Alignment of Dietary Patterns With the Dietary Guidelines for Americans 2015–2020 and Risk of All-Cause and Cause-Specific Mortality in the Women’s Health Initiative Observational Study

Abstract

Poor diet quality is a leading risk factor for death in the United States. We examined the association between Healthy Eating Index-2015 (HEI-2015) scores and death from all causes, cardiovascular disease (CVD), cancer, Alzheimer disease, and dementia not otherwise specified (NOS) among postmenopausal women in the Women’s Health Initiative Observational Study (1993–2017). This analysis included 59,388 participants who completed a food frequency questionnaire and were free of cancer, CVD, and diabetes at enrollment. Stratified Cox proportional hazards models were fit using person-years from enrollment as the underlying time metric. We estimated multivariable adjusted hazard ratios and 95% confidence intervals for risk of death associated with HEI-2015 quintiles, with higher scores reflecting more optimal diet quality. Over a median of 18.2 years, 9,679 total deaths 3,303 cancer deaths, 2,362 CVD deaths, and 488 deaths from Alzheimer disease and dementia NOS occurred. Compared with those with lower scores, women with higher HEI-2015 scores had an 18% lower risk of all-cause death and 21% lower risk of cancer death. HEI-2015 scores were not associated with death due to CVD, Alzheimer disease, and dementia NOS. Consuming a diet aligned with 2015–2020 US dietary guidelines may have beneficial impacts for preventing overall causes of death and death from cancer.

Abbreviations

BMI

body mass index

CI

confidence interval

CVD

cardiovascular disease

HEI

Healthy Eating Index

HR

hazard ratio

MET-hour

metabolic equivalent task hours

NOS

not otherwise specified

WHI

Women’s Health Initiative

WHI OS

Women’s Health Initiative Observational Study

Poor diet is the leading risk factor for death and the third leading risk factor for disability-adjusted life years in the United States (1). Worldwide, in 2017, 11 million deaths and 255 million disability-adjusted life years were attributable to poor dietary risk factors (2). Diet quality indices, like the Healthy Eating Index (HEI), (3, 4) which measures diet quality in terms of conformance to the Dietary Guidelines for Americans (5), are increasingly applied to prospective cohort data to capture the complexity of total diet (6). Reductions in mortality risk associated with higher scores on diet quality indices, including the HEI-2010, have been investigated in large prospective US cohorts, including the Women’s Health Initiative (WHI) Observational Study (WHI OS), the National Institutes of Health–AARP Diet and Health Study, and the Multiethnic Cohort Study, as part of the Dietary Patterns Methods Project (7–10). Findings based on such indices can readily be translated into relevant public health messages. Since our last report (7), the 2015–2020 US Dietary Guidelines for Americans were released, and 1 of the 4 indices we had explored, the HEI, was updated (4, 11, 12) to align with the new recommendations, including defined limits for added sugars and saturated fats. Furthermore, during this additional follow-up time in the WHI OS, the number of deaths doubled due to the aging of cohort members.

In tandem with a series of updated analyses in the aforementioned cohorts (11, 13), our objective was to examine associations of the HEI-2015 with all-cause and cause-specific (i.e., cancer, cardiovascular disease (CVD), Alzheimer’s disease, and dementia not otherwise specified (NOS)) mortality. We also explored how any associations observed varied by income, given the potential for income to modify both access to a high-quality diet and the underlying risk and causes of death.

METHODS

WHI has been described previously (14–16). Briefly, between 1993 and 1998, at 40 clinical centers throughout the United States, postmenopausal women who were 50–79 years old at study entry were recruited into either a clinical trials component (n = 68,132) or the WHI OS (n = 93,676 women). Participants were invited and consented to continued follow-up in extension studies from 2005 to 2010 and from 2010 to 2020. Follow-up in this analysis was through February 2017. Procedures and protocols were approved by institutional review boards at all participating institutions. A standardized written protocol, centralized training of staff, and quality assurance metrics supervised by the WHI Clinical Coordinating Center were used to ensure uniform data collection.

The sample for the present study was drawn from the 93,676 women participating in the WHI OS. We excluded women with (in this order): incomplete diet data (n = 96), implausible energy intakes of <600 kcals/day or >5,000 kcals/day (n = 3,570), baseline history of CVD or cancer other than nonmelanoma skin cancer (n = 25,794), baseline diabetes or missing diabetes information (n = 2,871), missing survival-time data (n = 286), and missing information on BMI and smoking status (n = 1,671). Our sample for analysis included 59,388 women.

