Cognitive performance in a South African cohort of people with HIV and comorbid major depressive disorder

Abstract

Cognitive performance in people with HIV (PWH) may be affected by brain injury attributable to the infection itself, by other medical and psychiatric comorbidities (including major depressive disorder; MDD), and by psychosocial factors (e.g., education, food insecurity). We investigated effects of these variables on cognitive performance in a South African cohort of PWH with comorbid MDD and incomplete adherence to antiretroviral therapy (ART). We also examined (a) associations of depression severity with cognitive performance, and (b) whether improvement in depression led to improved cognitive performance. Participants (N = 105) completed baseline neuropsychological, psychiatric, and sociodemographic assessments. Subsequently, 33 were assigned to a cognitive-behavioural therapy for ART adherence and depression (CBT-AD) and 72 to standard-of-care treatment. Eight months post-baseline, 81 (n_CBT-AD = 29) repeated the assessments. We investigated (a) baseline associations between sociodemographic, medical, and psychiatric variables and cognitive performance, (b) whether, from baseline to follow-up, depression and cognitive performance improved significantly more in CBT-AD participants, and (c) associations between post-intervention improvements in depression and cognitive performance. At baseline, less education (β = 0.62) and greater food insecurity (β = −0.20) predicted poorer overall cognitive performance; more severe depression predicted impairment in the attention/working memory domain only (β = −0.25). From baseline to follow-up, depression decreased significantly more in CBT-AD participants (p = .017). Improvement over time in depression and cognitive performance was not significantly associated except in the attention/working memory domain (p = .026). Overall, factors associated with cognitive performance were unrelated to brain injury. We conclude that clinicians examining PWH presenting with cognitive difficulties must assess depression, and that researchers investigating cognitive impairment in PWH must collect information on psychosocial factors.

Introduction

Various factors contribute to the low cognitive performance often observed in people with HIV (PWH). This is because cognition in PWH can be affected by a range of insults to neurological function, including direct injury to the brain from the infection itself and/or from other medical and psychiatric comorbidities (Hong and Banks 2015; Nightingale and Winston 2017; Winston and Spudich 2020).

Comorbid and co-occurring medical conditions that are highly prevalent in populations of PWH and that contribute to cognitive impairment include cerebrovascular disease (Vinikoor et al. 2013), substance use disorders (Millar et al. 2017), central nervous system (CNS) opportunistic infections (Saloner et al. 2019), and neurological insults such as head injuries (Lin et al. 2011). Among psychiatric conditions that are notable for the same reasons, major depressive disorder (MDD) is the most significant. Depression is more prevalent in PWH than in the general population (Freeman et al. 2008; Lofgren et al. 2020; Rezaei et al. 2019). Broadly speaking, a comorbid diagnosis of MDD in PWH negatively affects performance across the cognitive domains of motor function, processing speed, attention/working memory, learning and memory, and executive function (Bragança and Palha 2011; Fellows et al. 2013; Goggin et al. 1997; Rock et al. 2014; Rubin and Maki 2019). More severe depression is associated with greater impairment in these domains (McDermott and Ebmeier 2009; Paolillo et al. 2020; Shimizu et al. 2011).

Two distinct reasons may explain why PWH with comorbid MDD display these cognitive impairments. First, in cases where there is no identifiable pathological mechanism, the depression itself may give rise to cognitive deficits (Kang et al. 2014; Kiloh 1961). Second, the depression and cognitive impairment may share a common underlying pathological mechanism (neuroinflammation, caused by HIV infection of the CNS; Del Guerra et al. 2013; Fellows et al. 2013; Leonard and Maes 2012; Miller et al. 2009).

Incomplete adherence to antiretroviral therapy (ART) can also contribute to poor cognitive performance in PWH. Lack of adherence can lead to worse HIV disease outcomes (i.e., higher HIV RNA viral load, lower current and nadir CD4 count) and, consequently, a greater risk of cognitive impairment. In addition, ART can be associated with neurotoxicity. Efavirenz, particularly, has neuropsychiatric side effects (Cysique and Becker 2017; Saylor et al. 2016; Winston and Spudich 2020).

Aside from medical and psychiatric comorbidities, psychosocial factors (e.g., educational level, socioeconomic status, and food insecurity) can also strongly influence cognitive performance in PWH (Cysique and Becker 2015; Dreyer et al. 2021b; Kabuba et al. 2018; Nightingale et al. 2021; Watson et al. 2019; Winston and Spudich 2020).

Despite the numerous and varied factors that might contribute to poor cognitive performance in PWH, interventions targeted at improving that performance are rare: Most extant intervention programs tend to focus on improving HIV disease outcomes via management of ART (Alford and Vera 2018; Winston and Spudich 2020). Of particular interest here is that no studies to date have investigated whether treating depression improves cognitive performance in PWH and whether depression-related cognitive deficits are reversible in PWH. This is a notable knowledge gap especially given that cognitive-behavioral therapy for adherence and depression (CBT-AD) has an accumulated evidence base, including for PWH in South Africa (Mendez et al. 2021; Safren et al. 2014, 2021, 2009; Sherr et al. 2011). If cognitive impairment is non-organic (i.e., caused by the depression itself, without an identifiable pathological mechanism), successful treatment of the depressive disorder with CBT-AD should improve cognitive outcomes (Connors et al. 2019).

