Ethnic/Racial Disparities in Longitudinal Neurocognitive Decline in People with HIV

Abstract

Background:

To examine longitudinal neurocognitive decline among Latino, non-Latino Black, and non-Latino White people with HIV (PWH) and factors that might explain ethnic/racial disparities in neurocognitive decline.

Methods:

499 PWH (13.8% Latino, 42.7% Black, 43.5% White; baseline age: M=43.5) from the CNS HIV Anti-Retroviral Therapy Effects Research (CHARTER) study completed neurocognitive, neuromedical, and laboratory assessments every 6–12 months with up to five years of follow-up. Longitudinal neurocognitive change was determined via published regression-based norms. Survival analyses investigated the relationship between ethnicity/race and neurocognitive change, and baseline and time-dependent variables that might explain ethnic/racial disparities in neurocognitive decline, including socio-demographic, HIV-disease, medical, psychiatric, and substance use characteristics.

Results:

In Cox proportional hazard models, hazard ratios for neurocognitive decline were increased for Latino compared to White PWH (HR=2.25, 95% CI=1.35–3.73, p=0.002), and Latino compared to Black PWH (HR=1.86, 95% CI=1.14–3.04, p=0.013), with no significant differences between Black and White PWH (p=0.40). Comorbidities, including cardiometabolic factors and more severe neurocognitive comorbidity classification, accounted for 33.6% of the excess hazard for Latino compared to White PWH, decreasing the hazard ratio associated with Latino ethnicity (HR=1.83, 95% CI=1.06–3.16, p=0.03), but did not fully account for elevated risk of decline.

Conclusion:

Latino PWH may be at higher risk of early neurocognitive decline compared to Black and White PWH. Comorbidities accounted for some, but not all, of this increased risk among Latino PWH. Future research examining institutional, sociocultural, and biomedical factors, including structural discrimination and age-related biomarkers, may further explain the observed disparities.

Introduction

With the advent of antiretroviral therapy (ART), increasing numbers of people with HIV (PWH) with reliable access to ART are reaching older age¹. Despite treatment advances, HIV continues to be a major public health concern, especially among Latino/a/x/Hispanic (hereafter referred to as Latino) and non-Latino Black/African American (hereafter referred to as Black) communities in the United States. Latino and Black people have three and eight times the rate of HIV infection compared to non-Latino Whites (hereafter referred to as White)². Latino and Black PWH also face systematic disadvantage at multiple points in the HIV disease treatment cascade compared to White PWH³, and show higher rates of morbidity and mortality^4–6.

HIV-associated neurocognitive impairment (NCI) occurs in 20–50% of PWH⁷, is more prevalent among older PWH⁸, and is associated with compromised everyday functioning and early mortality^9,10. Cross-sectional studies show elevated risk for NCI among some Latino^11–14 and Black PWH¹⁵ compared to White PWH, and one longitudinal study of PWH identified Latino ethnicity as a predictor of neurocognitive decline¹⁶. Yet, the underlying reasons for these disparities remain unknown. Latino and Black PWH experience unequal access to HIV care, and tend to present with worse HIV disease burden and more medical comorbidities^3,4,6, which may increase risk for neurocognitive decline. In a recent cross-sectional study, worse historical HIV disease burden (i.e., nadir CD4) partially explained neurocognitive disparities between Latino and White PWH¹⁴, but much of the difference remained unexplained. Yet, no studies to date have investigated ethnic/racial disparities in longitudinal neurocognitive decline among Latino and Black PWH. We aimed to examine: (1) longitudinal neurocognitive decline among Latino, Black, and White PWH; and (2) factors that might explain any observed disparities. We hypothesized that Latino and Black PWH may be at higher risk for neurocognitive decline than White PWH, and increased HIV disease burden and comorbidities would partially explain any ethnic/racial disparities observed.

