Diet Quality and Body Mass Index Over 20 Years in the Multiethnic Cohort

Abstract

Background.

With increasing rates of overweight/obesity and disparities by ethnicity, it is important to understand the role of diet in ameliorating this health problem.

Objective.

The current study examined the relation of diet quality as measured by the Healthy Eating Index (HEI)-2015 with BMI and obesity among participants of the Multiethnic Cohort (MEC) in cross-sectional analyses at 3 time points (T-1 to T-3) over 20 years.

Design.

In a subset of 1,860 MEC participants, three cross-sectional analyses at cohort entry (1993–1996, T-1) and follow-ups in 2003–2008 (T-2) and 2013–2016 (T-3) were performed.

Participant/setting.

The cohort consists of African American, Native Hawaiian, Japanese American, Latino, and White adults in Hawaii and California with a mean age of 48 years at T-1.

Main outcome measure.

BMI/weight status in relation to diet quality.

Statistical analysis.

Linear and multinomial logistic regressions were applied to analyze the relation of diet quality with BMI and obesity while adjusting for known confounders.

Results.

HEI-2015 increased by 6.1 and 5.1 units for men and women from T-1 to T-3; the respective values for BMI were 1.5 and 2.4 kg/m². Diet quality was inversely associated with BMI across time: BMI was lower by −0.47, −0.72, and −0.92 units for every 10-point increase in HEI-2015 scores at T-1, T-2, and T-3 (p<0.0001 for all). During the 20 years, the association was consistently high among Japanese American participants (−0.79, −0.87, −1.02) and weakest in African American cohort members (−0.34, −0.37,−0.40). Higher diet quality was related to lower odds of having obesity at all 3 time points with prevalence odds ratios of 0.72, 0.57, and 0.60.

Conclusions.

These findings suggest that consuming a high-quality diet is related to a lower BMI and rates of overweight/obesity but with the strongest association at an older age. To understand the ethnic differences, investigations of dietary habits/behaviors and/or fat distribution patterns will be needed in the future.

Introduction

Based on the continuing increase in obesity prevalence¹, it has been predicted that almost 50% of adults in the US will develop obesity by 2030². This is a serious public health issue due to the elevated risk of type 2 diabetes, cardiovascular disease, and several types of cancer associated with obesity^{3, 4}. With disparities in obesity rates by ethnic background, primarily Native Hawaiian, African American, and Latino populations experience a high burden of morbidity and mortality due to excess body weight^5–7. As one approach to ameliorate this ongoing health problem, the promotion of healthful dietary patterns has been investigated. The Healthy Eating Index-2015 (HEI-2015) is the most up-to-date measure of diet quality; it is based on adherence to 13 dietary components based on the recommendations of the 2015–2020 Dietary Guidelines for Americans (DGA)^8–10. However, few studies have examined concurrent diet quality and BMI as part of aging over time and previous evidence for different ethnic groups is limited^11–13. In an analysis of the Multiethnic Cohort Study (MEC) with 53,977 African American, Native Hawaiian, Japanese American, Latino, and White participants¹², a change in diet quality scores from low to high over 10 years was associated with less weight gain (by 0.55–1.17 kg in men and 0.62–1.31 kg in women) as compared to stable diet quality. This inverse association was present in most ethnic groups and stronger in younger cohort members and those with higher body mass index (BMI). A cross-sectional analysis among 10,930 non-Hispanic White, non-Hispanic black, Mexican American, other Hispanic, and other non-Hispanic participants from the Third National Health and Nutrition Examination Survey showed that individuals in the poor (score of 50 or less, 17.8%) vs. good (81 or more, 10.7%) diet quality category were almost twice as likely to have obesity¹¹. A longitudinal analysis of 6236 adults in the Multi-Ethnic Study of Atherosclerosis study reported an inverse association of HEI versions from different years with obesity primarily among White participants, while the association was weaker among Chinese American and not seen in Hispanic and Black participants¹³. The objective of the current study was to examine the association of diet quality with BMI at three time points among participants of the Adiposity Phenotype Study (APS)¹⁴, a subset of the MEC to address the hypothesis that a higher diet quality as measured by the HEI-2015 is associated with a lower probability of overweight and obesity across different ethnic groups. The analysis addresses the following questions: How do mean BMI and diet quality change in a population at different time points during a 20-year period? Does the association between diet quality and obesity vary as individuals transition from midlife to older age? Is there a difference in the relation of diet quality and BMI by ethnic group?

