Adherence to dietary patterns and risk of incident dementia: Findings from the Atherosclerosis Risk in Communities Study

Abstract

Background:

Previous studies have suggested that adherence to healthy dietary patterns during late life may be associated with improved cognition. However, few studies have examined the association between healthy dietary patterns during midlife and incident dementia.

Objective:

Our study aimed to determine the association between adherence to healthy dietary patterns at midlife and incident dementia.

Methods:

We included 13,630 adults from the Atherosclerosis Risk in Communities (ARIC) Study in our prospective analysis. We used food frequency questionnaire responses to calculate 4 dietary scores: Healthy Eating Index-2015 (HEI-2015), Alternative Healthy Eating Index-2010 (AHEI-2010), alternate Mediterranean (aMed) diet, and Dietary Approaches to Stop Hypertension (DASH). Participants were followed until the end of 2017 for incident dementia. Cox regression models adjusted for covariates were used to estimate risk of incident dementia by quintile of dietary scores.

Results:

Over a median of 27 years, there were 2,352 cases of incident dementia documented. Compared with participants in quintile 1 of HEI-2015, participants in quintile 5 (healthiest) had a 14% lower risk of incident dementia (hazard ratio, HR: 0.86, 95% confidence interval, CI: 0.74–0.99). There were no significant associations of incident dementia with the AHEI-2010, aMed, or DASH scores. There were no significant interactions by sex, age, race, education, physical activity, hypertension, or obesity.

Conclusion:

Adherence to the HEI-2015, but not the other dietary scores, during midlife was associated with lower risk of incident dementia. Further research is needed to elucidate whether timing of a healthy diet may influence dementia risk.

INTRODUCTION

It has been estimated that 47 million people worldwide lived with dementia in 2016 [1]. With an aging population, this number is projected to increase to at least 131 million people by the year 2050 [1]. The economic burden of dementia is also growing, and the total estimated cost of dementia worldwide is $818 billion and is expected to increase.

Prevention of dementia is critical due to the lack of effective treatments. Diet is a low-cost, modifiable, and potentially effective primary prevention strategy to reduce the risk and burden of dementia. While many studies have examined the association of individual nutrients and foods with cognition [2–4], fewer studies have examined dementia risk in relation to dietary patterns, which account for nutrient interactions and represent real-life behavior. Increasing evidence suggests that adherence to a Mediterranean-style diet may be associated with a reduced risk of cognitive decline [5–7]. This dietary pattern has been found to be associated with improvements in global cognition and several cognitive domains, such as episodic memory, working memory, delayed recall, immediate recall, attention, processing speed, and verbal fluency [8]. The cognitive benefit of the Mediterranean diet may work through cardiovascular mechanisms, as the diet has been found to be associated with reduced risk of cardiovascular disease and potentially obesity, hypertension, and dyslipidemia [9–12].

Although many of these previous studies examined cognitive function based on cognitive testing, fewer have assessed incident dementia. The outcome of dementia has the advantage over cognitive decline or function in that it captures cases that may be missed from cognitive assessments that rely on in-person visits. Furthermore, previous studies were conducted mostly among older populations and had short follow-up periods. Therefore, these studies may have missed a biologically relevant time window, as dementia has a long preclinical period. A recent study found that among 8,225 participants who were dementia-free, there was no association between adherence to a healthy dietary pattern during midlife and risk of subsequent dementia over a median follow-up period of 25 years [13]. Conversely, the Mediterranean-DASH Intervention for Neurodegenerative Delay (MIND) diet, Mediterranean diet, and Dietary Approaches to Stop Hypertension (DASH) diet have been found to be associated with lower risk of developing incident Alzheimer’s disease [14].

Other healthy dietary patterns include the Healthy Eating Index (HEI), Alternative Healthy Eating Index (AHEI), and DASH diet. These dietary patterns have some overlapping components and different scoring methods. They have been found to be associated with lower risk of cardiovascular disease and therefore may also reduce the risk of dementia as vascular risk factors have been found to be associated with dementia [15–18].

We aimed to examine whether adherence to healthy dietary patterns, assessed using the HEI-2015, AHEI-2010, alternate Mediterranean (aMed) diet, and DASH scores, was associated with incident dementia in the Atherosclerosis Risk in Communities (ARIC) study.

