Trajectory Analysis of Cognitive Outcomes in Children with Perinatal HIV

Abstract

Background:

Children with perinatal HIV (pHIV) may display distinct long-term cognitive phenotypes. We used group based trajectory modeling to identify clusters of children with pHIV following similar developmental trajectories and predictors of belonging to select cognitive trajectory groups.

Methods:

Participants included children, aged 4 to 17, with pHIV in Thailand and Cambodia. Cognitive measures included translated versions of Intelligence Quotient tests, Color Trails Tests, and Beery-Buktenica Developmental Test of Visual-Motor Integration conducted semiannually over three to six years. Best fit of trajectory groups was determined using maximum likelihood estimation. Multivariate logistic regression identified baseline factors associated with belonging to the lowest scoring trajectory group.

Results:

Group based trajectory analyses revealed a 3-cluster classification for each cognitive test, labeled as high, medium and low scoring groups. Most trajectory group scores remained stable across age. Verbal IQ declined in all three trajectory groups and the high scoring group for Children’s Color Trails Test 1 & 2 showed an increase in scores across age. Children in the lowest scoring trajectory group were more likely to present at an older age and report lower household income.

Conclusion:

Group based trajectory modeling succinctly classifies cohort heterogeneity in cognitive outcomes in pHIV. Most trajectories remained stable across age suggesting that cognitive potential is likely determined at an early age with the exception of a small subgroup of children who displayed developmental gains in select cognitive domains and may represent those with better cognitive reserve. Poverty and longer duration of untreated HIV may predispose children with pHIV to suboptimal cognitive development.

Introduction

Cognitive impairment is common in children with perinatal HIV (pHIV) even in those who are virally suppressed on antiretroviral therapy (ART)¹. Working memory, processing speed and executive function domains are primarily affected in children living with HIV². While the cross-sectional profile of cognitive impairment in children with HIV has been well described³, less is known about the developmental trajectory of children and adolescents living with pHIV⁴. Classification of subgroups of children with pHIV based on longitudinal cognitive performance may elucidate baseline factors associated with increased risk of poor cognitive development and provide prognostic guidance for clinicians and families caring for children with pHIV. Currently, definitions of cognitive impairment vary among studies, limiting our ability to accurately characterize the prevalence of cognitive impairment in children living with HIV globally, especially in cohorts residing in resource-limited settings. In addition, traditional analytic models are not well suited to disentangle complex clinical phenotypes of cognitive development in children with pHIV.

Group-based trajectory modeling addresses these issues by providing an alternative approach to identify homogeneous clusters of individuals that follow consistent paths over time without a priori assignment of criteria for cognitive impairment⁵. Longitudinal data can be summarized in a visually transparent fashion and baseline factors of subgroups with differing developmental potential can be compared in the absence of normative data using group based trajectory modeling. Previous studies have utilized group based trajectory analysis (GBTA) to examine cognitive subtypes in adults with HIV. These studies displayed heterogeneity in cognitive trajectories in adults with horizontally acquired HIV and revealed that older age and longer duration of HIV infection are predictors of cognitive decline in this cohort⁶.

Our previous work utilized multivariate linear regression analyses to compare cognitive scores at select time points in a pediatric HIV cohort in Thailand and Cambodia and revealed significantly worse cognitive performance in the pHIV group compared to uninfected controls regardless of ART treatment history⁷. Here, we expand on these findings by employing group based trajectory modeling to identify subgroups of children with pHIV who display distinct longitudinal cognitive profiles. Then, we used multivariate analyses to examine baseline characteristics across clusters to determine potential risk factors associated with subgroup designation. Identification of early factors associated with poor cognitive outcomes in children with pHIV may provide the basis for screening and intervention in this vulnerable population.