Methods

Study Population.

In 1993–1996, the MEC recruited 215,251 men and women from 5 ethnic groups in Hawaii and California to study diet and cancer. In order to include as many respondents as possible in the cohort, persons of mixed ancestry were assigned to one category according to the following priority ranking: African American, Native Hawaiian, Latino, Japanese American, White, and Other¹⁵. Participants were primarily of Japanese American (26.4%), White (22.9%), Latino (22.0%), African American (16.3%), and Native Hawaiian (6.5%) descent¹⁵. Since cohort entry, participants have responded to multiple follow-up questionnaires. The current analysis is based on a subset of the MEC, the APS, which recruited 1,861 MEC participants 60–77 years of age for detailed imaging studies in 2013–2016¹⁴. Recruitment was stratified by sex, ethnicity, and six BMI categories (18.5–21.9, 22–24.9, 25–26.9, 27–29.9, 30–34.9, and 35–40 kg/m²) to ensure equal representation of all body weight categories. Exclusion criteria included current BMI outside 18.5–40 kg/m² to fit imaging equipment, current/recent (<2 years) smoking, soft or metal implants, insulin/thyroid medications, or serious health conditions that may impact study participation. The overall participation rate was 23% of individuals who were believed to be alive at the time and eligible according to the latest follow-up information in 2003–2008.

For the 1,861APS participants, data from cohort entry (1993–1996, T-1) and 10-year follow-up (2003–2008, T-2) were retrieved and added to the information collected in 2013–2016 (T-3). As 244 individuals did not complete the follow-up questionnaire in 2003–2008 and due to missing BMI and/or diet information (3 at T-1, 79 at T-2, and 1 at T-3), the final sample sizes at T-1, T-2, and T-3 were 1858, 1538, and 1860, respectively. The Institutional Review Boards at the University of Hawaii and the University of Southern California approved the protocols for these investigations and all participants provided written informed consent.

Data Collection.

At all three time points, the participants completed the same 26-page self-administered questionnaire asking for demographic characteristics (birth date, sex, ethnic background of self and parents, birth place of self and parents, marital status), anthropometric measures, i.e., self-reported height and weight, lifestyle risk factors, and a quantitative food frequency questionnaire (QFFQ)¹⁶. Self-reported information estimating average time spent in sleep and different activities on a typical day was used to estimate hours of moderate and vigorous activity per day. Smoking status was assessed as never, current, and past. At T-3, participants also attended a clinic visit when height and weight measured, donated blood and stool samples, and completed MRI and DXA scans¹⁴.

Dietary Assessment.

The QFFQ was administered by paper and asked about dietary intake during the past year^{15, 16}. It consisted of over 180 food items with 8 frequency options and 3 portion sizes and was developed based on 3-day measured food records from men and women belonging to 5 ethnic groups¹⁶. For each group, the contribution of individual food items to the total intake of nutrients of major interest was computed. Then, the foods were ordered from highest to lowest in contribution to intake for each nutrient and the minimum set that yielded at least 85% of the intake was determined. The final items included in the QFFQ accounted for more than 85% of the intake for major nutrients and specific food items unique to the diets of each ethnic group. Calibration of the QFFQ using three 24-hr recalls showed satisfactory correlations (0.57–0.74 for nutrient densities). The 10-year-QFFQ was generally similar to the baseline-QFFQ but had several, mostly minor, modifications¹⁷. For instance, the examples for each item on the baseline-QFFQ were updated to reflect the foods most commonly reported in the original calibration study. Foods were added that had become recently available, e.g., cereal bars, additional Latino foods, e.g., burritos, enchiladas or chilaquiles instead of burritos and products that supplied nutrients of interest, e.g., soy milk for isoflavones. In addition, to avoid increasing the length of the questionnaire, some items were combined, e.g., spareribs and short ribs. For participants who reported total energy intakes outside the range of 500–8000 kcal in the baseline-QFFQ¹⁶, their diet was considered invalid and they were excluded from all analyses as they are implausible.

HEI-2015.