Methods

Participants were enrolled in the CNS HIV Anti-Retroviral Therapy Effects Research (CHARTER) study, a NIH-funded longitudinal study of PWH at six academic medical centers across the U.S. located in San Diego, California; St. Louis, Missouri; New York, New York; Galveston, Texas; Baltimore, Maryland; and Seattle, Washington. Study visits took place every 6–12 months between 2002 and 2015, and included comprehensive, standardized neurocognitive, neuromedical, and psychiatric assessments, a blood draw, and lumbar puncture, which have been described elsewhere¹⁷. Participants were fluent in English, as ascertained by self-report and determined by study staff. Inclusion criteria for the current analyses were: (1) self-identification as Latino/a or Hispanic, non-Latino Black or African American, or non-Latino White; and (2) having neurocognitive data available over 3–7 study visits. The procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1975, as revised in 2000.

Neurocognitive assessment and change over time.

All participants completed a comprehensive neurocognitive test battery in English assessing seven domains: verbal fluency, attention/working memory, processing speed, learning, memory, executive function, and complex motor skills, which have been described previously¹⁸. To examine neurocognitive change over time, z scores were generated for each domain based on published normative data, and changes were averaged to create a summary regression-based change score (sRCS)¹⁹. These scores control for the effect of repeated testing (normal test–retest variability, practice effects, and statistical artifacts) and baseline neurocognitive function on test performance in the normative sample, and included demographic adjustments for age, education, sex, and ethnicity/race (non-Hispanic White compared to other ethnic/racial groups)¹⁹.

Consistent with previous analyses¹⁶, neurocognitive change was ascertained for each participant at each follow-up visit relative to baseline performance based on the distribution of the normative sample, with the top 5% of the sRCS classified as “improve,” the middle 90% as “stable,” and the bottom 5% as “decline.” Overall change status for each participant over the study period was determined as: “improve” (if a participant had at least one “improve” status and no “decline” status); “decline” (if a participant had at least one “decline” status and no “improve” status), and (c) “stable” (if a participant had “stable” status across all visits). Thus, participants classified as experiencing “decline” showed a notable neurocognitive decline at some point over follow-up and never showed improvement. Three participants out of an original sample of 502 had both “improve” and “decline” across study follow-up and were excluded.

Socio-demographics.

Demographic information was obtained via self-report and followed NIH guidelines, which note: Latino or Hispanic refers to a person of Mexican, Puerto Rican, Cuban, South or Central American, or other Spanish-speaking culture of origin/descent, regardless of race²⁰. Latino or Hispanic participants were asked to indicate their country of origin/descent. A distinct variable, country of birth, was assessed by self-report, and dichotomized by those born in and outside of the 50 U.S. states. If a participant identified as both Latino and Black (e.g., Afro-Latino), or Latino and White, they were classified as Latino for current analyses.

Socioeconomic status (SES) was characterized by the Two Factor Hollingshead Index, a weighted average of years of education and current or longest held occupation ²¹. Hollingshead Index scores range from 11 to 69 with lower scores indicating higher SES. Reading level, considered a proxy for educational quality among English-speakers in the U.S., was assessed with the Wide Range Achievement Test (WRAT)²².

Neuromedical and laboratory.

Participants completed a structured neuromedical examination, including collection of blood, urine, and cerebrospinal fluid (CSF). HIV infection was determined via ELISA with Western blot confirmation. Routine clinical chemistry panels, complete blood counts, rapid plasma reagin, hepatitis C virus antibody, and CD4 T cells assays were performed. HIV RNA levels were measured in plasma and CSF by reverse transcriptase PCR (Roche Amplicor, v. 1.5, lower limit of quantitation 50 copies/mL). A central nervous system (CNS) Penetration Effectiveness (CPE) score was included as an estimate of the degree of CNS penetration of current ART regimen^23,24. The Veterans Aging Cohort Study (VACS) Index, a composite marker of HIV disease severity including age, traditional HIV biomarkers, and comorbidity biomarkers, was calculated as described previously²⁵. Higher VACS Index scores indicate worse health and the top tertile of scores was classified as “High VACS Index.”