At enrollment, participants self-reported demographic characteristics, health behaviors, and medical histories, using self-administered standardized questionnaires. In WHI’s Measurement Precision Study, questionnaire items on demographics and medical conditions were shown to be reliable (weighted κ > 0.8) (15). Use of postmenopausal hormone therapies (specifically, unopposed estrogen and/or estrogen plus progestin) via pills or patches was self-reported and we classified women as never, past, or current users. We calculated neighborhood socioeconomic status score, an index based on census-tract socioeconomic status and neighborhood deprivation variables, with higher scores indicating less deprivation (17).

Self-reported recreational physical activity was measured using the WHI brief physical activity inventory, which has been shown to be reliable (range of weighted κ, 0.67–0.71) (15) and valid when compared to accelerometer data (r = 0.73) (18) For each participant, we calculated metabolic equivalent task hours (MET-hour) per week of recreational physical activity and categorized physical activity level (0 MET-hour/week; 0.1–3 MET-hour/week; 3.1–8.9 MET-hour/week; ≥9 MET-hour/week; or missing data), as described in detail previously in WHI (19).

At the clinic visit, trained staff measured each participant’s weight and height using a standardized protocol. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared; we categorized BMI as follows: <18.5; 18.5–24.9; 25–29.9; 30–34.9; 35–39.9; ≥40.

Diet was assessed at enrollment using a self-administered food frequency questionnaire developed and validated specifically for WHI (20). The WHI food frequency questionnaire was designed to capture foods relevant for multiethnic and geographically diverse population groups. Previous research has demonstrated its reliability (for all nutrients, r = 0.76) when compared to the mean of 8 days of dietary intake from a 4-day food record and four 24-hour dietary recalls (20).

We measured diet quality with the HEI-2015 (12), which was created by the US Department of Agriculture and the National Cancer Institute, and aligns with the 2015–2020 US Dietary Guidelines for Americans. In brief, the HEI-2015 contains 13 components that sum to a total maximum score of 100 points, including 9 adequacy components and 4 moderation components (3). Of note, the previous HEI-2010 empty-calories component was replaced with 2 discrete categories in the HEI-2015—“Added Sugars” and “Saturated Fats”—to better align with the newly quantified added sugars recommendation. Recent evaluation of the HEI-2015 demonstrated evidence supportive of its construct validity, reliability, and criterion validity (11). We used WHI’s dietary data from the baseline food frequency questionnaire and the MyPyramid Equivalents database to construct variables related to the HEI, as described previously in WHI (21).

Vital status of participants was collected through annual clinical center follow-up of participants and proxies. In addition, periodic searches of the National Death Index were conducted. Cause of death was determined by medical record and death certificate review at the WHI Clinical Coordinating Center with oversight from the WHI physician-adjudicators and outcomes committee.

Statistical analysis

We generated descriptive statistics for demographics and lifestyle characteristics overall and by quintiles of the HEI-2015. Hazard ratios, 95% confidence intervals, and P values were computed from Cox regression models stratified by 5-year age at baseline (50–54, 55–59, 60–64, 65–69, 70–74, and ≥75 years), BMI, race/ethnicity (American Indian or Alaskan Native; Asian or Pacific Islander; Black or African American; Hispanic/Latina; White; other), and smoking (never, past, or current). Outcomes included all deaths and deaths due to cancer, CVD, Alzheimer disease, or dementia NOS. We used multivariable adjusted models, including linear covariates for age, energy, and alcohol, and categorical covariates of education, income, marital status, recreational physical activity, and postmenopausal hormone therapy. Neighborhood socioeconomic status was examined but not informative for models and was not retained.

For any observed associations for cancer- and CVD-associated deaths, we explored effect modification by income by including a cross-product term of HEI and income in our stratified Cox model. We performed a sensitivity analyses to assess 1) reverse causality by removing deaths occurring in the first 2 years of follow-up, 2) residual confounding by restricting the sample to women without known hypertension, and 3) residual confounding by smoking by restricting the sample to never smokers. We also calculated Fine and Gray subdistribution hazards ratios to account for competing risks (22). Analyses were conducted using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina).