Hence, the current study had two primary aims. First, we sought to investigate the contribution of HIV-related factors, medical and psychiatric comorbidities, and psychosocial variables to cognitive performance in a sample of PWH and comorbid major depressive disorder, who have incomplete ART adherence. Identifying potentially modifiable contributors to cognitive impairment in PWH could aid treatment strategies and help improve the functioning of PWH. Second, we sought to determine whether (a) depression severity in that sample is associated with impaired cognitive performance, and (b) improvement in depression over time, as a consequence of exposure to either CBT-AD or standard of care, leads to improved cognitive test performance.

Method

Participants and setting

Participants were 105 PWH with MDD who had failed first-line ART. Those who were not virally suppressed (HIV RNA viral load > 400 copies/mL) at baseline (n = 72) were part of the sample of a large randomized controlled trial of a cognitive-behavioral treatment for ART adherence and depression (CBT-AD; Joska et al. 2020; Safren et al. 2021). Of those 72, 33 had been assigned to the treatment arm and 39 to a standard of care condition. As part of this study, an additional 33 participants who were not part of the trial were also assigned to a standard of care condition. Hence, in this study, we had 33 participants in the CBT-AD group and 72 participants in the standard of care group.

Of the total sample of 105 participants, 81 (29 in the CBT-AD group, 52 in the standard of care group) were assessed again 8 months later (see Fig. 1).

Fig. 1.

Flowchart depicting stages of the study protocols

Data collection occurred at two primary care community clinics in Khayelitsha, a peri-urban community in Cape Town, South Africa. Khayelitsha was established under the principle of racial segregation executed by the apartheid regime. As a consequence of this legacy, today almost all of its residents are Black African and it is one of the poorest areas of Cape Town. Most adult residents of Khayelitsha speak isiXhosa as a first language, fewer than one-third of adult residents have completed high school, and there are high levels of HIV infection, crime, and unemployment (Crush et al. 2012; Nleya and Thompson 2009; Smit et al. 2016; Stern et al. 2017; City of Cape Town 2013).

Inclusion criteria for this study were (a) ≥ 18 years of age, (b) HIV-seropositive status (confirmed via medical record), (c) current diagnosis of MDD (according to the Mini International Neuropsychiatric Interview [MINI; Sheehan 2014]), and (d) failed first-line ART (identified by the community clinic as not having collected ART for > 3 months).

We did not exclude participants with medical and psychiatric comorbidities and/or other factors that could influence cognitive performance because we wanted the sample population to be representative of the clinical population of interest (i.e., PWH with MDD and incomplete ART adherence). The only exclusion criteria were (a) active and untreated major mental illness (i.e., psychosis or mania) that would interfere with participation, (b) inability or unwillingness to provide informed consent, and (c) lack of sufficient fluency in English or isiXhosa. Participants using antidepressants were eligible even if they met criteria for a current depressive episode; however, they had to have been on a stable antidepressant regimen and dose for at least 2 months.

All participants provided written informed consent. The study protocol was approved by the University of Cape Town (UCT) Faculty of Health Sciences Human Research Ethics Committee and the University of Miami Institutional Review Board.

Materials

Participants completed a neuropsychological test battery and measures of sociodemographic, HIV disease, and psychiatric characteristics. We also collected information not routinely gathered in HIV studies: psychosocial data, including those related to socioeconomic status (e.g., income, food insecurity), and medical history (neurological and cerebrovascular risk factors).

Neuropsychological assessment

The neuropsychological battery comprised 12 standardized tests, each of which assessed performance in one of the seven cognitive domains commonly affected by HIV (Grant 2008). This battery of tests has demonstrated adequate psychometric properties in this setting (Joska et al. 2011; Nyamayaro et al. 2020).

The domains, tests, and outcome variables were (1) executive functioning, as measured by the Color Trails Test 2 (CTT2), with the specific outcome variable being completion time; Wisconsin Card Sorting Test (WCST)—perseverative errors; (2) verbal learning and memory, Hopkins Verbal Learning Test-Revised (HVLT-R)—total across the three immediate recall trials, total on the delayed recall trial; (3) visuospatial learning and memory, Brief Visuospatial Memory Test-Revised (BVMT-R)—total across the immediate recall trials, total on the delayed recall trial; (4) verbal fluency, category fluency test—total number of animals/total number of fruits and vegetables named in 1 min; (5) attention/working memory, Wechsler Memory Scale-Third Edition (WMS-III) Spatial Span subtest—total raw score; Wechsler Adult Intelligence Scale-Third Edition (WAIS-III) Digit Span subtest—total raw score; (6) processing speed, CTT1–completion time; WAIS-III Digit Symbol Coding subtest—total raw score; WAIS-III Symbol Search—total raw score; (7) motor skills, Grooved Pegboard Test (GPT) dominant (DH) and nondominant hand (NDH)—completion time; Finger Tapping Test DH and NDH—completion time.

Tests were administered in either English or isiXhosa, depending on the participant’s preference, by a bilingual neuropsychology technician.

Measures of sociodemographic variables

Participants self-reported basic sociodemographic information (i.e., gender, age, highest level of education, monthly household income, primary language, and employment status) as well as details of their school performance (e.g., whether they had ever been held back or repeated a year in school, whether they were fully literate).

