Diet quality (DQ) has profound and long-term consequences on cognitive function(Reference Alosco, Spitznagel and Raz1–Reference Petersson and Philippou3). An emerging literature is reporting protective benefits of some dietary factors (such as vitamins D and E, PUFA, etc.) against cognitive decline as well as delayed onset and progression of Alzheimer’s disease (AD)(Reference Gu and Scarmeas4–Reference Yannakoulia, Kontogianni and Scarmeas6). Vitamin D has been implicated in cognitive decline due to possible neuronal loss with reduced number of vitamin D receptors in brain regions like the hippocampus and AD risk because of lower hippocampal vitamin D receptor mRNA(Reference Wilson, Houston and Kilpatrick7). PUFA (and their precursors) have numerous beneficial effects for improved brain health and cognition via optimal neurotransmission, better cell survival and reducing neuroinflammation in addition to influencing fluid intelligence, memory, gray and white matter volume and related microstructures(Reference Richard, Laughlin and Kritz-Silverstein8). Epidemiological evidence demonstrates a role for dietary intervention in the primary prevention of chronic diseases, even in old age(Reference Charlton9). Increasing evidence implicates certain dietary patterns as beneficial to brain health(Reference Alosco, Spitznagel and Raz1,Reference Solfrizzi, Panza and Frisardi5,Reference Dacks, Shineman and Fillit10) . For instance, the Mediterranean diet, typically characterised by higher intakes of fruit, vegetables, whole grains, fish, unsaturated fatty acids and moderate alcohol consumption, is important for its role in preserving cognitive health(Reference Moore, Hughes and Ward11). A systematic review from 2016 found memory (i.e. delayed recognition, long-term and working memory), executive function and visual constructs benefited from Mediterranean diets(Reference Hardman, Kennedy and Macpherson2). However, the study population was predominantly White across the board, with a couple of exceptions that included Hispanic participants. Another recent review looking into the Mediterranean, Dietary Approaches to Stop Hypertension (DASH) and Mediterranean-DASH Intervention for Neurodegenerative Delay (MIND) diets suggested that higher adherence to these diets is associated with less cognitive decline and a lower risk of AD(Reference van den Brink, Brouwer-Brolsma and Berendsen12). Diet that is very similar to a Mediterranean diet in composition, also widely used and recommended in the USA, is the DASH(Reference Appel, Moore and Obarzanek13) diet. While most primary studies on DQ and cognition focused on one or possibly two DQ measures, this study aimed to expand the current literature on three validated scoring systems to measure DQ, establishing a comprehensive approach.
There is a paucity of research on association studies investigating diet and cognitive performance/decline among different racial groups in the progression of AD(Reference Graff-Radford, Besser and Crook14–Reference Weuve, Barnes and Mendes de Leon18). In fact, most research on the relation of race to cognitive function in AD has been cross-sectional(Reference Barnes, Wilson and Li19). Longitudinal studies assess rates of cognitive decline, but few have examined the association between cognitive decline(Reference Wilson, Li and Aggarwal20) and DQ with genetic risk for AD. African Americans (AA) in particular suffer from higher incidence rates of AD, perhaps due to undiscovered genetic factors, disproportionately higher rates of risk factor diseases(Reference Hajjar, Wharton and Mack21) (such as diabetes and stroke), biological or environmental exposures that erode ‘cognitive reserve’ which may protect against or accelerate disease expression or detection bias of existing testing methods(Reference Bachman, Stuckey and Ebeling22). They also struggle to adhere to a healthy diet more than their White counterparts(Reference Hossain, Beydoun and Kuczmarski23–Reference Richards Adams, Figueroa and Hatsu25).
In the present study, we examined the cross-sectional and longitudinal relationships of DQ and cognition in a socio-economically diverse sample of AA middle-aged adults. We hypothesised that initial better DQ would be associated with higher baseline cognitive functioning. We also examined whether those relationships differ across sex and by increasing genetic risk for AD.