The HEI-2015 assesses diet quality by measuring of adherence to recommendations of the 2015–2020 DGA¹⁸. It consists of 13 dietary components with 9 adequacy components (Total Vegetables, Greens and Beans, Total Fruits, Whole Fruits, Whole Grains, Dairy, Total Protein Foods, Seafood and Plant Proteins, Fatty Acids) and 4 moderation components (Refined Grains, Sodium, Added Sugars, and Saturated Fats). The maximum score for the HEI-2015 is 100^{10, 18}. Total Vegetables, Greens and Beans, Total Fruits, Whole Fruits, Total Protein Foods, and Seafood and Plant Proteins were scored on a 5 point scale while Whole Grains, Dairy, Fatty Acids, Sodium, Refined Grains, Added Sugars, and Saturated Fats each accounted for 10 points. With three exceptions, the components are scored on a density per 1,000 calories: Fatty Acids is computed as the ratio of unsaturated to saturated fatty acids, while Added Sugars and Saturated Fats are calculated as percent of total energy intake. The calculation of the HEI-2015 applied a standardized, guidance based food grouping method based on estimated intakes obtained from the three QFFQs at T1-T3 using the simple HEI scoring algorithm method^{9, 18, 19}. Reported intake amounts of foods and beverages were converted into a uniform system of nutritionally meaningful groups by merging data with MyPyramid Equivalents Database (MPED)^13–14 and calculating components using MPEDs²⁰.

Anthropometric Measures.

BMI was calculated at T-1 and T-2 based on self-reported weight and height using the formula weight in kg divided by the square of height in meters. At T-3, BMI was computed from height (Heightronic model 235A at UH; SECA model 240 at USC) and weight (Scale-Tronix model 5102 at UH; Health o meter Professional ProPlus at USC) measurements performed in duplicate at clinic visit¹⁴. BMI was categorized into underweight/healthy weight, overweight, and obesity (<25, 25-<30, and ≥30 kg/m²).

Statistical Analysis.

For each time point, descriptive statistics (means and SD) were computed and differences in repeated HEI-2015 scores and BMI over time were evaluated using mixed models to account for within-person variances. Pearson correlation coefficients were computed to describe the relation of HEI-2015 scores and BMI across time and with each other. In a cross-sectional approach, the association between diet quality and BMI/obesity was analyzed separately at T-1, T-2, and T-3 by applying general linear regression. Models for the total study sample and by ethnic group were estimated with BMI as a dependent variable and HEI-2015 score (in 10-point increments) as independent predictor, both as continuous variables. Based on previous analyses related to adiposity in the MEC^{21, 22}, the following covariates were included in all models: age, smoking status (never, past, current), sex, physical activity (<1, 1+ hrs/day), and alcohol intake as drinks (<1/month, ≥1 drink/month - <1 drink/day, 1+/day) as assessed by the corresponding questionnaire and ethnicity (African American, Native Hawaiian, Japanese American, Latino, and White). Each component score was also modeled separately using the same approach and covariates. Polytomous multinomial logistic regression was applied to estimate prevalence odds ratios (POR) as pairwise binary logistic regressions of BMI categories with underweight/healthy weight as reference category, per 10-point increment of the HEI-2015 score. To test for possible differences by sex, we added an interaction term to the main models at each time point and performed stratified analyses for men and women. Separate estimates were also obtained for each ethnic group. The level of statistical significance was set at p<0.05. All analyses were conducted using SAS 9.4 software (Cary, NC)²³.

Results

The analysis population consisted of 1,858 MEC participants at T-1, 1,538 participants at T-2, and 1,860 participants at T-3 (Table 1). The mean ages of participants were 48.5 years for men and 48.2 years for women at T-1, 59.9 years for men and 59.7 years for women at T-2, and 69.3 years for men and 69.1 years for women at T-3. Mean HEI-2015 scores were higher by 6.1 units for men and 5.1 units for women after 20 years at T-3 than T-1 while mean BMI values were higher by 1.5 kg/m² and 2.4 kg/m², respectively. Men had 2–3 point lower HEI-2015 scores and a slightly higher BMI than women. Diet quality and BMI values were significantly associated with each other across time points. Correlations of HEI-2015 scores for the same individuals at T-1 with T-2 were r=0.57, for T-1 with T-3 r=0.51, and for T-2 with T-3 r=0.63 (all p<0.0001). HEI-2015 scores correlated strongly with BMI at T-1, T-2, and T-3 (0.84, 0.76, 0.87; p<0.0001). Total energy intake per day decreased by a mean of 420 Kcal from T-1 to T-3, but the association with the HEI-2015 scores at T-3 was weak (r=0.055; p=0.02).

Table 1.