Measures of HIV disease variables

HIV viral load and current CD4 cell number were extracted from the medical records. If participants did not have recent (1 month) testing, we collected blood samples (see Joska et al. 2020 for additional detail). ART regimens (i.e., reinitiated on first line, second line, or third line) were also extracted from the participant’s medical record. Participants also self-reported whether their nadir CD4 count had ever been below 100 cells/μL.

Measures of psychiatric variables

Psychiatric disorders were diagnosed using the MINI structured diagnostic interview (Sheehan 2014). This interview was conducted by a psychiatric nurse and supervised by a clinical psychologist. The Alcohol Use Disorders Identification Test (AUDIT; cut off > 20; Saunders et al. 1993) was used to indicate high-risk alcohol use, and the Hamilton Rating Scale for Depression (HAM-D; Hamilton 1960; Williams et al. 2008) was used to assess depression severity. HAM-D scores between 0 and 7 indicate no depression; 8–16, mild depression; 17–23, moderate depression; ≥ 24, severe depression (Zimmerman et al. 2013).

Measures of psychosocial and socioeconomic variables

The Quality of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q; Endicott et al. 1993) assessed satisfaction in daily life. A modified version of the Adult AIDS Clinical Trials Group [ACTG] SF-21 (Wu et al. 1997) assessed health-related quality of life (i.e., related to physical, social, and cognitive functioning and emotional well-being). The Household Food Insecurity Access Scale (HFIAS; Coates et al. 2007) measured household food insecurity. Each of these was completed by the participant with the help of the research team.

Measures of medical history

Participants were classified as having a significant history of neurological problems if they reported ever having experienced one or more of the following neurological events: a closed or open head injury with loss of consciousness > 30 min; a stroke; a coma; epilepsy; a seizure without a diagnosis of epilepsy; and bacterial meningitis.

Participants were classified as having a significant history of vascular risk factors if they reported two or more of the following vascular risk factors: any heart problem (such as coronary artery disease, heart arrhythmia or other heart diseases); heart attack; diagnosis of hypertension (irrespective of whether they were on medication or not); diagnosis of diabetes; and history of having smoked cigarettes.

Procedure

Individuals who remained eligible for study participation after the screening procedures were scheduled for the set of baseline visits. The baseline neuropsychological assessment, along with the measures of medical history, happened on a separate visit to the rest of the measures (these two visits were separated by approximately 2 weeks). After the baseline visits, participants randomized to the CBT-AD condition received the intervention described by Joska et al. (2020) and Safren et al. (2021). Briefly, the intervention was organized across five modules, with one each covering psychoeducation about depression, motivational interviewing, problem solving, behavioral activation, and medication adherence (Safren et al. 1999). It was delivered over eight sessions by trained nurses, using a task-sharing approach.

Participants in the comparison group received standard of care offered at the community clinic, along with a letter of referral to the medical officer listing the psychiatric and cognitive disorders for which they had met diagnostic criteria.

All participants were scheduled for a follow-up visit 8 months post-baseline. At that visit, they were readministered the neuropsychological test battery and the HAM-D.

Statistical analysis

We used R version 4.1.2 (2021–11–01) and RStudio Version 2021.09.0 to complete all analyses, with the threshold for statistical significance set at α = 0.05.

First, we processed and standardized the neuropsychological data. Normative standards for the neuropsychological tests were based on control data collected by two previous studies in the UCT HIV Mental Health Research Unit (Gouse & Robbins, personal communication, June 2018; Westgarth-Taylor & Joska, personal communication, November 2017). To assure similarity across key demographic (age, ethnicity, language, education), psychosocial, and socioeconomic characteristics, these data were collected between 2008 and 2016 from healthy community-dwelling individuals (N = 233) who presented at the same community clinics in Khayelitsha from which the current sample was recruited. In the studies that collected the control data, participant inclusion criteria were (1) HIV seronegative, (2) ≥ 18 years of age, and (3) at least 5 years of formal education. Exclusion criteria were (1) major psychiatric conditions, (2) neurological disease that could affect brain integrity, (3) lifetime history of head injury resulting in loss of consciousness > 30 min, and (4) current substance use disorder.

We used the control data to calculate demographically corrected z-scores, using standard regression-based norming processes. The z-scores were then converted to demographically corrected T-scores (M = 50, SD = 10). If participants had z-scores greater than 5 SD below the mean, the conversion to a T-score resulted in negative T-score. In these cases, we assigned a score of zero, the lowest possible T-score to maintain the clinical significance of the low performance. Neuropsychological data were summarized into domain and global T-scores by taking the average of T-scores within each domain and then across domain T-scores.

Second, we calculated sample descriptive statistics (values taken at the baseline assessment). T-tests (or welch two sample t tests when groups had unequal variance) and chi-square analyses were used to investigate differences in baseline characteristics between the CBT-AD and standard of care groups.

Third, we used univariate linear regressions to investigate the association, at baseline, between each of the sociodemographic, medical, and psychiatric variables and overall cognitive functioning (as estimated by the global T-score). Initially, strength of associations were determined using Pearson and point-biserial correlation coefficients. Subsequently, variables significantly associated (p < 0.05) with global T-scores were entered into multivariate linear regression models to determine which best explained cognitive test performance. A backwards stepwise approach was used for model building, where the variables with the smallest t value were sequentially removed from the model. Each time a variable was removed, the new model was compared to the previous model using a chi-square test to ensure that that removal had no significant effect on the model. For each model, Cook’s D investigated influential outliers. We examined any point with a Cook’s D over 4/n, where n is the number of observations. Outliers were deleted from the analyses if the model was improved without them.