Materials and methods
Database
Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) is a prospective cohort study initiated in 2004 to investigate health disparities in medical, metabolic and cognitive outcomes in a socio-economically diverse sample of Whites and AA (30–64 years old at baseline) recruited from selected neighbourhoods in Baltimore, MD. Initial data were collected in two phases (visit 1: 2004–2009). Phase 1 consisted of screening, recruitment, first 24-h dietary recall and household interviews in participants’ homes. Phase 2 consisted of the second 24-h dietary recall and physical examinations in mobile medical research vehicles. The first follow-up examinations were performed approximately 5 years later (visit 2: 2009–2013; mean follow-up time of 4·62 (se 0·95) years; range: 0·42–8·20) at which two 24-h dietary recalls were also collected. Neuropsychological tests were administered at both visits on the medical research vehicles. The numbers of participants with at least one of the eleven cognitive test scores at visits 1 or 2, dietary and covariate data at baseline ranged from 123 to 228 (k = 1·70–1·95 observations per participant), which yielded 5–15 % (k = 1·0–1·7) missing observations.
Written informed consent was obtained from all participants at each visit during which they were provided with a protocol booklet in layman’s terms and a video that described all procedures and future re-contacts. HANDLS study was ethically approved by the National Institutes of Health, National Institute of Environmental Health Sciences Institutional Review Board.
Study sample
The initial HANDLS sample was recruited at visit 1 (n 3720) with complete data on demographics. In this study, we excluded participants from European ancestry (n 1522) due to non-availability of genetic data in this group and examined only AA participants (n 2198). Restricting our sample to participants over 50 years (n 979) and then excluding participants with missing data on valid cognitive tests, dietary assessment and genetic polymorphisms yielded a sample of 342 at visit 1 and 268 at visit 2. We restricted our sample to ‘over 50 years’ for a greater variability in cognitive decline measures across both racial groups compared. We calculated change in DQ over time for visit 1 (n 244) and visit 2 (n 249). After excluding participants with incomplete covariate data, our final sample for analyses was 228 for visit 1 and 230 for visit 2 (Fig. 1). This sample selectivity was adjusted for using a two-stage Heckman selection model(Reference Yosef Hochberg26).
Fig. 1. Participant flowchart. Numbers of participants varied from 123 to 228 because there were missing data on cognitive tests (k = 1·70−1·95 observations/participant).
Cognitive measures
A cognitive battery of tests was administered to participants consisting of Mini-Mental State Examination; California Verbal Learning Test-List A (CVLT-List A); CVLT-Free Recall Long Delay; Benton Visual Retention Test (BVRT); Brief Test of Attention; Trailmaking Test A (Trails A); Trailmaking Test B (Trails B); Digits Span Forward Test; Digits Span Backward Test; Clock Command Test; Identical Pictures Test; Card Rotation Test and Animal Fluency Test. Details of these tests are available in Appendix 1 in online Supplementary Material. Except for the BVRT and the Trailmaking Tests, higher scores reflect better cognition. For BVRT and Trailmaking Tests parts A and B, better performance on BVRT was measured by fewer errors; the Trailmaking Tests were measured by faster performance. Cognitive performance test scores at baseline (visit 1), follow-up (visit 2) and change between visits, by sex, for HANDLS participants >50 years are presented in online Supplementary Table S1.
Genetic data
In total, 1024 HANDLS participants were successfully genotyped to 907 763 SNP at the equivalent of Illumina 1M array coverage. Sample exclusion criteria were (1) call rate <95 %, discordance between self-reported sex and sex estimated from X-chromosome heterogeneity, cryptic relatedness, discordance between self-reported African ancestry and ancestry confirmed by genetic data. SNP exclusion criteria were (1) Hardy–Weinberg equilibrium P-value <10–7, (2) minor allele frequency <0·01 and (3) call rate <95 %. Genotype quality control and data management were conducted using PLINKv1.06 (PMID: 17701901). Cryptic relatedness was estimated via pairwise identity by descent analyses in PLINK and confirmed using RELPAIR (PMID: 11032786). HANDLS participant genotypes were imputed using MACH/minimac version 2.0 (https://genome.sph.umich.edu/wiki/Minimac) based on combined haplotype data for the 1000 Genomes Populations project phase 3 version 5 multi-ethnic reference panel. Our final sample includes subjects with complete genetic data as they are further narrowed down by the availability of complete dietary, cognitive and covariate information.
