Dietary pattern analysis is a relatively new method to quantify diet quality, monitor changes in population-based diet quality, and assess the relationships between diet quality and health-related outcomes [1
]. Analyses of individual nutrients and foods are still important; however, as we consume foods in combination, the synergistic effect of food on health must be considered [2
]. Studying diet patterns also allows researchers to promote dietary recommendations in the form of patterns, which may be easier for the general public to interpret and follow [5
Dietary indices are important tools used to assess dietary patterns. They may be derived from theoretically based scoring systems or a priori approaches, which are guided by evidence-based research [2
]. For example, the Healthy Eating Index (HEI) is constructed to reflect the evidence-based recommendations of the Dietary Guidelines for Americans (DGA) and to evaluate conformance to these recommendations [6
]. Every five years the Dietary Guidelines Advisory Committee reviews current nutrition research and provides an Advisory Report to the United States Department of Agriculture (USDA) and Health and Human Services (HHS) [4
]. Guided by this report, the USDA and HHS publish the Dietary Guidelines for Americans (DGA), which provide evidence-based food and beverage recommendations to the public [4
]. Since the implementation of the original HEI in 1995, development and application of successive HEIs to assess and monitor dietary status have been ongoing [3
]. The HEI-2005 introduced density-based scoring standards, which were continued in the HEI-2010 and HEI-2015. Additional changes to scoring standards were implemented between each new HEI to reflect updates in the DGA [3
To add to the evidence for using dietary patterns to guide dietary advice, the National Cancer Institute (NCI) initiated the Dietary Patterns Methods Project (DPMP) [13
]. This is a collaborative project between three large cohort studies, including the NIH-AARP Diet and Health Study (AARP) [16
], the Multiethnic Cohort Study (MEC) [18
] and the Women’s Health Initiative Observational Study (WHI-OS) [20
]. These groups adopted overlapping and complementary research questions and conducted standardized diet analysis using four dietary indices, including the HEI-2010, the Alternative Healthy Eating Index-2010 (AHEI-2010), the alternate Mediterranean Diet (aMED), and the Dietary Approaches to Stop Hypertension (DASH) [15
]. A summary of the results revealed that men and women with higher quality diets (higher index scores) were at significantly lower risk (11–28%) of death from all-causes, cardiovascular disease (CVD), and cancer compared to people with lower quality diets (lower index scores) [15
]. This was true for mortality assessments made with each dietary index, with the exception of women from the WHI-OS cohort, where the AHEI-2010 score was not associated with cancer death [15
The HEI-2015 was introduced to reflect the DGA 2015–2020 [12
]. The HEI-2015 is comprised of 13 components and the maximum HEI score is 100 points [12
]. There are nine adequacy components (foods to eat enough of) including Total Fruits, Whole Fruits, Total Vegetables, Greens and Beans, Whole Grains, Dairy, Total Protein Foods, Seafood and Plant Proteins, and Fatty Acids. There are four moderation components (foods to limit) including Refined Grains, Sodium, Added Sugars, and Saturated Fats. Most components are scored on a density basis, that is, amounts per 1000 kcal of intake [12
]. The Fatty Acids component is scored using the ratio of poly- and mono-unsaturated fatty acids (PUFAs and MUFAs) to saturated fatty acids (SFAs), and Added Sugars and Saturated Fats are scored as a percentage of total energy intake. The key differences between the HEI-2010 and the HEI-2015 include the scoring for legumes and the Empty Calories component. In the HEI-2010, legumes were allocated into two components and have expanded into four components with the HEI-2015, including Total Vegetables, Greens and Beans, Total Protein Foods, and Seafood and Plant Proteins [12
]. In the HEI-2015, the Empty Calories single component was replaced with the two components of Saturated Fat and Added Sugars [12
]. Alcohol is no longer included in the Empty Calories component score; instead, the energy (kcal) from alcohol is now added to the total energy intake per day [12
To assess the efficacy of the new HEI score, the AARP, MEC, and the WHI-OS each completed a standardized mortality assessment using the HEI-2015. The aim of this study is to examine the association between the HEI-2015 and mortality from all-cause, CVD, and cancer in the MEC, which represents a large cohort of adult men and women from five distinct ethnic groups residing in Hawaii and Los Angeles (LA).
