Plant-Based Dietary Patterns and Risk of Alzheimer Disease and Related Dementias in the Multiethnic Cohort Study

Abstract

Background and Objectives

Plant-based diets have been linked to slower cognitive decline, but data on long-term dietary changes and from diverse populations are limited. The primary aim of this study was to examine plant-based dietary patterns and their change over time in relation to Alzheimer disease and related dementias (ADRDs).

Methods

This prospective longitudinal analysis of the Multiethnic Cohort Study, based in Hawaii and California (primarily Los Angeles County), included data on African American, Japanese American, Latino, Native Hawaiian, and White participants who completed food frequency questionnaires at baseline (1993–1996; age 45–75 years) and at 10-year follow-up (2003–2008) and whose Medicare claims were linked to identify incident ADRDs. A priori indices for the overall plant-based diet index (PDI), the healthful plant-based diet index (hPDI), and the unhealthful plant-based diet index (uPDI) were analyzed in Cox regression models for ADRD.

Results

The analysis included 92,849 participants (mean age 59.2 years, 55.1% female, 21,478 with ADRDs) for the baseline diet and 45,065 participants (8,360 with ADRDs) for the 10-year dietary change. For the baseline diet, comparing the highest vs lowest quintile, PDI and hPDI were associated with 12% (hazard ratio [HR] 0.88; 95% CI 0.85–0.92) and 7% (HR 0.93; 95% CI 0.89–0.97) lower risks of ADRD, respectively, whereas uPDI was related to a 6% higher risk (HR 1.06; 95% CI 1.01–1.10). For the dietary change over time, the strongest association with ADRD was observed for uPDI rather than for PDI or hPDI. Compared with those with a stable score (<0.5 SD change), participants with a large increase in uPDI (≥1 SD) showed a 25% higher risk (HR 1.25; 95% CI 1.15–1.36) and those with a large decrease in uPDI showed an 11% lower risk (HR 0.89; 95% CI 0.84–0.94). The associations between the plant-based diet indices and ADRD were generally similar by age group (<60 vs ≥60 years at baseline), race and ethnicity, or APOE ε4 carrier status.

Discussion

These findings suggest that adopting plant-based diets, specifically refraining from low-quality plant-based diets, even at an older age, is associated with a lower risk of ADRDs.

Introduction

Plant-based diets have been beneficial in reducing the risk of cardiometabolic diseases^1-5 and may have similar benefits for the prevention of Alzheimer disease and related dementias (ADRDs),⁶ given the established role of cardiometabolic conditions in ADRD etiology.⁷

Vegetarian and vegan diets in past research excluded some or all animal foods and did not differentiate the quality of plant foods. To assess the health benefits of plant foods consumed at varying proportions of the diet and to account for the quality of plant foods, plant-based diet indices have been developed, namely, the overall plant-based diet index (PDI), the healthful plant-based diet index (hPDI), and the unhealthful plant-based diet index (uPDI).^8,9 Several studies of the plant-based diet indices have shown beneficial associations for slower cognitive decline,¹⁰ providing more consistent evidence than strict vegetarian and vegan diets.¹¹

Yet, data are still limited and inconsistent on the association of the plant-based diet indices with dementia outcomes. A study in the UK Biobank reported that dementia risk was reduced by 18% with hPDI and increased by 29% with uPDI.¹² However, the Rotterdam Study found no overall association between the plant-based diet indices and dementia risk.¹³ These analyses were based on a 1-time assessment of diet and did not examine dietary changes over time. There is also a need to expand the studies to multiple racial and ethnic populations, where the analysis of a wider range of dietary practices may inform us of the underlying associations and likely reasons for large racial and ethnic disparities in dementia/ADRDs.

In the Multiethnic Cohort (MEC) Study, we have observed large racial and ethnic differences in ADRD risk: compared with the White group, risks were 37%–39% higher in African American, 21%–30% higher in Native Hawaiian, similar in Latino, and 11%–36% lower in Asian American, depending on sex and covariate adjustment.^14,15 We also have found that healthy dietary patterns, measured as adherence to dietary recommendations, and their improvement over time were associated with a reduced risk of ADRD.¹⁶ In this study, we examined the association of plant-based dietary patterns assessed by the a priori indices and their 10-year changes with subsequent ADRD risk and whether the associations differed by sex, age, race and ethnicity, and APOE ε4 status.

Methods

Study Population

The MEC Study was designed to evaluate lifestyle and genetic factors of cancer and other chronic diseases across several common racial and ethnic groups in the United States.¹⁷ Cohort members were recruited in Hawaii and California (mainly Los Angeles County) between 1993 and 1996 based on drivers’ license files as a primary sampling source and with voters’ registration file and Health Care Financing Administration file as additional sources. African American, Japanese American, Latino, Native Hawaiian, and White men and women aged 45–75 years were targets for recruitment and entered the cohort by completing a mailed, comprehensive questionnaire. Respondents identified their racial or ethnic background on the questionnaire using the following list of categories: Black or African American, Chinese, Filipino, Hawaiian, Japanese (includes Okinawan), Korean, Mexican or other Hispanic, White or Caucasian, or other. A second comprehensive questionnaire was mailed to all participants who were alive at the 10-year follow-up survey between 2003 and 2008. The cohort members were linked to Medicare Fee-for-Service (FFS) claims data (1999–2019) to identify chronic diseases and to state death files and the National Death Index to ascertain vital status. This study was limited to the MEC participants of the 5 targeted racial and ethnic groups who have been Medicare FFS beneficiaries.