Genetic risk score calculation
Previously reported genetic variants at specific genetic loci implicated with phenotypes of AD were used for genetic risk score calculation (online Supplementary Table S2). Of the 130 reported genetic variants, seventy-seven had valid SNP identifier. Seventy out of seventy-seven SNP had imputed genotype data in the HANDLS study. After excluding two SNP with poor imputation quality score (R 2 < 0·30), there were sixty-eight SNP for the final analysis. These SNP were then screened for significant associations with the Mini-Mental State Examination from the published literature. This was primarily because few studies used more than two tests (including Mini-Mental State Examination) to measure cognitive performance. Only twelve of the sixty-eight showed a significant association with baseline cognitive performance, across sex, age, race and geographical location(Reference Antonell, Balasa and Oliva27–Reference Tisato, Zuliani and Vigliano34). The genotype dosages of the risk alleles of these twelve SNP were used for the calculation of the HANDLS AD genetic risk score (hAlzScore). The online Supplementary Table S2 describes those SNP. Table 1 presents with individual SNP and hAlzScore correlation. The SNP were located on the following genes: transferrin, TF (n 1); cystatin 3, CST3 (n 1); presenilin 1, PSEN1 (n 1); prion protein, PRNP (n 1); insulin degrading enzyme, IDE (n 1); transcription factor A, mitochondrial, TFAM (n 1); APOE (n 2); angiotensin I converting enzyme, ACE (n 2); glyceraldehyde-3-phosphate dehydrogenase, GAPDH (n 1) and cholinergic receptor nicotinic beta 2 subunit, CHRNB2 (n 1).
Table 1. Individual SNP* correlations with hAlzScore (genetic risk score for Alzheimer’s disease (AD) in Healthy Aging in Neighborhoods of Diversity Across the Life Span (HANDLS))†
TF, transferrin; CST3, cystatin 3; PSEN1, presenilin 1; PRNP, prion protein; IDE, insulin degrading enzyme; TFAM, transcription factor A, mitochondrial; ACE, angiotensin I converting enzyme; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; CHRNB2, cholinergic receptor nicotinic beta 2 subunit.
* SNP dosages that were reverse-coded to create the hAlzScore, due to alternative allele increasing the risk of AD: rs165932, rs1799990, rs449647, rs4806173 and rs4845378.
† Genes/SNP: TF (rs1049296_C), CST3 (rs1064039_A), PSEN1 (rs165932_T), PRNP (rs1799990_A), IDE (rs2251101_T), TFAM (rs2306604_C), APOE-ϵ2 (rs405509_A), ACE (rs4291_A), ACE (rs4343_A), APOE-ϵ2 (rs449647_A), GAPDHS (rs4806173_C) and CHRNB2 (rs4845378_G).
Diet quality assessment
Method
All 24-h dietary recalls were collected using the United States Department of Agriculture computerised automated multiple-pass method(Reference Moshfegh, Rhodes and Baer35). The automated multiple-pass method was designed to provide prompts throughout all five steps of the recall to capture all the foods and drinks consumed throughout the previous day. The steps are described in detail elsewhere(Reference Appel, Moore and Obarzanek13). Trained interviewers provided an illustrated food model booklet, rulers and measuring cups and spoons to participants to help them estimate accurate quantities of foods and beverages consumed. The approximate time between recalls was 4–10 d. Each recall was coded using the United States Department of Agriculture Survey Net data processing system, matching foods consumed with codes in the Food and Nutrient Database for Dietary Studies(36). Of the 3720 participants examined at visit 1, 2177 individuals and at visit 2, 2140 persons completed two 24-h dietary recalls.
