Trajectories of depressive symptoms, neurocognitive function and viral suppression with antiretroviral therapy among youth with HIV over 36 months

Abstract

Background:

Depression and neurocognitive impairment are highly prevalent among persons living with HIV and associated with poorer clinical outcomes; however, longitudinal studies of depression-neurocognition relationships in youth living with HIV (YLWH), and the role of antiretroviral therapy (ART), are lacking. This study tested whether (1) depressive symptomatology, across somatic, cognitive, and affective symptom domains, improved with ART, and (2) more severe depressive symptoms at baseline were associated with poorer neurocognitive function and (3) poorer HIV suppression.

Setting:

Data were collected from 181 YLWH (18-24 years) who were treatment-naïve, a subset of whom (n=116) initiated ART.

Methods:

Participants were categorized into elevated (DS) or non-elevated (non-DS) depressive symptom groups at entry (Beck Depression Inventory-II ≥14) and followed for 36 months. Neurocognition (five-domain battery) and depressive symptoms were repeatedly assessed. Longitudinal models examined depressive symptomatology, neurocognition, and odds of HIV non-suppression by group.

Results:

Greater improvements in depressive symptoms were observed in the DS group over 36 months (beta=−0.14, [−0.24,−0.03]), particularly within cognitive and affective domains. Verbal learning performance increased in the DS group (beta=0.13, [0.01,0.24]), whereas psychomotor function improved somewhat in the non-DS group (beta=−0.10, [−0.22,0.00]). Adjusted for ART adherence, odds of HIV non-suppression did not significantly differ by group (odds ratio=0.22, [0.04,1.23]); however, greater somatic symptoms at study entry were associated with increased risk of non-suppression over time (odds ratio=2.33 [1.07,5.68]).

Conclusion:

Depressive symptoms were associated with differential neurocognitive trajectories, and somatic depressive symptoms at baseline may predict poorer subsequent HIV suppression. Identifying and treating depressive symptoms at ART initiation may benefit neurocognitive and clinical outcomes in YLWH.

Introduction

Clinical depression is the most prevalent psychiatric disorder affecting persons living with HIV (PLWH) in the United States¹. Depressive symptomatology is known to increase as HIV progresses, and when left untreated can hasten poor clinical outcomes and increase mortality². Depression also reduces adherence to antiretroviral therapy (ART)³, which limits its efficacy in suppressing HIV. Thus, identifying and treating depressive symptoms among PLWH, particularly upon initiation of ART, is of central importance to improving clinical outcomes.

Increased depression risk among PLWH is likely a result of both biological (i.e., HIV infection-related) and psychosocial factors⁴, such as social stigmatization and substance use. Among non-HIV-infected persons, depressive symptomatology manifests across multiple symptom domains, encompassing somatic (e.g., fatigue, sleep disturbances), cognitive (e.g., guilt, sadness), and/or affective (e.g., anhedonia) features, with somatic complaints more strongly influenced by immunologic changes relative to other domains⁵. By extension, in PLWH, somatic symptomatology may be more attributable to immunologic sequelae of HIV than other biological mediators^6,7. While certain ART regimens may influence depressive symptoms^8,9, the extent to which ART alters depressive symptoms in a domain-specific manner is unclear. Furthermore, relationships between ART and depressive symptom domains in younger populations (i.e., youth living with HIV; YLWH) with less advanced disease remain underexplored.

Depressive symptoms are also associated with neurocognitive impairment, which is prevalent among PLWH across the lifespan¹⁰ and may, in fact, be worsened by comorbid depression, according to cross-sectional analyses^11,12. However, longitudinal studies of depression-neurocognition relationships in PLWH are limited and have yielded conflicting results^13,14. Differences in study population, such as age, time since diagnosis, or ART status, all of which affect cognitive function among adult PLWH¹⁵, or in methodology used to measure depression (e.g., depressive symptom scores versus clinical diagnosis) and neurocognitive function (e.g., global versus domain-specific) may account for the lack of consensus. Importantly, the cognitive benefits of ART for YLWH, who have significant risk for impairment and present with a distinct pattern of neurocognitive functioning, are less clear^16,17, and longitudinal studies examining depression-neurocognition associations in YLWH are severely lacking. Given the high prevalence of neurocognitive deficits and depression in YLWH^17,18, understanding the impact of depressed mood on neurocognitive function longitudinally is essential for optimizing cognitive health and mitigating adverse clinical outcomes in YLWH who initiate ART.