Dietary Assessment and Covariates

A quantitative food frequency questionnaire (QFFQ) assessed participants’ usual diet during the previous year at cohort baseline in 1993–1996 and at 10-year follow-up in 2003–2008. The QFFQ on more than 180 food items was developed based on 3-day measured food records¹⁷ and validated against 3 unannounced 24-hour dietary recalls in a calibration study showing satisfactory correlations for nutrient densities (0.57–0.74).¹⁸ The 10-year follow-up QFFQ also showed a good validity (average correlation 0.58) and a high reproducibility of the baseline QFFQ (average correlation 0.73) in a second calibration study. ¹⁹ The baseline and 10-year follow-up questionnaires also collected sociodemographic data and health-related and lifestyle-related information, including education, marital status, medical and medication history, smoking history, sleep and physical activity, and body weight and height.

Plant-Based Diet Indices

Three plant-based diet indices, PDI, hPDI, and uPDI, were calculated based on the food intake data from the QFFQ, following the scoring method previously developed.^8,9 In the MEC Study, 16 food groups comprised all 3 plant-based diet indices.^20-22 These food groups were categorized into 7 healthy plant foods (whole grains, fruits, vegetables, vegetable oils, nuts, legumes, and tea/coffee), 4 less healthy plant foods (refined grains, fruit juices, potatoes, and added sugars), and 5 animal foods (animal fats, dairy, eggs, fish/seafood, and meat), based on their reported associations with health outcomes, such as cardiovascular disease, type 2 diabetes, certain cancers, and intermediate conditions.^8,9 For each of the 16 food groups used to calculate all 3 indices, daily intake in grams per 1,000 kcal was divided into sex-specific quintiles and assigned a score from 1 to 5. The total score for each index was calculated as the sum of the scores across 16 food groups. The theoretically plausible range is 16–80 for all 3 indices. However, the scoring systems are different across the indices. For PDI, higher quintiles of both healthy and unhealthy plant foods were given higher scores. For hPDI, higher quintiles of healthy plant foods received higher scores, whereas higher quintiles of less healthy plant foods received lower scores. For uPDI, higher quintiles of less healthy plant foods received higher scores, whereas higher quintiles of healthy plant foods received lower scores. For all 3 indices, animal foods were scored inversely.

Outcome Ascertainment

ADRDs were identified using the ICD-9 and ICD-10 codes in the Medicare FFS claims data, based on 2 or more separate claims; the FFS claims data are available for research since 1999. For the ADRD definition, as previously described, ¹⁴ we combined methods recommended by Medicare,²³ an earlier analysis of dementia subtypes based on Medicare FFS data,²⁴ and the Alzheimer’s Association²⁵ to exclude certain dementias that are not considered as ADRDs (alcohol-induced dementia, drug-induced dementia, Jakob-Creutzfeldt disease, and Huntington disease) and include common ADRD subtypes, such as Lewy body dementia (LBD or dementia with Lewy bodies). We categorize the ADRD subtypes (ICD-9/ICD-10 codes) into AD-only (331.0/G30.0, G30.1, G30.8, and G30.9), AD of mixed etiology (AD with any other subtypes), vascular dementia (VD)-only (290.40, 290.41, 290.42, and 290.43/F01.50), LBD-only (331.82/G31.83), frontotemporal dementia ICD-9 codes (numbers only)/ICD-10 codes (with an alphabet prefix) (331.11 and 331.19/G31.01 and G31.09), and degenerative dementia not otherwise specified (NOS) (290.0, 290.20, 290.21, 290.3, 290.10, 290.11, 290.12, 290.13, 290.9, and 294.20/F03.90; 294.21/F03.91; 294.10/F02.80; 294.11/F02.81; and 331.2/G31.1). We did not include frontotemporal dementia in the stratified ADRD subtype analysis because of small numbers.

Statistical Analysis

We compared selected characteristics of participants in the lowest, middle, and highest quintiles of the plant-based diet indices. Cox proportional hazards models with age as the time metric were applied to estimate hazard ratios (HRs) and 95% CIs for ADRD risk associated with the plant-based diet indices in quintiles. Follow-up for incident ADRDs began 2 years after either the earliest Medicare FFS linkage (January 1, 1999, for individuals who were already on Medicare) or the individual Medicare enrollment (for beneficiaries who have signed up since 1999), with the 2-year latency applied to capture stable Medicare FFS beneficiaries.¹⁴ Follow-up ended on the first claim date for ADRD, the date of death, or the analysis censor date (December 31, 2019), whichever came first. Tests for trend in associations across the diet quintiles were performed by modeling the sex-specific, race-specific, and ethnicity-specific median values within each quintile as a continuous variable. The proportional hazards assumption was verified by the Schoenfeld residual method.²⁶