Study Design

Participants included children enrolled in the Pediatric Randomized Early versus Deferred Initiation in Cambodia and Thailand (PREDICT) study, a randomized trial evaluating the impact of timing of ART initiation on HIV disease progression and co-morbidities conducted at 7 sites across Thailand and 2 sites in Cambodia beginning in 2006⁸. ART-naïve children, 1–12 years old, with CD4% between 15–24% and without history of AIDS defining (CDC category C) illness were enrolled in the PREDICT study. Information regarding sex, age, ethnicity, CD4 nadir, caregiver status, caregiver education, and household income was collected using standardized study questionnaires. Caregiver status was classified as living with at least one biologic parent, living with other relatives, or living in an orphanage. Caregiver education was coded as below primary (less than 9 years) or high school and above (9 years or more). Caregiver self-reported yearly household income was stratified as below average, average or above average based on national data. For participants residing in an orphanage, household income data was recorded as a missing variable. Laboratory and clinical evaluations were conducted every 12 weeks and included general clinical history and physical examination, plasma CD4 count and percentage. Plasma HIV viral load was quantified every 24 weeks. Children were randomized to initiate ART at study entry (immediate arm) or according to the national guidelines operative at the time (CD4 < 15% or CDC category C event, deferred arm). First line ART regimen included zidovudine, lamivudine and nevirapine. Informed consent from caregivers and assent from children ≥ 7 years old were obtained in writing prior to study enrollment. The study was approved by Thai and Cambodian national and local Institutional Review Boards. After completion of the PREDICT study (ISRCTN00234091), most caregiver-children dyads elected to enroll in a subsequent longitudinal neurocognitive cohort study, Resilience study. Data included in this study were obtained from both the PREDICT and Resilience studies.

Neuropsychologic testing

Neuropsychologic testing was completed semiannually by Thai psychologists or Thai and Cambodian trained nurses as a sub-study of the main PREDICT trial beginning in 2008 and as part of the Resilience study beginning in 2014. Trained nurses were certified after correctly completing and scoring a minimum of 10 participants per test under supervision. Measures included translated versions of the Wechsler Intelligence Scale for Children, Third Edition (WISC-III)⁹ for ages 6 years and older or Wechsler Preschool and Primary Scale of Intelligence, Third Edition (WPPSI-III)¹⁰ for ages 4 to 5 years, Beery-Buktenica Developmental Test of Visual-Motor Integration (Beery VMI)¹¹, and Children’s Color Trails Test 1 and 2 (CTT1 and CTT2)¹². Of note, only Thai versions of the WISC-III and WPPSI-III were available, and therefore, only Thai children completed IQ testing¹². Berry VMI and CTT instructions were translated and back-translated into Thai and Khmer languages by bilingual translators⁶. Age cutoffs differed between tests as follows: WPPSI-III for ages 4 to 5 years, WISC-III for ages 6 and older, Beery VMI for ages 2 years and older and CTT for ages 8 years and older. Thus, the number of children included and average duration of individual follow-up in each test-trajectory group model differed. Average follow-up for neurocognitive testing ranged from 3 to 6 years. External quality assurance was performed through randomly selected video recording, observation and scoring review by senior US and Thai neuropsychologists⁷. Raw scores for all tests were transformed to scaled scores, all within established score ranges, using US-based norms per referenced manuals. Standardized scores were used to incorporate the impact of age and sex on cognitive performance across the wide age range in our cohort.

Statistical analyses

We applied a modified version of group based trajectory modeling to identify subgroups of children following similar cognitive trajectories across age using a STATA based program developed by Jones and Nagin⁵. Traditional GBTA identifies clusters of participants following different clinical courses over time with a clearly defined starting point (i.e. week 0 of a study or specific age at enrollment). Our modified approach evaluated children enrolled at different ages over variable intervals of follow-up to display cohort heterogeneity in cognition across age. Given the variability in age at enrollment and follow-up time, our modifications allowed us to display age related changes in cognitive scores without reducing the power of our study (i.e. excluding the youngest or oldest subgroups or those with short follow-up times).

Neuropsychologic tests were analyzed independently. Cognitive scores for each trajectory group were analyzed as a function of age using maximum likelihood estimation as described by Nagin et al⁵, adjusting for the time varying covariates of plasma HIV-RNA (log-transformed) and treatment status (on or off ART). Joint estimation of the parameters describing trajectory shape and time varying covariates of treatment status and viral load allowed us to account for the influence of variable treatment initiation and resulting viral suppression on trajectory course. Best fit of trajectory group number and shape was determined using Bayesian Information Criterion¹⁴. Individuals were assigned to the group for which posterior probability of membership was highest. Interactions between age and trajectory groups were assessed using random effects linear regression to confirm that trajectory groups displayed statistically distinct cognitive outcomes across the age range. Posterior probabilities and odds ratios were used to assess the adequacy or “fit” of the model. Group homogeneity was determined by minimum fit requirements of group average posterior probability > 0.7 and odds of correct classification > 4.5. If trajectory groups met only one of two requirements, they were still included in the overall model, but not analyzed independently in risk factor analyses.