The current study examined longitudinal changes in depressive symptomatology and neurocognitive functioning in YLWH initiating ART. Our prior work with this cohort identified improvements in motor, complex executive, and attentional/working memory neurocognitive domains across 3 years, which were unrelated to ART treatment or depressive symptomatology¹⁶. Thus, here we tested first whether depressive symptomatology improved during ART in a symptom-specific manner and the role of high depressive symptom burden at baseline (i.e., DS group) on neurocognitive domain-specific changes. Second, we evaluated whether the DS group demonstrated differential changes in neurocognitive functioning over time, and the extent to which depressive symptomatology co-varied with particular domains of neurocognitive performance. Lastly, we sought to determine whether individuals in the DS group had poorer HIV suppression across 3 years of ART.

Methods

Study participants

A total of 181 participants were enrolled in ATN 071, ‘Neurocognitive Assessment in Youth Initiating ART,’ between April 2008 and July 2010, a 3-year longitudinal study conducted at 22 urban sites throughout the US including Puerto Rico (ClinicalTrials.gov Identifier NCT00683579). Enrollment included YLWH aged 18-24 years with behaviorally-acquired HIV infection and no prior ART except limited ART to prevent mother-to-child HIV transmission. Inclusion/exclusion criteria and ART regimens were previously described¹⁶, including efavirenz usage. Treatments groups were as follows: 1) fifty-six participants who initiated ART with CD4⁺ T cell counts >350 cells/uL as part of their participation in a study of early ART (ATN 061); 2) sixty-five participants with CD4⁺ T cell counts >350 cells/uL who delayed treatment until CD4 counts declined to <350 cells/uL; and 3) sixty participants with CD4⁺ T cells counts <350 cells/uL who initiated ART at entry. Thirty-one participants were lost to follow-up at week 144; thus, 150 (84.5%) participants were retained at study exit. Neurocognitive and depressive symptom data at both entry (week 0) and exit (week 144) were available from 133 and 139 participants, respectively. Plasma RNA viral load levels, CD4⁺ T cell counts, information on neuropsychiatric and other medical diagnoses, ART adherence, socioeconomics, demographics, and medications were obtained from participant interviews at each study visit and from clinic records as previously described¹⁷. ART adherence was measured by 7-day self-report of percent doses taken.

Depressive symptomatology

Depressive symptoms were assessed using the 21-item Beck Depression Inventory, 2^nd Edition (BDI-II) at all visits. BDI-II is scored by summing the item ratings, which use a 4-point scale (0-3), with a total symptom score range of 0 to 63. Original validation of psychometric properties of the BDI-II in psychiatric outpatients¹⁹ determined cut-off scores as follows: 0-13 minimal, 14-19 mild, 20-28 moderate, 29-63 severe. Thus, in our analyses, individuals with BDI-II scores ≥14 at baseline (i.e., mild symptoms or greater) were categorized as having significant depressive symptoms (referred to as the “DS group”). In other words, DS group membership was set at baseline, though BDI-II scores were assessed over time. Factor analyses of BDI-II typically identify a 2-factor structure encompassing ‘somatic’ and ‘cognitive-affective’ symptom domains²⁰. Recently, however, a 3-factor model was shown to best differentiate BDI-II scores of PLWH, specifically into somatic, cognitive, and affective subdomains²¹. The somatic subdomain (BDI-S) is derived from items assessing fatigue, sleep, energy, and appetite; the cognitive subdomain (BDI-C) captures depressive cogitations, including self-worth, pessimism, guilt, and suicidality, and the affective subdomain (BDI-A) includes anhedonia, crying, and apathy items.

