, respectively, for a total of 71%.

A two-step CA was performed with participants with complete data in all three domain z-scores (n = 295). The CA revealed a two-cluster solution. Bayesian Information Criterion (BIC) values decreased markedly from the first to the second cluster, with diminishing reductions through the fifth cluster, and increased thereafter (Supplemental Table 3). The ratio of distance measures and the silhouette measure of cohesion and separation were highest for the two-cluster solution. The first cluster represented 54% of the sample (n = 158) and demonstrated impaired cognition [S/EF: M = -1.07 (SD = 0.67); VM: M = -1.49 (SD = 0.96); VF: M = -0.68 (SD = 0.79)], with scores approximately one SD below that of the second cluster (n = 137), which had intact cognition [S/EF: M = -0.06 (SD = 0.66); VM: M = -0.18 (SD = 0.74); VF: M = 0.55 (SD = 0.71)]. Baseline global cognitive status was not associated with CD4 nadir (Table 1) or with retention within the study (Supplemental Table 1).

CD4 nadir and baseline cognitive status in relation to cognitive trajectories

In participants with multiple visits (n = 132), a LMM was initially used to investigate any relationship between CD4 nadir and global cognition over time (Table 2). A trend towards statistical significance was observed for the association between CD4 nadir (p = 0.071) and the trajectory of global cognitive functioning. Baseline global cognitive status was also associated with change in global cognition over time (p < 0.001). Significant main effects of baseline global cognitive status and time were also present (both p < 0.001).

Table 2 Fixed effects from linear mixed model for global cognition

To further explore the cognitive domains driving these findings, LMMs assessed the association between CD4 nadir and longitudinal trajectories of each cognitive domain (Fig. 1; Tables 3, 4 and 5). CD4 nadir was associated with change in VF over time (p = 0.020 for the interaction of CD4 nadir * years since nadir), but not with change in S/EF (p = 0.123) or VM (p = 0.911). Specifically, the CD4 nadir < 200 group improved in VF [+ 0.035 SD/year, SE = 0.01, t(121) = 3.09, p = 0.002], while the CD4 nadir > 200 group was stable [-0.006 SD/year, SE = 0.02, t(148) = -0.572, p = 0.568]. There was also a main effect of CD4 nadir, such that the CD4 nadir > 200 group had a significantly higher VF z-score than the CD4 nadir < 200 group when collapsing across time points (p = 0.050). The CD4 < 200 group initially began with a lower VF z-score and increased, eventually equaling that of the CD4 nadir > 200 group roughly a decade post-nadir, and then exceeded it.

Fig. 1

CD4 nadir was associated with the trajectory of VF, but not S/EF or VM. Baseline global cognitive status was associated with the trajectory of S/EF and VM, but not of VF. Graphs depict domain-specific cognitive trajectories by CD4 nadir (left) and baseline global cognitive status (right). P values represent the effect of CD4 nadir * time since nadir (left) or the effect of baseline global cognitive status * time since nadir (right). CD4n = CD4 nadir

Table 3 Fixed effects from linear mixed model for Speed/Executive FunctionTable 4 Fixed effects from linear mixed model for Visuospatial MemoryTable 5 Fixed effects from linear mixed model for Verbal Fluency

Baseline global cognitive status was associated with change in S/EF (p = 0.016 for the interaction of cognitive status * years since nadir) and VM (p < 0.001) over time, but not with VF change (p = 0.296). For S/EF, the sample overall improved over time (p = 0.006 for the main effect of years since nadir), but this effect differed by baseline global cognitive status. Namely, the baseline-impaired cognitive group improved in S/EF [+ 0.04 SD/year, SE = 0.01, t(143) = 5.62, p < 0.001], while the baseline-intact cognitive group was stable [+ 0.01 SD/year, SE = 0.01, t(138) = 1.40, p = 0.165]. For VM, the baseline-impaired cognitive group improved [+ 0.05 SD/year, SE = 0.01, t(122) = 4.62, p < 0.001], while the intact group declined subtly [-0.03 SD/year, SE = 0.01, t(144) = -2.35, p = 0.020]. Finally, there were significant main effects of baseline global cognitive status for all three domains (all p < 0.001), with better cognition across visits in those with intact baseline cognitive status.

CD4 nadir and baseline cognitive status in relation to reliable cognitive change

To aid in clinical interpretation of cognitive trajectories, domain-specific RCIs were computed to adjust for factors such as practice effects and regression to the mean. RCI computations used the test with the highest PCA loading and available RCI norms (Woods et al. 2006; VF: Letter Fluency; S/EF: Grooved Pegboard Non-dominant; VM: BVMT-R Delayed Recall). In the overall sample, rates of reliable cognitive improvement and decline, respectively, were 9.4% and 13.3% for VF, 30.6% and 28.1% for S/EF, and 3.8% and 30.8% for VM. Thus, the net cognitive trajectory was stable in the VF and S/EF domains and declining in the VM domain. Domain-specific change categories were not related to inter-test interval (VF: τ = -0.08, p = 0.280; S/EF: τ = -0.04, p = 0.598; VM: τ = -0.05, p = 0.505) or number of repeated assessments (VF: τ = 0.10, p = 0.237; S/EF: τ = -0.01, p = 0.875; VM: τ = -0.06, p = 0.438).

Given LMM results, we examined reliable VF change in relation to CD4 nadir and reliable change in S/EF and VM in relation to baseline global cognitive status. VF improvement was slightly more common in the low nadir group (12.5%) than the high nadir group (7.5%; Fisher’s p = 0.364), while VF decline was slightly more common in the high nadir group (15.0%) than the low nadir group (10.4%; χ2 p = 0.459), but neither comparison was statistically significant. In line with LMM results, S/EF improvement was significantly more common in the baseline-impaired cognitive group (40%) than the baseline-intact group (22%; χ2 p = 0.034), while S/EF decline was similarly common in the two groups (baseline-impaired: 28.3%; baseline-intact: 28.8%, χ2 p = 0.954). Finally, the rate of VM improvement was 6% in the baseline-impaired cognitive group and 2% in the baseline-intact group (Fisher’s p = 0.369), while VM decline was significantly more common in the baseline-intact group (41.7%) than the baseline-impaired group (20.9%; χ2 p = 0.011).

CD4 trajectories as a potential mechanism of VF change

Given the primary findings, CD4 dynamics over time were analyzed as a potential mechanism of the association between CD4 nadir and VF change. CD4 nadir was examined in relation to clinically meaningful change in current CD4 (i.e., improvement from < 500 to > 500 cells/μl vs non-improvement), which in turn was examined in relation to the RCI change category (i.e., improvement, stability, decline) for VF. There was a significant association between CD4 nadir and CD4 change, such that 24.0% (12/50) of the CD4 nadir < 200 group versus only 2.5% (2/81) of the CD4 nadir > 200 group demonstrated an improving CD4 trajectory, [x2(1, N = 131) = 15.02, p < 0.001]. Regarding CD4 change in relation to reliable VF change, 21.4% (3/14) of participants with improving CD4 trajectories demonstrated VF improvement, versus only 8.0% (9/113) of participants with unimproved CD4 trajectories, but this difference was not statistically significant [Exact p = 0.129]. Similarly, 7.1% (1/14) of participants with improving CD4 trajectories and 13.3% (15/113) of participants with unimproved CD4 trajectories demonstrated VF decline [Exact p = 1.00].