Comparing those with the lowest quality diets to those with the highest quality, the reduction in risk of mortality from all-cause, CVD, and cancer was 21%, 24%, and 20%, respectively, for men and 21%, 25%, and 16%, respectively, for women. In the HEI-2010 mortality analysis with the MEC, the reduction in the risk of mortality from all-cause, CVD, and cancer was 25%, 26%, and 24%, respectively, for men, and 21%, 23%, and 11%, respectively, for women [19
]. Therefore, HRs have slightly improved for women and marginally lowered for men between the HEI-2015 and HEI-2010 mortality analyses with the MEC. When the HEI-2015 scoring standards were applied to the MEC sample with fewer mortality cases, from the follow-up period of the HEI-2010 mortality analysis, the HR results were also very similar. The reduction in risk of mortality from all-cause, CVD, and cancer was 22%, 22%, and 20%, respectively, for men and 21%, 23%, and 11%, respectively, for women. For the DPMP, standardized mortality assessments were also conducted with the HEI-2010 and the WHI-OS, and AARP cohorts. The HEI-2010 mortality analysis with the WHI-OS was conducted with 63,115 US women, of whom 83% identified as white [21
]. The results of this study showed a reduction in risk between quintile 1: quintile 5, with HRs of 24%, 22%, and 23% for all-cause, CVD, and cancer mortality, respectively. The HEI-2010 mortality analysis for the AARP contained 242,321 men and 182,342 women from six US states and over 90% of participants identified as white [17
]. The same mortality analysis applied to the AARP showed a reduction in risk of 22%, 15%, and 24% for all-cause, CVD, and cancer mortality, respectively, for men and 23%, 21%, and 18%, respectively, for women. The results from the HEI-2015 mortality analysis with the MEC and the HEI-2010 mortality analysis with the WHI-OS and AARP are very similar, more so than the mortality results from the HEI-2010 analyses with the MEC. For example, the reduction in risk of cancer mortality for women in the MEC, WHI-OS, and AARP were 11%, 23%, and 18%, respectively, using the HEI-2010. The reduction in risk of cancer mortality for women in the MEC changed to 16% in the HEI-2015 analysis. A recent meta-analysis on the association between the HEI-2005/HEI 2010, AHEI, and DASH and health outcomes for both men and women had similar findings to this HEI-2015 analysis with the MEC [32
]. Consistency between the HEI-2010 analyses and this HEI-2015 analysis reinforces the use of HEI as an assessment of diet quality.
The analysis that removed one component at a time did not change the protective association of the remaining HEI components. Therefore, no one HEI component made an independent significant contribution to the total score (Figure 1
). The Saturated Fat component did distinguish itself, as its removal changed the HRs slightly towards the null for all three mortality outcomes among men only. Previous versions of the HEI included Saturated Fat as part of the empty calories component comprised of solid fat, added sugars, and alcohol [3
]. The HEI-2015 offers the first opportunity to evaluate the Added Sugars and Saturated Fat components independently, however alcohol is now part of total energy intake. Removal of the Added Sugars component had no influence on moving the hazard ratios. On the other hand, the component whose removal consistently changed the HRs away from the null was Refined Grains for all-cause mortality, CVD mortality, and cancer mortality among men only, although this shift was minor and did not change the overall results. The by-component models using HEI-2010 among the members of the AARP cohort reported the unexpected finding of an increase in risk for all-cause mortality among men and women with higher scores for the Refined Grain component (indicating lower consumption) [16
]. Future analysis might consider reconciling these observations across the HEI-2010 and HEI-2015 and the cohorts involved in the DPMP. These results support and reinforce the multidimensionality of the HEI and the representation of diet quality using a wide array of components.