Additionally, we assessed change in standardized cognitive test scores as participants aged using random effects linear regression and the score at the earliest time point (youngest age) at which the test was conducted as a reference group. Change in scores at each age in years was assessed sequentially and was considered significant if linear coefficients for change from baseline test score across consecutive age groups met the threshold of p < 0.05. Magnitude of change in scores was determined by comparing the score at the oldest age at which linear coefficients remained significant to the score at youngest age (reference) for each trajectory group. Negative changes in scores were characterized as a decline in cognition and positive changes in scores as cognitive improvement. Trajectories not meeting this criterion were considered to reflect stable scores across age. Finally, multiple logistic regression was used to identify baseline factors associated with belonging to the lowest scoring trajectory group for each cognitive test. Dichotomized variables of CD4 nadir (< or ≥ 350 cells/mL), treatment arm (immediate versus deferred ART), household income (below average versus average or above average), biologic parent as primary caregiver, caregiver education (under 9 years of schooling or 9 years of schooling and above), and age at study entry (< or ≥ 8 years) were assessed as baseline risk factors. Variables in univariate analyses achieving threshold of p < 0.15 were retained in multivariate models; factors associated with group classification were considered significant at p < .05.

Results

Among the 286 children enrolled in the cognitive sub-study from 2008 to 2014, all completed the Beery VMI, 264 (92%) completed the CTT 1 and 2, and 165 (58%) completed WISC-III and WPPSI-III verbal and performance IQ testing. Demographic and treatment information are listed in Table 1. Over half of children were on ART at the start of the neurocognitive sub-study and the majority (87%) were on treatment at the end of the PREDICT study in 2011. Subsequently, in the Resilience study, all remaining participants started ART.

Table 1.

Demographic and treatment variables

	Intelligence Quotient test1	Color Trails test2	Beery VMI
n	165	264	286
Age at baseline ND test [Median (IQR)]	8.2 (6.3–10.3)	8.7 (8.3–10.0)	7.0 (4.6–9.1)
Sex [n (percentage)]
Male	62 (37.6)	110 (41.7)	121 (42.3)
Female	103 (62.4)	154 (58.3)	165 (57.7)
Ethnicity [n (percentage)]
Thai	165 (100.0)	161 (61.0)	170 (59.4)
Cambodia	0 (0.00)	103 (39.0)	116 (40.6)
On ART at start ND test [n (percentage)]
No	68 (41.2)	88 (33.3)	139 (48.6)
Yes	97 (58.8)	176 (66.7)	147 (51.4)
On ART at last ND test [n (percentage)]
No	25 (15.2)	35 (13.3)	37 (12.9)
Yes	140 (84.9)	229 (86.7)	249 (87.1)
Week on ART at first ND test [Median (IQR)]	45.0 (17.9–72.0)	102.0 (57.1–242.0)	36.0(11.9–60.0)
Week of follow-up (first ND to last ND) [Median (IQR)]	279.9 (217.7–314.7)	200.4 (79.4–283.6)	297.1(278.1–316.1)
Primary caregiver [n (percentage)]
Biological Parent	63 (38.2)	166 (62.9)	183 (64.0)
Other relative	65 (39.4)	63 (23.9)	65 (22.7)
Orphanage	37 (22.4)	35 (13.3)	38 (13.3)
Household Income [n (percentage)]
Low income	67 (40.6)	147 (55.7)	162 (56.6)
Average or above average	57 (34.6)	78 (29.6)	82 (28.7)
Unknown	41 (24.9)	39 (14.8)	42 (14.7)

¹Only Thai children completed Intelligence Quotient testing due to language constraints for the WISC-III and WPPSI tests

²Fewer children completed Color Trails Test (CTT) compared to the Beery-Buktenica Developmental Test of Visual-Motor Integration (Beery VMI) due to differences in age restrictions for each test

