Examining cognitive decline across black and white participants in the Harvard Aging Brain Study

Abstract

Background:

Black Americans are approximately twice as likely to develop dementia as compared to White Americans and the magnitude of this disparity is often attributed to a variety of factors that include psychosocial and vascular risk factors. However, less is known about the potential contribution of Alzheimer’s disease pathological differences.

Objective:

To examine potential differences in cross-sectional and longitudinal cognitive performance in black and white participants who were clinically normal at baseline.

Methods:

296 participants (48 African-American/black participants) underwent MRI and amyloid PET at baseline. Linear mixed models were used to examine the main effects of race, years of education, reading ability, Framingham Heart Study cardiovascular risk score (FHS-CVD), white matter hyperintensities (WMH), and amyloid (Aβ) burden on the Preclinical Alzheimer Cognitive Composite-5 (PACC5).

Results:

Lower levels of educational attainment and reading ability were found for blacks compared to whites. By contrast, no differences in FHS-CVD, WMH, or Aβ were found by racial group. Baseline differences in PACC5 score were attenuated after adjusting for educational factors, vascular factors and Aβ, but remained lower for blacks compared to whites (β= −0.24, p= 0.014). Further, blacks demonstrated a faster rate of PACC5 decline longitudinally compared to whites (β: −0.055, p= 0.025) after adjusting for covariates.

Conclusion:

Accounting for educational factors, vascular factors, and Aβ burden diminished, but did not eliminate, racial differences in PACC5 performance longitudinally. Understanding potential differences in longitudinal cognitive outcomes by race may be important for upcoming secondary prevention trials.

1. Introduction

Black Americans are approximately twice as likely to develop dementia as compared to White Americans [1–4] and the magnitude of this disparity is often attributed to a combination of socioeconomic, psychosocial, and vascular risk factors [5–10]. Indeed, factors such as literacy and quality of education over the lifespan [11–13], as well as vascular markers such as hypertension and diabetes [14,15] may account for a large proportion of cross-sectional racial differences in cognitive performance.

Whether the manifestation of brain pathology such as cerebrovascular disease or Alzheimer’s disease (AD) plaques and tangles differs between blacks and whites is not entirely clear, as molecular and imaging biomarker data in diverse populations remains limited. Several studies have shown greater burden of cerebrovascular disease in blacks compared to whites [15,16, 17] and that subclinical cerebrovascular disease, such as white matter hyperintensities (WMH), more strongly impact cognitive performance in blacks than whites [16,18]. However, with regards to molecular biomarkers of AD even less is known. Two investigations using cerebrospinal fluid (CSF) have found lower levels of cerebrospinal total tau and phosphorylated tau in blacks, despite similar levels of CSF amyloid-β42 (Aβ42) levels [18,19]. Using Aβ positron emission tomography (PET), findings have been mixed, as one study found higher Aβ burden in blacks [20], whereas no differences were found in another study [19].

How these factors independently or interactively contribute to the higher incidence of dementia in blacks remains unclear, as many longitudinal studies do not show differences in the rate of cognitive decline by race, after accounting for baseline disparities [5,21–23]. Thus, higher rates of dementia in blacks may be a direct consequence of lower baseline cognitive performance, such that blacks are disproportionately closer to reaching the threshold of cognitive impairment compared to whites [3,15]. For this reason, longitudinal investigation becomes highly valuable, as individuals serve as their own control, reducing the impact of baseline group differences [5].

Unfortunately, studies examining longitudinal cognitive outcomes in black participants, with AD and vascular biomarkers, are lacking. This is particularly the case for individuals who are clinically normal, as many prior studies have enrolled black participants with clinical impairment, who tend to present to the memory clinic later in the disease process with more severe symptoms compared to their white counterparts [4]. Further, despite a growing emphasis on the recruitment of racial/ethnic minorities in research studies, the proportion of participants of color remains low, making it even harder to tease apart the likely interaction between psychosocial, educational, and biological factors [24, 25]. On the other hand, the few studies that have enrolled minority participants, have highly selected samples, as cognitive cut-offs and medical exclusions to determine eligibility for the study disproportionately exclude minority participants from studies.