All models were adjusted for sex, race and ethnicity, average Medicare usage as outpatient claims (average number of claims <1, ≥1 per year), and history of diabetes (yes, no) as strata variables, while covariates included age at cohort entry (years), age at Medicare surveillance start (years), average Medicare usage as inpatient claims (average number of claims <1, ≥1 per year), education (years), marital status (married, not married), history of hypertension (yes, no), heart disease (yes, no) or stroke (yes, no), smoking history (never, former, current), body mass index (BMI, kg/m²), physical activity (hours spent on moderate-to-vigorous activities per day), sleep duration (6–8 h/d or not), total energy intake (log transformed kcal/d), and alcohol consumption (g/d). All models were also adjusted for the baseline neighborhood socioeconomic status (SES, upper 2 vs lower quintiles) using a robust sandwich variance estimate, aggregated over the 1990 Census block groups, to account for potential area-level correlations.²⁷

In the dietary change analysis, to make the plant-based diet indices comparable between the 2 surveys, we applied the quintile cutpoints from the baseline indices to calculate scores at 10-year follow-up. Based on the SD of score change in each index, we categorized participants into those with a large decline (≥1 SD decrease), a moderate decline (0.5–<1 SD decrease), a stable pattern (<0.5 SD change), a moderate increase (0.5–<1 SD increase), and a large increase (≥1 SD increase), while further adjusting for baseline scores. We also examined the 16 component food groups for 10-year change in categories of a large decrease (moved down 2 or more quintiles over time), a moderate decrease (moved down 1 quintile), no change (stayed in the same quintile), a moderate increase (moved up 1 quintile), and a large increase (moved up 2 or more quintiles).

Heterogeneity in the diet-ADRD association across subgroups of participants was tested based on the Wald statistics for the cross-product term between the given dietary trend variable and the subgroup indicator. Restricted cubic splines were used to nonparametrically examine a possible nonlinear relationship between the indices and ADRD risk.²⁸ All p values were 2-sided, and all statistical analyses were performed using SAS version 9.4 (SAS Institutes, Inc., Cary, NC).

Standard Protocol Approvals, Registrations, and Patient Consents

The study was approved by the institutional review boards (IRBs) of the University of Hawaii (CHS9575) and the University of Southern California (HS-17-00714). The IRBs considered informed consent to be implied through the return of the baseline questionnaire, which was mailed to potential participants along with a cover letter explaining the study.

Data Availability

Deidentified participant data for the findings of this study will be made available on request, pending application to and approval by the Multiethnic Cohort Research Committee (uhcancercenter.org/for-researchers/mec-data-sharing).

Results

Among a total of more than 215,000 MEC participants, 117,031 participants (55.5% female) were Medicare FFS beneficiaries of the 5 targeted racial and ethnic groups. At baseline, those in the Medicare subset were more likely to be younger (59.3 vs 61.0 years), more likely to be Japanese American or White, more educated, more physically active, and more likely to live in a neighborhood with higher SES, while being less likely to have cardiometabolic conditions (eTable 1). As is customary in Medicare analyses of ADRDs,²⁹ we excluded participants who were younger than 64 years at the start of Medicare surveillance (n = 4,748; mostly enrolled for disabilities), who had less than 2 years on Medicare FFS (n = 2,750; transient beneficiaries), who had an ADRD claim within the first 2 years of MEC-Medicare linkage (n = 2,147; likely prevalent cases), and who reported AD on the follow-up questionnaire in 1998–2002 (n = 222). We also excluded individuals who reported an implausible diet based on energy and macronutrient intakes (n = 3,922).³⁰ Specifically, we excluded all individuals whose total energy intake fell beyond the range of mean ± 3 robust SDs, computed based on the middle 80%, and repeated a similar process for fat, protein, and carbohydrate intakes. After further excluding those with missing covariates (n = 10,393), a total of 92,849 participants remained in the final analysis (eFigure 1).

During an average follow-up of 10.9 years, 21,478 individuals with incident ADRDs were identified among 92,849 participants (mean age 59.2 years at cohort baseline, 55.1% female). Those in the highest quintile of PDI or hPDI were more likely to be older, more educated, married, physically active, nonsmokers, less obese, sleeping an optimal number of hours, and consuming less energy than those in the lowest quintile (Table 1). By contrast, participants in the highest quintile of uPDI tended to be younger, less educated, less likely to have diabetes, and more likely to have higher energy and alcohol intakes than those in the lowest quintile.

Table 1.

Characteristics of Participants in the Bottom, Middle, and Top Quintiles of 3 Plant-Based Diet Indices in the Multiethnic Cohort Study