Medical comorbidities assessed included hepatitis C virus (HCV), markers of cardiometabolic health, and neurocognitive comorbidity status. Cardiometabolic markers included body mass index (BMI), systolic and diastolic blood pressure, levels of total cholesterol, high-density lipoprotein (HDL), low-density lipoprotein (LDL), and triglycerides, and self-report of diabetes, hypertension, and hyperlipidemia. Non-fasting levels of plasma total cholesterol, HDL, LDL, and triglycerides were collected, processed, and assayed by standard certified laboratory methods. Other biomarkers examined included: aspartate aminotransferase (AST) (a marker of liver damage)²⁶; serum total protein (composite measure of albumin and globulin protein in blood); and hematocrit (percentage of oxygen-rich red blood cells in blood).

Neurocognitive comorbidity status was ascertained as described previously²⁷, and based on self-reported history of non-HIV developmental, cerebrovascular, psychiatric, substance use, and neuromedical conditions that may confer risk for NCI. These neuromedical and neuropsychiatric comorbidities are common with PWH were categorized as incidental (normal/mild), contributing (moderate), or confounding (severe)¹⁷.

Psychiatric and substance use.

Current (last 30 days) and past histories of major depressive and substance use disorders were assessed using the Composite International Diagnostic Interview (CIDI)²⁸. History of injection drug use was ascertained by self-report, and recent substance use was identified via urine toxicology on the day of testing.

Statistical Analysis

Baseline socio-demographic, HIV disease, medical comorbidity, psychiatric, and substance use characteristics were compared across Latino, Black, and White PWH using ANOVAs for continuous variables, and Chi-Square or Fisher’s exact tests for categorical variables. For variables which showed significant differences at p < 0.05 across ethnic/racial groups, pairwise comparisons were conducted with Tukey’s Honest Significant Difference (HSD) tests for continuous variables and Chi-Square or Fisher’s exact tests for categorical variables.

To examine differences in neurocognitive change status by ethnic/racial group (Aim 1), we used a Chi-Square test comparing Latino, Black, and White PWH by neurocognitive change group (decline, stable, improve). We examined time to neurocognitive decline (versus “no decline”) across Latino, Black, and White PWH with Kaplan-Meier survival curves and log-rank tests. To examine factors that might explain racial/ethnic differences in neurocognitive decline (Aim 2), we ran two Cox proportional hazard models. We examined hazard ratios for time to neurocognitive decline by ethnicity/race (Model 1). In addition to ethnicity/race, we included other relevant baseline and time-dependent predictors; predictors collected at multiple study visits were modelled as time-dependent (Model 2). To select these predictors, we first examined ethnic/racial group differences in all socio-demographic, HIV disease, medical, psychiatric, and substance use factors in Table 1. Predictors that differed by ethnic/racial group at p < 0.20 were entered into a stepwise regression model predicting time to neurocognitive decline, and we used Akaike Information Criterion (AIC) with backward elimination to reduce model predictors and select a final optimized model (Model 2). To quantify the contribution of predictors to ethnic/racial differences in neurocognitive decline in our final models, we calculated reductions in hazard ratios tied to race/ethnicity from Model 1 to Model 2, consistent with prior published studies of racial cognitive health disparities²⁹. We did not consider baseline neurocognitive performance nor study site as predictors, given our cognitive change scores account for baseline performance, and study site was highly confounded with the distribution of ethnic/racial groups in the U.S.

Table 1.