With these caveats in mind, we sought to investigate baseline and longitudinal differences in cognitive performance between clinically normal black and white participants with consideration of multiple relevant covariates including educational factors, vascular markers, and Aβ burden. Participants from the Harvard Aging Brain Study who underwent longitudinal cognitive evaluation with the preclinical Alzheimer’s cognitive composite-5 (PACC5)[26,27], a composite honed to detect subtle cognitive changes in preclinical AD, were investigated. We hypothesized that after adjusting for these factors, racial differences in baseline PACC5 performance would be attenuated and rate of longitudinal decline would be comparable by race.

2. Methods

2.1. Participants

Two hundred and ninety-six participants from the Harvard Aging Brain Study (age at baseline: 72.5±7.1, range= 50–89; years of education: 15.8 ± 3.0, 61.8% women, 15.8% African American/black) were studied using protocols and informed consent procedures approved by the Partners Healthcare Human Research Committee (see Table 1). The Harvard Aging Brain study is an ongoing observational study [28]. Participants were clinically normal at baseline based on clinician consensus with consideration of Clinical Dementia Rating (CDR)[29] score with global score 0, age- and education-adjusted Mini Mental State Exam (MMSE)[30] of 25 and above, scores exceeding age- and education-adjusted cut-offs on the 30-Minute Delayed Recall of the Logical Memory Story A [31] and a Geriatric Depression Scale (GDS)[32] score of less than 11. None of the participants had a history of alcoholism, drug abuse, head trauma or current serious medical or psychiatric illness. Racial group membership was determined via participant self-report, with non-exclusive options modeled after the United States’ federal Office of Management and Budget guidelines. Of the 48 participants who self-identified as black, 38 were born in the United States, 1 was born in Canada, and the remaining 9 were born in the Caribbean (e.g., Haiti, Barbados, Trinidad and Tobago).

Table 1.

Mean (SDs) baseline demographic information as a function of race.

Measure	White	Black	P value
N(female%)*	249 (58.6)	48 (78.7)	0.009
Age (years)	72.6 (7.2)	72.0 (6.3)	0.57
Education (years)	16.1 (3.0)	14.4 (2.9)	<0.001
American National Adult Reading Test	122.9 (7.3)	112.8 (11.3)	<0.001
Hollingshead Scale	25.8	33.02	<0.001
Born outside US/Canada (%)*	8.8	18.8	<0.001
Mini-Mental State Exam	29.1	28.4	<0.001
Digit Symbol Coding	47.8	40.1	<0.001
Logical Memory	15.5	13.5	<0.001
Free and Cued Total	31.8	31.2	0.29
Categories	43.9	37.7	<0.001
FHS-CVD risk score	30.7 (18.8)	33.3 (17.3)	0.38
APOE ε4+ (%)*	30.0	25.5	0.56
Amyloid-β cortical aggregate	1.16 (0.2)	1.12 (0.1)	0.15
WMH (log transformed, volume mm3)	7.5 (0.9)	7.8 (1.0)	0.16

^*Chi-squared analysis. FHS-CVD: Framingham Health Study Cardiovascular Disease risk score, APOE= apolipoprotein E, WMH= white matter hyperintensities

2.2. Cognitive outcome measure

Cognitive performance—the PACC5.

A description of the derivation of the Preclinical Alzheimer’s Cognitive Composite- version 5 score has been previously reported [26,27]. Briefly, the factor was comprised of scores from (1) Logical Memory Delayed Recall, (2) MMSE Total score, (3) Wechsler Adult Intelligence Scale-Revised (WAIS-R) Digit Symbol coding, (4) the Free and Cued Selective Reminding Test (FCSRT), in which total recall + free recall were combined, and (5) Category Fluency. Measures were z-transformed based on the baseline mean and standard deviation and averaged. Participants were administered the PACC5 on an annual basis for an average 4.7 years (range: 1–7 years).

2.3. Factors of interest

Premorbid reading ability.

We used ability level as estimated by the American National Adult Reading Test (AMNART).

Hollingshead Four Factor Index of Socioeconomic Status.

The Hollingshead Four Factor Index of Socioecominic status is an unpublished scaled based on marital status, employment status, education level, and occupational attainment [33]. Education level and occcupational attainment are each based on a 7 point scale with higher scores reflecting lower education and lower occuapational attainment, respectively.