Fourth, to investigate the association, at baseline, between depression severity (as measured by HAM-D score) and performance within each cognitive domain (as measured by the domain T-score), we used univariate linear regression modeling and Pearson and point-biserial correlations.

Fifth, to determine whether the CBT-AD treatment improved depression severity and cognitive test performance and whether improvements in depression severity were associated with improvements in cognitive test performance over time, we used linear mixed-effects modeling fit by maximum likelihood. The first model had HAM-D scores as an outcome to determine if, from baseline to follow-up, depression scores had improved significantly more in the group assigned to receive CBT-AD. We then built models with the global domain T-score and each cognitive domain T-score as a separate outcome variables to investigate whether, from baseline to follow-up, (1) cognitive test performance of participants assigned to the CBT-AD condition had improved significantly more than that of those assigned to the standard of care condition, and (2) improvement in HAM-D scores in both groups was associated with improvement in cognitive test performance at follow-up.

Results

Table 1 presents the sample’s demographic and clinical characteristics. Most participants were women, a statistic representative of the South African PWH population (George et al. 2019). isiXhosa was the primary language of almost all participants (93%). The sample’s median monthly household income was USD100, 89% were unemployed, and 68% had experienced severe food insecurity.

Table 1.

Sample demographic and clinical variables at baseline: descriptive statistics (N = 105)

Variable	M (SD)	f (%)
Sociodemographic
Sex (female)		76 (72.38%)
Age (yrs)	39.79 (9.14)
Education (yrs completed)	9.30 (2.45)
Monthly household income (ZAR)	1600 (0–2900)a
HFIASb	12.74 (6.90)
ACTG SF-21	46.65 (16.18)
Q-LES-Q	41.83 (12.85)
Held back in school		72 (68.57%)
Literacyc		91 (86.67%)
Medical
Log10 HIV viral load	3.56 (1.44)
Current absolute CD4	248.49 (209.38)
History of neurological events		37 (35.24%)
Vascular risk		16 (15.24%)
ART regimens
Reinitiated		56 (53.85%)
Second		47 (45.19%)
Third		1 (0.96%)
HIV RNA viral suppressiond		25 (23.81%)
Self-reported nadir CD4 count < 100 cells/ml		68 (64.76%)
Psychiatric
HAM-D	25.63 (7.10)
High-risk alcohol usee		30 (28.85%)
Past diagnosis of MDD		96 (93.20%)
History of recurrent MDD		22 (21.15%)

ZAR South African rands, HFIAS Household Food Insecurity Access Scale, ACTG SF-21 Adult AIDS Clinical Trials Group SF-21, Q-LES-Q Quality of Life Enjoyment and Satisfaction Questionnaire, ART antiretroviral therapy, HAM-D Hamilton Rating Scale for Depression, MDD major depressive disorder

^aMedian (interquartile range)

^bHigher score indicates greater food insecurity

^cPercentage of participants who could read and write

^dHIV RNA viral load < 400 copies/mL

^eIndicated if AUDIT score > 20

Regarding educational characteristics, most participants (85%) had not completed high school (in South Africa, this is 12 years capped by a major exit examination); 13% of the sample self-reported that they could not read nor write; and 69% had been held back at least 1 year due to poor academic performance.

Regarding HIV disease variables, two-thirds of the sample reported a nadir CD4 count of below 100 cells/μL. Just over half of participants had been reinitiated on first-line treatment ART; 45% had been prescribed second-line ART.

Regarding other medical history, 35% of participants had a history of significant neurological events and 15% met the criteria for having significant cardiovascular disease risk factors.

Regarding psychiatric history, all participants had a primary diagnosis of current MDD, 93% of the participants met criteria for a diagnosis of MDD once in the past and 21% had a history of recurrent MDD. The average HAM-D score fell within the “severe depression” range (M = 25.63, SD = 7.10) at baseline (Zimmerman et al. 2013), with 61% of participants endorsing current suicidal ideation and/or a lifetime history of making a suicide attempt.

Regarding other psychiatric disorders (as per MINI diagnosis), 41% of participants met criteria for current alcohol use disorder, with the mean AUDIT scores indicating harmful alcohol consumption (Saunders et al. 1993). Finally, 8% of the participants met criteria for panic disorder, 6% for agoraphobia, 4% for post-traumatic stress disorder, 2% for social anxiety disorder, and 2% for substance use disorder. No participants met criteria for bipolar mood disorder, obsessive–compulsive disorder, or generalized anxiety disorder.

Regarding differences in baseline demographic and clinical characteristics between the CBT-AD and standard of care group, there were no significant group differences in HAM-D scores (p = 0.197) and other baseline characteristics (ps > 0.056), except for Q-LES-Q scores (p = 0.013) which were significantly higher in the CBT-AD group (M = 46.37, SD = 11.22) compared to the standard of care group (M = 39.71, SD = 13.09). There were also no significant group differences in baseline global or cognitive domain T-scores (ps > 0.161). Viral non-suppression was an entry criterion for the CBD-AD group, whereas 34.72% of the standard of care group were virally suppressed at baseline.