MATERIALS AND METHODS

Study population

Participants in this analysis were men and women from the ARIC study, a community-based prospective cohort of 15,792 participants from 4 U.S. communities (Washington County, Maryland; Forsyth County, North Carolina; Jackson, Mississippi; and Minneapolis suburbs, Minnesota) [19]. Participants were aged 45–64 years old when they were recruited in 1987–1989 (visit 1). They were also seen in-person in 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), 2011–2013 (visit 5), 2016–2017 (visit 6), 2018–2019 (visit 7), visit 8 is in progress, and had annual follow-up telephone interviews. The Institutional Review Boards of participating institutions approved the ARIC study and participants provided written informed consent at each site.

In the present analysis, participants who were neither black nor white (n=48) or black and from Minneapolis, Minnesota or Washington County, Maryland (n=55) were excluded due to small numbers resulting in a lack of adequate representation of these groups within these study centers. Additionally, participants who did not fill out more than 15 food frequency questionnaire (FFQ) responses (n=46), had extreme energy intakes (women: <500 or >3,500 kcal/d; men: <700 or >4,500 kcal/d [n=268]), had a history of stroke at baseline (n=267), were missing data on covariates of interest (n=1,234), or did not have sufficient data to calculate all 4 dietary scores (n=244), were excluded. The final study population included 13,630 participants.

Diet assessment and computation of diet scores

Diet was assessed at visit 1 (1987–1989) and visit 3 (1993–1995). The 66-item semi-quantitative FFQ, modified from the Willett questionnaire [20], has been previously validated in ARIC [21, 22]. Trained interviewers asked participants how often, on average, they consumed food items of a given portion size in the preceding year. Total energy intake and nutrient intakes were calculated using data from the US Department of Agriculture. Only visit 1 data was used if participants developed incident dementia prior to visit 3 or did not have dietary data available at visit 3; otherwise, the averages of food and nutrient responses from visits 1 and 3 were used [23]. Total energy intake was calculated using the same method.

The HEI-2015 score reflects the recommendations from the 2015–2020 US Dietary Guidelines for Americans [24]. It consists of 13 food and nutrient components that are scored based on energy-adjusted cutoffs (Supplemental Table 1). Possible scores range from 0 to 100.

The AHEI-2010 score includes foods and nutrients that have been found to be associated with risk of major chronic diseases [25]. There are 11 components that each range from 0 to 10, allowing for a total maximum score of 110 points (Supplemental Table 1).

The aMed score reflects food and nutrients common in a Mediterranean-style diet and has been adapted for the US population [26]. There are 9 components that are scored 0 or 1 based on sex-specific cutoffs; total possible score ranges from 0 to 9 (Supplemental Table 1).

The DASH score was created based on foods that were emphasized or discouraged in the DASH trial [27]. There are 8 components that are scored 1 to 5 based on quintiles of each component (Supplemental Table 1). The total score ranges from 8 to 40.

For our main analysis, we categorized participants by quintile of each score to assess trends and potential dose-response associations between diet and dementia.

Dementia ascertainment

Methods for ascertaining dementia in ARIC have been described in depth elsewhere [15, 28]. In brief, dementia was defined by at least one of the following criteria: 1) adjudicated dementia based on a complete neuropsychological battery, informant interviews and expert review at ARIC Neurocognitive Study visits (visits 5 and 6); 2) dementia determined from telephone interviews with participant or informant; or 3) dementia based on a prior hospitalization or death code for dementia. If participants met more than one of the criteria, the earliest date of dementia was used. Due to a possible lag in ascertainment of dementia identified by informant interviews, hospitalization, and death, six months was subtracted from the date dementia was ascertained by these methods.