Vascular risk factors.

An office-based Framingham Heart Study general cardiovascular disease (FHS-CVD) risk score [34] was computed as has been previously reported by our group [35]. Briefly, the FHS-CVD risk score was calculated on baseline data and is a weighted sum of age, sex, antihypertensive treatment (yes/no), systolic blood pressure (millimeters of mercury), body mass index, history of diabetes (yes/no), and current cigarette smoking status (yes/no). The FHS-CVD risk score provides a 10-year probability of future cardiovascular events.

MRI acquisition.

Participants underwent MRI on a Siemens TrioTIM 3 T scanner equipped with a 12-channel phased-array whole-head coil. Head motion was restrained with a foam pillow and extendable padded head clamps. Earplugs were used to attenuate scanner noise. High-resolution 3D T1-weighted multi-echo magnetization-prepared, rapid acquisition gradient echo anatomical images were collected with the following parameters: TR = 2200 ms; multi-echo TEs = 1.54 ms, 3.36 ms, 5.18 ms, and 7 ms; FA = 7°, 4× acceleration, 1.2 × 1.2 × 1.2 mm voxels.

Pittsburgh compound B (PIB)-PET imaging.

C11-PIB was synthesized and administered at Massachusetts General Hospital (MGH, Siemens ECAT EXACT HR1 scanner)[36]. Distribution volume ratio (DVR) images were created with Logan plotting (40–60 min, cerebellar reference). A global cortical aggregate including frontal, lateral, and retrosplenial regions was calculated for each participant, which was subsequently used to dichotomize participants into amyloid negative (Aβ-) and amyloid positive (Aβ+) groups using a cutoff of 1.185 using a Gaussian mixture model[37]. All statistical models used continuous Aβ, but Aβ was dichotomized for illustrative purposes in the figures.

White matter hyperintensities.

Baseline white matter hyperintensity burden (WMH) was assessed using fluid attenuation inversion recovery (FLAIR) images (repetition time = 6000 milliseconds; echo time = 454 milliseconds; inversion time = 2100 milliseconds 1 × 1 × 1.5 mm voxels; 2x acceleration). As previously described [38], WMH were identified from each individual’s FLAIR image using a previously validated algorithm [39]. The total WMH volume (in cubic millimeters) was extracted using a mask defined by the Johns Hopkins University White Matter Atlas[40]. WMH volumes were log-transformed to better approximate a normal distribution of values.

Apolipoprotein E (APOE).

A blood sample was collected for direct genotyping of APOE (heterozygotes and homozygotes for the ε4 were collapsed into the one category, with all ε4 haplotypes included).

2.4. Analytic approach.

Linear mixed-effects models (nlme package, R version 3.5.2) with random effects of intercept and slope were used to assess race as a predictor of longitudinal PACC5 decline. Main effects and interaction terms by time were examined for each predictor. To facilitate comparison of cognitive decline by race, all continuous variables were centered to the sample mean and black participants were the reference group. Our base model included race, age, and sex to predict longitudinal PACC5 (model 1: PACC5 ~ race × time + age × time + sex × time). Next, we added years of education, and AMNART as predictors to determine whether accounting for educational and factors diminished any race differences on longitudinal PACC5 (model 2: PACC5 ~ Model 1 + education × time + AMNART × time). Subsequently, we added FHS-CVD, WMH, and continuous Aβ burden as predictors to determine whether accounting for biomarkers impacted race differences on PACC5 (model 3: PACC5 ~ Model 2 + FHS-CVD × time + WMH × time + Aβ × time). We also ran models including baseline PACC5 performance, occupational attainment, country of origin (i.e., non-U.S. born), and APOE genotype, which yielded similar results to main models (see Supplement table 1). Additionally, findings were similar in a 3:1 sample matched on AMNART score (n= 144 white participants, 48 black participants) using the MatchIt package in R (see Supplement table 2).

Secondary analyses included models investigating the interaction with race and each biomarker by time separately to predict PACC5 (models 4–6). Additional secondary analyses investigated the 5 subcomponents of the PACC5 as dependent variables (i.e., Logical Memory, MMSE, FCSRT, Digit Symbol, Category Fluency) in separate models including sex, education level, AMNART, FHS-CVD, WMH, and Aβ as covariates.