Healthy Eating Index 2010
Food-based DQ was also evaluated with the Healthy Eating Index 2010 (HEI-2010). The National Cancer Institute’s Applied Research website provided the basic steps for calculating the HEI-2010 component and total scores and statistical codes for 24-h dietary recalls(37). A detailed description of the procedure used for this study is available on the HANDLS website(38). The HEI-2010 includes twelve components, nine of which assess adequacy of the diet and the remaining three should be consumed in moderation. The nine components are: (1) total fruit; (2) total vegetables; (3) whole fruit; (4) greens and beans; (5) whole grains; (6) dairy products; (7) total protein foods; (8) seafood and plant proteins and (9) fatty acids. Refined grains, Na and empty energy content reflect components that should be consumed in moderation(Reference Guenther, Kirkpatrick and Reedy39). Component and total HEI-2010 scores were calculated for each recall day and were averaged to obtain the mean for both days combined.
Dietary Approaches to Stop Hypertension
The score for DASH diet adherence, based on nine nutrients, was determined for each participant using the formula reported by Mellen et al.(Reference Mellen, Gao and Vitolins40). The nine target nutrients were total fat, saturated fat, protein, fibre, cholesterol, Ca, Mg, Na and K. Micronutrient goals were expressed per 1000 kcal. The total DASH score was generated by the sum of all nutrient targets met. If the participant achieved the DASH target for a nutrient, a value of 1 was assigned, and if the intermediate target for a nutrient was achieved, a value of 0·5 was assigned. A value of zero was assigned if neither target was met. The maximum DASH score was 9; individuals meeting approximately half of the DASH targets (DASH score = 4·5) were considered DASH adherent(Reference Mellen, Gao and Vitolins40).
Mean adequacy ratio
DQ was also assessed using nutrient adequacy ratio (NAR) for seventeen micronutrients and mean adequacy ratio (MAR) scores(Reference Murphy, Foote and Wilkens41,Reference Fanelli Kuczmarski, Mason and Beydoun42) . The NAR score was determined by dividing each participant’s daily intake of a micronutrient divided by the RDA for that micronutrient. The micronutrients were vitamins A, C, D, E, B6, B12, folate, Fe, thiamin, riboflavin, niacin, Cu, Zn, Ca, Mg, P and Se. The RDA was adjusted for participants’ ages and sexes, and vitamin C was adjusted for smokers(43). The NAR score was converted into percentage with values exceeding 100 truncated to 100. The formula used to calculate the MAR score was: MAR = (∑NAR scores)/17(Reference Fanelli Kuczmarski, Bodt and Stave Shupe44). NAR and MAR were calculated separately for each daily intake and then averaged. MAR scores represented nutrient-based DQ since they were based on intakes of foods and beverages and no supplements.
Diet quality score
Two principal components analyses(Reference Jake Lever45) were conducted whereby baseline DQ indices (HEI-2010, DASH and MAR) as well as their annual rates of change were reduced into two measures, namely DQ and DQ Change (ΔDQ), respectively, using the Kaiser rule for component extraction (Eigen value > 1) and examining the scree plot. In both cases, 46–54 % of the total variance was explained by the single component(Reference Jake Lever45). Those measures were used in the main analysis, for data reduction purposes.
Covariates
Socio-demographic, lifestyle and health-related potential confounders
All regression models adjusted for socio-demographic factors, namely age, sex, race, educational levels (less than high school coded as ‘0’; high school coded as ‘1’; and more than high school coded as ‘2’) and poverty status (below v. above 125 % the federal poverty line). Additional adjustment factors include BMI (kg/m2), current drug use (‘opiates, marijuana or cocaine’ = 1 v. not = 0) and current smoking status (‘never or former smoker’ = 0 v. ‘current smoker’ = 1). These models were further adjusted for self-reported history of type 2 diabetes, hypertension, dyslipidaemia, CVD (stroke, congestive heart failure, non-fatal myocardial infarction or atrial fibrillation), inflammatory disease (multiple sclerosis, systemic lupus, gout, rheumatoid arthritis, psoriasis, thyroid disorder and Crohn’s disease) and use of non-steroidal anti-inflammatory drugs (prescription and over-the-counter) during visit 1.