Characteristicsa	Total	Overall plant-based diet index			Healthful plant-based diet index			Unhealthful plant-based diet index
		Q1	Q3	Q5	Q1	Q3	Q5	Q1	Q3	Q5
No. of participants	92,849	20,904	19,997	17,233	20,546	17,297	19,799	19,022	18,860	17,462
Age at cohort entry, y, mean ± SD	59.2 ± 8.4	57.7 ± 8.3	59.3 ± 8.4	60.9 ± 8.3	56.8 ± 8.3	59.3 ± 8.3	61.4 ± 8.1	60.4 ± 8.2	59.3 ± 8.4	57.8 ± 8.5
Age at Medicare surveillance initiation, y, mean ± SD	69.6 ± 4.8	68.8 ± 4.4	69.6 ± 4.8	70.5 ± 5.1	68.5 ± 4.2	69.6 ± 4.8	70.6 ± 5.1	70.1 ± 4.9	69.6 ± 4.8	69.0 ± 4.5
Female, n (%)	51,185 (55.1)	11,448 (54.8)	11,172 (55.9)	9,253 (53.7)	11,582 (56.4)	9,330 (53.9)	11,058 (55.9)	10,511 (55.3)	10,309 (54.7)	9,664 (55.3)
Race/ethnicity, n (%)
African American	11,277 (12.1)	3,356 (16.1)	2,345 (11.7)	1,563 (9.1)	2,896 (14.1)	2,036 (11.8)	2,077 (10.5)	2,544 (13.4)	2,282 (12.1)	1,871 (10.7)
Japanese American	31,157 (33.6)	5,357 (25.6)	7,074 (35.4)	6,859 (39.8)	5,888 (28.7)	5,601 (32.4)	7,949 (40.1)	5,730 (30.1)	6,236 (33.1)	6,570 (37.6)
Latino	17,348 (18.7)	3,760 (18.0)	3,707 (18.5)	3,168 (18.4)	3,452 (16.8)	3,567 (20.6)	3,384 (17.1)	3,254 (17.1)	3,641 (19.3)	3,234 (18.5)
Native Hawaiian	7,024 (7.6)	2,321 (11.1)	1,405 (7.0)	808 (4.7)	2,215 (10.8)	1,220 (7.1)	1,085 (5.5)	1,295 (6.8)	1,384 (7.3)	1,581 (9.1)
White	26,043 (28.0)	6,110 (29.2)	5,466 (27.3)	4,835 (28.1)	6,095 (29.7)	4,873 (28.2)	5,304 (26.8)	6,199 (32.6)	5,317 (28.2)	4,206 (24.1)
Medicare use over follow-up, n (%)
Inpatient service (≥1/y)	3,641 (3.9)	992 (4.7)	797 (4.0)	526 (3.1)	931 (4.5)	650 (3.8)	655 (3.3)	764 (4.0)	678 (3.6)	731 (4.2)
Outpatient service (≥1/y)	54,591 (58.8)	12,047 (57.6)	11,677 (58.4)	10,397 (60.3)	11,651 (56.7)	10,056 (58.1)	12,101 (61.1)	11,726 (61.6)	11,099 (58.8)	9,900 (56.7)
Education (>12th grade), n (%)	56,959 (61.3)	12,360 (59.1)	12,371 (61.9)	10,870 (63.1)	12,846 (62.5)	10,467 (60.5)	12,255 (61.9)	12,417 (65.3)	11,518 (61.1)	10,134 (58.0)
Higher nSES, n (%)	44,888 (48.3)	9,410 (45.0)	9,808 (49.0)	8,713 (50.6)	9,871 (48.0)	8,328 (48.1)	9,813 (49.6)	9,319 (49.0)	9,176 (48.7)	8,391 (48.1)
Married, n (%)	64,506 (69.5)	13,580 (65.0)	14,124 (70.6)	12,469 (72.4)	13,866 (67.5)	12,155 (70.3)	13,946 (70.4)	13,113 (68.9)	13,295 (70.5)	11,927 (68.3)
History of hypertension, n (%)	33,777 (36.4)	7,795 (37.3)	7,250 (36.3)	6,122 (35.5)	7,340 (35.7)	6,354 (36.7)	7,186 (36.3)	6,971 (36.6)	6,867 (36.4)	6,344 (36.3)
History of heart disease, n (%)	6,412 (6.9)	1,385 (6.6)	1,298 (6.5)	1,356 (7.9)	1,199 (5.8)	1,176 (6.8)	1,577 (8.0)	1,438 (7.6)	1,336 (7.1)	1,053 (6.0)
History of stroke, n (%)	1,774 (1.9)	418 (2.0)	328 (1.6)	350 (2.0)	348 (1.7)	305 (1.8)	437 (2.2)	389 (2.0)	367 (1.9)	297 (1.7)
History of diabetes, n (%)	8,657 (9.3)	2,292 (11.0)	1,863 (9.3)	1,279 (7.4)	1,527 (7.4)	1,575 (9.1)	2,238 (11.3)	2,698 (14.2)	1,616 (8.6)	1,080 (6.2)
Physical activity (≥30 min/d), n (%)	59,550 (64.1)	12,509 (59.8)	12,932 (64.7)	11,777 (68.3)	12,541 (61.0)	11,021 (63.7)	13,541 (68.4)	12,996 (68.3)	12,096 (64.1)	10,368 (59.4)
Current smoking, n (%)	13,321 (14.3)	4,343 (20.8)	2,611 (13.1)	1,544 (9.0)	3,976 (19.4)	2,529 (14.6)	1,823 (9.2)	2,331 (12.3)	2,572 (13.6)	2,904 (16.6)
Obesity (BMI ≥30 kg/m2), n (%)	16,661 (17.9)	5,055 (24.2)	3,463 (17.3)	1,936 (11.2)	4,788 (23.3)	3,147 (18.2)	2,395 (12.1)	3,337 (17.5)	3,294 (17.5)	3,343 (19.1)
Optimal sleep (6–8 h/d), n (%)	77,184 (83.1)	16,758 (80.2)	16,744 (83.7)	14,693 (85.3)	16,743 (81.5)	14,370 (83.1)	16,692 (84.3)	16,095 (84.6)	15,721 (83.4)	14,191 (81.3)
Energy intake, kcal/d, mean ± SD	2,160 ± 1,015	2,245 ± 1,119	2,152 ± 994	2,067 ± 893	2,295 ± 1,095	2,176 ± 1,022	1,998 ± 884	2,064 ± 935	2,184 ± 1,036	2,205 ± 1,051
Alcohol intake, g/d, mean ± SD	8.9 ± 23.7	16.1 ± 37.6	7.5 ± 17.9	4.5 ± 11.1	7.8 ± 18.4	9.5 ± 24.6	8.8 ± 25.7	7.0 ± 14.1	8.1 ± 18.4	12.5 ± 37.7

Abbreviations: BMI = body mass index; nSES = neighborhood socioeconomic status.