Characteristics and Diet Quality Scores of Study Participants in the Multiethnic Cohort at Three Points in Time

Variable	T-1 (1993–1996)		T-2 (2003–2008)		T-3 (2013–2016)
	Men	Women	Men	Women	Men	Women
Number	922	936	766	772	923	937
	Means ± SD
Age, years	48.5 ± 2.6	48.2 ± 2.6	59.9 ± 2.8	59.7 ± 2.9	69.3 ± 2.8	69.1 ± 2.7
Body mass index, kg/m2	26.4 ± 3.6	25.7 ± 4.3	27.1 ± 3.9	26.8 ± 4.6	27.9 ± 4.4	28.1 ± 5.2
Total energy intake (kcal/day)	2485±1069	2058±973	2080±878	1761±799	2035±968	1730±899
HEI-2015a Total Score	64.2 ± 10.2	67.3 ± 10.2	69.3 ± 10.1	72.3 ± 10.3	70.3 ± 10.5	72.4 ± 9.9
Adequacy Components
Total Vegetables	3.9 ± 1.0	4.3 ± 0.9	4.3 ± 0.9	4.5 ± 0.8	4.3 ± 1.0	4.6 ± 0.8
Greens and Beans	3.5 ± 1.4	3.9 ± 1.3	3.7 ± 1.3	4.1 ± 1.2	3.8 ± 1.3	4.2 ± 1.2
Total Fruits	3.4 ± 1.6	3.8 ± 1.5	3.8 ± 1.4	4.1 ± 1.3	3.9 ± 1.4	4.2 ± 1.2
Whole Fruits	4.0 ± 1.4	4.3 ± 1.2	4.4 ± 1.2	4.6 ± 1.0	4.5 ± 1.2	4.7 ± 0.9
Whole Grains	4.4 ± 3.0	5.2 ± 3.1	5.3 ± 3.1	5.7 ± 3.0	5.6 ± 3.1	5.5 ± 3.1
Dairy	3.7 ± 2.3	4.5 ± 2.5	4.1 ± 2.5	5.1 ± 2.6	4.3 ± 2.6	4.8 ± 2.6
Total Protein Foods	4.6 ± 0.7	4.6 ± 0.8	4.9 ± 0.5	4.7 ± 0.7	4.8 ± 0.5	4.8 ± 0.6
Seafood & Plant Proteins	4.4 ± 1.0	4.4 ± 1.0	4.8 ± 0.6	4.7 ± 0.8	4.7 ± 0.7	4.7 ± 0.7
Fatty Acids	6.3 ± 2.4	6.2 ± 2.4	6.4 ± 2.4	6.3 ± 2.5	6.2 ± 2.6	6.2 ± 2.7
Moderation Components
Sodium	4.5 ± 2.9	4.0 ± 2.9	4.3 ± 3.0	4.5 ± 3.1	4.7 ± 3.0	4.7 ± 3.2
Refined Grains	5.4 ± 3.1	5.9 ± 3.1	7.0 ± 2.9	7.8 ± 2.8	7.6 ± 2.7	8.1 ± 2.5
Saturated Fats	7.6 ± 2.4	7.5 ± 2.3	7.2 ± 2.5	7.1 ± 2.5	6.6 ± 2.6	6.7 ± 2.6
Added Sugars	8.5 ± 2.2	8.6 ± 1.9	9.1 ± 1.5	9.0 ± 1.4	9.3 ± 1.4	9.2 ± 1.3
	N (%)
Ethnicity
African American	132 (14.3)	184 (19.7)	93 (12.1)	127 (16.5)	133 (14.4)	184 (19.6)
Native Hawaiian	145 (15.7)	162 (17.3)	111 (14.5)	124 (16.1)	145 (15.7)	161(17.2)
Japanese American	230 (25.0)	203 (21.7)	212 (27.7)	195 (25.3)	230 (24.9)	204 (21.8)
Latino	201 (21.8)	190 (20.3)	155 (20.3)	147 (19.0)	201 (21.8)	191 (20.4)
White	214 (23.2)	197 (21.0)	195 25.5)	179 (23.2)	214 (23.2)	197 (21.0)
Alcohol intakeb
<1 drink/month	297 (32.2)	491 (52.5)	255 (33.3)	413 (53.5)	311 (33.7)	504 (53.8)
≥1 drink/month - <1 drink/day	349 (37.9)	316 (33.8)	330 (43.8)	296 (38.3)	381 (41.3)	320 (34.2)
≥1 drink/day	244 (26.5)	105 (11.2)	181 (23.6)	63 (8.2)	216 (23.4)	95 (10.1)
Physical activityc
< 1 hr/day	470 (51.0)	533 (56.9)	262 (34.2)	335 (43.4)	354 (38.4)	437 (46.6)
≥1 hr/day	450 (48.8)	392 (41.9)	498 (65.0)	433 (56.1)	567 (61.4)	499 (53.3)
Smoking status
Never	461 (50.0)	613 (65.5)	374 (48.8)	500 (64.8)	485 (52.65)	656 (70.0)
Past	391 (42.4)	263 (28.1)	386 (50.4)	261 (33.8)	438 (47.5)	21 (30.0)
Current	66 (7.2)	52 (5.6)	1 (0.1)	0 (0.00)	0 (0)	0 (0)