Associations of sociodemographic, medical, and psychiatric variables with cognitive performance at baseline

The results of the univariate associational analyses (measured using linear regression; see Table 2) showed that older age, fewer years of education, and greater food insecurity were significantly associated with lower global T-scores at baseline. There were no significant associations between global T-scores and any of the medical or psychiatric variables, including HIV disease variables and depression severity.

Table 2.

Univariate associations of sociodemographic, medical, and psychiatric variables with global T-score at baseline (N = 105)

Variable	Estimate	95% CI	p	ESE
Sociodemographic
Age (yrs)a	−0.20	−0.33 to −0.07	.002**	−0.29
Sex (female)b	−0.23	−2.99 – 2.52	.867	−0.02
Education (yrs completed)a	0.64	0.15 – 1.13	.011*	0.25
Monthly household income (ZAR) a, d	< −.001	< −.001 – < .001	.699	−0.11
HFIASa, c	−0.19	−0.36 to − 0.01	.040*	−0.20
ACTG SF-21a	−0.02	−0.10 – 0.06	.577	−0.06
Q-LES-Qa	−0.06	−0.15 – 0.04	.242	−0.12
Held back in schoolb	0.09	−2.56 – 2.75	.944	0.01
Literacyb	3.06	−0.52 – 6.64	.093	0.16
Medical
Log10 HIV viral loada	−0.17	−1.03 – 0.70	.703	−0.04
Current absolute CD4a	< −.001	−0.01 – 0.01	.872	−0.02
History of neurological events b	−2.28	−4.82 – 0.26	.079	−0.17
Vascular riskb	−1.01	−4.44 – 2.41	.559	−0.06
ART regimensb				−0.06
Second vs. reinitiated	−0.25	−2.75 – 2.25	.845
Third vs. reinitiated	−8.43	−21.19 – 4.33	.193
Self-reported nadir CD4 count b	−0.76	−3.34 – 1.81	.557	0.06
Psychiatric
HAM-Da	< .001	−0.18 – 0.17	.994	< −.001
High-risk alcohol useb, d, e	2.33	−0.36 – 5.03	.089	0.19
Past diagnosis of MDDb	−0.04	−5.03 – 4.95	.988	0.14
History of recurrent MDDb	0.76	−2.29 – 3.80	.622	0.05

95% CI 95% confidence interval, ESE effect size estimate, ZAR South African rands, HFIAS Household Food Insecurity Access Scale, ACTG SF-21 Adult AIDS Clinical Trials Group SF-21, Q-LES-Q Quality of Life Enjoyment and Satisfaction Questionnaire, ART antiretroviral therapy, HAM-D Hamilton Rating Scale for Depression, MDD major depressive disorder

^*p < .05;

^**p < .01;

^***p < .001

^aContinuous variable, therefore ESE calculated using Pearson’s correlation coefficient

^bCategorical variable, therefore ESE calculated using the point-biserial correlation coefficient

^cHigher score indicates greater food insecurity

^dHigh-risk alcohol use indicated if AUDIT score > 20

^eOne participant removed from analysis because data value was an influential outlier (Cook’s D = 0.06)

When building the multivariate linear regression models to determine the best model of baseline global cognitive performance in this sample, we entered the three significant variables individually associated with global T-score: age, education (years completed), and HFIAS score.

The final model (adjusted R² = 0.10, p = 0.002) included education and HFIAS score. For every year of education completed, global T-score increased by 0.62 points (CI: 0.08, 1.15; p = 0.025; r = 0.25), on average. For every one unit increase in HFIAS score, global T-score decreased by 0.20 points (CI: −0.38, −0.03; p = 0.020; r = −0.20), on average.

The model was not significantly different from the fully specified model (adjusted R² = 0.12, p = 0.083), indicating a good model fit. Data from two participants were removed from the analyses because they contained influential outliers (Cooks’ D = 0.23 and 0.21).

Association between depression severity and cognitive performance at baseline

At baseline, depression severity was significantly associated with performance in the attention/working memory domain, with increasing depression severity related to worse attention/working memory. On average, for every one point increase in HAM-D score, the T-score in the attention/working memory domain decreased by 0.25 points (CI: −0.45, −0.05; p = 0.016; r = −0.24).

HAM-D score was not significantly associated with performance in any other cognitive domain: Motor (estimate = 0.08; CI: −0.23, 0.39; p = 0.615; r = 0.05); information processing speed (estimate = −0.01; CI: −0.28, 0.27; p = 0.967; r = −0.004); verbal fluency (estimate = 0.06; CI: −0.15, 0.28; p = 0.572; r = 0.06); auditory-verbal learning and memory (estimate = −0.03; CI: −0.30, 0.24; p = 0.810; r = −0.02); visuospatial learning and memory (estimate = −0.04; CI: −0.35, 0.26; p = 0.780; r = −0.03); executive function (estimate = 0.19; CI: −0.09, 0.46; p = 0.177; r = 0.13).