^aThe characteristics are from the cohort baseline data, except for Medicare usage, which is characterized by the annual average number of inpatient and outpatient claims over the duration of each participant’s Medicare use.

The plant-based diet indices at cohort baseline were each significantly associated with ADRD risk in both men and women (Table 2). PDI and hPDI were inversely associated with ADRD, with participants in the highest quintile showing 12% (HR 0.88; 95% CI 0.85–0.92) and 7% (HR 0.93; 95% CI 0.89–0.97) lower risks, respectively, compared with those in the lowest quintile. uPDI was associated with an increased risk of 6% (HR 1.06; 95% CI 1.01–1.10) for the highest vs lowest quintile comparison. The associations were all dose-dependent (all p-trend ≤ 0.006) and linear in restricted cubic splines (eFigure 2).

Table 2.

Plant-Based Diet Indices at Cohort Baseline and ADRD Risk in the Multiethnic Cohort Study, 1999–2019

	All (n = 92,849)		Men (n = 41,664)		Women (n = 51,185)		p-heterogeneity
	ADRD	HR (95% CI)a	ADRD	HR (95% CI)a	ADRD	HR (95% CI)a
Overall plant-based diet index
Q1: ≤44	4,327	1.00 (ref)	1,776	1.00 (ref)	2,551	1.00 (ref)
Q2: 45–47	3,867	0.95 (0.91–1.00)	1,579	0.96 (0.90–1.02)	2,288	0.95 (0.90–1.01)
Q3: 48–50	4,605	0.92 (0.88–0.96)	1,857	0.91 (0.86–0.98)	2,748	0.92 (0.87–0.98)
Q4: 51–53	4,257	0.93 (0.89–0.97)	1,770	0.96 (0.90–1.02)	2,487	0.92 (0.87–0.97)
Q5: ≥54	4,422	0.88 (0.85–0.92)	1,935	0.91 (0.85–0.97)	2,487	0.87 (0.82–0.92)
p-trend		<0.001		0.01		<0.001	0.31
Healthful plant-based diet index
Q1: ≤43	3,933	1.00 (ref)	1,565	1.00 (ref)	2,368	1.00 (ref)
Q2: 44–46	3,341	0.95 (0.91–1.00)	1,438	0.98 (0.91–1.05)	1,903	0.94 (0.88–1.00)
Q3: 47–49	3,921	0.93 (0.89–0.98)	1,662	0.95 (0.89–1.02)	2,259	0.92 (0.87–0.98)
Q4: 50–53	4,774	0.91 (0.88–0.95)	1,978	0.94 (0.88–1.01)	2,796	0.89 (0.84–0.95)
Q5: ≥54	5,509	0.93 (0.89–0.97)	2,274	0.96 (0.90–1.03)	3,235	0.91 (0.86–0.96)
p-trend		0.001		0.26		0.001	0.36
Unhealthful plant-based diet index
Q1: ≤43	4,822	1.00 (ref)	2,007	1.00 (ref)	2,815	1.00 (ref)
Q2: 44–46	4,040	1.00 (0.96–1.04)	1,675	0.99 (0.93–1.06)	2,365	1.01 (0.95–1.06)
Q3: 47–49	4,325	1.00 (0.96–1.04)	1,808	0.98 (0.92–1.04)	2,517	1.01 (0.96–1.07)
Q4: 50–53	4,733	1.03 (0.99–1.07)	1,929	0.99 (0.93–1.06)	2,804	1.05 (1.00–1.11)
Q5: ≥54	3,558	1.06 (1.01–1.10)	1,498	1.08 (1.01–1.15)	2,060	1.05 (0.99–1.11)
p-trend		0.006		0.06		0.04	0.86

Abbreviations: ADRD = Alzheimer disease and related dementia.

^aThe associations were adjusted for age at cohort baseline, age at Medicare surveillance initiation, sex, race/ethnicity, Medicare usage, education, neighborhood socioeconomic status, marital status, history of hypertension, history of heart disease, history of stroke, history of diabetes, smoking status, body mass index, physical activity, sleep duration, total energy intake, and alcohol consumption.