^aHealthy Eating Index-2015

^bInformation for alcohol intake was missing for N=56 at T-1 and N=33 at T-3

^cModerate to vigorous activity

Among all ethnic groups except African American and White participants, both HEI-2015 scores and BMI were higher as participants aged (Table 2). In the overall cohort, the difference in HEI-2015 scores was 7.6% between T-1 and T-2 and another 0.7% to T-3. Mean BMI values differed across ethnic groups, however, the means increased in all ethnic groups as participants aged; all trends from T1 to T3 were statistically significant (p<0.0001). Comparisons of the unadjusted distribution of HEI-2015 scores by BMI status and ethnicity indicated a trend of lower scores for participants with overweight/obesity than those with underweight/healthy weight across the five ethnic groups (Figure 1).

Table 2.

Mean Healthy Eating Index-2015 Scores and BMI Values by Ethnic Group in the Multiethnic Cohort Study, 1993–2016

Ethnicity	Variable	T-1 (1993–1996)	T-2 (2003–2008)	T-3 (2013–2016)	Difference
		Mean ± SD	Mean ± SD	Mean ± SD	p-valuea
African American	HEI-2015b	68.6 ± 10.7	72.4 ± 10.1	72.2 ± 10.1	<0.0001
	BMI	27.1 ± 4.3	28.3 ± 4.4	29.3 ± 5.1	<0.0001
Native Hawaiian	HEI-2015b	65.6 ± 10.3	69.9 ± 10.9	70.8 ± 10.3	<0.0001
	BMI	27.0 ± 4.4	28.1 ± 4.7	28.8 ± 5.0	<0.0001
Japanese American	HEI-2015b	62.7 ± 9.7	70.0 ± 10.5	71.2 ± 10.6	<0.0001
	BMI	25.0 ± 3.4	25.7 ± 3.6	26.2 ± 4.1	<0.0001
Latino	HEI-2015b	64.3 ± 9.5	68.1 ± 10.0	69.7 ± 10.2	<0.0001
	BMI	26.9 ± 3.8	28.0 ± 4.0	29.1 ± 4.8	<0.0001
White	HEI-2015b	68.5 ± 10.1	73.3 ± 9.5	72.8 ± 9.8	<0.0001
	BMI	24.8 ± 3.5	25.9 ± 4.1	27.0 ± 4.6	<0.0001
All	HEI-2015b	65.8 ± 10.3	70.8 ± 10.3	71.3 ± 10.3	<0.0001
	BMI	26.0 ± 4.0	26.9 ± 4.3	28.0 ± 4.8	<0.0001

^aP-value of difference by time obtained through general linear regression using mixed models without adjustment

^bHealthy Eating Index-2015

Figure 1.

Mean Healthy Eating Index-2015 Scores by BMI Status and Ethnicity at Three Time Points, Multiethnic Cohort Study, 1993–2016. The horizontal line within the box represents the median of the HEI-2015 score, the circle in the box shows the mean, the bottom and the top of the boxes mark the first and the third quartile, the whiskers represent the minimum and maximum values, and the little circles beyond the whiskers show outliers.

The total HEI-2015 score was inversely associated with BMI (Table 3) across all time points as participants aged; BMI was lower by −0.47, −0.72, and −0.92 units for every 10-point increase in scores at T-1, T-2, and T-3 (p<0.0001 for all). Although the respective interaction terms by sex did not reach statistical significance, the estimates appeared greater for women than men: −0.65 vs. −0.31 (p=0.08) at T-1, −0.87 vs. −0.59 (p=0.15) at T-2, and −1.14 vs. −0.75 (p=0.05) at T-3. Adherence to the adequacy components Total Fruits, Whole Fruits, Whole Grains, and Greens and Beans, was inversely associated to BMI with lower values of 0.38, 0.34, 0.18, and 0.22 at T-3, respectively, for each 1-point increase in the component score. On the other hand, the Total Protein Foods component was associated with higher BMI, i.e., per 1-point greater adherence, the BMI was 0.64 units higher at T-3. The Total Vegetables, Dairy, Seafood and Plant Proteins components were not associated with BMI. Of the moderation components, higher scores, which indicate lower intake, for Sodium, Refined Grains, and Saturated Fats but not Added Sugars were related to lower BMI at all three time points.