Effects of CBT‑AD treatment on depression severity and cognitive performance over time

The first model investigated if, from baseline to follow-up, HAM-D scores had improved significantly more in the group assigned to receive CBT-AD than in the standard of care group. Analyses detected a significant group × timepoint interaction (estimate = −5.35; CI: −9.75, −0.96; p = 0.017), indicating that those in the CBT-AD group improved by an estimated 5.35 points more than those in standard of care group. Analyses also detected a significant main effect for timepoint (estimate = −9.67; CI: −12.24, −7.10; p < 0.001), indicating that, in overall sample (i.e., regardless of group assignment) and on average, HAM-D scores improved by an estimated 9.67 points from baseline (M = 25.63, SD = 7.10) to follow-up (M = 13.96, SD = 9.28). Analyses detected no significant main effect for group (estimate = −1.93; CI: −5.13, 1.27; p = 0.235).

The next set of models investigated whether, from baseline to follow-up, global and cognitive domain T-scores had improved significantly more in the group assigned to receive CBT-AD than in the standard of care group. This hypothesis was not confirmed, with analyses detecting no significant group × timepoint interaction (see Table 3).

Table 3.

Effects of CBT-AD treatment on cognitive performance over time: linear mixed effects regression model (N = 105)

	Estimate	95% CI	p
Outcome = Motor domain T-score
Predictors
Group (CBT-AD versus standard of care)	−2.15	−6.54 – 2.25	.337
Timepoint (baseline versus follow-up)	3.67	1.97 – 5.36	< .001*
Group × Timepoint interaction	−0.94	−3.79 – 1.91	.516
Outcome = Information Processing Speed domain T-score
Predictors
Group (CBT-AD versus standard of care)	−1.07	−5.16 – 3.01	.605
Timepoint (baseline versus follow-up)	3.75	1.62 – 5.87	.001*
Group × Timepoint interaction	1.18	−2.40 – 4.76	.516
Outcome =Attention/Working Memory domain T-score
Predictors
Group (CBT-AD versus standard of care)	0.30	−2.81 – 3.40	.850
Timepoint (baseline versus follow-up)	1.79	0.28 – 3.29	.020*
Group × Timepoint interaction	1.81	−0.72 – 4.35	.160
Outcome = Verbal Fluency domain T-score
Predictors
Group (CBT-AD versus standard of care)	−2.32	−5.59 – 0.94	.161
Timepoint (baseline versus follow-up)	1.71	−0.003 – 3.42	.049*
Group × Timepoint interaction	2.18	− 0.69 – 5.06	.136
Outcome = Auditory-Verbal Learning and Memory domain T-score
Predictors
Group (CBT-AD versus standard of care)	−1.94	−6.17 – 2.29	.368
Timepoint (baseline versus follow-up)	5.01	2.98 – 7.04	< .001*
Group ×Timepoint interaction	2.31	−1.10 – 5.72	.183
Outcome = Visuospatial Learning and Memory domain T-score
Predictors
Group (CBT-AD versus standard of care)	−1.02	−5.89 – 3.85	.679
Timepoint (baseline versus follow-up)	3.48	1.43 – 5.53	.001*
Group × Timepoint interaction	−0.14	−3.59 – 3.30	.934
Outcome = Executive Function domain T-score
Predictors
Group (CBT-AD versus standard of care)	−0.68	−4.76 – 3.41	.744
Timepoint (baseline versus follow-up)	3.13	0.66 – 5.60	.013*
Group × Timepoint interaction	−0.64	−4.82 – 3.54	.763
Outcome = Global T-score
Predictors
Group (CBT-AD versus standard of care)	−1.27	−3.87 – 1.34	.338
Timepoint (baseline versus follow-up)	3.37	2.58 – 4.16	< .001*
Group × Timepoint interaction	0.79	−0.54 – 2.12	.241

HAM-D Hamilton Rating Scale for Depression, CBT-AD cognitive-behavioral therapy for adherence and depression

^*p < .05;

^**p < .01;

^***p < .001

Finally, we investigated whether baseline-to-follow-up improvement in HAM-D scores was associated with baseline-to-follow-up improvement in global and cognitive domain T-scores. In these models, we adjusted for baseline-to-follow-up changes in viral suppression (viral suppression × timepoint interaction).

Analyses detected no significant main effect for HAM-D score (−0.09 < estimates < 0.11; CIs: −0.30, 0.34; ps > 0.290) or significant HAM-D × timepoint interaction (−0.14 < estimates < 0.13; CIs: −0.37, 0.38; ps > 0.179), except in the attention/working memory domain.

Regarding the attention/working memory domain, the analysis detected a significant main effect for HAM-D score (estimate = −0.24; CI: −0.39, −0.10; p = 0.001) and a significant HAM-D × timepoint interaction (estimate = 0.23; CI: 0.05, 0.41; p = 0.013). Perusal of the data presented in Fig. 2 suggests that this significant result may be accounted for by the fact that, at baseline, depression severity was associated with worse cognitive performance; however, at follow-up (i.e., when depression had improved), this association was much weaker and no longer statistically significant.

Fig. 2.

Association between depression severity and attention/working memory domain performance at both time points (N = 105)

In this final set of models, the main effect for timepoint (baseline versus follow-up) for all global and cognitive domain T-scores was not significant except in the cases of global T-score (estimate = 3.07; CI: 0.76, 5.38; p = 0.009) and verbal fluency domain T-score (estimate = 5.10; CI: 0.33, 9.86; p = 0.036); for the other outcome variables, −2.68 < estimates < 5.45; CI: −6.79, 11.34; ps > 0.063.