3. Results

3.1. Demographics

Demographic information as a function of race is reported in Table 1. Relative to whites, blacks reported fewer years of education (t = 3.6, p = .0006), had lower scores on the AMNART (t= 5.6, p < .0001)), and were more likely female (χ²(1) = 6.8, p = 0.009). There were no racial-group differences with respect to baseline age (t= 0.6, p = .56), APOE ε4 status (χ²(1) = 0.34, p = .56), FHS-CVD risk score = (t = −0.8, p= 0.45), level of continuous Aβ (t = 1.5, p= 0.15), or log-transformed WMH volume (t= −1.4, p = 0.17). The average time of follow-up across the sample was 4.9 years, with an average of 4.4 years of follow-up for blacks and 5.0 years of follow-up for whites.

3.2. Cross-sectional associations between race and cognitive performance

To examine baseline differences in PACC5 performance by race, we examined whether race was associated with PACC5 performance using linear regression after adjusting for age, sex, education level and reading ability. Even after for controlling for these covariates we found that black participants demonstrated lower scores on PACC5 compared to whites (β= −0.28, p= 0.002). After also adjusting for FHS-CVD, WMH, and continuous Aβ, black participants remained lower in their performance on the PACC5 (β= −0.24, p= 0.014).

3.3. Race to predict longitudinal cognitive decline

Next, we turned to longitudinal models to examine potential differences in rate of decline by race. Our base model included sex and age to predict longitudinal PACC5 performance. In Model 1 (Table 2), there was a significant race by time interaction meaning that black participants declined at a rate that was 0.05 standard deviations/year greater than whites with the same age and sex, 95% CI [-0.10, −0.005]. Adding baseline PACC5 to this model did not significantly alter the findings.

Table 2.

Summary of linear mixed models examining racial group differences in PACC5 performance

Model Term	Standardized Estimate	t Value	P value
Model 1
Age × time	−0.008 ± 0.001	−6.9	<0.001
Sex × time	0.016 ± 0.018	0.88	0.376
Race × time	−0.06 ± 0.025	−2.46	0.014
Model 2
Age × time	−0.006 ± 0.001	−4.33	<0.001
Sex × time	0.014 ± 0.018	0.74	0.467
Education × time	−0.001 ± 0.003	−0.22	0.825
AMNART × time	−0.001 ± 0.003	−0.41	0.679
Race × time	−0.062 ± 0.027	−2.26	0.024
Model 3
Age × time	−0.002 ± 0.002	−1.51	0.132
Sex × time	0.059 ± 0.022	2.67	0.008
Education × time	−0.001 ± 0.003	−2.25	0.025
AMNART × time	−0.000±0.001	−0.33	0.739
WMH × time	−0.001±0.009	−0.153	0.879
Aβ × time	−0.335±0.039	−8.543	<0.001
FHS-CVD × time	−0.002±0.001	−3.51	<0.001
Race × time	−0.055±0.025	−2.25	0.0248

Note: Beta values are listed (± standard error). Age and education are centered to mean of sample. Reference is black group. AMNART=American National Adult Reading Test, FHS-CVD = Framingham Heart Study Cardiovascular disease risk score, WMH = White Matter Hyperintensities, Aβ = continuous amyloid

3.4. Adjusting for educational factors

In Model 2 (Table 2), there was a significant race by time interaction that persisted after controlling for educational factors with a steeper rate of PACC5 decline over time for blacks compared to whites (β= −0.062, t = −2.26, 95% CI [−0.12, −0.008]). This finding suggests that educational factors did not fully account for differences in slope of PACC5 performance. Adding baseline PACC5 to this model did not significantly alter the findings.

3.5. Consideration of cerebrovascular and AD biomarkers

After adjusting for FHS-CVD, WMH, continuous Aβ in the model and their interactions with time, a slope difference by race remained (β: −0.06, t = −2.25, 95% CI [−0.10, −0.007]) (see Table 2) (Figure 1).

Figure 1.

Longitudinal cognitive trajectories of participants by race

PACC=Preclinical Alzheimer’s cognitive composite (z score), time is in years. Model controlling for demographic factors, educational factors, vascular factors, and amyloid. Continuous variables centered to sample mean.