Statistical analysis
Stata 15.0 (StataCorp) was used to conduct all analyses. First, participants’ characteristics, including covariates and exposures, were compared by sex using t tests for continuous variables and χ 2 tests for categorical variables. Second, several mixed-effects regression models on continuous initial DQ and ΔDQ scores calculated from total scores of HEI-2010, MAR and individual components were conducted to test associations with cognitive performance measures, while adjusting for potential confounders. We used linear mixed-effects models to characterise the overall pattern of change in cognitive function and to examine the relation of a specific predictor (e.g. DQ or hAlzScore) to initial level of cognitive function and annual rate of change. In this approach, both initial level of cognition and individual rate of change are explicitly modelled as sources of random variability and possible correlates of how rapidly cognition changes. Everyone is assumed to follow the mean path of the group except for random effects which cause the initial level of cognition to be higher or lower and the rate of change to be faster or slower. Thus, we added a random effect for the intercept and another for the slope. Specifically, each model included years elapsed between visits (TIME), exposures/covariate main effects and two-way interaction terms between TIME and exposures/covariates. We assumed the unavailability of outcomes to be missing at random(Reference Ibrahim and Molenberghs46). Sex-specific associations were examined through stratified analyses separately among men and women. Effect modification by sex was formally tested for effects of hAlzScore/DQ/ΔDQ on baseline cognitive performance (two-way interaction terms) and on cognitive change over time (three-way interaction terms). These models were adjusted for covariates (see Covariates section) that include socio-demographics, lifestyle and health-related factors. Scores for Trails A and B were log-transformed before modelling due to the extreme distribution of both. All other cognitive tests were not skewed.
Three sets of models were tested: (1) models with hAlzScore as the main predictor, for cross-sectional and longitudinal cognitive performance, (2) models with DQ and ΔDQ as the main predictors for cross-sectional and longitudinal cognitive performance and (3) models with DQ and ΔDQ interacting with hAlzScore to determine cross-sectional and longitudinal cognitive performance. In addition, to test for clinical significance, the exposures and outcomes were transformed into z-scores. They were then run in the same mixed models in lieu of the unstandardised variables, and the effect sizes were noted. An effect size >0·2 was considered strong, while >0·1 was moderate.
To account for multiple testing, given that there were two exposures, type I error was reduced to 0·05/2 = 0·025 for main effects and for interaction terms for the mixed-effects regression models. Three-way interaction terms were deemed statistically significant at an α-error level of 0·05.
Results
Descriptive findings are outlined in Table 2. Women had higher HEI-2010 and DASH scores than men represented by means across visits (48·0 and 2·3 v. 43·6 and 1·4, P = 0·03 and P = 0·004), respectively. Other notable differences include current smoking status, current use of illicit drugs and BMI. Table 3 displays findings from the linear mixed-effects regression models for hAlzScore on cognitive test performance over time. After adjustment for multiple testing, none of the tests was associated with hAlzScore longitudinally, except Clock Command in men (0·04 (se 0·01), P = 0·01), showing a protective effect. However, hAlzScore was significantly associated with a decline in CVLT-DFR (−0·41 (se 0·14), P = 0·004) in men and BVRT (0·69 (se 0·26), P = 0·009) in women. Other longitudinal effects were inconsistent overall and within sex. Table 4 presents the associations between DQ and cognitive change by time. None of the tests survived multiple testing, except Trails B in women: longitudinal association with ΔDQ reflecting a worsening of performance (−0·04 (se 0·01), P = 0·01). We also conducted a sensitivity analysis with total energy intake (data not shown) that did not affect our current findings.
Table 2. Characteristics of Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) study participants (>50 years) by sex (mean across waves) (Mean values with their standard errors; percentages)
HEI-2010, Healthy Eating Index 2010; DASH, Dietary Approaches to Stop Hypertension; MAR, mean adequacy ratio; hAlzScore, HANDLS Alzheimer’s risk score; PCA, principal component analysis; HS, high school; PIR, poverty income ratio; NSAID, non-steroidal anti-inflammatory drugs.
* P value was based on independent-samples t test when row variable is continuous and χ 2 test when row variable is categorical.
† CVD includes self-reported stroke, congestive heart failure, non-fatal myocardial infarction or atrial fibrillation.