Table 3.

Associations of the Health Eating Index-2015 and its Components with BMI in the Multiethnic Cohort Study, 1993–2016^a

	T-1 (1993–1996)		T-2 (2003–2008)		T-3 (2013–2016)
	Estimate	P-value	Estimate	P-value	Estimate	P-value
Total HEI-2015a Scoreb	−0.47	<0.0001	−0.72	<0.0001	−0.92	<0.0001
HEI-2015 Dietary Componentsc
Total Vegetables	0.02	0.80	−0.04	0.74	−0.24	0.05
Greens and Beans	−0.11	0.09	−0.14	0.08	−0.22	0.01
Total Fruits	−0.34	<0.0001	−0.38	<0.0001	−0.38	<0.0001
Whole Fruits	−0.33	<0.0001	−0.30	0.001	−0.34	0.001
Whole Grains	−0.13	<0.0001	−0.10	0.004	−0.18	<0.0001
Dairy	0.00	0.93	−0.04	0.37	−0.05	0.23
Total Protein Foods	0.47	0.0001	0.54	0.002	0.64	0.0004
Seafood and Plant Proteins	−0.04	0.63	−0.08	0.59	−0.17	0.26
Fatty Acids	−0.01	0.75	−0.13	0.002	−0.21	<0.0001
Sodium	−0.08	0.02	−0.18	<.0001	−0.18	<0.0001
Refined Grains	−0.14	<0.0001	−0.09	0.02	−0.15	0.0006
Saturated Fats	−0.13	0.002	−0.33	<.0001	−0.33	<0.0001
Added Sugars	0.05	0.25	0.11	0.13	0.05	0.53

^aObtained through general linear models adjusted for ethnicity, age, sex, smoking status, alcohol intake, and physical activity per 10 points of the HEI-2015 score

^bHealthy Eating Index-2015

^cIndividual component scores were modeled separately as associations per 1 point change in component score with the same covariates as for the Total HEI-2015 Score

In models stratified by ethnic group (Table 4), higher diet quality was generally inversely related to BMI. The association between HEI-2015 and BMI grew stronger from T-1 to T-2 in all groups except Native Hawaiian participants who showed the strongest association at T-1 (−0.72). Across time, the associations were consistently high among those with Japanese American ancestry (−0.79, −0.87, −1.02), followed by White participants (−0.31, −0.84, −1.27), while they were weakest and not significant in African American cohort members (−0.34, −0.37,−0.40).

Table 4.

Association of the HEI-2015 Score with BMI by Ethnic Group in the Multiethnic Cohort Study, 1993–2016^a

	T-1 (1993–1996)		T-2 (2003–2008)		T-3 (2013–2016)
	Estimate	P-value	Estimate	P-value	Estimate	P-value
African American	−0.34	0.16	−0.37	0.23	−0.40	0.16
Latino	−0.27	0.20	−0.80	0.0005	−1.06	<0.0001
Japanese American	−0.79	<0.0001	−0.87	<0.0001	−1.02	<0.0001
Native Hawaiian	−0.72	0.005	−0.58	0.04	−0.58	0.04
White	−0.31	<0.0001	−0.84	0.0001	−1.27	<0.0001

^aObtained through general linear models adjusted for age, sex, smoking status, alcohol intake, and physical activity per 10 points of the Healthy Eating Index-2015 score

Better diet quality was associated with lower odds of having overweight or obesity at all three time points (Figure 2). In the multinomial logistic regression models, the PORs for overweight and obesity per 10 point increase in HEI-2015 were 0.87 (95% CL 0.79, 0.97) and 0.72 (95% CL: 0.62, 0.83) at T-1, respectively. The strongest association, i.e., the lowest POR for having overweight or obesity were seen at T-2 with respective values of 0.72 (95% CL: 0.64, 0.82) and 0.57 (95% CL: 0.49, 0.67). The associations were strongest among Japanese American, Latino, and White participants although the differences in PORs were modest and the confidence intervals overlapped. In the Native Hawaiian group, the PORs were only significant for obesity at T-1 and T-2, whereas no significant associations were detected among those with African American ancestry at any time.