Higher diet quality was associated with improved mortality outcomes in this analysis among a multiethnic population of men and women. All of the HEI-2015 components contributed to the association of diet quality and mortality. Given this, dietary components needing improvement for people with the lowest quality diets could be emphasized in public health messages. The components with the largest differences in median scores between quintile 1 and quintile 5 were identified as Total Fruits, Whole Fruits, Greens and Beans, Whole Grains, and Refined Grains for men and Total Fruits, Whole Grains, and Refined Grains for women (Figure 2
). The results of the present analysis suggest increasing the intake of foods that fall into these components may improve mortality outcomes for people with the lowest quality diets. Data from the 2007–2010 National Health and Nutrition Examination Survey (NHANES) reported that men and women 31 years and older in the US do not meet the requirements for fruits, vegetables, and whole grains and exceed the recommended intake of refined grains [4
]. The message on increasing the intake of fruits, vegetables, and whole grains to improve health outcomes is consistent with results from both the MEC and NHANES [4
The components with the lowest median scores were the same for people in quintile 1 and quintile 5. These components were Dairy, Fatty Acids, and Sodium, with median scores of less than 50%. Improving scores for Dairy, Fatty Acids, and Sodium components may help to improve mortality outcomes for people with the lowest quality diets. The current mortality analysis does not provide evidence on whether increasing Dairy, Fatty Acids, and Sodium component scores for people with the highest quality diets will offer any additional protection. The DGA reports, on average, that adults in the US are not achieving their recommended intakes of dairy and oils, and average intakes of saturated fats and sodium are met or exceeded [4
]. The MEC results and the DGAs support evidence that people with low quality diets should increase their intakes of these foods to improve health outcomes [4
]. The current dietary guidelines also promote replacing SFAs with PUFAs to reduce CVD-related deaths and decreasing sodium intake to reduce CVD events [4
]. Based on this evidence from the DGAs and the MEC, people with the highest quality diets may have improved mortality outcomes if they meet recommended intakes of SFAs, PUFAs, and sodium.
The median component scores for men and women in the MEC in quintile 1 and quintile 5 were at 100% (a perfect score) for the Total Protein Foods component. Similarly, results from NHANES 2007–2010 support that mean intakes of meat, poultry, and eggs for adults 31 years and older are at recommended levels for women and at or above recommended levels for men [4
]. The standard for a perfect score for Total Protein Foods is ≥2.5 oz per 1000 kcal per day. There is no upper limit for Total Protein Foods using this scoring standard; therefore, we do not know if consuming more than 2.5 oz of Total Protein Foods per 1000 kcal per day further improves or worsens mortality outcomes. Also, we do not know if consuming Total Protein Foods in excess replaces intake of foods found in the other 12 components, which would lower these component scores. Having no upper limit for component scoring standards may be a limitation of the HEI.
The median component scores for quintile 1 and quintile 5 for Added Sugars for men and women in the MEC were each above 88%. In comparison, the average intake of added sugar for adults in NHANES were all above the recommended maximum limit [4
]. Previous research on the MEC found that Japanese Americans had the greatest percentage of people who met the DGA recommendations for added sugars [34
]. The MEC has a large proportion of Japanese Americans (26.4%), which may explain why median component scores for Added Sugars indicate a low intake in the MEC compared to excessive average intakes in NHANES. For the Seafood and Plant Proteins component, the median scores among the MEC men and women in both the 1st and 5th quintiles were above 80%. Results reported from NHANES 2007-2010 indicated intakes of nuts, seeds, and soy products were at or above recommended intakes among adults, whereas seafood intakes were below recommendations [4
]. Previous research shows that Native Hawaiians and Japanese Americans in the MEC have more servings of fish per day than other ethnic groups [34
]. These two ethnic groups make up 1/3 of the MEC sample; therefore, their seafood intake may contribute to the median component scores of over 80% for Seafood and Plant Proteins.
The overall mean HEI-2015 score among men and women in the MEC, as estimated using the FFQ, was 65 and 69, respectively. Among men, Native Hawaiian men had the lowest score at 63 compared to African American and white men, with scores of 67. For women, the range was 71 among African American and white women to 67 among Native Hawaiian women. At the time of this paper, no published information about HEI-2015 scores among adults or children had been published. For any comparison, HEI-2015 scores derived from a FFQ would be preferable, as was done by Liese et al. [15
]. Although the mean scores by ethnic group were almost all statistically significantly different, the range of scores was small, i.e., four points for both men and women. An examination of component scores between ethnic groups to further explore variation in dietary exposures is warranted.
