Changes in Diet Quality over 10 Years Are Associated with Baseline Sociodemographic and Lifestyle Factors in the Multiethnic Cohort Study

ABSTRACT

Background

Trends in diet quality among US adults indicate a steady improvement, but data on longitudinal individual-level changes in diet quality are still limited.

Objective

We examined changes in diet quality over 10 y and sought to determine whether baseline sociodemographic and lifestyle factors predicted the changes in a multiethnic population.

Methods

Data were from 63,255 African American, Native Hawaiian, Japanese American, Latino, and white men and women (45–75 y old at baseline) in the Multiethnic Cohort, who completed a quantitative food frequency questionnaire at baseline (1993–1996) and 10-y follow-up (2003–2007) and had no prevalent cancer or heart disease at either survey. Overall diet quality was measured by use of the Healthy Eating Index–2015 (HEI-2015), the Alternative Healthy Eating Index–2010 (AHEI-2010), the alternate Mediterranean Diet score, and the Dietary Approaches to Stop Hypertension (DASH) score. We used a general linear model with adjustment for covariates to compare diet quality changes by baseline characteristics in men and women separately.

Results

Overall diet quality improved over 10 y by 3.2 points in men and 2.9 in women assessed using the HEI-2015, although scores for some components worsened (saturated and trans fats, indicating increased intake) or remained unchanged at a low quality level (whole grains, dairy, and sodium). In multivariable models where changes in HEI-2015, AHEI-2010, and DASH were harmonized to a 100-point score, greater increases in scores in both men and women were found for Japanese American ethnicity (increase by 0.5–4.7 in the 3 scores, P < 0.03), higher education (by 0.5–1.5, P ≤ 0.001), normal weight (BMI 18.5 to <25, by 0.6–2.5, P ≤ 0.01), nonsmoking (by 1.5–2.7, P < 0.001), higher moderate/vigorous physical activity level (by 0.3–0.8, P ≤ 0.04), and multivitamin use (by 0.4–0.7, P < 0.001) at baseline.

Conclusions

Sociodemographic and lifestyle factors, closely associated with diet quality, also predicted subsequent changes in diet quality over time in this multiethnic population.

Introduction

A healthful, high-quality diet is critical for preventing noncommunicable diseases later in life (1–3). However, according to 1999–2010 NHANES data, the overall diet quality of many US adults, measured by a predefined index, remained poor across all age groups, despite the trend of a steady improvement in diet quality scores over the 12-y period (4). Patterns of change in diet quality over time may vary by personal characteristics. Indeed, longitudinal studies found that age, race/ethnicity, education level, and lifestyle factors, including physical activity, smoking, obesity status, and menopausal hormone therapy use in women, were associated with change in diet quality (5–8). In addition, a few US cohorts reported that improvement in diet quality was associated with a lower risk of chronic disease (9, 10) and mortality (6, 11) and with concurrent changes in body weight (12) and body composition (13) in a desirable direction. However, data on long-term changes in diet quality among adults are still limited, especially from ethnically diverse populations.

In our previous analysis in the Multiethnic Cohort (MEC) (14), we found that diet quality at baseline assessed by 4 predefined indexes was associated with sociodemographic and lifestyle factors, and that the associations with several factors varied between men and women: being widowed, being a previous smoker, and having a low BMI were related to lower dietary scores in men but not in women. In the present study, we investigated the patterns of diet quality changes from baseline, when the participants were middle to older aged, to a 10-y follow-up survey and the associations of baseline sociodemographic and lifestyle factors with diet quality changes.

Methods

Study population

The MEC was established to study diet and chronic disease in Hawaii and California, primarily Los Angeles County (15). In 1993–1996, more than 215,000 participants aged 45–75 y completed a comprehensive, 26-page self-administered questionnaire that asked questions on demographic factors, dietary habits, other lifestyle factors, medical conditions, and family history of cancer. Participants were mainly African American, Native Hawaiian, Japanese American, Latino, and white due to targeted recruitment. The institutional review boards of the University of Hawaii and the University of Southern California approved the study protocol. In 2003–2007, 98,214 participants completed a 10-y follow-up survey, a repeat of the 26-page questionnaire. For the current analyses, we excluded participants who were not members of 1 of the 5 racial/ethnic groups (n = 5246) or who had extreme diets based on energy and macronutrient intakes at baseline or follow-up (n = 5930). Specifically, we computed a robust SD based on the truncated normal distribution after excluding the top and bottom 10% tails of the log energy distribution. Then, we excluded all individuals with energy values out of the ranges of the mean ± 3 robust SDs. We applied similar approaches to exclude individuals with extreme fat, protein, or carbohydrate intakes. Based on these exclusions, we removed individuals with a large number of missing food items on their questionnaire responses. We further excluded participants who reported heart attack, angina at baseline or follow-up (n = 11,248), or cancer at baseline, and those who had invasive cancer linked to tumor registries up to the date of the follow-up questionnaire (n = 12,535). Based on these exclusion criteria, a total of 63,255 participants remained for the analysis.