Several analyses were performed to compare subpopulations or ADRD subtypes. In race-specific and ethnicity-specific analyses (eTable 2), the risk reduction with PDI and hPDI was found in all groups except Native Hawaiian, while the risk increase with uPDI was significant only in the White group. However, these differences were not statistically significant (p-heterogeneity ≥ 0.10). In the subgroup with APOE genotype (n = 38,302; eTable 3), the diet-ADRD associations were slightly weaker than in the entire cohort listed in Table 2 and did not seem to differ by ε4 status (p-heterogeneity ≥ 0.20). When considering deaths as a competing risk, the associations were attenuated and the HR was only significant for PDI among women (HR 0.92; 95% CI 0.87–0.98, p-trend = 0.01) (eTable 4). In stratified analysis by baseline BMI, the associations were similar between lower (<25.6 kg/m²) and higher BMI groups (p-heterogeneity ≥ 0.13) (eTable 5). Considering common subtypes of ADRD, inverse associations with PDI or hPDI were apparent for dementia NOS, AD-only, and VD-only, whereas increased risk with uPDI was only found for AD of mixed etiology (eTable 6).

The longitudinal diet analysis was restricted to 45,065 participants who completed both the baseline and 10-year follow-up questionnaires. Compared with those who did not complete the follow-up survey, participants in the analysis were more likely to be younger, Japanese American or White, more educated, married, and living in a neighborhood with higher SES, and less likely to report cardiometabolic conditions at cohort baseline (eTable 7). Changes in hPDI or uPDI over 10 years were strongly associated with subsequent ADRD risk, as presented in restricted cubic spline analyses (Figure 1). Compared with a stable diet as the reference, a moderate decline in PDI (HR 1.09; 95% CI 1.03–1.16) and a large decline in hPDI (HR 1.17; 95% CI 1.09–1.26), as well as a large increase in uPDI (HR 1.25; 95% CI 1.15–1.36), were associated with a higher risk of ADRD, whereas a lower risk was seen with a large decline in uPDI (HR 0.89; 95% CI 0.84–0.94) (Table 3). Declines in PDI were associated with an increased risk in the older group (≥60 years at cohort baseline and initial dietary assessment), but not in the younger group (45–59 years); however, the difference did not reach significance (p-heterogeneity = 0.19). Similar patterns were observed for hPDI and uPDI in both age groups (p-heterogeneity ≥ 0.37). In a sensitivity analysis excluding 617 participants with ADRDs identified within the first 2 years after the 10-year follow-up, the results remained similar (eTable 8). In a subgroup with APOE genotype (n = 26,829, eTable 9), the associations were similar as in the entire cohort listed in Table 3 and did not statistically differ by ε4 status (p-heterogeneity ≥ 0.12).

Figure 1. Association Between Changes in Plant-Based Diet Indices Over 10 Years and Subsequent ADRD Risk, Based on Restricted Cubic Splines, in the Multiethnic Cohort Study, 2003–2019.

The models were adjusted for age at 10-year follow-up, age at Medicare surveillance start, sex, race/ethnicity, Medicare usage, education, neighborhood socioeconomic status, marital status, history of hypertension, history of heart disease, history of a stroke, history of diabetes, smoking status, body mass index, physical activity, sleep duration, total energy intake, alcohol consumption, and the corresponding plant-based diet index score at cohort baseline. ADRD = Alzheimer disease and related dementia; hPDI = healthful plant-based diet index; PDI = overall plant-based diet index; uPDI = unhealthful plant-based diet index.

Table 3.

Changes in Plant-Based Diet Indices Over 10 Years and Subsequent ADRD Risk in the Multiethnic Cohort Study, 2003–2019

	All (N = 45,065)		Age at cohort baseline				p-heterogeneity
			<60 y (n = 26,502)		≥60 y (n = 18,563)
	ADRD	HR (95% CI)a	ADRD	HR (95% CI)a	ADRD	HR (95% CI)a
Overall plant-based diet index
Large decline	1,960	1.06 (1.00–1.13)	433	0.94 (0.83–1.07)	1,527	1.10 (1.02–1.18)
Moderate decline	1,615	1.09 (1.03–1.16)	398	1.10 (0.98–1.24)	1,217	1.08 (1.01–1.17)
Stable	2,690	1.00 (ref)	668	1.00 (ref)	2,022	1.00 (ref)
Moderate increase	1,095	1.05 (0.98–1.13)	285	1.01 (0.87–1.17)	810	1.07 (0.98–1.16)
Large increase	1,000	1.04 (0.96–1.12)	278	1.04 (0.90–1.20)	722	1.04 (0.94–1.13)
p-trend		0.31		0.56		0.15	0.19
Healthful plant-based diet index
Large decline	1,134	1.17 (1.09–1.26)	244	1.15 (1.01–1.33)	890	1.19 (1.09–1.29)
Moderate decline	1,075	1.03 (0.96–1.10)	210	0.91 (0.78–1.05)	865	1.07 (0.99–1.15)
Stable	3,465	1.00 (ref)	848	1.00 (ref)	2,617	1.00 (ref)
Moderate increase	1,232	0.99 (0.93–1.06)	325	0.96 (0.84–1.09)	907	1.00 (0.93–1.08)
Large increase	1,454	1.00 (0.94–1.07)	435	0.98 (0.87–1.11)	1,019	1.01 (0.94–1.09)
p-trend		0.002		0.27		0.003	0.78
Unhealthful plant-based diet index
Large decline	2,507	0.89 (0.84–0.94)	683	0.83 (0.74–0.93)	1,824	0.91 (0.86–0.97)
Moderate decline	1,399	0.92 (0.86–0.97)	364	0.94 (0.82–1.07)	1,035	0.91 (0.85–0.97)
Stable	3,073	1.00 (ref)	724	1.00 (ref)	2,349	1.00 (ref)
Moderate increase	714	1.07 (0.98–1.17)	145	1.06 (0.89–1.25)	569	1.08 (0.97–1.20)
Large increase	667	1.25 (1.15–1.36)	146	1.43 (1.18–1.74)	521	1.22 (1.11–1.34)
p-trend		<0.001		<0.001		<0.001	0.37

Abbreviations: ADRD = Alzheimer disease and related dementia.