Neurocognitive testing

The assessment battery included five neurocognitive performance domains known to be vulnerable to HIV infection, namely (1) psychomotor/motor, (2) attention/working memory, (3) executive function, (4) verbal learning, and (5) visuospatial memory, as described previously^16,17. Briefly, standard scores using published normative data, with adjustments for age and, where available, race, Hispanic ethnicity, education, and/or gender, were computed based upon the following instruments: psychomotor/motor scale scores were derived from the Digit Symbol (Wechsler Adult Intelligence Scale; WAIS-III²²), Grooved Pegboard²³, and Timed Gait²⁴ tests; attention/working memory scale scores from the WAIS-III Digit Span and Letter-Number Sequencing subtests; executive function scale scores from Trailmaking Test Part B^25,26, Controlled Oral Word Association²³, and Stroop Test Interference Trial^27,28; verbal learning scale scores from the Hopkins Verbal Learning Test-Revised^27,29; and visuospatial memory scale scores from the Brief Visuospatial Memory Test-Revised^27,30. Participants were followed for approximately 3 years, completing study evaluations at weeks 0, 24, 48, 96, and 144. Domains 1-3 were evaluated at all visits, domains 4 and 5 assessed only at weeks 0 and 144. Standard scores based on published normative databases were computed and converted to z-scores to allow combining tests using a common metric, such that higher scores indicate better performance. Potential neurocognitive confounders (e.g., poor effort, cannabis use on day of visit) identified at each study visit were exclusion criteria for corresponding participant-visit data in analyses.

Statistical analyses

Means and standard deviations are reported for demographic and clinical variables, and summary statistics calculated at baseline were compared between DS groups using the nonparametric Kruskal-Wallis test, t-test, or Chi-squared test, where appropriate. Viral load values were coded as detected or not-detected (33.1% of total), based on the assay lower limit of detection (50 copies/mL). Viral non-suppression was defined as either having detectable virus at study completion, or inconsistent suppression across the study (detectable virus at ≥2 visits). To assess longitudinal changes in depressive symptoms and neurocognitive performance, linear mixed-effects models (LMM) were fit using lme4 ³¹ with random intercepts varying among participant and study site, and which included (1) time (weeks from study entry) and (2) a time-by-DS group interaction as fixed effects. Model covariates included participant age, race (3-level factor representing major racial composition: African American, Caucasian, or Other/Mixed/Unspecified), gender, education level (3-level ordered factor: incomplete high school/GED, completed high school/GED, or beyond high school/GED), annual income (3-level ordered factor: <$3,000, $3,000-$12,000, or >$12,000), current employment status (2-level factor: employed/unemployed), illicit drug use (Y/N for stimulants, cocaine, opiates, dissociatives, or hallucinogens ≥weekly), cannabis use (Y/N ≥weekly use), detectable viral load (Y/N), absolute CD4⁺ T-cell count, and treatment group. Numeric variables were standardized in all models. To assess the effects of clinical factors, sociodemographic variables, and time on BDI-II total and subdomain scores, four LMMs were fit as described above. BDI-II total and subdomain scores were log-transformed to improve normality of residuals, assessed by quantile-quantile plot. Five LMMs were fit as described above, each with neurocognitive domain score as dependent variable. One logistic regression model tested the extent to which DS group was associated with risk of viral non-suppression, as defined above. In this model, which included only those who initiated ART at study entry (i.e., treatment arms 1 and 3; N=116), ART adherence was computed by calculating each individual’s mean weekly percent adherence across the study, and continuous viral load was used, as all subjects had >50 counts at baseline.

Regression diagnostics were conducted by identifying outlier cases in each model based on Bonferroni-corrected t-tests of studentized residuals and identifying cases of high influence based on Cook’s Distance and hat (leverage) values, omitting these cases, and re-evaluating model estimates. Removal of six outlying observations from the psychomotor/motor index model and seven observations from the complex executive index model shifted interaction term estimates (Table 3); however, these cases did not possess attributes that may have biased the result (e.g., lost to follow-up, study site, treatment arm, clinical or sociodemographic characteristics). Standardized betas (β) and bootstrapped 95% confidence intervals (500 iterations) for LMMs were used to estimate the expected change in depressive symptoms and cognitive performance per one-unit change in fixed effects of interest, and estimated conditional means and bootstrapped 95% CIs were computed using effects ³² and lme4, respectively, and reported between DS groups. Bootstrapped CIs (2000 iterations) were also computed for exponentiated log-odds (i.e., odds ratios; OR) of predictor variables in the logistic model. Analyses conducted in R 3.6.0 ³³.