Dietary data

Diet was assessed at baseline and the 10-y follow-up by using a quantitative food frequency questionnaire (QFFQ) to assess participants’ usual intake for >180 food items during the past 12 mo. The baseline QFFQ was developed from 3-d measured food records (15). Daily intakes of foods and nutrients were calculated using a food composition table specific to the MEC. A calibration study showed satisfactory correlations for nutrients as densities (0.57–0.74) between the QFFQ and three 24-hour recalls for all ethnic and sex groups being studied (16). For the 10-y follow-up survey, the QFFQ was updated with changes in the design, food lists to include new food products, and examples given for each item. In a second calibration study, we found high correlations for nutrient densities (0.70–0.74) between the baseline and 10-y follow-up QFFQs. As part of the Dietary Patterns Methods Project (DPMP) (17), 4 predefined diet quality indexes (DQIs) were selected because of their particular relevance to dietary guidlines that had been commonly used in US populations: the Healthy Eating Index–2015 (HEI-2015; theoretical range of 0 to 100 points with 13 components), the Alternative Healthy Eating Index–2010 (AHEI-2010, 0 to 110 points with 11 components), the alternate Mediterranean Diet score (aMED, 0 to 9 points with 9 components), and the Dietary Approaches to Stop Hypertension (DASH) score (8 to 40 points with 8 components). The HEI-2015 evaluates conformance to the 2015–2020 Dietary Guidelines for Americans (18). This DQI replaced the HEI-2010 initially computed in the DPMP. The AHEI-2010 (19) and aMED (20) identify dietary patterns consistently associated with lower risk of chronic disease. The DASH index reflects adherence to a DASH-style diet designed to reduce blood pressure, which is high in fruits and vegetables, moderate in low-fat dairy products, and low in animal protein but high in plant protein (21). For all DQIs, higher scores reflect a higher-quality diet. Three of the DQIs use population-specific cutpoints in order to score the diet quality: the sodium component of AHEI-2010 (based on deciles) and all components of aMED (medians, except the alcohol component with fixed ranges) and DASH (quintiles). In the current analysis, to make DQIs comparable between the 2 surveys, we applied the cutpoints from the baseline diet to compute scores at 10-year follow-up for the distribution-dependent components.

Sociodemographic and lifestyle factors

Among the information collected at baseline, we considered 11 sociodemographic and lifestyle factors for men and 12 for women: age (45–54, 55–64, or 65–75 y), race/ethnicity (white, African American, Native Hawaiian, Japanese American, or Latino), education (≤high school, vocational/some college, or ≥graduated college), marital status (married, separated/divorced, widowed, or never married), BMI (underweight: <18.5; normal: 18.5 to <25; overweight: 25 to <30; obese: 30 to <35; obese: ≥35 kg/m²), smoking status (never, past, or current), physical activity (<0.5, 0.5–1.3, or >1.3 h/d spent in moderate or vigorous activity, based on tertiles), multivitamin use (no or yes), family history of cancer (no or yes), energy intake (430–1670, 1671–2380, or 2381–8670 kcal/d, based on tertiles), alcohol intake (none, >0 to <1, 1 to <2, or ≥2 drinks/d), and menopausal hormone therapy use (never or ever) for women only. These characteristics were chosen based on the previous study in the MEC, in which all of them were associated with diet quality at baseline (14).

Statistical analysis

We computed DQI changes by subtracting values at baseline from those at 10-y follow-up. Since the scales for DQIs are varied, we presented the change per 100 points, calculated as DQI change × 100/theoretical range for index. The theoretical range is 100 points for HEI-2015, 110 for AHEI-2010, 9 for aMED, and 32 for DASH. The variability in the changes in aMED between the 2 surveys is very small, and because there is a limited discrete number of score changes (range: 0–9), diet quality changes in subgroups were only computed for the HEI-2015, AHEI-2010, and DASH.

We computed covariate-adjusted mean changes in DQIs with 95% CIs in subgroups defined by baseline characteristics using the SAS general linear models procedure with a least square means statement; adjustment was made for baseline scores (continuous) and all other subgroup categories. We computed P values for comparisons of mean changes within reference groups, adjusting for multiple comparison using the Tukey method and overall P values for the global test or P values for linear trends across subgroups for each covariate with 3 or more levels. All analyses were performed using SAS version 9.4 (SAS Institute, Inc.). All tests were 2 sided with statistical significance set at P < 0.05.

Results

Sociodemographic and lifestyle characteristics at baseline are presented in Table 1 for MEC participants eligible for this analysis. The participants averaged 57.3 ± 8.3 y of age at baseline. Japanese Americans comprised the largest proportion, while African Americans and Native Hawaiians were the smallest groups. The prevalence of obesity (BMI ≥30) was 15.1% for men and 17.4% for women.

TABLE 1.

Baseline characteristics of 63,255 men and women who completed both baseline and 10-y follow-up food frequency questionnaires in the Multiethnic Cohort Study¹

Baseline characteristics	Men	Women
n	27,001	36,254
Age, y
45–54	42.8	41.2
55–64	35.2	35.2
65–75	22.0	23.7
Race/ethnicity
African American	7.7	12.7
Native Hawaiian	7.4	7.7
Japanese American	36.0	33.3
Latino	20.5	18.3
White	28.4	28.0
Education
≤High school	29.7	36.2
Vocational/some college	30.4	31.3
≥Graduated college	39.9	32.4
Marital status
Married	79.1	65.5
Separated/divorced	11.3	17.7
Widowed	2.4	10.8
Never married	7.3	6.0
BMI
<18.5	0.4	2.4
18.5 to <25	37.0	50.0
25 to <30	47.5	30.3
30 to <35	12.0	11.7
≥35	3.1	5.7
Smoking status
Never	35.3	59.4
Past	49.9	28.8
Current	14.9	11.8
Physical activity, h/d2	1.07 (1.46)	0.71 (1.07)
Multivitamin use	48.3	54.4
Family history of cancer	35.9	39.8
History of high blood pressure	32.1	29.1
History of diabetes	7.0	6.0
History of stroke	1.0	0.9
Ever use of menopausal hormone therapy	—	49.5
Energy intake, kcal/d	2431 ± 1036	1946 ± 869
Alcohol intake, g/d	3.6 (18.3)	0 (2.2)

¹Values are means ± SDs or percentages, unless otherwise indicated.