^aThe associations were adjusted for age at 10-year follow-up, age at Medicare surveillance initiation, sex, race/ethnicity, Medicare usage, education, neighborhood socioeconomic status, marital status, history of hypertension, history of heart disease, history of stroke, history of diabetes, smoking status, body mass index, physical activity, sleep duration, total energy intake, alcohol consumption, and the corresponding plant-based diet index score at cohort baseline. Large decline: ≥1 SD decrease; moderate decline: 0.5–<1 SD decrease; stable: <0.5 SD change; moderate increase: 0.5–<1 SD increase; large increase: ≥1 SD increase.

In racial and ethnic comparisons, the risk increase with a large decline in hPDI reached statistical significance only in Japanese American, Latino, and White groups, but overall, there was no significant heterogeneity in the association (eTable 10). By contrast, although the uPDI-ADRD association was in a positive direction in all racial and ethnic groups, the magnitude differed (p-heterogeneity = 0.03). The risk increase associated with a large increase in uptake of uPDI was greater in the Native Hawaiian (HR 1.69; 95% CI 1.13–2.54) group than in African American (1.29; 1.03–1.62), White (1.26; 1.08–1.47), Japanese American (1.23; 1.06–1.42), or Latino (1.15; 0.92–1.44) groups. In a competing risk model (eTable 11), HRs for changes in hPDI or uPDI were attenuated, especially in the older group (p-heterogeneity = 0.04 for uPDI), whereas HRs for a large increase in uPDI remained significant in both age groups. The association between changes in the plant-based diets and subsequent ADRD risk did not vary by baseline BMI (p-heterogeneity ≥ 0.22) (eTable 12). An increased risk with a large decline in hPDI and a large increase in uPDI was found for dementia NOS, AD-only, and AD of mixed etiology, whereas risk reduction with a large decline in uPDI was only observed for VD-only (eTable 13).

We examined the 16 food groups, which comprise the 3 plant-based diet indices in different scoring schemes, for ADRD risks associated with the intake change over 10 years in a mutually adjusted model (Figure 2). A large decrease in 4 of the 7 healthy plant food groups (whole grains, vegetable oils, nuts, and tea/coffee) was associated with a higher risk of ADRD by 11%–15%, whereas a large increase in fruits was related to an 11% higher risk when comparing with no change (detailed results in eTable 14). Of the 4 less healthy plant food groups, increased consumption of added sugars was associated with a 12% higher risk of ADRD (HR 1.12; 95% CI 1.03–1.22, p-trend <0.001) and was the only component food group associated with a statistically significant risk increase. Of the 5 animal food groups, a large decrease in egg consumption was associated with a 12% higher risk of ADRD (HR 1.12; 95% CI 1.04–1.21), with a linear trend (p-trend <0.001). Decreases in animal fats or meat consumption were also related to a 10%–11% higher risk of ADRD without a linear trend.

Figure 2. Association Between Changes in Consumption of Component Food Groups Over 10 Years and Subsequent ADRD Risk in the Multiethnic Cohort Study, 2003–2019.

The model was adjusted for age at 10-year follow-up, age at Medicare surveillance initiation, sex, race/ethnicity, Medicare usage, education, neighborhood socioeconomic status, marital status, history of hypertension, history of heart disease, history of stroke, history of diabetes, smoking status, body mass index, physical activity, sleep duration, total energy intake, alcohol consumption, and component score (1–5) at cohort baseline. All 16 component changes are mutually adjusted for, by being simultaneously entered in the model. Red, large decrease; orange, moderate decrease; gray, no change (reference); lime, moderate increase; green, large increase. ADRD = Alzheimer disease and related dementia.

Discussion

In a large multiethnic population, adherence to an overall plant-based dietary pattern at mid to late adulthood was associated with a 12% lower risk of ADRD. Indices that discern the quality of the plant-based diets showed more modest associations: 7% lower risk for healthy patterns (hPDI) and 6% higher risk for unhealthy patterns (uPDI). On the contrary, changes in quality-specific plant-based dietary patterns over 10 years were strongly associated with the subsequent risk of ADRD: 17% higher risk with decline in hPDI and 25% higher risk with increases in uPDI compared with a stable diet, similarly for older age (≥60 years at cohort entry) as in the younger age group. An exploratory analysis of the component food groups for the plant-based diet indices showed that changes over time in most healthy plant foods, added sugars among unhealthy plant foods, and animal fats, eggs, and meat among animal foods were associated with ADRD risk. A higher risk of ADRD associated with increases in uPDI was found in all racial and ethnic groups, except the Latino group. Otherwise, the associations between plant-based diet indices and ADRD were generally similar by age, sex, race and ethnicity, APOE ε4 status, or BMI, suggesting benefits in preventing ADRD by improving the quality of plant-based diets.