Figures 1–3 show trajectory models for each cognitive test after adjusting for time-varying covariates of treatment status and viral load. For each neurocognitive test, group based trajectory analysis revealed a 3-cluster trajectory classification, representing high, medium and low scoring groups across age (Figures 1–3). Thirteen of the 15 trajectory groups met both minimum thresholds requirements for individual group assignment (≥ 0.7 for posterior probability and/or ≥ 4.5 for odds ratio). The remaining two trajectory subgroups met one of the two criteria. For the medium trajectory group in CTT1, posterior probability was 0.77 and odds of correct classification was 2.57. For the medium trajectory group in CTT2, posterior probability was 0.67 and odds of correct classification was 4.86. We identified similar trends across all trajectory groups for select cognitive tests. Verbal IQ scores demonstrated a significant decline by approximately 10-points across childhood and into adolescence for all trajectory groups (Figure 1B). Beery VMI scores and Performance IQ scores remained relatively stable across age (Figure 1A and Figure 3). The direction of change in Verbal and Performance IQ scores for ages 4–5 years (WISC-III) and 6 years and over (WPPSI-III) did not differ in independent analysis. The high scoring trajectory groups for CTT 1 & 2 showed a 10-point or greater significant increase in scores across older childhood and early adolescence, while the low and medium scoring trajectory groups showed no change in scores across age (Figure 2). Means and standard deviations of cognitive scores across age for each trajectory group, as well as linear coefficients and p-values for change in scores compared sequentially using baseline neurocognitive test scores as the reference value are provided (see table, Supplemental Digital Content 2).

Figure 1: Performance and Verbal Intelligence Quotient (IQ) trajectory groups.

Trajectory subgroups, characterized as high, medium and low, for performance and verbal IQ test and percentage of children of children belonging to each subgroup. Solid lines display mean scores and dashed lines display confidence intervals for each trajectory group across age. A. Scores in each subgroup for performance IQ remained relatively stable across age. B. Scores in all trajectory groups for verbal IQ decreased across age by approximately 10 points.

Figure 3: Beery-Buktenica Developmental Test of Visual-Motor Integration (Beery VMI) trajectory groups.

Beery VMI trajectory subgroups characterized as high, medium and low performance and the percentage of children belonging to each subgroup. Solid lines display mean scores and dashed lines display confidence intervals for each trajectory group across age. Scores of Beery VMI remained stable across age in all trajectory groups.

Figure 2: Color Trail Tests trajectory groups.

A. Trajectory subgroups, characterized as high, medium and low, for Children’s Color Trails Test 1 and 2 and percentage of children of children belonging to each subgroup. Solid lines display mean scores and dashed lines display confidence intervals for each trajectory group across age. A. Scores in the low and medium trajectory groups remained stable, while scores in high group increased by 10 points across age for Childen’s Color Trails Test 1. B. Scores in the low and medium subgroups remained stable, while scores in the high performance group increased by 11 points across age for Children’s Color Trails Test 2.

Multivariate logistic regression analyses identified risk factors predictive of low cognitive trajectory group status (Table 2). Average or above average household income was associated with reduced odds of belonging to the lowest cognitive trajectory groups across multiple cognitive tests (OR 0.33–0.41; 95%CI (0.17–0.65) for CTT1 and (0.22–0.72) for Beery VMI p<0.005). For the Beery VMI test, children not living with their biologic parent had reduced odds of belonging to the lowest scoring trajectory group (OR: 0.42, 95%CI (0.22–0.82), p=0.01). For Performance IQ, older age (≥ 8 years) at time of presentation was associated with increased odds of low cognitive trajectory group status (OR 2.72; 95%CI (1.31–5.65), p=0.01). No baseline risk factors were significantly associated with belonging to the lowest scoring verbal IQ or CTT2 trajectory groups. HIV-related variables of CD4 nadir (CD4 count <350 cells/mm³) and treatment arm (immediate versus deferred ART initiation) were not associated with belonging to a select trajectory group (Table 2).

Table 2.

Logistic regression analyses of baseline risk factors for belonging to low scoring trajectory group.