Figure 2.

Association of Diet Quality with BMI at T-1, T-2, and T-3 by Ethnicity, Multiethnic Cohort Study, 1993–2016. Prevalence odds ratios and 95% confidence limits for 10 points of the HEI-2015^a score were obtained through multinomial logistic regression adjusted for race/ethnicity, age, sex, smoking status, alcohol intake, and physical activity.

^aHEI-2015 = Healthy Eating Index-2015; ^bPOR =Prevalence Odds Ratios; ^cLCL = Lower 95% Confidence Limit; ^dUCL = Upper 95% Confidence Limit

Discussion

Based on data collected in a prospective cohort over close to 20 years, participants were 13–23% and 28–40% less likely to be classified as overweight or obese for each 10-point increase in HEI-2015 scores at three time points representing mean ages of 48, 60, and 69 years. The strongest association of the HEI-2015 Total Score and its components with BMI was seen at the oldest age, possibly due to the lower energy intake that was weakly associated with the HEI-2015 scores at T-3. The HEI separates diet quality from diet quantity, but it does not account for over-consuming calories¹⁰. Although caution is warranted given that the results are group means and not changes in individuals, the increase of BMI despite higher HEI-2015 scores and most of its adequacy components over time may be due to lower energy needs related to aging²⁴. Analyses stratified by ethnicity indicated inverse associations between diet quality and BMI for all five groups although they were not statistically significant for African American participants. These observations support the hypothesis that a better diet quality is associated with a lower prevalence of overweight or obesity across ethnic groups despite differences in the strength of association. The discrepant findings for African American participants may be due to their higher HEI-2015 scores at T-1, the translation of certain ethnic-specific foods into diet quality components, or different energy requirements due to specific weight patterns in this group.

The current findings agree with a MEC report that investigated 4 dietary indices: HEI-2015, Alternative Healthy Eating Index (AHEI)-2010, alternate Mediterranean diet score (aMED), and Dietary Approaches to Stop Hypertension (DASH). It showed an inverse relation of diet quality with weight gain over 10 years in most subgroups¹². Although the sample size of the current study is much smaller (1,860 vs. 53,977), it examined an additional time point 8 years later to determine that a high quality diet remains associated with lower BMI at an older age.

Inverse associations between diet quality and BMI change have been reported using multiple approaches of defining diet quality^25–31. The Resilience for Eating and Activity Despite Inequality study found that a 1 point change in the Australian Dietary Guideline Index was inversely associated with a minimum of 0.012 kg/m² lower BMI gain²⁵. In the prospective EPIC-PANACEA project, change from the lowest to the highest tertile of the Mediterranean Diet (MED) score reduced the likelihood of developing overweight or obesity by 10% over 5 years²⁶. The Seguimiento Universidad de Navarra cohort reported that the risk of gaining weight in the first 2 years of follow-up was lowest for those who adhered to the MED diet. The PORs of gaining ≥3 kg or ≥5 kg were 0.80 (95% CI: 0.70, 0.92) and 0.76 (95% CI: 0.62, 0.92)²⁷. A 15-year longitudinal study in Australia showed that men with the highest diet quality had the lowest gain in BMI (0.05 v. 0.11 kg/m² per year, p=0.01) but reported no significant association in women²⁹. In the Framingham Offspring Study, women with the highest diet quality index score (DQI) as compared to the lowest category gained 3.3 lbs. vs. 8.0 lbs. over 8 years; the respective values for men were 2.7 lbs. and 5.1 lbs³⁰. In a French prospective cohort adherence to a high-quality diet as assessed by six different dietary scores was associated with lower weight gain in men with POR values from 0.63 to 0.72²⁸. Participants in the Health Professionals Follow-Up Study and Nurses and Health Study gained less weight over 4 years when diet quality increased during that time³¹. A cross-sectional analysis among Canadian residents also supports the hypothesis of lower body weight with better diet quality³².

Of the few investigations including populations other than White^{33, 34}, the Coronary Artery Risk Development in Young Adults Study reported an overall 25% lower risk of weight gain (HR: 0.75; 95% CI: 0.65, 0.87) for a higher DQI³³. In contrast to our non-significant findings in the African American group, higher diet quality was associated with a 10% lower risk of gaining 10 kg in White but a 15% higher risk in African American participants with baseline obesity (POR:0.92; 95% CL: 0.81, 1.04) and (POR:1.15; 95% CL: 1.05, 1.23), respectively³³. The authors suggested this may be due to different food choices across groups as a high DQI score does not necessarily reflect the intake of the same food products. A cross-sectional study in Chinese adults detected an inverse association of the AHEI-2010, DQI, and DASH with obesity; respective PORs per 1-SD were 0.71 (95% CI: 0.51, 0.99), 0.63 (95% CI: 0.46, 0.86), and 0.57 (95% CI: 0.38, 0.84)³⁴.