Note that the main effect for viral suppression for all global and cognitive domain T-scores was not significant (−0.10 < estimates < 1.8; CIs: −3.99, 5.15; ps > 0.120), neither was the viral suppression × timepoint interaction (−3.26 < estimates < 1.36; CIs: −6.17, 6.19; ps > 0.079).

Discussion

In this study, a sample of 105 incompletely ART adherent South African PWH with comorbid major depressive disorder, all of whom were from socioeconomically disadvantaged backgrounds, completed a baseline sociodemographic, psychiatric, and neuropsychological assessment. Some (n = 33) were then randomly assigned to a CBT-AD condition while the rest (n = 72) were assigned to standard-of-care treatment. Eight months post-baseline, 81 (n_CBT-AD = 29) repeated the assessment.

Results indicated that, at baseline, participants with less education and more food insecurity delivered overall poorer performance on the cognitive test battery. At that timepoint, HIV disease factors and medical/psychiatric comorbidities were not significant predictors of global cognitive functioning; however, depression severity was significantly associated with poorer performance in the domain of attention/working memory. This significant effect in that cognitive domain was not present at follow-up, when the cohort overall was less depressed. Regarding this improvement in depression, it should be noted that depression improved significantly more in the group assigned to receive CBT-AD than those in the comparison group.

HIV disease factors might be expected to predict cognitive performance in a sample with poorly controlled viral replication: many previous studies have reported such associations (Ellis et al. 2011; Heaton et al. 2011; Jumare et al. 2018; Robertson et al. 2007; Sacktor et al. 2002; Starace et al. 2002). However, recent trends in the literature suggest that the effects of HIV on cognitive performance may be weaker than the effects of socioeconomic and psychosocial variables, especially in samples drawn from low- and middle-income countries (LMICs) or from socioeconomically disadvantaged populations (Cysique and Becker 2015; Do et al. 2018; Dreyer et al. 2021b; Haddow et al. 2018; Maki et al. 2015; Vo et al. 2013; Winston et al. 2013).

One of these socioeconomic/psychosocial variables is level of educational attainment. The current data demonstrating its effects on cognitive performance are consistent with a longstanding and strong line of research indicating that individuals with lower levels of education perform more poorly on standardized cognitive tests than those with higher levels (Lenehan et al. 2015; Strauss et al. 2006).

However, data regarding the effects of the other significant socioeconomic/psychosocial variable in our analyses, food insecurity, are not as prevalent. This is despite the fact that food insecurity in LMICs is a significant problem (Crush et al. 2012). For instance, in South Africa 68–89% of people living in peri-urban communities outside Cape Town are estimated to have moderate-to-severe food insecurity (Battersby 2011; Lee et al. 2021; Misselhorn and Hendriks 2017). Beyond the obvious physical health implications, food insecurity has been associated with cognitive impairment (Gao et al. 2009; Koyanagi et al. 2019). It is unclear, however, whether the mechanism underlying this association is related to purely nutritional pathways or whether food insecurity is a proxy for broader structural inequalities that can affect cognitive performance via a host of complex social, educational, socioeconomic, and health pathways (Burch et al. 2016; Farah 2017; Hobkirk et al. 2017; Nightingale et al. 2021; Watson et al. 2019; Weiser et al. 2009, 2012, 2011).

Despite reportedly high rates of food insecurity in LMIC-resident PWH (Anema et al. 2009), there are no studies that have investigated relations between food insecurity and cognitive performance among PWH. The current study showed that greater food insecurity is associated with weaker global cognitive performance in LMIC-resident PWH. Previous studies investigating this relationship in US-based samples of PWH found that food insecurity was associated with weaker global cognitive performance, more specifically, poorer performance in the domains of motor function, information processing speed, and learning and memory (Hobkirk et al. 2017; Tamargo et al. 2021).

Regarding associations between cognitive performance and depression, our analyses suggested that higher HAM-D scores (indicating greater depression severity) were associated with weaker performance in the domain of attention/working memory at baseline, when the sample’s mean HAM-D score fell in the severely depressed range (Zimmerman et al. 2013). However, at follow-up, when the mean HAM-D score fell within the mild range, the association no longer met the threshold for statistical significance. This finding is consistent with several previous studies suggesting that more severe depression is associated with greater cognitive impairment (McDermott and Ebmeier 2009; Shimizu et al. 2011), and, perhaps more notably, with Cysique et al. (2016), who found that global cognitive functioning was only impaired in PWH with chronic and clinically unstable depression. With specific regard to attention/working memory and depression in PWH, the literature is equivocal: Some studies report that comorbid depression impairs performance in this domain (Bragança and Palha 2011; Fellows et al. 2013; Gibbie et al. 2006; Goggin et al. 1997; Harrison et al. 2017; Rubin et al. 2014), whereas others report no such relationship (Akolo et al. 2014; Bryant et al. 2015; Cysique et al. 2007; Haddow et al. 2018).