Assessment of covariates

Age, sex, race, education, and physical activity were self-reported at visit 1 through an interviewer-administered questionnaire. Race and center were combined into an interaction variable to account for the unequal distribution of blacks and whites by each study site. Physical activity was measured via the Baecke questionnaire and exercise during leisure time was scored from 1 to 5 [29, 30]. Apolipoprotein (APOE) ε4 genotype and carriers were defined based on the number of ε4 alleles present (0, 1, or 2). Smoking status, alcohol status, and clinical variables were recorded at baseline. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Obesity was defined as BMI ≥30 kg/m². Plasma total cholesterol level was measured using enzymatic methods [31, 32]. Systolic blood pressure was calculated as the average of the second and third measurements, which were taken after 5 minutes of rest using a random-zero sphygmomanometer. Anti-hypertensive medication use and prevalent coronary heart disease were self-reported. Diabetes was defined as fasting glucose level ≥126 mg/dL, non-fasting glucose level ≥200 mg/dL, self-report of physician diagnosis, or use of anti-diabetic medications.

Cognitive testing, performed at visit 2, used the Delayed Word Recall Test, the Digit Symbol Substitution Test, and the Word Fluency Test. A z-score at visit 2 was calculated by averaging the 3 test-specific z-scores.

Statistical analysis

Descriptive statistics were used to compare baseline characteristics of study participants across quintiles of each dietary score. Cox regression models were used to quantify associations between dietary scores and incident dementia by estimating hazard ratios (HR) and 95% confidence intervals (CI). Follow-up time accumulated from visit 1 to the event date, date of death, or December 31, 2017, whichever occurred first. The proportional hazard assumption was assessed via Schoenfeld residuals test. Model 1 was adjusted for age, sex, race-center, total energy intake, education level, APOE ε4 genotype, smoking status, physical activity, and alcohol status (for HEI-2015 and DASH scores since they do not include alcohol in their scoring). Model 2 was further adjusted for BMI, total cholesterol, systolic blood pressure, antihypertensive medication use, diabetes status, and history of coronary heart disease. Model 3 was additionally adjusted for 3-test cognitive score at visit 2. Model 3 was considered our primary model. P for trend was assessed using the median score of each quintile. We used likelihood ratio tests to test for interactions by potential effect modifiers (sex, age, race, education, physical activity, hypertension, and obesity). We also implemented competing risk models to account for the competing risk of death [33]. Stata version 14 was used for all analyses (StataCorp, College Station, Texas.)

RESULTS

Baseline characteristics of participants by quintile of HEI-2015 score are shown in Table 1. Participants who had higher HEI-2015 scores, indicating healthier dietary intake, were more likely to be female, have a college education, be a never smoker, never drinker, be more physically active, have a lower total energy intake, and have a higher baseline cognitive 3-test z-score compared with participants with the lowest HEI-2015 scores. Those in higher quintiles of HEI-2015 were also more likely to have diabetes, take antihypertensive medication, and have a history of CHD compared with participants in quintile 1. Characteristics were mostly similar by quintile of AHEI-2010 (Supplemental Table 2) and aMed (Supplemental Table 3), with some exceptions. Some noteworthy exceptions were that those at higher levels of adherence to AHEI-2010 were more likely to be current drinkers, consumed higher levels of total energy intake, and were not different on the basis of anti-hypertensive medication use (Supplemental Table 2). Those in higher quintiles of aMed were not different according to sex, were more likely to consume alcohol, and consumed higher levels of total energy intake (Supplemental Table 3). Characteristics were very similar by quintile of DASH (Supplemental Table 4).

Table 1.

Baseline characteristics of ARIC participants by quintile of HEI-2015 score^a.