RESULTS

Over a median of 18.2 years of follow-up, 9,679 total deaths occurred, including 3,303 cancer deaths, 2,362 CVD deaths, and 488 deaths attributed to Alzheimer disease or dementia NOS. As shown in Tables 1 and 2, compared with women in the lowest quintile (quintile 1), women in quintile 5 had lower intake of added sugars (7.9% vs. 11.5% of calories consumed) and saturated fats (7.2% vs. 13.2% of calories consumed) as expected, but similar mean intakes of sodium (1.66 vs. 1.65 g/1,000 kcals), with none close to the ideal score. Women in quintile 5 were also more likely to have never smoked (53% vs. 48%), be White (89% vs. 80%), be more physically active (16.9 vs. 7.9 MET-hour/week), and have lower BMIs (25 vs. 29) than those in quintile 1.

Table 1.

Descriptive Characteristics, by Quintiles of Index Scores for the Healthy Eating Index-2015, of Participants (n = 59,388) in the Women’s Health Initiative Observational Study, 1993–February 2017^a

Characteristic	Q1 (n = 11,871)	Q2 (n = 11,880)	Q3 (n = 11,888)	Q4 (n = 11,882)	Q5 (n = 11,867)
Range of Healthy Eating Index-2015 pointsb	20.5–58.5	58.5–65.4	65.4–70.7	70.7–76.1	76.1–96.3
Age, years	62 (0.07)	63 (0.07)	63 (0.07)	63 (0.07)	64 (0.07)
Body mass indexc	29 (0.06)	27 (0.05)	27 (0.05)	26 (0.05)	25 (0.04)
MET-h/wk physical activity	7.9 (0.1)	10.9 (0.1)	13.0 (0.1)	14.9 (0.1)	16.9 (0.1)
Neighborhood socioeconomic status	74.5 (0.09)	76.1 (0.08)	76.8 (0.08)	77.2 (0.08)	77.5 (0.07)
Energy intake, kcal	1,778.8 (6.7)	1,614.9 (5.6)	1,529.7 (5.0)	1,475.1 (4.6)	1,433.3 (4.2)
Alcohol intake, g	4.4 (0.09)	5.7 (0.11)	6.4 (0.11)	6.5 (0.11)	6.5 (0.11)
Total fruits, cup equivalents/1,000 kcald	0.56 (0.004)	0.93 (0.005)	1.18 (0.006)	1.40 (0.006)	1.62 (0.007)
Whole fruits, cup equivalents/1,000 kcale	0.35 (0.003)	0.64 (0.005)	0.85 (0.006)	1.04 (0.006)	1.24 (0.006)
Total vegetables, cup equivalents/1,000 kcalf	0.77 (0.003)	0.99 (0.004)	1.13 (0.005)	1.26 (0.005)	1.37 (0.005)
Greens and beans, cup equivalents/1,000 kcalf	0.07 (0.0006)	0.10 (0.0009)	0.12 (0.001)	0.15 (0.001)	0.18 (0.001)
Whole grains, cup equivalents/1,000 kcal	0.44 (0.004)	0.64 (0.004)	0.78 (0.005)	0.93 (0.005)	1.19 (0.006)
Dairy, cup equivalents/1,000 kcalg	0.86 (0.005)	1.00 (0.006)	1.07 (0.006)	1.16 (0.006)	1.29 (0.006)
Total protein foods, cup equivalents/1,000 kcalf	2.72 (0.01)	2.76 (0.01)	2.76 (0.009)	2.76 (0.009)	2.70 (0.008)
Seafood and plant protein, cup equivalents/1,000 kcalh	0.59 (0.004)	0.77 (0.005)	0.88 (0.005)	1.00 (0.006)	1.18 (0.006)
Fatty acidsi	1.63 (0.003)	1.73 (0.004)	1.81 (0.004)	1.91 (0.004)	2.11 (0.004)
Refined grains, cup equivalents/1,000 kcal	2.76 (0.009)	2.49 (0.008)	2.30 (0.007)	2.10 (0.006)	1.81 (0.005)
Sodium, g/1,000 kcal	1.65 (0.003)	1.68 (0.003)	1.70 (0.003)	1.70 (0.003)	1.66 (0.003)
Added sugarsj	11.48 (0.06)	9.89 (0.04)	9.20 (0.04)	8.70 (0.03)	7.93 (0.03)
Saturated fatsj	13.2 (0.03)	11.1 (0.03)	9.7 (0.02)	8.5 (0.02)	7.2 (0.02)

Abbreviations: MET-h, metabolic equivalent of task hours; Q, quintile

^a Data are reported as mean (standard error) unless otherwise indicated.