Results

Participant characteristics and depressive symptomatology

At visit 1 (study entry), 72 (39.8%) participants had elevated depressive symptoms (DS group: BDI-II total score ≥14) and 18 (9.9%) had a concurrent diagnosis of clinical depression. Individuals in the DS group were more likely to have clinical depression (X²=6.03, p<0.05) and less education (X²=6.14, p<0.05) than individuals in the non-DS group, and were more likely to be unemployed (X²=8.18, p<0.01) and regularly use tobacco (X²=7.05, p<0.01) and cannabis (X²=4.02, p<0.05). Five individuals with clinical depression were on antidepressant pharmacotherapy at visit 1, and 5 more initiated treatment at some point during the study. Statistically significant differences in age, race, gender, household income, alcohol use, CD4⁺ T cell counts, viral load, and treatment arm assignment were not observed between DS and non-DS groups (Table 1). However, ART medication adherence throughout the course of the study was lower in the DS group (weekday: 84±2% versus 91±1%, Kruskal-Wallis H=14.3, p<0.001; weekend: 79±3% versus 89±2%, H=9.39, p<0.01), who were also more likely to be lost to follow-up (X²=4.17, p=0.04). Notably, ART adherence did not differ across the three treatment groups (weekday: X²=1.30, p=0.52; weekend: X²=1.59, p=0.45).

Table 1.

Study participant characteristics at baseline.

	Non-DS (N=109)	DS (n=72)
	BDI < 14	BDI ≥ 14	t/X2
Age	20.9 (1.8)	21.0 (1.7)	0.91
Race (AA/Other/C)	75/17/17	46/19/7	3.82
Education (<HS/HS/>HS)	23/35/51	27/16/29	6.14*
Gender (M/F)	90/19	56/16	0.37
Income (<3k, 3-12k, >12k)	52/30/25	41/19/9	3.25
Employment (Y/N)	56/52	21/51	8.18**
Tobacco Use (Y/N)	34/73	38/34	7.05**
Alcohol Use (Y/N)	30/78	20/52	0.00
Clinical Depression (Y/N)	6/103	12/60	6.03*
Cannabis Use (Y/N)	27/81	29/43	4.02*
Viral Load (median copies/mL blood)	16441	15900	0.14
CD4 T cell count (cells/uL blood)	472 (204)	487 (239)	−0.41
Study Arm (1/2/3)	33/41/35	23/24/25	0.35
BDI-II Total Score	6.77 (3.9)	24.2 (8.8)	16.5**
BDI-II Somatic Domain Score	3.69 (2.4)	10.5 (4.3)	11.9**
BDI-II Cognitive Domain Score	1.79 (1.7)	8.8 (4.3)	16.8**
BDI-II Affective Domain Score	1.29 (1.5)	4.9 (2.7)	12.3**
Psychomotor/Motor Domain	−0.80 (0.89)	−1.10 (1.64)	1.2
Complex Executive Domain	−0.18 (0.74)	−0.54 (1.04)	2.23*
Attention Domain	−0.25 (0.76)	−0.43 (0.71)	1.42
Verbal Learning Domain	−1.31 (1.05)	−1.87 (0.97)	3.16**
Non-Verbal Memory Domain	−0.76 (1.25)	−1.01 (1.30)	1.16

Significance tests based on two-tailed, independent-samples T-tests, Kruskal-Wallis, or X² tests. DS=Depressive symptom group; BDI=Beck Depression Inventory-II; AA=African American; C=Caucasian; HS=High School; Substance use was dichotomized by weekly usage. Neurocognitive domain z-scores were adjusted using published standard scores for age, gender, race, and education, as previously reported. Values reported as mean (SD).

^*p <0.05

^**p <0.01.

