Associations between race, APOE genotype, cognition, and mortality among urban middle-aged white and African American adults

We examined associations between cognition and mortality and how these relationships vary by race and Apolipoprotein E (APOE) genotype, in a longitudinal study of 2346 middle-aged White and African American adults (30–64 years at baseline) from the Healthy Aging in Neighborhoods of Diversity across the Life Span cohort study. Baseline cognition spanned global mental status, and several domains obtained using principal components analysis (PCA; PCA1: verbal memory/fluency; PCA2: attention/working memory; PCA3: executive function/visuo-spatial abilities). Cox regression models evaluated associations between cognition and all-cause and cardiovascular disease (CVD)-mortality. Interactions between cognition and APOE2 as well as APOE4 allelic dose were tested, and race was a key effect modifier. Higher APOE4 dose was associated with increased CVD-mortality (hazard ratio [HR] per allele = 1.37; 95% CI 1.01–1.86, p = 0.041); APOE2 dosage’s association with CVD-mortality was non-significant (HR = 0.60; 95% CI 0.35–1.03, p = 0.065). Higher PCA3 was associated with lower all-cause (HR = 0.93; 95% CI 0.87–0.99, p = 0.030) and CVD (HR = 0.85; 95% CI 0.77–0.95, p = 0.001) mortality risks, the latter association being more pronounced among Whites. PCA2 interacted synergistically with APOE2 dosage, reducing risks for all-cause mortality (PCA2 × APOE2: − 0.33 ± 0.13, p = 0.010) and CVD mortality (PCA2 × APOE2: − 0.73 ± 0.31, p = 0.019). In conclusion, greater executive function/visuo-spatial abilities were associated with reduced CVD-specific mortality, particularly among Whites. Greater “attention/working memory” coupled with higher APOE2 dosage was linked with reduced all-cause and CVD mortality risks.


Digit Span Forward and Backward (DS-F and DS-B)
The Wechsler Adult Intelligence Scale, Revised 4 Digit Span Forward and Backward primarily capture attention and working memory, a component of executive function. The tests were administered according to the manual's instructions. The outcome variable was the total score, which was the total number of correct answers for each test.

Category Fluency
Category fluency 5,6 is a measure of semantic verbal fluency, where participants are asked to generate as many animals as possible within a 60 second duration. Higher scores indicate better category fluency. The outcome variable was the total number of correctly generated words (i.e., words that were not intrusions and perseverations).

Brief Test of Attention (BTA)
For the BTA 7 , a test of divided auditory attention, the examiner administered up to 10 trials of letters and numbers (4-18 items) that increased in length with each trial. Only the numbers portion of the test was administered. For each trial, participants were asked to disregard the number of letters read, while tracking how many numbers were recited. They were also told to keep their hands in fists to avoid finger counting. The outcome variable was the total number of correct trials.

Trail Making Tests A and B (TRAILS A and B)
The Trail Making Tests A and B 8 primarily capture attention and executive functioning, respectively. The main executive function subdomain that TRAILS B captures is set-shifting and cognitive control. Both trials also measure visuo-motor scanning and processing speed.
Participants were asked to draw a line between consecutive numbers (TRAILS A) and alternate between numbers and letters (TRAILS B) as quickly as they could. They were informed that they were being timed. The examiner pointed out errors that were then corrected by the participant.
Errors were captured via increased time. Scores for TRAILS A and B reflected seconds to completion, where higher scores indicate poorer performance.

Clock Drawing Test -Clock to Command (CDT)
The Clock Drawing Test 9 is a measure of visuo-spatial abilities, that also captures elements of memory and executive function. Participants are instructed to draw a clock, put in all of the numbers, and set the hands to 10 minutes past 11. Performance is based off correct drawings of the clock face (0-2), numbers (0-4) and hands (0-4). Scores ranged from 0 to 10, with higher scores indicating better performance. Participants who did not score a perfect score on the command portion of the test were also asked to copy a clock with the hands set to 10 minutes after 11.

Wide Range Achievement Test -3 rd Edition: Word and Letter Reading Subtest (WRAT)
The WRAT Word and Letter Reading Subtest 10 is a test of reading ability that is often used as a proxy for literacy and quality of education. Participants were instructed to correctly read a list of 50 words that increased in difficulty. If the first five words were not correctly pronounced, letter reading was also administered. Standard instructions were used with the tan form. The outcome variable used was the total number of correctly pronounced words.