²Hours spent in moderate or vigorous activity per day.

Mean DQIs and changes between baseline and 10-y follow-up surveys are presented in Table 2. Mean DQIs were higher in women than men except for those determined with the aMED (P < 0.001). Over 10 y, mean DQI changes were greater in men than in women except for those determined with the aMED (P' ≤ 0.001).

TABLE 2.

Dietary quality indexes at baseline and 10-y follow-up surveys in the Multiethnic Cohort Study¹

	Baseline	10-y follow-up	Change2	Change per 100 points3
Men (n = 27,001)
HEI-2015	65.6 ± 10.3	68.8 ± 10.6	3.2 ± 9.8	3.2 ± 9.8
AHEI-2010	64.5 ± 9.9	67.1 ± 10.4	2.5 ± 10.0	2.3 ± 9.1
aMED	4.3 ± 1.8	4.4 ± 1.8	0.1 ± 1.9	1.1 ± 20.6
DASH	24.0 ± 4.5	25.2 ± 4.4	1.2 ± 4.0	3.7 ± 12.6
Women (n = 36,254)
HEI-2015	69.4 ± 10.3	72.3 ± 10.6	2.9 ± 9.9	2.9 ± 9.9
AHEI-2010	65.7 ± 9.3	67.9 ± 9.9	2.2 ± 9.5	2.0 ± 8.6
aMED	4.2 ± 1.8	4.3 ± 1.8	0.1 ± 1.9	1.1 ± 20.6
DASH	24.2 ± 4.4	25.3 ± 4.4	1.1 ± 4.0	3.4 ± 12.6

¹Values are means ± SDs. All means were significantly different between men and women, P ≤ 0.001, except for the changes in the aMED, by t-test. AEHI, Alternative Healthy Eating Index; aMED, alternate Mediterranean Diet score; DASH, Dietary Approaches to Stop Hypertension; HEI, Healthy Eating Index.

²Change = score at 10-y follow-up – score at baseline.

³Change per 100 points = (score at 10-y follow-up – score at baseline) × 100/theoretical range. The theoretical range is 100 points for HEI-2015, 110 for AHEI-2010, 9 for aMED, and 32 for DASH.

Supplemental Figure 1 displays component scores for the HEI-2015 at baseline and 10-y follow-up simultaneously as percentages of the maximum scores (5 or 10 points per component). For all components, mean scores remained similar or slightly increased (improved) over 10 y in both men and women, except for the saturated fat component (moderation component with higher score indicating lower intake), for which the scores slightly decreased (worsened). Scores for the whole-grain (men: 51.6%; women: 59.1%), dairy (men: 42.8%; women: 51.5%), and sodium (men: 42.6%; women: 43.7%) components remained low relative to the other components. In the AHEI-2010, scores for the trans fat component decreased, indicating higher consumption in both men and women, but were still >9 points. For AHEI-2010 and DASH, 2 components, sweetened beverages and sodium, contributed most to the improvements in the scores, while the refined grains component contributed most to the improvement in HEI-2015 scores. Otherwise there were no substantial changes in any of the other components (Supplemental Table 1), although the changes were statistically significant for most of the components mainly due to the large sample size. The component score for whole grains in the AHEI-2010 remained very low both in men (2.8 points) and women (3.3 points). Among the same age groups, in cross-sectional comparisons between baseline and 10-y follow-up results, the mean DQIs were slightly higher for most age groups in both men and women for the HEI-2015 and AHEI-2010 (Supplemental Table 2). For example, men aged 55–64 y at baseline and at follow-up had mean HEI-2015 scores of 66.0 and 67.9 and mean AHEI-2010 scores of 64.7 and 66.8, respectively.