‡ Inflammatory conditions include multiple sclerosis, systemic lupus, gout, rheumatoid arthritis, psoriasis, thyroid disorder and Crohn’s disease.
§ Includes over the counter and prescription drugs in that category.
Table 3. Coefficient estimates for associations between cognitive test performance and Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) Alzheimer’s risk score (hAlzScore) by time, for HANDLS participants >50 years of age with complete and reliable cognitive test scores derived from mixed-effects linear regression models (β-Coefficients and standard errors)‡
MMSE, Mini-Mental State Examination; k, total number of observations/total number of groups per test; CVLT-List A, California Verbal Learning test – List A; CVLT-DFR, California Verbal Learning Test-Long-Delayed Free Recall; BVRT, Benton Visual Retention Test.
* P < 0·10, ** P < 0·05, *** P < 0·01.
Significant interaction with sex: † P < 0·10, †† P < 0·5.
‡ Continuous covariates were mean-centred.
Table 4. Coefficient estimates for associations between diet quality and cognitive change by time, for Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) participants >50 years of age with complete and reliable cognitive test scores derived from mixed-effects linear regression models (β-Coefficients and standard errors)†††
MMSE, Mini-Mental State Examination; CVLT-List A, California Verbal Learning test-List A; CVLT-DFR, California Verbal Learning Test-Delayed Free Recall; BVRT, Benton Visual Retention Test.
* P < 0·10, ** P < 0·05, *** P < 0·01.
Significant interaction between time and sex: † P < 0·10, †† P < 0·05.
Significant interaction between sex and diet: ‡ P < 0·10, ‡‡ P < 0·05.
Significant interaction between sex and diet (change): § P < 0·10, §§ P < 0·05.
Significant interaction between sex and diet (change) and time: || P < 0·10, |||| P < 0·05.
¶ Significant interaction between sex and diet and time (P < 0·10).
††† Continuous covariates were mean-centred.
‡‡‡ Represents change in diet quality over time (about 5 years from baseline).
§§§ Represents diet quality at baseline (time 0).
Online Supplementary Table S3 shows cross-sectional (baseline v. baseline) and longitudinal (baseline v. change, change v. change) associations between cognitive test scores and hAlzScore, and DQ (DQ and ΔDQ). Annual rate of change in the CVLT-List A was associated with an interaction between ΔDQ and hAlzScore in the total population (Time × ΔDQ × hAlzScore: 0·15 (se 0·06), P = 0·008) as well as in women (Fig. 2) (0·21 (se 0·08), P = 0·006), indicating protective effects of DQ at higher AD risk levels. No other associations were statistically significant after correcting for multiple testing.
Fig. 2. Predictive margins for California Verbal Learning test-List A by time, across levels of diet change, among women (n 122) with highest genetic risk for Alzheimer’s disease: mixed-effects linear regression models. 
, Diet change (mean − 1sd); 
, Diet change (mean + 1sd); 
, diet change (mean).
Finally, to tease apart the dietary index/indices driving the findings, we conducted two additional sensitivity analyses with just main findings from our principal component analyses. The results for DQ in cognition are presented in online Supplementary Table S4, while the results of gene × diet interactions are presented in online Supplementary Table S5. We found that all three indices had significant contributions to Trails B test scores over time. However, only HEI-2010 and DASH scores influenced the gene × diet interactions for CVLT-List A and delayed free recall.
Discussion
This study prospectively examined the relationship between change in DQ and a genetic risk for AD on cognition in urban-dwelling AA adults. Our findings indicated that improvements in DQ over time were associated with a slower rate of decline on a test of verbal memory particularly among AA women with higher genetic risk for AD (Fig. 2). The association was not present in men but persisted overall in mixed-sex analyses. No cross-sectional associations (initial diet and related findings) were detected in our present analyses, except for Trails A and B in women only.
AD is a progressive cognitive decline that diminishes social and occupational functioning. AD is typically characterised by memory deficits, cognitive deterioration, functional impairment in activities of daily living and neuropsychiatric symptoms(Reference Holston47). It has been poorly identified and assessed in AA(Reference Wilson, Capuano and Sytsma48,Reference Corsi, Di Raimo and Di Lorenzo49) , resulting in an escalating public health crisis as reflected by an increased prevalence of the disease in AA.