	HEI-2015 Quintile
Characteristic	Q1 n=2,726	Q2 n=2,726	Q3 n=2,726	Q4 n=2,726	Q5 n=2,726	Pb
HEI-2015 score	60 ± 4	67 ± 1	72 ± 1	76 ± 1	83 ± 3	<0.001
Age, years	54 ± 6	54 ± 6	54 ± 6	54 ± 6	55 ± 6	<0.001
Female, %	41	51	56	62	68	<0.001
Black, %	25	28	28	24	23	<0.001
College education, %	25	32	38	40	46	<0.001
Smoking status, %						<0.001
Never smoker	30	39	44	47	49
Former smoker	30	31	32	32	36
Current smoker	40	29	24	21	15
Alcohol status, %						<0.001
Never drinker	20	24	26	27	28
Former drinker	22	18	17	18	17
Current drinker	58	58	57	55	55
Physical activity index (1–5)c	2.2 ± 0.7	2.4 ± 0.8	2.4 ± 0.8	2.5 ± 0.8	2.6 ± 0.8	<0.001
APOE ε4 (1–2 alleles), %	29	29	31	30	34	0.004
BMI, kg/m2	27 ± 5	28 ± 5	28 ± 6	28 ± 5	27 ± 5	0.03
Total cholesterol, mg/dL	213 ± 42	213 ± 41	213 ± 40	216 ± 42	217 ± 44	<0.001
Diabetes, %	9	11	12	12	13	<0.001
Systolic blood pressure, mmHg	121 ± 19	122 ± 19	121 ± 18	121 ± 18	120 ± 18	0.2
Antihypertensive medication, %	25	27	32	32	32	<0.001
Coronary heart disease, %	4	4	4	4	6	<0.001
Average total energy intake, kcal/d	1,617 ± 593	1,566 ± 541	1,450 ± 496	1,434 ± 483	1,326 ± 414	<0.001
Visit 2 global cognitive z-score	−0.1 ± 1	−0.1 ± 1	0 ± 1	0.1 ± 1	0.2 ± 1	<0.001

^aValues for categorical variables are given as percentage; for continuous variables, mean ± standard deviation. Higher quintiles indicate greater adherence to a healthy dietary pattern.

^bCategorical variables were analyzed using chi-square test. Continuous variables were analyzed using analysis of variance (ANOVA) test.

^cPhysical activity index score was calculated based on intensity and time of sport and exercise during leisure time; 1-lowest and 5-highest.

APOE, apolipoprotein E; ARIC, Atherosclerosis Risk in Communities study; BMI, body mass index; CHD, coronary heart disease; HEI, Healthy Eating Index; mg/dL; milligrams per deciliter; mmHg, millimeters of mercury; Q, quintile.

There were 2,352 cases of incident dementia over a median follow-up of 27 years. In model 1, there was a significant association between HEI-2015 score and risk of incident dementia comparing quintile 5 to quintile 1 (HR: 0.86, 95% CI: 0.75–0.99) (Table 2). The association of the other quintiles (2, 3, 4) of HEI-2015 with dementia compared with quintile 1 were not significant, but a test for trend was statistically significant. After additionally adjusting for clinical covariates in Model 2, the association remained significant. In model 3, which additionally adjusted for the 3-test cognitive score at visit 2, participants in quintile 5 had a 14% lower risk of dementia compared with participants in quintile 1 (HR: 0.86, 95% CI: 0.74–0.99), but the P for trend was non-significant (P=0.1). There were no significant associations for AHEI-2010, aMed, or DASH scores and incident dementia (Table 2). Comparing quintile 5 to quintile 1 in model 3, the association between AHEI-2010 (HR: 1.04, 95% CI: 0.91–1.20), aMed (HR: 1.01, 95% CI: 0.88–1.16), and DASH (HR: 1.10, 95% CI: 0.96–1.26) scores were non-significant. In competing risk models to account for the competing risk of death, per standard deviation increase in each score, the hazard ratio was 0.95, CI% 0.87–1.04 for HEI-2015, 1.06, 95% CI: 0.98–1.16 for AHEI-2010, 1.01, 95% CI: 0.92–1.11 for aMed, and 1.00, 95% CI: 0.92–1.09 for DASH.

Table 2.

Hazard ratio (95% CI) of dementia according to quintile of HEI-2015, AHEI-2010, aMed, and DASH scores^a.