Table 3 presents mean changes per 100 points in 3 of the DQIs over 10 y by baseline characteristics in men, with adjustment for all of the covariates listed in the table. Baseline DQI scores were inversely associated with changes in diet quality over time in both men and women. Therefore, all models were further adjusted for baseline DQI scores. Almost all subgroups for the 11 characteristics showed an increase in all DQIs, except for African American (no significant change) and Latino (0.9-point decrease) men in the AHEI-2010. Overall, Japanese American men had greater increases across the DQIs than white men (by 1.3 points for HEI-2015 and 2.2 points for AHEI-2015, P < 0.001), while white men showed a greater increase in the DASH score only (by 0.8–2.5, compared with African American, Native Hawaiian, and Latino men, P ≤ 0.003). Mean increases in the scores were greater in the participants in younger age groups (45–64 compared with 65–75 y) by 0.5–1.3 point (P ≤ 0.001), except for DASH scores, in which older age groups (55–75 compared with 45–54 y) showed greater increases by 0.5–0.8 point (P ≤ 0.045). Overall, DQI increases were greater in more educated groups (compared with ≤high school by 0.7–1.2, P ≤ 0.001), the normal or overweight group (except for AHEI-2010, by 0.7–1.3, P ≤ 0.014), never smokers (never and past smokers for AHEI-2010 compared with current smokers by 1.5–2.7, P < 0.001), more physically active men (>1.3 compared with < 0.5 h/d by 0.7, P < 0.001), and multivitamin users (by 0.4–0.7, P < 0.001). Alcoholic beverage drinkers at baseline showed greater increases in the scores compared with nondrinkers (by 0.5–2.4, P ≤ 0.048). The overall tendency for HEI-2015 changes by baseline characteristics was generally observed in African American, Japanese American, Latino, and white men, but less consistently observed in Native Hawaiian men (Supplemental Table 3). Mean changes in the HEI-2015 among Native Hawaiians only differed by smoking status (current compared with never and past smokers) and physical activity level (the highest and middle compared with the lowest groups). Similarly, most subgroups in women showed increases in the 3 DQIs, except for African American (no change in the AHEI-2010 and DASH) and Latina (1 point decrease in the AHEI-2010) women and current smokers (no change in all DQIs, Table 4). In women, DQI increases over 10 y were greater in younger age groups (45–64 compared with 65–75 y by 0.5–1.5 points, P ≤ 0.009), more educated groups (compared with ≤high school by 0.5–1.5, P < 0.001), the normal-weight group (by 0.6–2.5, P < 0.001), never smokers (never and past smokers for AHEI-2010 compared with current smokers by 1.7–2.5, P < 0.001), more physically active groups (>1.3 compared with <0.5 h/d by 0.5–0.8, P < 0.001), and multivitamin users (by 0.6–0.7, P < 0.001). Marital status, family history of cancer, and menopausal hormone therapy use were not significantly associated with DQI changes in women. Unlike in men, women who had 2 or more drinks of alcoholic beverages at baseline showed smaller increases in the HEI-2015 (by 0.9, P = 0.001). In racial/ethnic-specific analyses for HEI-2015 changes in women (Supplemental Table 4), Native Hawaiians showed less consistent trends than the other groups. For example, mean HEI-2015 changes did not vary by BMI group in Native Hawaiians, while normal-weight women at baseline in the other racial/ethnic groups showed greater increases than overweight or obese women.

TABLE 3.

Diet quality index changes by baseline characteristics among 27,001 men in the Multiethnic Cohort Study¹