Examining gene variations may be one pioneering method to explain the pathophysiological and clinical symptoms observed in persons with AD, a multifactorial disorder. The pathogenesis of AD in AA elders may be related to the amyloid-β cascade and pathogenesis of neuropsychiatric symptoms. Several neuroanatomical structures and neurotransmitters are shared in the pathogenesis of AD and neuropsychiatric symptoms such as schizophrenia, major depression and personality alterations. These derive from abnormalities in the limbic system and frontal and temporoparietal regions with altered function of the serotonergic, noradrenergic and cholinergic systems in the brain. Collectively, these neurochemical and neuroanatomical changes can result in the clinical symptoms manifested in AA elders with AD. This theory of the pathogenesis of AD in AA elders with AD may also support the temporal nature of the clinical symptoms given the increased abnormalities in neurotransmitters and neuroanatomy specific to the limbic system. However, further analysis is warranted about this theory since it is based on the limited number of clinical symptoms reported and examined in AA elders with AD as well as indications of mixed pathologies(Reference Barnes, Leurgans and Aggarwal50).
In terms of the genetics of AD, ApoE ϵ4 increases the risk of both age-related cognitive decline and the transition from mild to severe cognitive impairment(Reference Mount, Ashley and Lah51). Moreover, there is evidence that AD patients who are ϵ4 carriers have a faster rate of cognitive decline(Reference Hendrie, Murrell and Baiyewu52,Reference Stewart, Russ and Richards53) , although the data are equivocal. A few studies have investigated this issue reporting that ϵ4 carriers exhibit a phenotype characterised by greater memory impairment(Reference Potter, Plassman and Burke54). In other words, AD patients who have memory complaints are significantly more likely to be ϵ4 carriers. In addition, greater memory deficits on formal neuropsychological testing have been observed in AD patients who are ϵ4 carriers. Studies on ApoE ϵ4 status and episodic memory have involved predominantly White samples except for Fillenbaum et al. who compared the effects of ϵ4 status on baseline cognitive functioning in AA v. White AD patients(Reference Barnes, Arvanitakis and Yu55). Our risk score in HANDLS (hAlzScore) contained two ApoE SNP which could elucidate the observed, long-term memory association in women. Although we have ApoE information on all 1024 genotyped HANDLS participants, the current analyses did not specifically focus on the overlapping subjects (those included in the final sample who also had complete ApoE data) as we continued with the risk score calculation. The lack of more current literature on racial difference in AD further justifies the need for studying ApoE in a unique study population such as ours.
Interestingly, and in contrast to our current finding, there were racial differences in cognitive abilities such that the ϵ4 allele was related to faster decline in semantic memory and working memory for Whites but not for AA.
Dietary modification may have the potential to reduce the risk of developing AD. A recent meta-analysis (n 34 168) showed that the highest Mediterranean diet score was associated with reduced risk of developing cognitive disorders (relative risk = 0·79, 95 % CI 0·70, 0·90)(Reference Wu and Sun56), while supplementation with olive oil or nuts was associated with improved cognitive function(Reference Rajaram, Valls-Pedret and Cofán57). A study that investigated a relationship between Southern diet (high in added fats, fried food, eggs, processed meats and sugar-sweetened beverages) and Prudent diet (rich in vegetables, fruit, cereals and legumes, whole grains, rice/pasta, fish, low-fat dairy products, poultry and water) in individuals at risk for AD found an association between Southern diet and reduced cognitive performance among AA(Reference Nutaitis, Tharwani and Serra58). In both Whites and AA adults, greater adherence to a Prudent dietary pattern was associated with better cognitive outcomes suggesting differential effects of diet on cognitive function in middle-aged individuals at high risk for AD. This suggests that diet could be a non-pharmaceutical tool to reduce cognitive decline in racially diverse populations(Reference Gorelick, Furie and Iadecola59). Mediterranean, DASH(Reference Wu, Song and Chen60–Reference Berendsen, Kang and van de Rest62) and MIND(Reference Omar63,Reference Cherian, Wang and Fakuda64) have all been linked to reduced risk of AD and lower cognitive decline in a recent publication(Reference van den Brink, Brouwer-Brolsma and Berendsen12). Suggested mechanisms include: olive serves as one of the building block components of MedDi and MIND diets and the exerted potential health beneficial might be suggested due to the presence of its bioactive constituents such as oleic acids and phenolic compounds in olives, for example, as well as the combined neuroprotective functions of the antioxidants, MUFA and PUFA.