Sample characteristics at baseline (N=499) by ethnic/racial group

Characteristic	Latino	Blacka	Whitea	Group differences	Group Comparison
	(n=69)	(n=213)	(n=217)	(p value)
No. of visits	5.5 (1.7)	5.9 (1.6)	5.6 (1.6)	0.10	–
Years of study follow-up	2.7 (1.2)	2.8 (1.0)	2.8 (1.2)	0.75	–
Socio-demographic
Age (years)	42.4 (7.5)	43.8 (7.7)	43.6 (8.9)	0.45
Sex (% men)	79.7%	68.1%	90.3%	<0.001	W>L, B
Education (years)	12.6 (2.4)	12.5 (2.5)	13.7 (2.6)	<0.001	W>L, B
Hollingshead (SES)d	42.1 (10.7)	44.8 (11.0)	36.9 (11.8)	<0.001	B,L>W
Reading level (WRAT)	93.4 (12.7)	85.5 (14.9)	102.1 (11.3)	<0.001	W>L>B
HIV Disease
AIDS diagnosis	69.6%	61.0%	53.9%	0.06	–
Duration of infection (years)	11.0 [5, 16]	10.7 [5, 14]	10.1 [3, 15]	0.50	–
Current CD4b	431 [274, 632]	419 [264, 610]	508 [337, 695]	0.005	W>B
Nadir CD4 b	70.0 [21, 311]	157 [26, 295]	197 [73, 320]	0.03	W>L
On ART	76.8%	69.0%	68.7%	0.42	–
% Detectable Plasma (on ART)	55.8%	42.5%	33.1%	0.01	L>W
% Detectable CSF (on ART)	18.0%	17.3%	8.9%	0.08	–
ART Regimen CPE	8.06 (1.76)	8.27 (2.16)	8.03 (2.00)	0.59	–
Duration of all ART (months)b	63.5 (47.1)	53.4 (48.1)	64.9 (57.8)	0.20	–
High VACS Index	23.5%	36.4%	19.4%	<0.001	B>W
Medical
Hepatitis C virus	14.5%	43.2%	19.8%	<0.001	B>W, L
Body Mass Index (kg/m2)	26.6	26.5	26.0	0.51	–
Systolic Blood Pressure (mm Hg)	124.4	124.0	124.6	0.93	–
Diastolic Blood Pressure (mm Hg)	76.8	79.0	76.7	0.06	–
Total Cholesterol (mg/dL)	183.7	173.3	181.4	0.06	–
HDL (mg/dL)	40.8	51.5	39.8	<0.001	B>L,W
LDL (mg/dL)	99.2	89.7	100.8	0.004	W,L>B
Triglycerides (mg/dL)	228.7	164.1	238.8	<0.001	W,L>B
Diabetes	7.3%	12.7%	7.4%	0.14	–
Hypertension	18.8%	18.8%	15.7%	0.65	–
Hyperlipidemia	7.3%	5.6%	14.3%	0.008	W>B,L
ASTc	36.9 (21.4)	42.1 (30.1)	36.1 (28.9)	0.08	–
Serum total protein	8.0 (0.8)	8.3 (0.8)	7.5 (0.7)	<0.001	B>L>W
Hematocrit	41.1 (4.66)	40.2 (3.9)	42.3 (4.4)	<0.001	W>B
Neurocognitive Comorbidity Status (%)				0.015	B>W
Incidental (normal/mild)	52.2%	52.6%	67.7%
Contributing (moderate)	33.3%	32.4%	23.0%
Confounding (severe)	14.5%	15.0%	9.22%
Psychiatric and Substance Use
Current Major Depressive Disorder	20.3%	10.8%	15.2%	0.12	–
Lifetime Major Depressive Disorder	50.7%	45.5%	59.5%	0.02	W>B
Current Any Substance Use Disorder	7.3%	9.4%	6.0%	0.41	–
Lifetime Alcohol Use Disorder	49.3%	54.0%	56.7%	0.56	–
Lifetime Cannabis Use Disorder	21.7%	31.5%	29.5%	0.29	–
Lifetime Meth Use Disorder	15.9%	1.9%	28.6%	<0.001	W>L>B
Lifetime Cocaine Use Disorder	37.7%	58.7%	25.4%	<0.001	B>L,W
Lifetime Opioid Use Disorder	13.0%	25.4%	8.3%	<.0.001	B>L,W
Injection drug use ever	26.1%	29.1%	27.6%	0.89	–
Urine toxicology (% positive)	11.8%	30.5%	6.0%	<0.001	B>W,L

Abbreviations: B, black; W, white; L, Latino; IQR, interquartile range; SES, socioeconomic status; WRAT, Wide Range Achievement Test; ART, antiretroviral therapy; CSF, cerebrospinal fluid; CPE, central nervous system penetration effectiveness score; AST, Aspartate aminotransferase; VACS, Veterans Aging Cohort Study Index Score; HDL, high-density lipoprotein; LDL, low-density lipoprotein

Notes: Values shown as Mean (SD), Median [IQR], or %

^anon-Latino

^bsquare-root transformed

^clog₁₀ transformation prior to comparison analysis

^dLower scores indicate higher socioeconomic status (SES)

Results

Participants were 499 PWH (69 Latino, 213 Black, 217 White) with an average baseline age of 43.5 (range = 18–68), 79.4% were men, and 63.1% were gay or bisexual. Among Latinos, 43.5% were Mexican, 24.6% Puerto Rican, 7.2% South or Central American, 4.3% Cuban, and 20.3% of “other Spanish culture/origin,” and 85.5% of Latinos were born within the 50 U.S. states.