Secondary analyses that investigated three-way interactions with biomarkers, race and time in separate models, revealed a significant race × Aβ × time interaction, such that blacks with elevated Aβ declined faster than whites with elevated Aβ: (β= −0.27, t= −1.99, 95% CI [−0.53, −0.004]) (Figure 2). By contrast, there was not a significant race × FHS-CVD × time interaction (β= −0.002, t= −1.26, 95% CI [-0.005, 0.001]), nor a significant race × WMH × time interaction (β=0.016, t= 0.63, 95% CI [-0.03, 0.06]).

Figure 2.

Longitudinal PACC performance as a function of race and amyloid level (by tertile).

ACC=Preclinical Alzheimer’s cognitive composite (z score), time is in years. Model controlling for demographic factors and educational factors. Continuous variables centered to sample mean.

3.6. Subcomponents of the PACC5

MMSE- Cross-sectional analyses using linear regression, did not reveal differences on MMSE performance by race after accounting for all the covariates (β= −0.31, t= −1.77, p= 0.077). Additionally, there were no differences in rate of decline on MMSE by race (β= −0.06, t= −1.25, 95% CI [-0.15, 0.03]) Logical Memory- Cross-sectional analyses did not reveal a main effect of race on Logical Memory performance after accounting for all the covariates (β= −0.94, t= −1.65, p= 0.109). Longitudinal models did not reveal any differences in Logical Memory performance over time by race (β= −0.12, t= −0.98, 95% CI [−0.36, 0.122]). Free and Cued Selective Reminding- On the FCSRT, there was not a cross-sectional association with race (β= 0.45, t= 0.46, p= 0.65). However, blacks performed worse longitudinally over time compared to whites (β= −0.53, t= −2.44, 95% CI [−0.95, −0.10]). Digit Symbol Coding- Cross-sectional analyses demonstrated lower performance for blacks compared to whites on Digit Symbol (β= −4.15 t= −2.42, p= 0.016), but there was not a significant race × time interaction (β= −0.23, t= −1.01, 95% CI [-0.7, 0.22]). Category Fluency- There was not a cross-sectional difference by race on Category fluency (β= −2.18, t= −1.35, p= 0.17). Additionally, the rate of decline on Category Fluency did not differ by race (beta= −0.24, t= −0.90, 95% CI [-0.75, 0.27]).

4. Discussion

In this study, we examined baseline and longitudinal differences on the preclinical Alzheimer’s cognitive composite between black and white participants from the Harvard Aging Brain study who were clinically normal at baseline. While differences on most cognitive tests were found to be lower in black participants, we found that after controlling for educational and socioeconomic factors, vascular factors, and Aβ burden, differences in baseline performance were attenuated. Nonetheless, rates of longitudinal cognitive decline were steeper for blacks compared to whites. However, this pattern was not consistently observed across all subcomponents of a cognitive composite; MMSE, Logical Memory, Digit Symbol and Category Fluency did not show any differences longitudinally by race, but the FCSRT showed steeper decline in black participants compared to white participants.

Our finding of lower baseline cognitive performance in blacks compared to whites is consistent with findings from multiple previous studies [4,10,11]. However, accounting for years of education and premorbid reading ability attenuated this baseline performance disparity. Further, our results showed that rate of cognitive decline remained steeper for blacks compared to whites, even after controlling for educational and sociocultural factors, vascular factors, and Aβ burden. These findings diverge from the current literature primarily showing similar rates of cognitive decline after adjusting for baseline disparities [3,14,20,22]. One important difference in the current study is that individuals were clinically normal at baseline, whereas previous studies often included individuals with Mild Cognitive Impairment and/or dementia. Determining whether rates of longitudinal decline by race may differ at various points along the AD continuum, will be something to further investigate with future studies. Additionally, while our black participants had similar levels of education compared to other studies investigating longitudinal decline in black participants [14,22], we did not examine quality of education [13], or other potential sources of sample bias that may have played a role in our findings. We did, however, also adjust for reading ability that may partially address this potential confound [11,12].