HEI-2015
	Q1: 61b n=2,726	Q2: 67 n=2,726	Q3: 72 n=2,726	Q4: 76 n=2,726	Q5: 82 n=2,726	Pc
No. events	445	447	474	474	512
Model 1	1 (ref.)	0.89 (0.78–1.02)	0.91 (0.80–1.04)	0.89 (0.78–1.02)	0.86 (0.75–0.99)	0.05
Model 2	1 (ref.)	0.88 (0.77–1.00)	0.89 (0.78–1.01)	0.85 (0.74–0.97)	0.84 (0.73–0.96)	0.01
Model 3	1 (ref.)	0.86 (0.75–1.00)	0.89 (0.77–1.02)	0.90 (0.78–1.04)	0.86 (0.74–0.99)	0.1
AHEI-2010
	Q1: 37 n=2,726	Q2: 45 n=2,726	Q3: 51 n=2,726	Q4: 58 n=2,726	Q5: 67 n=2,726	P
No. events	427	499	457	450	519
Model 1	1 (ref.)	1.05 (0.93–1.20)	0.97 (0.85–1.11)	0.90 (0.78–1.03)	1.00 (0.88–1.15)	0.4
Model 2	1 (ref.)	1.06 (0.93–1.20)	0.94 (0.84–1.09)	0.88 (0.77–1.00)	0.98 (0.86–1.12)	0.2
Model 3	1 (ref.)	1.06 (0.93–1.22)	0.99 (0.86–1.13)	0.92 (0.80–1.06)	1.04 (0.91–1.20)	0.9
aMed
	Q1: 2 n=2,545	Q2: 3 n=2,443	Q3: 4 n=2,780	Q4: 5 n=2,524	Q5: 7 n=3,338	P
No. events	404	402	467	454	625
Model 1	1 (ref.)	1.00 (0.87–1.15)	0.97 (0.85–1.11)	0.96 (0.84–1.10)	0.98 (0.86–1.11)	0.6
Model 2	1 (ref.)	0.98 (0.86–1.13)	0.95 (0.83–1.09)	0.92 (0.81–1.06)	0.95 (0.83–1.08)	0.3
Model 3	1 (ref.)	1.04 (0.90–1.20)	1.02 (0.88–1.17)	0.99 (0.86–1.15)	1.01 (0.88–1.16)	0.9
DASH
	Q1: 18 n=3,431	Q2: 22 n=2,710	Q3: 24 n=2,037	Q4: 27 n=2,781	Q5: 30 n=2,671	P
No. events	555	450	349	481	517
Model 1	1 (ref.)	1.03 (0.91–1.17)	1.04 (0.91–1.20)	0.95 (0.83–1.08)	1.05 (0.92–1.20)	0.9
Model 2	1 (ref.)	1.01 (0.89–1.15)	1.00 (0.87–1.14)	0.90 (0.79–1.02)	1.02 (0.89–1.16)	0.6
Model 3	1 (ref.)	1.06 (0.93–1.21)	1.03 (0.89–1.20)	0.97 (0.85–1.11)	1.10 (0.96–1.26)	0.5

^aCox proportional hazards models to estimate hazard ratios (HR) and 95% confidence intervals (CI). Model 1 was adjusted for age, sex, race-center, total energy intake, education level, APOE ε4 genotype, smoking status, physical activity, and alcohol status (for HEI-2015 and DASH scores). Model 2 was additionally adjusted for BMI, total cholesterol, systolic blood pressure, antihypertensive medication use, diabetes status, and history of coronary heart disease. Model 3 included model 2 covariates in addition to global cognitive score at visit 2. AHEI, Alternative Healthy Eating Index; aMed, alternate Mediterranean diet; DASH, Dietary Approaches to Stop Hypertension; HEI; Healthy Eating Index; no, number; ref, reference.

^bMedian score of tertile.

^cTrend was tested using the median value within each tertile.

There were no consistent interactions between the dietary scores and dementia by sex, age, race, education, physical activity, hypertension, or obesity (all P for interactions>0.05).

DISCUSSION

In our study population of 13,630 black and white participants from four U.S. communities, we found that a better HEI-2015 score was weakly associated with less incident dementia, but there were no significant associations for AHEI-2010, aMed, or DASH scores. For all four scores, there were no significant trends in dementia risk by quintile of dietary scores and the results were consistent by subgroups.

Our study results are consistent with other studies that also found a null association between dietary patterns and dementia. A previous analysis in the ARIC study, using derived dietary patterns from principal components analysis (a Western diet pattern and prudent diet pattern), found no significant associations with 20-year change in cognitive function or with incident dementia [34]. An analysis of 8,225 participants from the Whitehall II study found that there was no association between a healthy diet at midlife, assessed by the AHEI-2010 score and derived dietary patterns, and incident dementia over a median follow-up of 24.8 years [13]. Other studies conducted in Bordeaux, France and Minnesota, USA that assessed adherence to the Mediterranean diet did not find an association with incident dementia among dementia-free older adults [35, 36].