Adjusting for demographics, clinical factors, and ART adherence, longitudinal analysis revealed that depressive symptoms (log-BDI scores) generally decreased throughout the study (Fig. 1), with larger decreases for individuals within the DS group (β_week*DS=−0.14, 95% CI: [−0.24, −0.03]). Raw BDI scores decreased on average from 13.7 ± 10.6 at entry to 9.9 ± 11.1 at study exit (week 144), and the proportion of participants with elevated symptoms (BDI-II ≥14) was significantly reduced (39.8% to 23.8%; X²=9.51, p=0.002). Across the study, cannabis use was associated with greater depressive symptoms (Supplemental Table 1).

Figure 1.

Log-transformed depressive symptom domain scores across time. Individuals with elevated depressive symptom burden (DS; BDI-II score ≥14; closed circles) or minimal symptoms (non-DS; BDI-II score <14; open circles) at study entry. Domain scores based on three-factor model of BDI-II. Conditional mean log-BDI scores were adjusted for demographic and clinical covariates in linear mixed-effects models. 95% confidence intervals shown.

Depressive symptom domains improved over time in both groups (Fig. 1), though improvements in affective and cognitive domain scores were greater in the DS versus non-DS group (affective: β_week*DS=−0.22, 95% CI: [−0.31, −0.14]); cognitive: β_week*DS=−0.19, 95% CI: [−0.28, −0.09]; somatic: β_week*DS=−0.08, 95% CI: [−0.16, 0.01]; Supplemental Table 1). Study group assignment was unrelated to improvements in depressive symptoms. Forty participants (29%) recorded higher BDI scores by study exit, and 16 participants had received a new clinical diagnosis of depression. Weekend ART adherence was higher in individuals with worsening symptoms (89.2% versus 85.2%, H=4.08, p=0.04), with inconclusive differences in weekday adherence (90.0% versus 88.1%, H=1.56, p=0.21). Efavirenz (EFV) use, an antiretroviral associated with neuropsychiatric side effects⁹, was similar between individuals with worsening (n=40 total; 5 taking EFV) versus stable or improved symptoms (n=96 total; 17 taking EFV; X²=0.35, p=0.55).

Depressive symptoms moderate changes in neurocognitive performance

Previously, we reported elsewhere that neurocognitive function improved in motor, complex executive, and attentional/working memory domains, and were unrelated to ART treatment group¹⁶. Exploratory analyses of verbal learning and visuospatial memory performance had also suggested no differential ART treatment group effect. Here, we expanded upon our prior work by examining whether a high baseline depressive symptom burden moderated neurocognitive function across the 3-year study period by re-evaluating neurocognitive performance over time in these five domains as a function of DS group. At baseline (week 0), the DS group performed significantly worse in two of five domains (complex executive index: Z=−0.18±0.74 (non-depressive) versus −0.54±1.04 (depressive), t=−2.23, p=0.03; verbal learning: Z=−1.31±1.05 versus −1.87±0.97, t=−3.16, p=0.002, Table 1). However, by study exit (week 144), the DS group no longer performed statistically significantly worse on complex executive function than the non-DS group (Z=0.08±0.85 (non-depressive) versus −0.02±0.84 (depressive), t=0.62, p=0.54), but still exhibited poorer verbal learning (Z=−1.42±1.09 versus −1.83±1.14, t=−2.00, p=0.049). Across all visits, unadjusted mean scores were generally lower in the DS group (visuospatial memory: Z=−0.80±1.18 versus −1.13±1.26; motor/psychomotor: Z=−0.17±1.04 versus −0.51±1.17; attentional/working memory: Z=−0.15±0.82 versus −0.28±0.82; verbal learning: Z=−1.37±1.07 versus −1.84±1.04; complex executive: Z=0.15±0.82 versus 0.00±0.95).