	HEI-2015		AHEI-2010		DASH
Baseline characteristics	Change per 100 points	P 2	Change per 100 points	P 2	Change per 100 points	P 2
Age, y
45–54	2.5 (2.2, 2.8)	Ref.	1.6 (1.3, 1.9)	Ref.	2.0 (1.6, 2.4)	Ref.
55–64	2.6 (2.3, 2.9)	0.70	1.6 (1.3, 1.9)	0.99	2.8 (2.4, 3.2)	<0.001
65–75	1.3 (1.0, 1.7)	<0.001	1.1 (0.8, 1.4)	0.001	2.5 (2.0, 2.9)	0.045
P3	<0.001		<0.001		0.017
Race/ethnicity
White	1.8 (1.4, 2.1)	Ref.	1.6 (1.3, 1.8)	Ref.	3.4 (3.0, 3.8)	Ref.
African American	2.0 (1.6, 2.5)	0.71	-0.1 (-0.5, 0.3)	<0.001	0.9 (0.3, 1.5)	<0.001
Native Hawaiian	2.3 (1.8, 2.7)	0.13	2.8 (2.4, 3.2)	<0.001	2.2 (1.6, 2.7)	<0.001
Japanese American	3.1 (2.7, 3.4)	<0.001	3.8 (3.5, 4.1)	<0.001	3.0 (2.6, 3.4)	0.21
Latino	1.6 (1.2, 1.9)	0.81	-0.9 (-1.2, -0.6)	<0.001	2.6 (2.2, 3.1)	0.003
P3	<0.001		<0.001		<0.001
Education
≤High school	1.8 (1.4, 2.1)	Ref.	0.8 (0.5, 1.1)	Ref.	2.2 (1.8, 2.6)	Ref.
Vocational/some college	2.1 (1.8, 2.4)	0.08	1.5 (1.2, 1.8)	<0.001	2.3 (1.9, 2.7)	0.90
≥Graduated college	2.6 (2.3, 2.9)	<0.001	2.0 (1.7, 2.3)	<0.001	2.8 (2.4, 3.3)	0.001
P3	<0.001		<0.001		<0.001
Marital status
Married	2.5 (2.3, 2.8)	Ref.	1.7 (1.5, 1.9)	Ref.	2.7 (2.4, 3.0)	Ref.
Separated/divorced	1.9 (1.6, 2.3)	0.002	1.5 (1.1, 1.8)	0.38	2.3 (1.9, 2.8)	0.41
Widowed	1.8 (1.1, 2.5)	0.15	1.1 (0.4, 1.7)	0.19	1.7 (0.8, 2.6)	0.15
Never married	2.4 (1.9, 2.8)	0.83	1.5 (1.1, 1.9)	0.73	3.0 (2.4, 3.5)	0.75
P3	<0.001		0.08		0.045
Body mass index, kg/m2
18.5–<25	2.6 (2.3, 2.9)	Ref.	1.5 (1.3, 1.8)	Ref.	3.0 (2.7, 3.4)	Ref.
25–<30	2.5 (2.2, 2.7)	0.52	1.4 (1.2, 1.7)	0.73	2.8 (2.4, 3.1)	0.30
30–<35	1.9 (1.5, 2.2)	<0.001	1.3 (1.0, 1.7)	0.61	2.2 (1.7, 2.7)	0.003
≥35	1.6 (1.0, 2.3)	0.014	1.4 (0.9, 2.0)	0.99	1.7 (0.9, 2.5)	0.008
P3	<0.001		0.68		<0.001
Smoking status
Never	3.0 (2.7, 3.3)	Ref.	2.0 (1.7, 2.3)	Ref.	3.5 (3.1, 3.9)	Ref.
Past	2.5 (2.2, 2.8)	<0.001	1.9 (1.7, 2.2)	0.85	2.9 (2.5, 3.3)	<0.001
Current	0.9 (0.5, 1.3)	<0.001	0.4 (0.0, 0.7)	<0.001	0.8 (0.4, 1.3)	<0.001
P3	<0.001		<0.001		<0.001
Physical activity, h/d
<0.5	1.7 (1.4, 2.1)	Ref.	1.0 (0.7, 1.3)	Ref.	2.1 (1.6, 2.5)	Ref.
0.5–1.3	2.3 (2.0, 2.6)	<0.001	1.6 (1.3, 1.8)	<0.001	2.5 (2.1, 2.9)	0.038
>1.3	2.4 (2.1, 2.7)	<0.001	1.7 (1.4, 2.0)	<0.001	2.7 (2.3, 3.1)	<0.001
P3	<0.001		<0.001		<0.001
Multivitamin use
No	1.8 (1.5, 2.1)	Ref.	1.2 (0.9, 1.5)	Ref.	2.1 (1.7, 2.4)	Ref.
Yes	2.5 (2.2, 2.8)	<0.001	1.6 (1.4, 1.9)	<0.001	2.8 (2.4, 3.2)	<0.001
P3	<0.001		<0.001		<0.001
Family history of cancer
No	2.1 (1.8, 2.4)	Ref.	1.4 (1.2, 1.7)	Ref.	2.3 (2.0, 2.7)	Ref.
Yes	2.2 (1.9, 2.5)	0.20	1.4 (1.1, 1.7)	0.90	2.5 (2.1, 2.9)	0.23
P3	0.20		0.90		0.23
Energy intake, kcal/d
430–1670	2.2 (1.8, 2.5)	Ref.	1.4 (1.1, 1.7)	Ref.	2.4 (1.9, 2.8)	Ref.
1671–2380	2.3 (2.0, 2.6)	0.76	1.5 (1.2, 1.8)	0.71	2.6 (2.2, 3.0)	0.47
2381–8670	2.0 (1.7, 2.3)	0.38	1.5 (1.2, 1.7)	0.80	2.3 (2.0, 2.7)	0.99
P3	0.18		0.52		0.98
Alcohol intake, drink/d
None	1.7 (1.4, 2.1)	Ref.	0.4 (0.2, 0.7)	Ref.	2.0 (1.6, 2.4)	Ref.
>0–<1	2.3 (2.0, 2.6)	<0.001	1.2 (0.9, 1.5)	<0.001	2.5 (2.2, 2.9)	0.007
1–<2	2.3 (1.9, 2.7)	0.006	1.3 (0.9, 1.6)	<0.001	2.6 (2.1, 3.1)	0.06
≥2	2.2 (1.8, 2.6)	0.029	2.8 (2.5, 3.2)	<0.001	2.6 (2.1, 3.0)	0.048
P3	<0.001		<0.001		0.014

¹Values are means (95% CIs), adjusted for baseline DQI score and all covariates in the table. Change per 100 points = (score at 10-y follow-up – score at baseline) × 100/theoretical range. AEHI, Alternative Healthy Eating Index; DASH, Dietary Approaches to Stop Hypertension; HEI, Healthy Eating Index; Ref., reference.

²Comparison between categories with adjustment for multiple comparison.

³Overall comparison for race/ethnicity and marital status, and linear trend across subgroups for age, education, BMI, physical activity, energy intake, and alcohol intake.

TABLE 4.

DQI changes by baseline characteristics among 36,254 women in the Multiethnic Cohort Study¹