	Performance IQ			Verbal IQ						Color Trails Test 1						Color Trails Test 2			Berry VMI
	Univariate			Univariate			Multivariate			Univariate			Multivariate			Univariate			Univariate			Multivariate
	OR	95%CI	P	OR	95%CI	P	OR	95%CI	P	OR	95%CI	P	OR	95%CI	P	OR	95%CI	P	OR	95%CI	P	OR	95%CI	P
CD4 nadir (cells/mL)
CD4 <350	Ref			Ref			Ref			Ref						Ref			Ref
CD4 ≥350	1.07	(0.45–2.56)	0.88	2.16	(0.90–5.20)	0.09^	2.50	(1–6.28)	0.051	1.74	(0.78–3.90)	0.18				1.58	(0.56, 4.50)	0.39	0.63	(0.27–1.51)	0.30
Treatment arm
Immediate	Ref			Ref						Ref						Ref			Ref
Deferred	1.45	(0.75–2.82)	0.27	1.08	(0.52–2.23)	0.83				0.73	(0.39–1.35)	0.32				1.22	(0.54, 2.74)	0.64	0.90	(0.53–1.52)	0.69
Household Income
Below average	Ref			Ref			Ref			Ref						Ref			Ref			Ref
Average or above average	0.58	(0.27–1.26)	0.17	0.8	(0.36–1.76)	0.58^	0.71	(0.32–1.61)	0.42	0.33	(0.17–0.65)	0.001*	0.31	(0.16–0.62)	0.001*	0.68	(0.27, 1.68)	0.40	0.40	(0.22–0.72)	0.002^	0.41	(0.22–0.74)	0.003*
Unknown	0.73	(0.31–1.73)	0.48	3.14	(0.98–10.12)	0.06^	3.19	(0.98–10.4)	0.06	0.87	(0.32–2.35)	0.79	0.65	(0.23–1.88)	0.43	0.61	(0.2, 1.85)	0.38	1.35	(0.55–3.3)	0.51^	2.56	(0.95–6.94)	0.06
Primary caregiver
Parent	Ref			Ref						Ref						Ref			Ref			Ref
Other	1.52	(0.78–2.98)	0.22	1.43	(0.68–2.97)	0.34				1.12	(0.59–2.12)	0.73				0.56	(0.25, 1.27)	0.17	0.58	(0.34–0.99)	0.05^	0.42	(0.22–0.82)	0.01*
Age at study entry (years)
<8	Ref			Ref						Ref						Ref			Ref			Ref
≥8	2.72	(1.31–5.65)	0.01*	1.27	(0.60–2.68)	0.53				0.83	(0.45–1.54)	0.56				1.53	(0.64, 3.66)	0.34	0.43	(0.24–0.76)	0.004^	0.58	(0.31–1.08)	0.08
Education of caregiver
Primary or less	Ref			Ref						Ref			Ref			Ref			Ref
Higher than primary	1.06	(0.54 −2.11)	0.86	1.52	(0.70–3.29)	0.29				1.71	(0.89–3.27)	0.11^	1.79	(0.88–3.64)	0.11	1.67	(0.70–3.99)	0.25	1.25	(0.72–2.17)	0.42

^{^}Denotes variables included in multivariate analysis using the cutoff of p<0.15 as inclusion criteria

^*Denotes variables significantly associated with belonging to the low scoring trajectory group

Discussion

Group based trajectory analyses displayed significant heterogeneity in cognitive outcomes among our cohort of Southeast Asian children who are long-term survivors of perinatal HIV. Our current approach expands upon our prior analyses, which focused on comparison of mean standardized scores at select time points⁷, and demonstrates that though cognitive scores for most children remain relatively consistent over time, a subgroup of children in select domains experience change in cognitive scores across age. Additionally, we found that sociodemographic factors significantly influence cognitive development and trajectory group classification. GBTA revealed the diversity in longitudinal cognitive outcomes across childhood, which were succinctly classified into 3-cluster models, and allowed for assessment of baseline risk factors associated with persistently low cognitive scores.