Unlike large population-based studies that have found greater cardiovascular risk factors and microvascular ischemic disease in blacks compared to whites [41], we did not find that the FHS-CVD risk score or WMH load measured at baseline were higher in blacks than whites. Further, including these factors did not account for differences in cognitive decline by race. Previous studies suggest that the racial differences in cognitive and clinical outcomes may largely be explained by the increased prevalence of vascular risk factors in blacks, resulting in increased cerebrovascular disease [15]. While we did not observe a significant interaction between race and vascular factors on longitudinal cognitive decline, our ability to identify this relationship was limited by the restricted range of cerebrovascular disease in the sample (participants were excluded for symptomatic stroke or intracranial hemorrhage, extensive small vessel ischemic disease).

In contrast to the lack of observed impact of vascular factors, we found that blacks with elevated Aβ showed steeper cognitive decline compared to whites with elevated Aβ, despite no baseline differences in continuous Aβ burden. Findings are in keeping with a prior study that found that the cognitive trajectory was related to continuous Aβ only in black participants, although this study examined cognitive decline prior to the time that Aβ was measured and did not control for vascular risk factors [42]. Adding APOE as a covariate did not alter findings of racial differences in cognitive decline.

When individual subcomponents of the PACC5 were examined as the longitudinal outcome, divergent patterns emerged across measures. Several measures did not show differences by race cross-sectionally or over time (MMSE, Logical Memory, Category Fluency). By contrast, the FCSRT demonstrated steeper decline over time for blacks compared to white participants, but did not show baseline differences. On Digit Symbol Coding, blacks had lower scores at baseline compared to whites, but rate of decline did not differ by race. Previous studies from HABS have shown differences in longitudinal patterns across different cognitive measures on the PACC5 [23,24] according to Aβ status, and the FCSRT, in particular, is a test that has shown the earliest detectable decline. It remains unclear whether steeper decline on the FCSRT was observed in blacks compared to whites given the properties of this test or whether steeper decline reflects differences between groups in disease risk. Additionally, two of the measures that did not exhibit differences by race, were measures that were used to determine eligibility for the study (i.e., MMSE, Logical Memory). Thus, for all participants selected into the study, a certain standard of performance was required, which may have impacted the comparison of baseline performance on these measures across race. Further, we did not observe substantial differences in our results when accounting for baseline PACC5 performance, but we acknowledge that it is challenging to account for differences in performance without introducing additional bias [3]. Taken together, factors such as inclusion criteria, clinical status, measurement properties, and socio-cultural differences are important considerations when examining potential longitudinal differences across racial groups.

Several limitations to this study should be considered in interpreting the findings. Eligibility inclusion criteria are likely to have played a large role in the findings of this study. This is the case for both limiting the sample size of our black participants, as well as resulting in certain characteristics of black participants who were enrolled. Black participants in our sample were more likely female and more likely to vary in their country of origin, multilingualism, and cultural background. Participants were required to perform above cut-off on several cognitive measures that were validated on primarily white populations with limited considerations to cultural factors. Participants were required to have minimal vascular disease, thus potentially diminishing our power to identify relationships that could be observed with the FHS-CVD score and WMH. On the flipside, however, the lack of clear vascular differences between groups had the advantage of more directly comparing amyloid burden and cognitive decline in our study.

Taken together, we believe that these findings, despite all the limitations, may still be of value given the very limited data on AD biomarkers available in diverse older populations. Unfortunately, the proportion of black to white participants in this study is similar if not higher than many other studies that have amyloid PET imaging and longitudinal follow-up [19, 43], which in the case of the Knight Alzheimer Disease Center in St. Louis, took over a decade to recruit 173 African-Americans [5]. Additionally, while we acknowledge our sample is highly selected, the characteristics of our sample may be reflective of those who would be eligible for ongoing prevention trials [43]. Ultimately, greater success in recruiting individuals from diverse racial and ethnic groups will be critically important to fully elucidate whether rates of cognitive decline differ across the disease spectrum by race.

In summary, the current study investigated the effect of race on longitudinal cognitive decline. While adjusting for differences in educational factors attenuated the effects of baseline cognition by race, they did not fully account for longitudinal differences observed. Future efforts are sorely needed to further disentangle the mechanisms underlying potential racial differences in rates of cognitive decline, which in turn, may improve early detection, diagnosis and treatment of Alzheimer’s disease.