Our findings are consistent with previous reports that plant-based dietary patterns assessed by predefined indices are associated with sustaining cognitive function. A study from the Memory and Aging Project found that higher hPDI was associated with better cognitive function, whereas higher uPDI was associated with faster cognitive decline.³¹ Similar findings have been reported in China from prospective^32,33 and cross-sectional³⁴ studies. A cross-sectional analysis of the National Health and Nutrition Examination Survey data found an inverse association between hPDI and prevalent mild cognitive impairment.³⁵ The Chicago Health and Aging Project suggested racial differences; higher hPDI was associated with slower cognitive decline in African American participants, but not in White participants.⁶

Our results provide further evidence to the few studies conducted on plant-based diets and dementia risk. A small prospective study in Taiwan observed that vegetarians had a 33% lower risk of dementia compared with nonvegetarians.³⁶ A study in the UK Biobank reported that dementia risk was reduced by 18% with higher hPDI and increased by 29% with higher uPDI,¹² consistent with the stronger associations with uPDI vs hPDI we observed in the MEC Study. The Rotterdam Study found no overall association between plant-based diets and dementia risk but found different patterns by APOE ε4 carrier status: higher hPDI was associated with lower risk among ε4 carriers, while showing a U-shaped association among noncarriers.¹³ Our findings did not suggest different associations by APOE ε4.

Studies on temporal changes in plant-based dietary intakes in relation to dementia risk are rare. Although a 10-year change in PDI did not yield an association with ADRD, changes in hPDI and uPDI were strongly associated with risk, after adjustment for the baseline scores. These findings underline the importance of considering the quality of plant-based diets, not just consuming foods from nonanimal sources. Notably, among the food groups constituting the 3 plant-based diet indices, decreased consumption of most healthy plant foods was associated with a higher risk of ADRD, while increased intake of added sugars was the only unhealthy plant food category that conferred a higher risk. Eggs were the only animal food category for which increased consumption was associated with lower risk in our data, which is consistent with some observational data³⁷ (e.g., the Memory and Aging Project)^38,39 and a randomized trial,⁴⁰ but not others.^41,42

Plant-based diets are rich in vitamins, minerals, phytochemicals, and dietary fiber, which may help prevent cognitive decline, and are low in saturated fats and cholesterol, which contribute to a healthy blood lipid profile.^1,43,44 Strict diets that exclude any animal foods, such as vegan diets, may lead to deficiencies of vitamins B12 and D and minerals such as iron and zinc, unless dietary supplements are used. The plant-based dietary patterns examined in this study did not take into account the quality of animal foods, with all animal food groups scored inversely. However, it is important to note that older adults may still require a certain intake of animal-derived products to address increased protein needs. ⁴⁵ In this study, increased consumption of any animal food group over a 10-year follow-up was not associated with a higher risk of ADRD. A recent study from the Nurses’ Health Study and the Health Professional Follow-up Study suggests that dietary patterns rich in plant-based foods including moderate consumption of healthy animal-based foods, such as low-fat dairy products, may promote overall healthy aging. ⁴⁶

Strengths of this study include the population-based sampling for the cohort, the long-term follow-up with little loss to outmigration, the large number of participants from multiple racial and ethnic groups, and the comprehensive data on diet and potential confounders with repeated assessments. However, there are some limitations to note. ADRD was ascertained based on Medicare claims data and might not have been accurately captured. Nonetheless, studies have demonstrated that Medicare claims are largely concordant with clinical diagnoses of dementia and have shown improved consistency in recent decades.^47-49 The transition from ICD-9 to ICD-10 codes in Medicare might have led to inconsistency in our ADRD ascertainment,⁵⁰ although a recent study found a similar performance in defining dementia between the 2 versions.⁵¹ Dietary intake assessed by a food frequency questionnaire is subject to measurement error including random recall error. In addition, there may have been unknown or unmeasured risk factors and residual confounding, so the results may not accurately reflect the true associations. We suspect that these errors and omissions likely have attenuated the underlying associations toward null and, thus, would not explain our significant results. Our results may not be generalizable beyond the geographic and demographic populations represented in the MEC Study. Finally, despite the large overall sample size, the risk estimates for some subgroups should be interpreted with caution because of the smaller sample size, warranting replications in future research, including the racial and ethnic differences observed.

In conclusion, our findings suggest that increasing adherence to a healthful plant-based diet is associated with a reduced risk of ADRD, and that a strong opposite relationship exists for an unhealthful plant-based diet among middle-aged and older adults. These results indicate the importance of improving the quality of plant-based diets, even later in life, as a dietary strategy for dementia prevention.

Glossary

ADRD

Alzheimer disease and related dementia

BMI

body mass index

FFS

Fee-for-Service

hPDI

healthful plant-based diet index

HR

hazard ratio

ICD-9

International Classification of Diseases, Ninth Revision

ICD-10

International Classification of Diseases, 10th Revision

IRB

institutional review board

LBD

Lewy body dementia

MEC

Multiethnic Cohort

NOS

not otherwise specified

PDI

overall plant-based diet index

QFFQ

quantitative food frequency questionnaire

SES

socioeconomic status

uPDI

unhealthful plant-based diet index

VD

vascular dementia