All verbal IQ trajectory groups experienced a decline in scores indicating that verbal function may be a sensitive indicator of neurocognitive dysfunction as children with perinatal HIV age. Previous studies have also demonstrated high rates of language impairment in youth with perinatal HIV infection and prenatal HIV exposure and the persistence of language impairment over an 18-month follow-up period^16,17. Our study expands on these findings and demonstrates that verbal function, as measured by Wechsler Intelligence tests, declines as children with pHIV age even in those who demonstrate relatively high verbal IQ scores in early childhood. It is also likely that the use of US-based norms or other inherent test-specific cultural biases impart non-linear effects when applied to contemporary non-native English speaking populations. These effects may become more apparent as children age. The increase in performance in the high scoring trajectory group for Children’s Color Trails Tests 1 and 2 reflect improvement in processing speed, motor and executive function across age and suggest that a subgroup of children with preserved cognitive reserve may benefit from practice effect in these cognitive domains during adolescence. In contrast, low and medium scoring groups with stable scores over time are potentially representative of a subgroup of children who are unable to achieve these developmental gains due to HIV-related neuronal injury and resulting low cognitive reserve¹⁸. These results demonstrate the utility of GBTA in identifying subgroups of children with distinct cognitive profiles across the age spectrum. Further studies evaluating cognitive trajectories in HIV-uninfected Thai and Cambodian children are needed to determine the clinical significance of these findings.

Prior work has demonstrated the strong influence of socioeconomic factors on development in children with and without HIV^13,19,20. Our results indicate that certain environmental factors, especially poverty, are associated with poor cognitive outcomes independent of disease severity and treatment in children who are long-term survivors of HIV. Poverty is a known predictor of poor developmental outcomes in children residing in low and middle-income countries, including Thailand and Cambodia^21,22. Similar to our findings, a study of cognitive impairment in adolescents and young adults living with perinatal HIV in the UK demonstrated a correlation between poverty and low neurocognitive scores, but not with HIV-related factors in youth without prior AIDS defining illness³.

Interestingly, children living with their biologic parent(s) as the primary caregiver were more likely to belong to the lowest scoring cognitive group on the Berry VMI test, independent of other sociodemographic factors. These results contrast with other studies, which report better cognitive outcomes in children with pHIV who reside with their biologic parents^16,23. Mediating factors may account for variability in this relationship. Some examples include family density (i.e. number of children per caregiver) and caregiver stress, which may be more often experienced by HIV+ caregivers and is known to negatively influence cognitive and behavioral functioning in children living with HIV²⁴. A recent study found that higher levels of perceived neighborhood and violence related stress among caregivers was associated with lower cognitive scores in HIV-infected or exposed youth residing in New York City²⁵. Additional studies are needed to determine if the association between non-parental caregiver status and cognition found in our study is mediated by confounding psychosocial and demographic factors.

Timing of ART initiation in infants and children with pHIV variably influence developmental outcomes^4,7,26,27. Our study reveals that older age (≥8 years) at study enrollment, and therefore, later initiation of ART, is associated with worse performance on intelligence tests. Similar to the results in a recent study utilizing GBTA to evaluate cognitive trajectories in adults with horizontally acquired HIV⁶, our findings suggest that a longer duration of untreated HIV infection may be a risk factor for poor cognition in children with pHIV. Infants and young children with pHIV who initiated ART and were virally suppressed prior to 5 years of age demonstrated improvement in developmental scores in older childhood^4,27,28,29, whereas children who began ART at an older age (median age of 8 years) did not benefit from immediate initiation of ART⁷. These studies and ours suggest that initiation of ART in infancy or early childhood may be protective against suboptimal cognitive outcomes.

Our results should be interpreted with caution as we evaluated cognitive trajectories in children across various ages and intervals of time, and a limited number of children in our study were enrolled at the youngest (<5 years) and oldest (>14 years) ages in the parent protocol resulting in wider confidence intervals at the ends of the age spectrum. Although a strength of group based trajectory analyses is the lack of reliance on predetermined trajectory group criteria, a limitation of this approach is that analyses may result in trajectory values that cannot be interpreted. Additionally, the clinical implications of the current findings is difficult to interpret without an assessment of academic function or a comparison group of normative trajectories for HIV-uninfected Thai and Cambodian children. Our study shows that early initiation of ART may be neuroprotective in children with pHIV and interventions to improve cognitive outcomes focused on older children with pHIV and delayed initiation of ART are needed. Finally, some variables that might influence cognitive status were not available for examination in our study, including factors mediating the relationship between poverty and cognition. Regardless, our study identified the diversity in cognitive profiles of children with pHIV and predictors of sustained poor cognition in this vulnerable population. Longer follow-up including transition into older adolescence and assessment of cognitive trajectories in HIV-exposed, uninfected and HIV-unexposed, uninfected Thai and Cambodian children represent important areas of future work.