	HEI-2015		AHEI-2010		DASH
Baseline characteristics	Change per 100 points	P 2	Change per 100 points	P 2	Change per 100 points	P 2
Age, y
45–54	2.0 (1.8, 2.3)	Ref.	1.8 (1.6, 2.0)	Ref.	2.1 (1.7, 2.4)	Ref.
55–64	1.8 (1.5, 2.0)	0.06	1.3 (1.1, 1.5)	<0.001	1.8 (1.5, 2.1)	0.18
65–75	0.5 (0.2, 0.8)	<0.001	0.7 (0.4, 1.0)	<0.001	1.3 (0.9, 1.7)	<0.001
P3	<0.001		<0.001		<0.001
Race/ethnicity
White	0.9 (0.6, 1.1)	Ref.	1.6 (1.4, 1.8)	Ref.	2.9 (2.6, 3.2)	Ref.
African American	1.6 (1.3, 1.9)	<0.001	0.0 (-0.3, 0.3)	<0.001	0.1 (-0.3, 0.5)	<0.001
Native Hawaiian	1.3 (0.9, 1.7)	0.25	2.5 (2.2, 2.9)	<0.001	1.3 (0.8, 1.8)	<0.001
Japanese American	2.1 (1.8, 2.4)	<0.001	3.3 (3.1, 3.6)	<0.001	1.8 (1.5, 2.2)	<0.001
Latino	1.3 (1.0, 1.6)	0.08	-1.0 (-1.3, -0.7)	<0.001	2.4 (2.0, 2.8)	0.12
P3	<0.001		<0.001		<0.001
Education
≤High school	0.9 (0.6, 1.1)	Ref.	0.5 (0.3, 0.8)	Ref.	0.9 (0.6, 1.3)	Ref.
Vocational/some college	1.4 (1.1, 1.7)	<0.001	1.3 (1.1, 1.5)	<0.001	1.7 (1.4, 2.1)	<0.001
≥Graduated college	2.0 (1.7, 2.3)	<0.001	2.0 (1.8, 2.3)	<0.001	2.5 (2.1, 2.8)	<0.001
P3	<0.001		<0.001		<0.001
Marital status
Married	1.4 (1.2, 1.7)	Ref.	1.4 (1.2, 1.6)	Ref.	1.6 (1.3, 1.9)	Ref.
Separated/divorced	1.5 (1.2, 1.8)	0.96	1.4 (1.2, 1.6)	0.99	1.8 (1.4, 2.2)	0.61
Widowed	1.6 (1.2, 1.9)	0.86	1.3 (1.0, 1.6)	0.99	1.7 (1.3, 2.2)	0.87
Never married	1.2 (0.8, 1.7)	0.82	1.0 (0.7, 1.4)	0.17	1.7 (1.2, 2.3)	0.97
P3	0.62		0.22		0.6
BMI
18.5 to <25	2.5 (2.2, 2.7)	Ref.	2.0 (1.8, 2.2)	Ref.	3.2 (2.9, 3.5)	Ref.
25 to <30	1.6 (1.3, 1.8)	<0.001	1.4 (1.2, 1.6)	<0.001	2.0 (1.7, 2.3)	<0.001
30 to <35	1.2 (0.9, 1.5)	<0.001	1.0 (0.7, 1.3)	<0.001	1.0 (0.6, 1.5)	<0.001
≥35	0.4 (0.0, 0.9)	<0.001	0.8 (0.4, 1.2)	<0.001	0.7 (0.1, 1.2)	<0.001
P3	<0.001		<0.001		<0.001
Smoking status
Never	2.4 (2.2, 2.7)	Ref.	1.8 (1.6, 2.0)	Ref.	2.6 (2.3, 3.0)	Ref.
Past	1.9 (1.6, 2.1)	<0.001	1.9 (1.6, 2.1)	0.95	2.3 (2.0, 2.6)	0.051
Current	-0.0 (-0.3, 0.3)	<0.001	0.2 (-0.1, 0.4)	<0.001	0.2 (-0.3, 0.6)	<0.001
P3	<0.001		<0.001		<0.001
Physical activity, h/d
<0.5	1.2 (0.9, 1.4)	Ref.	0.9 (0.7, 1.1)	Ref.	1.4 (1.1, 1.7)	Ref.
0.5–1.3	1.5 (1.2, 1.7)	0.037	1.3 (1.1, 1.5)	<0.001	1.6 (1.2, 1.9)	0.49
>1.3	1.7 (1.4, 1.9)	<0.001	1.7 (1.5, 1.9)	<0.001	2.2 (1.8, 2.5)	<0.001
P3	<0.001		<0.001		<0.001
Multivitamin use
No	1.1 (0.9, 1.4)	Ref.	1.0 (0.8, 1.2)	Ref.	1.3 (1.0, 1.6)	Ref.
Yes	1.7 (1.5, 2.0)	<0.001	1.6 (1.4, 1.8)	<0.001	2.1 (1.8, 2.4)	<0.001
P3	<0.001		<0.001		<0.001
Family history of cancer
No	1.4 (1.1, 1.6)	Ref.	1.2 (1.0, 1.4)	Ref.	1.7 (1.4, 2.0)	Ref.
Yes	1.5 (1.2, 1.7)	0.35	1.3 (1.1, 1.6)	0.23	1.7 (1.4, 2.1)	0.54
P3	0.35		0.23		0.54
Menopausal hormone therapy use
Never	1.4 (1.1, 1.6)	Ref.	1.3 (1.1, 1.5)	Ref.	1.7 (1.4, 2.0)	Ref.
Ever	1.5 (1.2, 1.7)	0.26	1.2 (1.0, 1.5)	0.30	1.7 (1.4, 2.0)	0.90
P3	0.26		0.30		0.90
Energy intake, kcal/d
430–1670	1.5 (1.3, 1.8)	Ref.	1.0 (0.8, 1.2)	Ref.	1.7 (1.4, 2.1)	Ref.
1671–2380	1.5 (1.3, 1.8)	0.95	1.4 (1.2, 1.6)	<0.001	1.8 (1.5, 2.2)	0.73
2381–8670	1.2 (1.0, 1.5)	0.06	1.4 (1.2, 1.7)	<0.001	1.5 (1.2, 1.9)	0.49
P3	0.024		<0.001		0.26
Alcohol intake, drink/d
None	1.7 (1.5, 1.9)	Ref.	0.7 (0.6, 0.9)	Ref.	2.0 (1.7, 2.3)	Ref.
>0 to <1	1.7 (1.4, 1.9)	0.99	1.3 (1.1, 1.5)	<0.001	1.9 (1.6, 2.2)	0.96
1 to <2	1.6 (1.1, 2.0)	0.98	1.0 (0.6, 1.4)	0.48	1.6 (1.1, 2.2)	0.55
≥2	0.8 (0.3, 1.3)	0.001	2.1 (1.7, 2.6)	<0.001	1.3 (0.6, 1.9)	0.09
P3	0.003		<0.001		0.011