Sample characteristics by ethnicity/race

Table 1 shows baseline socio-demographic, HIV disease, medical, psychiatric and substance use characteristics by ethnic/racial group. White PWH were more likely to be men, had more years of education, and had higher SES than Latino and Black PWH. Reading level was highest among White, then Latino, followed by Black PWH. Regarding HIV disease characteristics, White PWH had higher current CD4 counts than Black PWH, and higher Nadir CD4 than Latino PWH. Among those on ART, Latino PWH had higher rates of detectable plasma HIV RNA than White PWH. Black PWH had higher VACS Index (indicative of worse health) than White PWH. Regarding medical comorbidities, Black PWH had higher rates of HCV, higher HDL cholesterol, lower LDL cholesterol, lower triglycerides, and higher serum total protein compared to Latino and White PWH. White PWH had higher rates of hyperlipidemia compared to Black and Latino PWH. Black PWH had lower hematocrit and worse neurocognitive comorbidity status (i.e. more serious comorbid non-HIV conditions, which confer increased risk for NCI) than White PWH. Regarding psychiatric and substance use characteristics, White PWH had higher rates of lifetime major depressive disorder than Black PWH, and higher lifetime methamphetamine use disorder than Latino and Black PWH. Black PWH had higher lifetime cocaine and opioid use disorder, and were more likely to have positive drug urine toxicology on the day of testing compared to White and Latino PWH.

Neurocognitive Change by ethnicity/race

Over study follow-up, 107 participants (21.4%) declined, 315 (63.1%) remained stable, and 77 (15.4%) improved in neurocognitive status. Figure 1 shows the proportion of participants by ethnicity/race who declined, remained stable, or improved. The overall model showed significant group differences (χ² = 11.84, df = 4, p = 0.019). In follow-up analyses, Latino PWH were at higher risk of neurocognitive decline (versus remaining stable or improving) compared to White (χ² = 10.24, df = 1, p = 0.001) and Black (χ² = 5.59, df = 1, p = 0.018) PWH, but the latter two groups did not differ (χ² = 0.91, df = 1, p = 0.34). The proportion of participants who improved did not differ by ethnic/racial group (ps > 0.68).

Figure 1.

Neurocognitive change (decline, stable, improve) by ethnicity/race. The overall model showed significant ethnic/racial differences in the proportion of participants who declined, remained stable, or improved neurocognitively (χ² = 11.84, df = 4, p = 0.019). Latino PWH were at higher risk of neurocognitive decline compared to White (χ² = 10.24, df = 1, p = 0.001) and Black (χ² = 5.59, df = 1, p = 0.018) PWH, while Black and White PWH did not differ (χ² = 0.91, df = 1, p = 0.34).

Figure 2 shows the pattern of decline by ethnicity/race over years of follow-up in a Kaplan-Meier survival analysis (p = 0.005). In pairwise log-rank tests, Latino PWH declined faster and had a higher proportion of neurocognitive decliners compared to White (HR = 2.24, 95% CI 1.24–4.05, p = 0.001) and Black (HR = 1.86, 95% CI 1.07–3.25, p = 0.01) PWH. Black and White PWH showed similar trajectories of neurocognitive decline over study follow-up (HR = 1.21, 95% CI 0.78–1.87, p = 0.40).

Figure 2.