Confidence in our findings is strengthened by several factors. First, we used a longitudinal design to ascertain temporality of these relationships while stratifying by sex that is important in cognitive decline. Second, we used a composite measure of eleven cognitive tests that assessed a range of cognitive abilities, reducing the opportunity for floor and ceiling effects and other sources of measurement error to affect results. Finally, the availability of a mean of repeated measurements of cognition per individual allowed us to simultaneously but separately model initial level of cognition and rate of change, thereby allowing us to more effectively adjust for the former while testing for sex differences in the latter.
Nevertheless, some study limitations should be noted. First, our final sample size after using multiple selection variables was rather small. We were also unable to determine the statistical power of our selected samples since the process in mixed models is more complex than in linear models and requires more assumptions(Reference Evangelos Kontopantelis, Parisi and Reeves65). It is also often estimated using simulations which are not always reliable. Second, although our models were adjusted for a wealth of potentially confounding covariates, causality cannot be inferred given the observational nature study and the possible role played by residual confounding. Third, outcome measures were only repeated up to twice over an average follow-up of 5 years, leaving room for improvement in studies with three or more time points. Fourth, although we performed our risk score calculation based on over 100 AD-related genes and reported SNP, hundreds of more SNP have been discovered since the Nature publication(Reference Bertram, McQueen and Mullin66), and we are unable to claim our list as comprehensive. Fifth, we excluded those <50 years to have greater variability in cognitive decline measures at the expense of statistical power with a larger sample size. Finally, no additional analyses were performed with complete ApoE allele status to further explore the associations.
This study aimed to investigate longitudinal associations of genetic risk for AD and DQ with cognitive outcomes, in a sample of <500 people. While we were well powered to do the study, we might have missed significant gene variations while creating our genetic risk score. It might be equally important to study who are <50 years in hopes of detecting some early changes that was outside the scope of this study. In addition, because of the projected growth of minority populations in the coming decades, larger multi-racial/ethnic studies of cognitive function in older people are needed.
Conclusions and implications
We conclude that among AA women with increased genetic risk for AD, a better-quality diet was associated with a slower rate of decline in verbal memory. It is evident that DQ and its change over time can impact memory in the long run, especially in people with higher risk for AD. Mechanistically speaking, the changes observed begin long before the detected impairments are manifest. While we cannot change the genetic risk for a disease, shifting to a better-quality diet offers possible long-term health benefits, as it has been well established in the literature. More studies are needed to investigate brain morphology and volume changes in relation to DQ, in an at-risk population for AD, over time.
Acknowledgements
This work was fully supported by the Intramural Research Program (grant no. ZIA-AG000195) of the National Institutes of Health, National Institute on Aging, NIA/NIH/IRP.
S. H. and M. A. B. designed the research (project conception, development of overall research plan, and study oversight); S. H. and M. A. B. conducted the research (hands-on conduct of the data analyses). S. H. prepared the manuscript (initial complete draft and all subsequent revisions). M. A. B., J. W., M. F. K., M. E. and A. B. Z. reviewed the content of the manuscript, partially prepared the manuscript, revised the manuscript and provided additional corrections. S. H. had primary responsibility for the final content. All authors read and approved the final manuscript.
The authors declare that there are no conflicts of interest.
The views expressed in this article are those of the author(s) and do not reflect the official policy of the Department of the Army/Navy/Air Force, Department of Defense, or the US Government.
Supplementary material
For supplementary material referred to in this article, please visit https://doi.org/10.1017/S0007114520001269