Center for Epidemiological Studies Depression Scale (CES-D)
The CES-D 11 is a 20-item measure of depressive symptomatology. Participants are asked to consider the frequency and severity of their symptoms over the last week. Scores ranged from 0 to 60. Scores of >16 indicated significant depressive symptoms and scores of >20 indicated a clinically significant amount of depressive symptoms.

Method S2: Mixed-effects regression models
The main multiple mixed-effects regression models can be summarized as follows: Multi-level models vs. Composite models Eq.

1.1-1.4
Where Yij is the outcome (Each cognitive test score measured at v1 and/or v2) for each individual "i" and visit "j"; is the level-1 intercept for individual i; is the level-1 slope for individual i; is the level-2 intercept of the random intercept ; is the level-2 intercept of the slope ; is a vector of fixed covariates for each individual i that are used to predict level-1 intercepts and slopes, which can include sociodemographic variables among others. In this analysis, mixed-effects regression models did not include exposures (Xij) or covariates (Zij). They were only used to predict empirical bayes estimators for baseline cognitive performance for each test, with TIME It is worth noting that the models were fit using the entire HANDLS cohort with complete data on either v1 or v2 on cognitive tests was used to improve reliability of predicted estimates. Finally, empirical Bayes estimators of the intercept were predicted from the simple model with no covariates by adding the fixed effect of the intercept (i.e. γ00) to its predicted individual-level random effect thus allowing for imputation of missing data for individuals with only 1 repeat. This baseline cognitive performance score for each test is heretofore labelled as CP.

Methods S3: Principal components analysis of cognitive performance scores
Following this estimation, baseline performance on each cognitive test score were entered into a principal components analysis (PCA) as measured variables 13 in which a number of common factors were extracted based on common variance, component loadings estimated and the residual variance labeled as uniqueness for each LARCC. The principal component analysis model can be summarized as follows:

CPi=
Where CPi is the standardized z-score for each predicted baseline cognitive performance test score, λij is the component loading for each CPi and each factor, Domainj is the standardized zscore for each factor j, and φi is the residual error, the squared value of which is the uniqueness.
The sum of squared factor loadings for each CPi is the communality or the common variance that is accounted for by the extracted factors.
An eigenvalue>1 rule was used and the scree plot was observed to determine the adequate number of extracted components that would produce the best model fit, particularly that would explain the greatest amount of variance in the data. The component loadings were then rotated using varimax orthogonal rotation and the factors were interpreted, and cognitive domains labeled accordingly, with cutoff point of 0.30 or more for significant loading. The component scores (z-scores) were predicted and used as markers of CPi for specific cognitive domains. Note that all CPi were entered in the direction of greater score → better performance. Thus, CPi of BVRT, TRAILS A and B were multiplied by -1, prior to inclusion in the PCA model (See

Figures S3 and S4
and Tables S3 and S4 for main results).

Results S1: Detailed results
The inclusion criteria for the three analytic samples (i.e., Sample 1, Sample 2, Sample 3 in Figure S1) yielded some differences in their characteristics relative to the initial study cohort (n=3,720     a Models included each of 4 cognitive performance variables separately as the main predictor for all-cause or CVD mortality and interacted this main predictor with ApoE4 dosage. The models were carried out in the overall population only. All models adjusted only for age, sex, race, poverty status, education and the WRAT-3 score using imputed data, in addition to other lifestyle and health-related factors, namely current drug use, current tobacco use, body mass index, self-rated health, co-morbidity index, HEI-2010, total energy intake, CES-D total score, and the inverse mills ratio.  a Models included each of 4 cognitive performance variables separately as the main predictor for all-cause or CVD mortality and interacted this main predictor with ApoE2 dosage. The models were carried out in the overall population only. All models adjusted only for age, sex, race, poverty status, education and the WRAT-3 score using imputed data, in addition to other lifestyle and health-related factors, namely current drug use, current tobacco use, body mass index, self-rated health, co-morbidity index, HEI-2010, total energy intake, CES-D total score, and the inverse mills ratio. a Models included each of 4 cognitive performance variables separately as the main predictor for all-cause or CVD mortality and interacted this main predictor with ApoE2 dosage. The models were carried out in the overall population only. All models adjusted only for age, sex, race, poverty status, education and the WRAT-3 score using imputed data, in addition to other lifestyle and health-related factors, namely current drug use, current tobacco use, body mass index, self-rated health, co-morbidity index, HEI-2010, total energy intake, CES-D total score, and the