¹Values are means (95% CIs), adjusted for baseline DQI score and all covariates in the table. Change per 100 points = (score at 10-y follow-up – score at baseline) × 100/theoretical range. AHEI, Alternative Healthy Eating Index; DASH, Dietary Approaches to Stop Hypertension; DQI, diet quality index; HEI, Healthy Eating Index.

²Comparison between categories with adjustment for multiple comparison.

³Overall comparison for race/ethnicity and marital status, and linear trend across subgroups for age, education, BMI, physical activity, energy intake, and alcohol intake.

Discussion

In this multiethnic population, overall diet quality, assessed by commonly used DQIs, slightly improved over 10 y, although the scores for some components of the DQIs decreased or remained low. Change patterns in the DQIs varied by sociodemographic and lifestyle characteristics at baseline. Japanese American ethnicity, higher education, normal BMI, nonsmoking, higher physical activity level, and multivitamin use at baseline were associated with a greater increase in the DQIs both in men and women. A similar tendency was observed in African Americans, Japanese Americans, Latinos, and whites in general, but less consistent in Native Hawaiians, when examining HEI-2015 changes. There were also sex- and/or DQI-specific changes. Widowed men tended to have a smaller increase in the DQIs than married men, while marital status was not associated with DQI change in women. Younger age groups both in men and women had a greater increase for all DQIs, except for the DASH in men showing an opposite trend. For the AHEI-2010, Latinos showed a slight decrease over time, and African Americans showed no change, whereas whites, Native Hawaiians, and Japanese Americans had increases. For the DASH, African Americans had no (women) or the smallest (men) increases.

Overall diet quality in US adults (aged 20–85 y) improved across the 12-y period in the NHANES, where cross-sectional assessment of diet quality was determined using the AHEI-2010 (4). The mean AHEI-2010 score increased from 39.9 in 1999–2000 to 46.8 in 2009–2010, but still remained low considering the maximum possible points of 110. Across the 12 y in the NHANES cross-sectional surveys, the non-Hispanic white group showed an improvement in the AHEI-2010 (without the trans fat component, P-trend < 0.001), while non-Hispanic black and Mexican American groups did not (P-trend ≥ 0.06) (4). Similarly in the MEC, whites had increased AHEI-2010 scores between the first (1993–1996) and follow-up (2003–2007) surveys, whereas African Americans and Latinos did not show an increase in both men and women. The improvement in diet quality over 10 y observed in the MEC may also reflect the general trend toward better diet in the United States, in addition to aging. Indeed, we found a slight improvement in the AHEI-2010 for most age groups using cross-sectional comparisons among the same age groups between baseline and the 10-y follow-up.

A main contributor to the improvement in the AHEI-2010 in the NHANES was the reduction in trans fat consumption (4). On the contrary, trans fat consumption in the MEC participants increased slightly over 10 y. Thus the component score for trans fat was reduced (from 9.9 to 9.1) but was still high. Saturated fat intake (as density) in the MEC participants slightly increased over time as assessed by the HEI-2015 (mean component score: 7.9 to 7.3), while the total HEI-2015 score improved from 67.8 to 70.8. In the NHANES 2011–2012 (≥2 y), the mean score was 6.1 for the saturated fat component and 56.6 for the total HEI-2015 score, but trends in diet quality over time were not reported (18). In the MEC, the scores for the whole-grain, dairy, and sodium components in the HEI-2015 remained relatively low. The component score for whole grains in the AHEI-2010 also remained very low in the MEC, while whole-grain consumption in US adults increased from 0.56 serving/d in 1999–2000 to 1.00 serving/d in 2011–2012 (22).

A few longitudinal studies have examined changes in diet quality over time among adults. In the Nurses' Health Study, the middle quintile (relatively no change) group of the AHEI-2010 changes showed a mean increase of 2.2 points over 12 y (6). The corresponding number in the Health Professionals Follow-up Study was a 3.4 point increase (6). In the Women's Health Initiative Observational Study, 12% of women had a ≥11-point increase in their AHEI-2010 scores over 3 y, while 8.8% had a ≥11 point or more decrease (13). The studies in these large US cohorts found an overall increase in diet quality over time, although detailed patterns have not been reported. In a cohort of young American adults (18–30 y at baseline), the diet quality score associated with cardiovascular disease risk increased from 61.4 to 71.1 (maximum possible score: 132) over 20 y with a greater increase among the black than in the white participants (23). A study in an Australian cohort (25–75 y at baseline) also reported an overall improvement in diet quality over 15 y (the Dietary Guideline Index: 71.9 to 76.2 in men and 80.6 to 83.8 in women; theoretical maximum score: 130) (5). The study found that younger age, higher occupational level in men, and physical activity and menopausal hormone therapy in women were independently associated with a greater increase in the score (5). In the MEC, we also observed a sex-specific association for marital status, by which DQI changes varied in men but not in women. An Australian cohort (aged ≥55 y) found that diet quality (the revised Dietary Guideline Index, DGI-2013) improved only in men over the 4 y of the study, but smoking was associated with a decrease in DGI-2013 both for men and women (8). Another Australian cohort of middle-aged women reported no substantial change in diet quality over time (the Australian Recommended Food Score: 32.6 in 2001 to 33.1 in 2013; theoretical maximum score: 74) (24). The sociodemographic and lifestyle factors related to changes in diet quality over time have been also found to predict diet quality in older adults (25) and various populations (26–28).