Identifying the characteristics of people with the lowest quality diets may help to further tailor nutrition education messages. Comparing the results across the WHI-OS, AARP, and MEC cohorts, college graduates were more likely to be classified as having a higher quality diet based on HEI-2010 [16
]. In the AARP and MEC cohorts, women had slightly higher diet quality scores than men [16
]. For both men and women in the MEC, a higher percentage of Japanese Americans, Native Hawaiians, and Latinos were in quintile 1 and a higher percentage of whites and African Americans were in quintile 5. The WHI-OS cohort is comprised of women and similar to the women in the MEC cohort, there was a higher percentage of whites in quintile 5 compared to quintile 1 and a higher percentage of Hispanic women in quintile 1 compared to quintile 5 [21
]. In the WHI-OS, a greater percentage of Black women were in quintile 1 compared to quintile 5, which is in contrast to the MEC, where a larger proportion of Black women were in quintile 5 [21
]. Thus, the quality of diet within any one group may vary by geographic area.
Of note, the highest versus the lowest HEI-2015 scores were consistently more protective for white and African Americans. The development of the HEI is guided by evidence-based research, and nutrition research is dominated by studies conducted among white and African American participants [36
]. This may account for why the HEI-2015 performs better for white and African Americans. In this analysis, Native Hawaiian men and women had a null association between HEI-2015 quintile 1: quintile 5 and all-cause, CVD, and cancer mortality. Native Hawaiians have the lowest sample size compared to all other ethnic groups in the MEC. Therefore, the power for the analyses of diet quality and mortality outcomes in Native Hawaiians may not be large enough to draw significant findings. Latino men and women also had null associations between HEI-2015 quintile 1: quintile 5 and CVD mortality. Latinos in the US have the lowest rate of CVD compared to other ethnic groups [40
]. These rates are mimicked in this current analysis, with Latino men and women making up 25% and 22% of this MEC sample, respectively, but having the second lowest rate of CVD mortality behind Native Hawaiians. Therefore, the relatively lower rate of CVD mortality in Latino men and women may be contributing to the null association between diet quality and CVD mortality. The associations between the HEI-2015 and mortality outcomes by ethnic group in the MEC are similar to those found in the HEI-2010 mortality analysis with the MEC [19
A limitation of this study was the use of a food frequency questionnaire (FFQ) to collect dietary information at baseline, which could introduce bias [22
]. However, this QFFQ was validated and calibrated in each ethnic-sex group and the correlations between the QFFQ and 24-hour dietary recalls were 0.55–0.74 for energy adjusted nutrients [22
]. Another limitation was dietary data only being assessed once at baseline; therefore, this analysis was not able to capture the influence of dietary changes on mortality outcomes. In addition, all demographic and anthropometric variables were self-reported, and we cannot rule out whether other factors not measured and controlled for, could have affected mortality outcomes; e.g., access to health care. Participants in the MEC were recruited from Hawaii and LA; therefore, results of this study may not be generalizable outside of these areas. Lastly, measurement error is an important consideration relevant to all self-reported behavioral variables. The simple models used to examine predictive validity do not address measurement error; however, efforts are underway to do so for future analyses.
The strengths of this study include the use of a large multiethnic sample that was followed prospectively for over 17 years and the use of a comprehensive QFFQ that was designed to capture ethnic specific foods, allowing for this multiethnic comparison. In addition, covariate data at baseline were collected for almost every participant, permitting the adjustment for multiple, salient risk factors. This study also applied the same standardized regression analysis previously used in diet quality and mortality analysis by the DPMP. Using this standardized approach allows comparisons to be made between the present study and previous and future DPMP analyses of dietary indices and mortality outcomes. For example, once finalized, comparisons can be made between the MEC, the WHI-OS and the AARP mortality analysis with the HEI-2015.