^b Values are expressed as a range.

^c Weight (kg)/height (m)².

^d Includes 100% fruit juice.

^e Includes all forms except juice.

^f Includes legumes (beans and peas).

^g Includes all milk products, including fluid milk, yogurt, and cheese, and fortified soy beverages.

^h Includes seafood, nuts, seeds, soy products (other than beverages), and legumes (beans and peas).

ⁱ Ratio of poly- and monounsaturated fatty acids to saturated fatty acids.

^j Reported as a percentage of calories consumed.

Table 2.

Descriptive Characteristics, by Quintiles of Index Scores for the Healthy Eating Index-2015, of Participants (n = 59,388) in the Women’s Health Initiative Observational Study, 1993–February 2017

Descriptive Characteristic	Q1 (n = 11,871), %	Q2 (n = 11,880), %	Q3 (n = 11,888), %	Q4 (n = 11,882), %	Q5 (n = 11,867), %
Race/ethnicity
Non-Hispanic White	80	83	86	87	89
Black	10	7	6	5	5
Hispanic	5	5	4	3	2
Other	5	5	4	5	4
Graduated college or higher level of education	31	41	46	50	54
Married or living as married	60	64	66	66	66
Never smokers	48	52	51	53	53
Postmenopausal hormone therapy
Never	44	39	37	36	35
Former	13	13	12	13	14
Current	42	48	51	51	51
Family income level
<$10,000	5	3	3	2	2
$10,000–19,999	13	10	9	8	8
$20,000–34,999	23	22	20	19	19
$35,000–49,999	19	19	19	19	19
$50,000–74,999	17	19	21	21	21
$75,000–99,999	7	9	10	11	11
$100,000–$149,999	5	6	8	8	9
≥$150,000	3	4	4	5	5
Unknown	7	7	7	7	7
Known hypertension	31	29	29	26	26

Abbreviation: Q, quintile.

As shown in Table 3, compared with women in quintile 1 of HEI-2015, women in quintile 5 had an 18% lower risk of all-cause mortality (hazard ratio (HR) = 0.82, 95% confidence interval (CI): 0.76, 0.87) and a 21% lower risk of death due to cancer (HR = 0.79, 95% CI: 0.70, 0.88), but not a lower risk of death due to CVD (HR = 0.94, 95% CI: 0.82, 1.08). No association with death from Alzheimer disease or dementia NOS was observed (HR = 1.09, 95% CI: 0.81, 1.48).

Table 3.

Multivariable Adjusted Hazard Ratios^a and 95% Confidence Intervals for Deaths^b According to Quintiles of the Healthy Eating Index-2015 Among 59,388 Women in the Women’s Health Initiative Observational Study, 1993–February 2017

HEI Quintile	No.	All-Cause Deaths			Cancer Deaths			CVD Deaths			Alzheimer or Dementia NOS Deaths
		No. of Any Deaths	Multivariable-Adjusted HR	95% CI	No. of Cancer Deaths	Multivariable-Adjusted HR	95% CI	No. of CVD Deaths	Multivariable-Adjusted HR	95% CI	No. of Alzheimer or Dementia Deaths	Multivariable-Adjusted HR	95% CI
1	11,871	2,022	1.00	Referent	719	1.00	Referent	477	1.00	Referent	78	Referent
2	11,880	1,931	0.94	0.88, 1.00	666	0.92	0.82, 1.02	479	1.02	0.90, 1.17	82	0.92	0.67, 1.26
3	11,888	1,877	0.88	0.83, 0.94	642	0.86	0.77, 0.96	474	0.99	0.87, 1.14	90	0.88	0.64, 1.21
4	11,882	1,863	0.84	0.78, 0.90	651	0.86	0.77, 0.97	418	0.85	0.73, 0.97	93	0.83	0.60, 1.14
5	11,867	1,986	0.82	0.76, 0.87	625	0.79	0.70, 0.88	514	0.94	0.82, 1.08	145	1.09	0.81, 1.48

Abbreviations: CI, confidence interval; CVD, cardiovascular disease; HEI, Healthy Eating Index; HR, hazard ratio; NOS, not otherwise specified.

^a Cox proportional hazard models were stratified by age in 5-year categories, body mass index, race/ethnicity, and smoking. Models were also adjusted for age (linear), energy, alcohol, education, income, marital status, physical activity, and postmenopausal hormone replacement therapy.