In the present study, some of the change patterns varied across the DQIs, likely due to the different scoring schemes of each DQI. The HEI-2015 is density-based (intake per 1000 kcal), while the AHEI-2010, aMED, and DASH are based on absolute intakes. For the HEI-2015 and AHEI-2010, component scores are continuous proportionally to intake levels once they are above the minimal levels (adequacy components) or below the maximum levels (moderation components). However, the DASH has 5 levels (1 to 5) and the aMED has only 2 (0 or 1) for each component, based on distributions of intakes. In addition, the HEI-2015 (0 to 100 points) and AHEI-2010 (0 to 110) have wider ranges of total scores compared with the DASH (8 to 40) and aMED (0 to 9), although we presented changes per 100 points to compare across the scores. Thus, the HEI-2015 and AHEI-2010 are more likely to be sensitive to small changes in dietary intakes over time, compared with the DASH and aMED. Particularly, aMED is too granular to reflect small changes in diet quality. Indeed, aMED scores for 22.5% of the participants remained the same, while their mean HEI-2015, AHEI-2010, and DASH scores improved by 2.7, 2.0, and 1.0 points over 10 y.

This study has a number of important strengths including a population-based prospective design, a large sample size, and participants with various racial/ethnic backgrounds. The comprehensive questionnaires collected a wide range of sociodemographic and lifestyle information, and eating habits using a validated QFFQ. The DQIs used in the current study were calculated using standardized methods with consensus in 3 large US cohorts, and demonstrated their predictability for health outcomes across (17) and within the cohorts (29–37).

In addition to measurement error inherent in FFQs, several limitations should be considered in interpreting the study findings. Although we excluded participants who had prevalent cancer or heart disease at either survey, we could not rule out dietary changes due to underlying illness or developing medical conditions during the 10-y follow-up. When we further adjusted for prevalent (33.8% at baseline) and incident (23.8% during follow-up) medical conditions, including high blood pressure, diabetes, and stroke, the patterns of changes in the DQIs remained similar. However, we were not able to consider other medical conditions, such as oral health and gastrointestinal problems, or other factors, such as cognitive ability in youth and over the life course (38, 39), that might affect diet quality. In addition, food compositions might change over time for certain foods, but we used the same food composition database for the baseline and 10-y follow-up questionnaires. The MEC participants are largely representative of the target population in Hawaii and Southern California as evidenced by the similarity in marital status and education as the 1990 census for those regions. However, selection bias due to the volunteerism may limit generalizability of our findings. In addition, participants who completed the follow-up QFFQ tended to be younger, Japanese American, white, never smokers, more educated, and less obese than with nonrespondents. Compared with general US adults, overall diet quality in the MEC participants appears to be higher, while changes in diet quality over time seem to be smaller, although direct comparison was not possible. Compared to the other 2 US cohorts, with which a standard method was developed for the DQI computation, the MEC participants had an approximately 10-point higher AHEI-2010 score at baseline, while the other 3 DQIs (HEI-2010, aMED, and DASH) were similar across the cohorts (17). Although socioeconomic status (SES) affects diet quality, income could not be considered in the analysis due to lack of information. Instead, education level served as a proxy for individual-level SES in the current analyses. When neighborhood SES was further adjusted for by linkage to census and geospatial data (40) in addition to education levels, the results for the DQI changes remained similar.

In conclusion, overall diet quality improved over 10 y in this multiethnic population. However, among the components in the dietary indexes examined, scores for saturated and trans fats worsened, indicating increased consumption, and those for whole grains, dairy, and sodium remained unchanged at a low quality level. Sociodemographic and lifestyle factors closely associated with diet quality, including race/ethnicity, education, body weight, smoking, physical activity, and multivitamin use, also predict subsequent changes in diet quality over time.

Notes

The Multiethnic Cohort Study is funded by grant U01 CA164973 from the National Cancer Institute (NCI). This study was partially supported by the NCI grants R03 CA223890, HHSN261201200423P, and P30 CA071789. MK was supported by a Support Program for Women in Science, Engineering and Technology grant from the National Research Foundation of Korea (NRF) (grant number 2019H1C3A1032224).

Author disclosures: The authors report no conflicts of interest.

Supplemental Figure 1 and Supplemental Tables 1–4 are available from the "Supplementary data" link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/jn/.

Abbreviations used: AHEI-2010, Alternative Healthy Eating Index–2010; aMED, alternate Mediterranean Diet score; DASH, Dietary Approaches to Stop Hypertension; DQI, diet quality index; HEI-2015, Healthy Eating Index–2015; MEC, Multiethnic Cohort; QFFQ, quantitative food frequency questionnaire; SES, socioeconomic status.