Neurocognitive decline over study follow-up by ethnicity/race. In a Kaplan-Meier survival analysis, the pattern of decline over years of follow-up differed significantly by ethnicity/race (p = 0.005). In pairwise log-rank tests, Latino PWH declined faster and had a higher proportion of neurocognitive decliners compared to White (HR = 2.24, 95% CI 1.24–4.05, p = 0.001) and Black (HR = 1.86, 95% CI 1.07–3.25, p = 0.01) PWH. Black and White PWH showed similar trajectories of neurocognitive decline over study follow-up (HR = 1.21, 95% CI 0.78–1.87, p = 0.40).

Examination of factors driving ethnic/racial differences in neurocognitive decline

Multivariable survival analyses examined whether baseline and time-dependent predictors might explain ethnic/racial differences in neurocognitive decline between Latino and White PWH, and Latino and Black PWH (Table 2). In unadjusted analyses (Table 2, Model 1A), hazard ratios for neurocognitive decline were elevated for Latino compared to White PWH (HR = 2.25, 95% CI 1.35–3.73, p = 0.002). All socio-demographic, HIV disease, medical, psychiatric, and substance use factors in Table 1 that differed across racial/ethnic groups at p < .0.20 were included in a stepwise regression model predicting neurocognitive decline among Latino and White PWH. Backwards AIC selected baseline (neurocognitive comorbidity status), and time-dependent (hematocrit, total cholesterol, and triglycerides) factors as optimal predictors, which were added to the final Cox model (Table 2, Model 2A). Lower hematocrit (HR = 1.08, p = 0.006), confounding neurocognitive comorbidity status (HR = 3.16, p = 0.004; HR = 3.38, p < 0.001), and lower total cholesterol (HR=1.01, p=0.04) were associated with earlier time to neurocognitive decline. With inclusion of these predictors, the excess hazard for early neurocognitive decline among Latino PWH decreased by 33.6%, but the hazard ratio associated with Latino ethnicity remained significant (HR = 1.83, 95% CI 1.06–3.16, p = 0.03).

Table 2.

Cox proportional hazard models for time to neurocognitive decline comparing Latino and White people with HIV (Model A) and Latinos and Black people with HIV (Model B)

Model A	Predictor	Reference	Risk	HR (95% CI)	P Value
Model 1A
	Ethnicity/race	White	Latino	2.25 [1.35, 3.73]	0.002
Model 2A
	Ethnicity/race	White	Latino	1.83 [1.06, 3.16]	0.03
	Hematocritc	1 unita	Lower	1.08 [1.02, 1.14]	0.006
	Comorbidity	Contributing	Confounding	3.16 [1.44, 6.93]	0.004
	Comorbidity	Incidental	Confounding	3.38 [1.69, 6.73]	<0.001
	Total Cholesterolc	1 unita	Lower	1.01 [1.00, 1.02]	0.04
	Triglyceridesc	1 unitb	Higher	1.00 [1.00, 1.03]	0.07
Model B	Predictor	Reference	Risk	HR (95% CI)	P Value
Model 1B
	Ethnicity/race	Black	Latino	1.86 [1.14, 3.04]	0.013
Model 2B
	Ethnicity/race	Black	Latino	2.39 [1.34, 4.19]	0.002
	Nadir CD4	1 sqrtb	Higher	1.04 [1.00, 1.07]	0.06
	Hematocritc	1 unita	Lower	1.08 [1.02, 1.14]	0.009
	Diabetesc	No	Yes	1.69 [0.85, 3.34]	0.13
	Urine toxicologyc	Negative	Positive	1.61 [0.91, 2.87]	0.10

Abbreviations: HR, hazard ratio; CI, confidence interval; sqrt, square root transformation; CD4, cluster of differentiation 4; Comorbidity, Neurocognitive Comorbidity Status; LT Meth SUD, lifetime methamphetamine substance use disorder

Notes:

^aHigher

^bLower

^cVariable modelled in a time-dependent manner

In unadjusted analyses (Table 2, Model 1B), hazard ratios for neurocognitive decline were elevated for Latino compared to Black PWH (HR = 1.86, 95% CI 1.14–3.04, p = 0.013). All factors in Table 1 that differed across racial/ethnic groups at p < .0.20 were included in a stepwise regression model predicting neurocognitive decline among Latino and Black PWH. Backwards AIC selected baseline (nadir CD4) and time-dependent (hematocrit, urine drug toxicology, and self-reported diabetes) factors as optimal predictors, which were added to the final Cox model (Table 2, Model 2B). Only lower hematocrit (HR = 1.08, p = 0.009) was significantly associated with earlier time to neurocognitive decline. With inclusion of these predictors, the elevated risk for early neurocognitive decline among Latino compared to Black PWH did not decrease, and remained significant (HR = 2.39, 95% CI 1.34–4.19, p = 0.002).