Formal longitudinal analyses that adjusted for clinical and demographic covariates, however, revealed that changes in verbal learning and motor/psychomotor performance were moderated by baseline depressive symptoms. Specifically, while individuals in the DS group performed worse on verbal learning at baseline (conditional mean Z-score=−1.82, 95% CI: [−2.20, −1.44]) versus non-DS individuals (−1.18 [−1.52, −0.84]), they had improved somewhat by study exit (Z=−1.65 [−2.03, −1.25]), whereas non-DS individuals showed worsening performance (−1.36 [−1.72, −0.99]; β_week*DS=0.13, 95% CI: [0.01, 0.24]; Fig. 2). Conversely, while the two groups performed similarly on motor/psychomotor tasks at baseline, the DS group improved significantly less than the non-DS group across the study (β_week*DS=−0.10, 95% CI: [−0.22, 0.00]; Fig 2). Trajectories across the other three neurocognitive domains did not significantly differ between groups (Supplemental Table 2). Visuospatial memory performance covaried somewhat with depressive symptoms in the cognitive domain (β_BDI-C=0.22 [0.00, 0.43]), though statistically significant associations among other depressive and neurocognitive domains were not observed (Supplemental Table 2). Viral non-detection (<50 copies/mL) at each visit was associated with better concurrent scores on motor/psychomotor (β=0.24 [0.06, 0.40]) and complex executive scales (β=0.20 [0.04, 0.35]).

Figure 2.

Conditional mean Z-scores in neurocognitive functional domains derived from linear mixed-effects models adjusted for demographic and clinical covariates. Individuals with elevated depressive symptom burden (DS; BDI-II score ≥14; closed circles) or minimal symptoms (non-DS; BDI-II score <14; open circles) at study entry (week 0). 95% confidence intervals shown.

Elevated baseline somatic depressive symptoms predict viral non-suppression

Forty-three (37%) of the 116 individuals who began ART at study entry exhibited viral non-suppression over the 3-year follow-up period. Non-suppressors were not over-represented by individuals in the DS group (n=18/43, X²=0.01, p=0.94), nor did logistic regression analysis provide conclusive evidence as to whether the odds of non-suppression were lower in the non-DS group (OR=0.22 [0.04, 1.23]). Somatic depressive symptoms at baseline, however, did confer significantly greater risk for non-suppression (OR=2.33 [1.07, 5.68]) after adjustment for baseline viral load, CD4⁺ T cell count, treatment arm assignment, ART adherence, and other factors (Supplemental Table 3), whereas associations between non-suppression and cognitive or affective symptoms were inconclusive (cognitive: OR=1.00 [0.42, 2.32]; affective: OR=1.53 [0.71, 3.35]). As expected, greater ART adherence was statistically significantly associated with decreased risk of non-suppression (OR=0.46 [0.25, 0.78]; Fig. 3). Notably, cannabis use was associated with lower risk of non-suppression (i.e., better viral suppression) (OR=0.24 [0.06, 0.79]; Fig. 3).

Figure 3.

Baseline risk factors for viral non-suppression during 3 years of ART. Logistic regression odds ratios and 95% confidence intervals shown. Minimal depressive symptom burden at entry (non-DS): BDI-II score <14; ART adherence: average self-reported weekly adherence across study period. Cognitive, affective, and somatic symptoms derived from BDI-II domain scores.

Discussion

Our longitudinal study of YLWH examined whether depressive symptomatology improved with ART and the extent to which depressive symptom burden was associated with poorer neurocognitive outcomes and poorer HIV suppression over three years of follow-up. Regardless of ART status or adherence, we found that depressive symptoms generally improved across the study, with greater improvements in cognitive and affective symptom domains for individuals with the highest symptom burden at entry (i.e., the DS group). Within our cohort, approximately 2 in 5 had elevated depressive symptoms at entry, with nearly 1 in 5 having received a diagnosis of major depression at some point prior to or during the study, largely consistent with major depression prevalence estimates among YLWH in high-income countries³⁴. At the end of three years, less than 25% had elevated depressive symptoms, though some individuals received clinical depression diagnoses and/or initiated antidepressant pharmacotherapy during the study, suggesting improved surveillance and treatment as two potential mechanisms of symptom resolution. However, given the small number of such participants in our cohort, and the preponderance of subclinical depressive symptoms (i.e., BDI-II<14) across the cohort, the causes of improved symptoms likely extend beyond increased clinical engagement.