The strengths of the current report include an ethnically diverse population with a follow-up of close to 20 years and the use of a validated QFFQ¹⁶, which was especially developed for the five different ethnic groups to include commonly consumed food items and dishes for all groups. Although the substance of the QFFQ remained the same, the original QFFQ was updated in 2003 to modify the food lists, amounts and examples or names for the food items¹⁷. As the dietary assessment method remained consistent across time, the results obtained at the three points in time are comparable. The HEI-2015 is aligned with the 2015–2020 DGA and has shown its efficacy in many studies^{9, 35}.

However, there are several limitations to this study. BMI was assessed by self-reported height and weight at T-1 and T-2 allowing for systematic error due to misreporting, but at T-3, height and weight were objectively measured, which often results in higher BMI values. It is possible that the stronger relation at T-3 is the result of the measured BMI, which is more accurate, as opposed to self-reports at T-1 and T-2³⁶. Despite a strong correlation between self-reported and anthropometric measures in a MEC subgroup, significantly greater misreporting was noted in female MEC participants with higher BMI³⁷ and in the Women’s Health Initiative³⁶. There may also be a differential underreporting between ethnic groups. The presence of under- and over-reporting of food intake may have also biased the findings although observations with implausible energy intake were excluded. It is possible that HEI-2015 scores among individuals with overweight or obesity were less than accurate than for persons with a healthy BMI due to the bias in reporting dietary intake¹⁶. A calibration study in the MEC detected significant differences by ethnicity when evaluating the performance of the QFFQ in characterizing intake¹⁶. As is well known, food frequency histories are not explicit for total dietary intake¹⁶; in the MEC, after energy adjustment, the correlations were satisfactory but not perfect¹⁶. Also, the calibration studies were conducted around the time of T-1; since then many factors have changed that may impact the current validity, e.g., recipes and components of commonly consumed food products and ready-to-eat meals as well as portion sizes at restaurants. The smaller sample size at T-2 due to non-respondents to the follow-up questionnaire may have affected the results for that time point. In addition, the current findings cannot be generalized to the population-at-large due to selection bias: people with BMIs below 18.5 and above 40 kg/m² and with serious health conditions were excluded, the participation rate was only 23%, and the MEC members had relatively high mean HEI-2015 scores as compared to values for the US³⁸. It is also important to keep in mind that a higher BMI or obesity is not necessarily associated with illness. Although obesity is recognized as a complex disease by the American Medical Asssociation³⁹, persons with BMIs above 25 or even 30 kg/m² may not experience adverse health consequences and the impact on morbidity and mortality varies across ethnic populations due to differences in body fat distribution and other factors^{14, 40–42}.

Conclusions

Results from the current study provide evidence in four out of five ethnic groups in the MEC for the hypothesis that a higher diet quality as measured by the HEI-2015 is potentially associated with a lower probability of being classified as overweight or obese from midlife to older age. Given the relatively small differences in BMI for an increase in 10 points of the HEI-2015, which requires major modifications in dietary intake, raises the question how clinically relevant the current findings are in obesity prevention. It is clear that additional lifestyle changes may be necessary to reduce overweight and obesity rates. Although this is not a novel finding, the current analysis presents findings specific to individuals from five ethnic groups, which may suggest new studies within particular groups to understand differences and appropriate dietary recommendations tailored to diverse eating habits and food preferences. To determine if maintenance of a high quality diet increases the chances of maintaining a healthy BMI, larger longitudinal studies will be needed.

Research Snapshot.

Research Question:

How is diet quality associated with BMI in different ethnic groups and over time?

Key Findings:

Within a subgroup of 1860 participants from five ethnic groups who are part of a large prospective cohort, three cross-sectional analyses across 20 years showed that diet quality was inversely related to body mass index (BMI). The association was strongest among individuals with Japanese American, Latino, White, and Native Hawaiian ancestry, but it was not significant in those with African American background. Despite increasing BMI with age, the protective association of diet quality was greatest at the last time point when the mean age approached 70 years.