Discussion

Present findings suggest that Latino PWH have a higher risk of early neurocognitive decline compared to White and Black PWH. Medical comorbidities, such as markers of cardiometabolic health, explained a portion of these disparities among Latino compared to White PWH, but much of the elevated risk for neurocognitive decline among Latinos remained unexplained after considering a host of HIV disease, psychiatric, and substance use characteristics, and a limited number of available socio-demographic factors. Many variables that differed by ethnicity/race at baseline were not significant predictors of neurocognitive decline. Our analyses used published, regression-based norms that allowed us to detect neurocognitive change in individual participants, while controlling for baseline neurocognitive performance, demographics, and effects of repeated testing.

Study results are consistent with previous cross-sectional findings of higher rates of NCI among some Latino groups compared to Whites^11,13,14. Our results add to this literature by demonstrating an ethnic/racial disparity in longitudinal neurocognitive decline among Latino PWH. In a previous longitudinal analysis of the CHARTER cohort, Latino ethnicity was a significant predictor of neurocognitive decline¹⁶, but this prior study did not include separate non-Latino White and Black comparison groups, did not examine factors that might partially explain neurocognitive disparities, and only included a small number of cardiometabolic factors, which are important predictors of neurocognitive decline in Latinos³⁰. The present study extends prior research by showing that risk of neurocognitive decline is elevated in Latino PWH compared both to White and Black PWH. This study is the first to directly investigate differences in HIV-associated neurocognitive outcomes between Latino and Black PWH. Our findings showing increased risk of neurocognitive decline among Latino compared to Black PWH, and no significant differences between Black and White PWH are somewhat unexpected, given that both Latino and Black PWH face structural disadvantage in the HIV care, and present with worse HIV disease burden. One prior cross-sectional study found higher rates of HIV-associated NCI among Black compared to White PWH¹⁵, yet it assessed neurocognition via the International HIV Dementia Scale, which is a brief measure with no normative adjustments and limited sensitivity for detecting mild forms of HIV-associated NCI³¹. Further, these two studies’ findings are not necessarily inconsistent, ethnic/racial disparities observed cross-sectionally may not extend to differences in longitudinal neurocognitive decline.

We were unable to identify a set of factors that fully explained the increased risk for neurocognitive decline among Latino PWH compared to White and Black PWH after considering several characteristics traditionally considered relevant to HIV-associated NCI. Our models selected by the AIC method for maximal predictive accuracy found that many variables that differed by ethnic/racial group at baseline were not independent predictors of time to neurocognitive decline. Risk for neurocognitive decline among Latino PWH remained elevated in our final models, suggesting that unassessed variables may play an important role in ethnic/racial differences in HIV-associated neurocognitive trajectories. Interestingly, after inclusion of relevant predictors in our final models, elevated risk for neurocognitive decline was not reduced among Latino compared to Black PWH, but was reduced considerably when comparing Latino and White PWH. Significant predictors in this latter model (i.e., hematocrit [lower percentage of oxygen-carrying red blood cells], confounding neurocognitive comorbidity status [presence of more severe non-HIV risks for neurocognitive impairment], and total cholesterol) were markers of comorbid conditions in HIV, rather than traditional HIV disease indicators. This finding highlights the import of medical comorbidities, in particular cardiometabolic factors, in neurocognitive disparities between Latino and White PWH.