Improvements in depressive symptoms were consistent with our prior finding that neurocognitive functioning also improved across time, irrespective of ART or clinical indicators of HIV-related disease (e.g., viral load, CD4⁺ T cell count). Depressive symptom improvement (and increases in cognitive performance) may be mediated by psychosocial and biological changes resulting from improved overall health during the study. For instance, management of opportunistic infection or access to social services may underlie improvements in depressive symptoms³⁵, and the fact that this cohort responded relatively well to ART (63% of those initiating ART had viral loads that were below detection at ≥3 follow-up visits) may indicate better physical health than PLWH in general. Larger improvements observed in the DS group within cognitive and affective, but not somatic, symptom domains may reflect overlapping somatic complaints with HIV itself, such as fatigue and appetite loss⁶. It is possible that differential changes in BDI-II scores reflected a floor effect of the instrument, though somatic scale scores were highest of the 3 domains. Consistent with the literature^3,36, ART adherence was, in fact, lower in the DS group, though time-varying depressive symptoms were not associated with ART adherence or viral load. Notably, individuals taking EFV-containing ART regimens did not exhibit poorer depressive symptom trajectories, which was consistent with our prior finding that EFV did not alter trajectories of neurocognitive functioning¹⁶. Participants lost to follow-up were slightly more likely to come from the DS group; thus, the larger improvements in BDI-II scores observed in the DS group may have been caused by biased estimates from data not missing at random. Nevertheless, we found that depressive burden either worsened or did not improve for a considerable subset of participants, which highlights the concerning fact that despite viral suppression, depressive symptoms often persist in YLWH and points to persistent biologic or psychosocial mediators of depressed mood in YLWH.

With respect to neurocognitive functioning, individuals in the DS group performed significantly worse on complex executive and verbal learning tests prior to ART than individuals in the non-DS group. Generally, the DS group performed worse than the non-DS group on all tests throughout the study, supporting the hypothesis that depressive symptoms were longitudinally associated with poorer neurocognition, and extending prior findings of depression-cognition associations from adults³⁷. Our longitudinal analyses indicated that changes in verbal learning and motor/psychomotor performance were moderated by DS group: individuals in the DS group increased verbal learning performance but decreased in motor/psychomotor function over time relative to non-DS individuals. A meta-analysis of depression-neurocognition relationships in youth found verbal learning and memory more strongly associated with depressive symptoms as compared to other neurocognitive domains³⁸. Thus, verbal learning increases may have been mediated by improved depressive symptoms in our sample. Psychomotor slowing is a core feature of depression³⁹, and is common among PLWH^14,40. Across symptom domains, somatic symptom improvements were relatively smaller in the DS group. Persistent somatic disturbances, such as poor sleep and fatigue, may therefore have been reflected in lower motor/psychomotor test scores; however, longitudinal associations between motor/psychomotor test scores and somatic depressive symptoms were not observed in covariate-adjusted analyses. Nevertheless, persistent somatic symptoms in YLWH, despite ART and viral suppression, may be an important treatment target for improving daily functioning in this population. We also found that non-detectable viral load was associated with better executive function and psychomotor/motor test performance across all visits, suggesting that peripheral viremia in YLWH reflects ongoing immune processes (and vice versa) that interact with the CNS to impact neurocognition^41,42.

Our analysis of baseline risk factors for HIV non-suppression revealed that individuals in the non-DS group had somewhat lower odds of non-suppression, adjusted for ART adherence, though this association did not reach statistical significance. Nonetheless, greater somatic depressive symptoms were associated with significantly increased odds of viral non-suppression, suggesting that the association between depressive symptoms and viral non-suppression extends beyond ART adherence. One possible mechanism for the observed somatic depressive symptom-HIV viremia relationship is the role of proinflammatory cytokines, which can be etiologic factors in depression, specifically somatic symptoms, due to their myriad neurobiological effects within the central nervous system (CNS). Elevated proinflammatory cytokines have been reported in PLWH with depressive symptoms versus those without^43,44, and are associated with poorer clinical outcomes in HIV⁴⁵ and weaker responses to ART in YLWH⁴⁶. Elevated cytokines are also associated with poorer responses to pharmacologic antidepressant treatment in non-HIV infected individuals⁴⁷, though antidepressants may actually decrease inflammation in PLWH⁴⁴, which could improve viral suppression. Furthermore, persistent immune activation, as evidenced by elevated soluble CD14 (sCD14), soluble intercellular adhesion molecule 1 (sICAM-1), and soluble vascular cellular adhesion molecule 1 (sVCAM-1), has been associated with poorer cognitive function in YLWH despite well-controlled viral suppression and normal CD4+ T cell counts⁴¹. Somatic depressive symptoms may therefore reflect an immune profile at greater risk for poorer ART responses, even after consideration of other HIV-associated biomarkers, but further exploration of these relationships in YLWH is warranted.