In our sample of PWH, depression severity was not associated with global cognitive performance or with performance in cognitive domains (e.g., information processing speed, verbal memory) that one might expect, based on previous studies, to be affected (see, e.g., Bryant et al. 2015; Fellows et al. 2013). The review by Rubin and Maki (2019) of 41 studies (most of them cross-sectional) examining associations between depression and cognition in PWH reported that the most reliably affected cognitive domains were motor function, information processing speed, learning and memory, and executive function. Only 3 of the 8, cross-sectional studies that measured attention/working memory, found a relationship between depression and attention/working memory.

We posit that reasons for the discrepancy between our depression-cognition results and those of previous studies might arise from differing sample characteristics. Our participants were recruited from socioeconomically disadvantaged neighborhoods, were all carrying a formal (via structured clinical interview) diagnosis of MDD on enrolment, with an average HAM-D score in the range conventionally described as “severely depressed” (Zimmerman et al. 2013). Moreover, most of our participants (72%) were female. In contrast, most HIV-cognition research is conducted in high-income settings with samples that are frequently all- or majority-male, and with depression measured only using a cutoff score on a symptom questionnaire such as the HAM-D or the Center for Epidemiologic Studies Depression Scale (CES-D; Radloff 1977). In fact, of the 41 studies reviewed by Rubin and Maki (2019), only 5 used structured clinical interviews to diagnose current MDD. Symptom questionnaires are screening instruments for overall depressive symptomology. As such, they are not conclusively diagnostic and are therefore not as accurate as clinical interviews in diagnosing MDD (Subica et al. 2014). Moreover, the relationship between depression and cognition may be different in PWH with a confirmed current diagnosis of MDD, compared to those experiencing depressive symptoms (Cysique et al. 2016). Sex differences (Dreyer et al. 2021a; Rubin et al. 2019) and socioeconomic hardship (Watson et al. 2019) could also play a role in the relationship between depression and cognitive performance in PWH.

Few longitudinal studies have investigated the effects of depression severity on cognitive performance. Studies using similar longitudinal designs as ours to investigate associations between depression and cognitive performance over time have produced mixed results (Cysique et al. 2007; Gibbie et al. 2006; Grant et al. 2014; Heaton et al. 2015; Paolillo et al. 2020; Vo et al. 2013). Two large longitudinal studies, the Multicenter AIDS Cohort Study (MACS; Vo et al. 2013) and CHARTER Study (Grant et al. 2014; Heaton et al. 2015), found that a diagnosis of current MDD/higher depressive symptoms were associated with declines in cognitive functioning. Additionally, Paolillo et al. (2020) found that a high cumulative burden of depression over time was associated with a steeper decline in cognitive functioning, compared to a low cumulative burden of depression. Cysique et al. (2007) found that a new current episode of MDD in men with HIV, who did not meet criteria for MDD at baseline, was not associated with changes in cognitive functioning within a 2-year period. Another study reported that, although depression declined and scores in several cognitive domains improved over a 2-year follow-up period, there was no significant association between the two (Gibbie et al. 2006). The same study did, however, find a similar result to ours in observing that depression scores were significantly inversely correlated with working memory performance at baseline.

This is the first study to describe the effects of CBT-AD on cognitive performance. Although the treatment was more effective than standard of care at relieving depressive symptomatology (confirming results from the primary outcome paper; Safren et al. 2021), this improvement did not generalize to cognitive performance. It is possible that the cognitive effects of the CBT-AD intervention could not be detected because the size of the subgroup receiving the CBT-AD treatment was small, meaning that potential effects of the treatment on cognitive performance may have gone undetected. Hence, results related to the effects of the intervention should be interpreted with caution.

The research presented here has several methodological limitations. First, the relatively small sample size means that the study may have been underpowered to detect potentially significant associations between various variables and cognitive performance. In addition, the findings drawn from this real-world clinical sample of PWH with depression and incomplete ART adherence who live in a socioeconomically disadvantaged setting in South Africa may not be generalisable to all samples of PWH with depression and incomplete ART adherence.

Conclusion

In this sample of people with HIV and incomplete ART adherence, recruited from socioeconomically disadvantaged settings in a low- or middle-income country, non-biological factors (educational level and food insecurity) were stronger predictors of global cognitive performance than the biological effects of HIV and other medical factors. This result, which emerged even in a group with poorly controlled HIV (i.e., whose brains would be especially vulnerable to biological effects of the infection), adds to the growing body of evidence that, in PWH, factors other than the disease are important determinants of cognitive performance. It is therefore imperative that future research collect information on psychosocial factors when assessing cognitive performance in PWH—collecting such data will allow studies to reduce the potential for inaccurate interpretation of cognitive test scores and consequent overestimation of the prevalence of cognitive impairment in this population.

We also found that severe depression in this sample of PWH is associated with poorer cognitive performance in the domain of attention and working memory. Because the strength of this association became markedly weaker when the level of depression improved, we suggest that depression in such samples of PWH could be a potentially modifiable risk factor for those presenting with attention and working memory dysfunction. Because attention and working memory are the gateway to information acquisition and serve as a necessary foundation for higher-level cognitive functioning, improvement in cognitive performance in this domain may improve the overall functioning of PWH (Parsons and Rizzo 2008).

Overall, factors associated with cognitive performance in this sample were likely not related to brain injury. It is important for clinicians to assess depression in PWH presenting with cognitive difficulties, and for HIV researchers to collect information on psychosocial factors in their studies of cognitive impairment.