Several intersecting structural/institutional, sociocultural, and biomedical variables may be important to consider in future ethnic/racial neurocognitive disparities research. Our current assessment of sociocultural constructs was limited to SES and reading level. Furthermore, the psychometrics of the measures we used to assess these constructs have yet to be established among Latinos. Sociocultural variables (e.g. acculturation, language use, health behaviors) and environmental factors (e.g. exposure to air pollution and other neurotoxicants, limited access to quality healthcare) not assessed in our study may contribute to the observed differences in HIV-associated NCI among Latino PWH^13,32,33. Furthermore, structural racism is known to be a root cause of racial/ethnic health inequities³⁴, and experiences of overlapping discrimination due to ethnicity/race, immigration, sexual orientation, and HIV status may be compounded among Latino PWH in the U.S.³⁵. Growing research shows combined discrimination predicts worse health outcomes among Latino PWH³⁶, and that exposure to trauma and stress are particularly detrimental to neurocognition in PWH³⁷.

Neighborhood and institutional-level factors have also been shown to contribute to higher rates of chronic diseases such as diabetes³⁸ that are more common among Latino compared to White PWH⁴. Our study ascertained diabetes only by self-report, which may have underestimated actual rates of disease among Latino and Black PWH due to disparities in healthcare access, quality, and literacy^4,39. Some of our measurements of other cardiometabolic risk indicators (i.e. non-fasting plasma measures of cholesterol, triglycerides) were notable predictors in our models comparing Latinos and Whites. Assessment of cardiometabolic and vascular risk factors via standard criteria⁴⁰ is crucial to future research on neurocognitive decline in aging PWH, given their association with HIV-associated cognitive impairment⁴⁰ and late-life dementias among Latinos⁴¹. Furthermore, given reports of accelerated neurocognitive aging in HIV⁴², including biomarkers linked to Alzheimer’s disease and HIV-related genetic factors associated with neurocognitive impairment⁴³ may be relevant to future work examining ethnic/racial neurocognitive disparities in PWH.

Strengths of this study include a large, geographically and demographically diverse cohort of PWH. Our Latino sample reported varied countries of origin. Neurocognitive assessment of seven domains at each study visit was robust, and neurocognitive change over time was determined via regression-based norms that adjust for expected change based on an individual’s baseline performance. Assessment of HIV disease characteristics, and psychiatric and substance use comorbidities were detailed, standardized across sites and visits, and were modeled in a time-dependent manner.

Our study has several limitations. Participants were assessed from 2002–2015, and findings may not generalize to contemporary cohorts of PWH who acquired HIV after ART became widely available in the U.S. Of note, however, Latino and Black PWH continue to face systematic disadvantage in accessing ART³. Our findings may also be influenced by selection bias, such that participants who were able to participate were healthier than those who were not, thus underestimating physical and cognitive difficulties among PWH. The lack of an HIV-negative comparison group precludes us from fully ascribing these disparities to HIV-associated underlying mechanisms. Sample sizes for neurocognitive decliners across the three ethnic/racial groups were modest, and the sample size of Latinos was considerably smaller than the White and Black groups. Further, there were fewer normative neurocognitive adjustments for Latinos compared to White and Black groups. While previous work indicates that country of origin/descent may be an important factor in investigations of neurocognitive difference among Latinos¹⁴, our sample sizes of Latino sub-groups, as well as of Afro-Latinos, were too small to assess their impact of neurocognitive trajectories longitudinally. Further, study criteria limited the Latino cohort to fluent English speakers. Future studies with larger samples of Latino PWH that better reflect the sociocultural, linguistic, geographic, and racial diversity of this ethnic group are needed.

The present study is the first to examine longitudinal neurocognitive disparities among Latino, Black, and White PWH, and represents an important step towards identifying factors that might be driving these inequities. Our findings suggest that management of cardiometabolic health among Latino PWH may be one critical pathway to reduce the elevated risk of neurocognitive decline observed in this population. Ongoing work by our lab incorporating factors such educational opportunity, socioeconomic disadvantage, and dementia-related biomarkers may shed further light onto these health disparities. In the ART era, HIV-associated neurocognitive deficits continue to contribute to challenges with medication adherence, employment, mental health, and social functioning¹⁰. Investigation and mitigation of factors that increase risk for early neurocognitive decline among ethnically and racially diverse PWH is an important task, and may help diminish neurocognitive disparities.