Lastly, we report a positive association between cannabis use and time-varying depressive symptoms, but not neurocognitive performance. Cannabis is used more frequently in PLWH compared to uninfected individuals, often to mitigate ART-related side effects and to manage psychosocial stressors, though its impacts on neurocognitive performance are unclear, with both protective⁴⁸ and detrimental effects⁴⁹ reported. The effects of cannabis use on clinical outcomes are also unclear⁵⁰, though we found cannabis use to be associated with decreased risk of viral non-suppression (i.e., better suppression), despite its positive association with depressive symptoms. Emerging evidence supports various immunomodulatory effects of cannabinoids (e.g., Δ⁹-tetrahydrocannabinol, cannabidiol) in ART-treated PLWH, including decreased T cell activation, inflammatory monocytes, and proinflammatory cytokine secretion (reviewed in Costiniuk and Jenabian, 2019) ⁵¹. However, randomized clinical trials to systematically assess whether cannabinoids are effective adjunctive agents to ART are lacking, and therefore no consensus has emerged regarding cannabis use and HIV viremia. Notably, concomitant use of tobacco, which was used by the majority of cannabis users in our study and exerts pro-inflammatory effects on ART-treated PLWH⁵², may have interfered with the potential immune-enhancing effects of cannabis. Future studies investigating cannabinoids on clinical outcomes in PLWH should therefore adjust for both depressive symptoms and tobacco smoking given their associations with immune function.

Our study had a number of strengths, such as its longitudinal design and comprehensive neurocognitive assessment; however, we acknowledge some limitations. First, our study population was predominately male, which may affect the generalizability of our findings. Among PLWH, men may exhibit a greater frequency of elevated depressive symptoms than women, but women with elevated depressive symptoms may have greater odds of impairment of executive control/inhibition compared to men⁵³. Although our study was not specifically designed to examine gender differences in depression-neurocognition relationships, we included gender as a covariate in all analyses and did not detect any statistically significant gender differences in depressive symptomatology or neurocognitive functioning. Second, our dichotomous classification of depressive symptom burden at baseline used a BDI-II total score cutoff of 14, rather than pre-existing clinical diagnosis of major depressive disorder/episode or a structured clinical interview. We opted for this approach in order to quantitatively assess granular, longitudinal changes in depressive symptoms, and due to the importance of subclinical depressive symptomatology, and its treatment, in outcomes related to medication adherence, quality of life, psychosocial functioning, and viral suppression. In addition, lower resourced individuals have comparatively less access mental health services, and thus clinical depression may have been underdiagnosed upon study entry. Lastly, two of five neurocognitive domains were assessed only at weeks 0 and 144, and our analysis of risk factors for viral non-suppression included a smaller number of participants, which increased uncertainty across our effect size estimates.

In summary, our findings highlight the complex relationships between depressive symptom burden and neurocognitive domain-specific changes in YLWH, and identify somatic symptoms conferring greater risk of viral non-suppression, potentially due to immune and/or inflammatory mechanisms known to correlate with somatic depression and HIV viremia in PLWH. Our results support the need for assessment and treatment of depressive symptoms, particularly somatic symptoms, upon ART initiation in order to optimize neurocognitive and clinical outcomes in YLWH, and for further investigation to clarify the mechanistic role of underlying inflammation in depression-cognition-viremia relationships among YLWH.