^b As of February 2017, death counts were as follows: 9,679 total deaths; 3,303 cancer deaths; 2,362 CVD deaths; 488 deaths from Alzheimer disease or dementia NOS; person-years for Q1–Q5, respectively, were as follows 172,707; 181,242; 185,070; 187,815; and 189,296.

Excluding women with <2 years of follow-up did not considerably change the quintile 5-to-quintile 1 hazard ratios observed for all-cause mortality (HR = 0.83, 95% CI: 0.77, 0.88), death due to cancer (HR = 0.80, 95% CI: 0.71, 0.90), or death due to CVD (HR = 0.94, 95% CI: 0.82, 1.09). Restricting the sample to never smokers also did not considerably change the quintile 5-to-quintile 1 hazard ratios for all-cause mortality (HR = 0.84, 95% CI: 0.76, 0.92), death due to cancer (HR = 0.79, 95% CI: 0.66, 0.96), or death due to CVD (HR = 1.00, 95% CI: 0.82, 1.22). We did not find variation in association by income status. When we limited the sample to women without known hypertension, the 95% confidence interval for the association of diet quality and CVD-associated deaths still crossed the null (HR = 0.86, 95% CI: 0.71, 1.04) (data not shown.).

Competing risk analyses yielded a similar null quintile 5-to-quintile 1 association for death due to CVD (subdistribution HR = 0.99, 95% CI: 0.86, 1.14) and similar inverse association for death due to cancer (subdistribution HR = 0.81, 95% CI: 0.72, 0.91) (data not shown).

DISCUSSION

This investigation of the HEI-2015 in the WHI OS extends and is consistent with previous findings in this cohort linking the HEI-2010 with reduced risk of death overall and after cancer (7). Our findings for these outcomes are consistent with findings from similar investigations in the National Institutes of Health–AARP Diet and Health Study and the Multiethnic Cohort (11, 13).

In the WHI OS, the HEI-2010 was associated with a reduced risk of death due to CVD over a median of 12.9 years of follow-up (7). In this analysis featuring extended follow-up (median, 18.2 years), we, unlike the National Institutes of Health–AARP Diet and Health Study and the Multiethnic Cohort (11, 13), did not find evidence of an association between the HEI-2015 and CVD deaths among this population of older women. Although we did not observe an association of diet quality and death due to Alzheimer disease or dementia NOS, given the biological plausibility of diet quality and its impact on cognitive function and brain structure (23–27), research exploring associations with a variety of dietary patterns could be useful.

We did not find effect modification by income status in this cohort. However, current theories about the association between income and diet quality may not apply to the WHI cohort (28). In the WHI OS, the birth year median (quintile 1, quintile 3) was 1932 (1926, 1938). It is possible that this cohort of healthy women born to families who survived the Great Depression are resourceful and could have the time (approximately half were retired at enrollment) and experience to prepare inexpensive, nutritious meals (28).

The WHI cohort has HEI scores similar to those of other US cohorts (9) and higher scores than the average American (29). Advantages of the present study include use of the multidimensional HEI-2015, which captures the potentially synergistic nature of multiple important dietary components (30). Other strengths include objectively measured height and weight, the large sample size, the prospective nature of evaluations, the long-term mortality follow-up, the central adjudication of deaths, and diversity by race/ethnicity, education, and income.

Study limitations include measurement error inherent to the food frequency questionnaire (31) and to other self-reported measures of health behaviors, like physical activity. Measurement error for energy intake is well recognized (32), but less is known about the extent and severity of measurement error for indices like the HEI-2015 presented in this report. This analysis also featured only 1 measure of diet, at baseline, which therefore requires the assumption that diet is relatively constant over time or exerts long-term effects. In addition, although we had detailed data allowing for careful control for the major confounders and to show that associations were unlikely to be artifacts of reverse causation, given the observational nature of this study, it remains possible that the associations observed could be explained by unmeasured confounders. In Fine and Gray’s subdistribution model, participants stay in the risk set if they have failed due to a competing event, which may not be fitting for mortality outcomes (33); however, this method is valid for producing cumulative incidence for outcomes affected by competing risks.

Our results from this large prospective analysis of ~ 60,000 postmenopausal women suggest that the consumption of a diet aligning with the 2015–2020 Dietary Guidelines for Americans may have beneficial impacts for preventing overall causes of death, particularly, death due to cancer.