However, our results contradict previous observational studies that have suggested an inverse association between healthy dietary patterns and cognitive decline. A meta-analysis of 32 studies from 25 unique cohorts concluded that most observational studies conducted found that adherence to a Mediterranean diet was associated with improved cognitive function, decreased risk of cognitive impairment, and decreased risk of dementia [5]. However, most previous studies were conducted among older populations (>65 years) and had short follow-up periods [37–43]. A recent review examined different dietary patterns [e.g. the Mediterranean diet, DASH diet, and Mediterranean/DASH Intervention for Neurodegenerative Delay (MIND) diet] and included 56 total studies [6]. Although many of the observational studies suggested that better adherence to these dietary patterns was associated with improved cognitive function, the evidence for dementia was too limited to draw firm conclusions. Furthermore, the results from randomized controlled trials were mixed [44–47].

Several reasons may explain why our findings differed from those of some previous studies. Many of the previous studies were conducted among older adults (>65 years) and assessed diet during late life, while our study assessed diet during midlife. Since diet was assessed during midlife and participants were followed for a median of 27 years, dietary behavior may have changed over time during the preclinical phase of dementia [48]. If participants were experiencing cognitive decline later in life, they may have changed from a healthy eating pattern to an unhealthy pattern perhaps due to the difficulty with preparing healthy foods because of impaired cognitive function. Therefore, the studies conducted among elderly adults may have found an inverse association between healthy diet and dementia due to reverse causation.

We found that the HEI-2015, but not the other scores, was associated with lower risk of dementia. The key differences between the HEI-2015 and the other patterns are the individual components for refined grains, all dairy, total protein, and added sugars. Refined grains and added sugars are scored negatively but are not penalized in the other scores. Fish is accounted for in both the total protein and seafood/plant protein components in the HEI-2015. In the aMed score, fish is its own category and it is not included in the AHEI-2010 or DASH scores. Fish is a high source of omega-3 fatty acids and vitamin D, and is hypothesized to be a key factor in previous associations between Mediterranean diet and improved cognition [49]. Omega-3 fatty acids reduce inflammation and oxidative stress, which are both associated with cognitive function [50, 51]. There is also evidence that severe deficiency of vitamin D, found in meat and fish sources, may increase risk of dementia [52, 53]. Olive oil and nuts, which contain high amounts of phenolic compounds, have been hypothesized to improve cognition through improving cerebrovascular blood flow and stimulation of neurogenesis [54]. The MIND diet, which emphasizes brain healthy foods such as berries, leafy green vegetables, fish, and olive oil, has been found to be associated with reduced risk of Alzheimer’s disease risk and cognitive decline (tertile 3 vs. tertile 1 HR: 0.47, 95% CI: 0.26–0.76) [14, 55]. Due to the questions in the ARIC FFQ not specifying berries or olive oil, we could not calculate the MIND score for comparison.

There were several limitations of our study. Diet was self-reported and therefore is subject to measurement error. However, the FFQ was administered by trained interviewers using a standard protocol. Diet was based on an average of dietary intake from visits 1 and 3, but we did not examine change in diet quality and risk of dementia. Future studies should examine how change in diet may be associated with dementia. There is also a possibility of residual confounding; however, covariates were measured using rigorous methods and were adjusted for in multivariable regression models. Additionally, there was potential misclassification of dementia cases since not all cases were adjudicated by the expert committee.

Our study also had several strengths. The study included a long follow-up time, which allowed us to evaluate the effects of midlife diet and capture more cases of incident dementia than previous studies given the lengthy preclinical phase of dementia. We leveraged repeated measurements of dietary intake by using the average of food and nutrient intakes from visits 1 and 3, allowing for a more precise measure. Lastly, we had rigorous ascertainment of dementia, and had a formal adjudication process for cases captured from in-person visits.

In summary, the results of our study suggest that adherence to the HEI-2015 dietary pattern during midlife may be associated with lower risk of dementia in later life. More research is warranted to further elucidate the relationship between midlife diet and incident dementia.