Accordance to the Dietary Approaches to Stop Hypertension diet pattern and cardiovascular disease in a British, population-based cohort

This study used data from participants who, at baseline, were aged between 39 and 79 years and lived in the East of England. The participants were members of the EPIC-Norfolk cohort study, which contains 25,639 people. Of these, 926 people were excluded because they did not have a completed FFQ, 34 people with missing data about baseline cardiovascular disease and diabetes, 1023 people with missing data in one or more covariate required in these analyses, and 1 person with a date of death recorded prior to their first health check, resulting in an analytical sample of 23,655. For analyses where the outcome studied was required to be an incident event we also excluded 863 people with self-reported IHD and 325 people with self-reported stroke at baseline, as necessary for the outcome. Figure 1 presents a flowchart showing how the analytical sample was reached and Table 1 compares the analytical sample and the excluded cases.

The EPIC-Norfolk study is part of a Europe-wide initiative initially developed to investigate the relationship between diet and cancer risk, which has since developed to include other clinical endpoints in addition to cancers [26]. Participants were recruited from the population aged between 39 and 79 years registered with general practices in the city of Norwich, its surrounding towns, and rural areas in Norfolk between the years 1993 and 1999. Those who were contacted and gave consent were invited to a health check, at which measurements and samples were taken for height, weight, body circumferences, blood pressure, urine and respiratory function. Dietary data were collected using a food frequency questionnaire (FFQ), described below, and other data were collected using a questionnaire on cigarette and alcohol consumption, physical activity, socioeconomic status (SES) and history of cardiovascular disease. The total EPIC-Norfolk cohort was found to be similar to the wider English population in 1993 in terms of height, weight, BMI, systolic blood pressure and total cholesterol [26]. Further details about the EPIC-Norfolk cohort have been published by Day et al. [26].

The typical food intake of study participants was assessed using a food frequency questionnaire adapted from an FFQ originally used in the Nurse’s Health Study, which has been adapted and validated for use in the UK [27]. This FFQ contains 130 line items and asks participants about the frequency at which they consumed each item across the previous year, from never or less than once per month to six or more times per day. Full details about this FFQ and how it was developed have been published previously by Welch et al. [28]. Estimated intake of each food or food group consumed was calculated using the FETA program, which combines the frequency of consumption reported in the FFQ with standardised portion data to estimate the intake of macro- and micro-nutrients [29]. FETA has been described previously in detail by Mulligan et al. [29].

The DASH accordance score was developed using the method set out by Fung et al. [19], which has been compared to three other methods for calculating accordance with the DASH diet pattern and found to produce similar results when measuring the association between DASH and colorectal cancer [20]. This method is based on an individual’s consumption of six food groups and two nutrients. Of these, five—fruit, vegetables, nuts and legumes, whole grains, and low-fat dairy—are encouraged and three—red and processed meats, salt, and non-milk extrinsic sugars (NMES)—are discouraged.

The energy-adjusted consumption of each of these groups was calculated using the residual method, previously described by Willett [30]. For the five encouraged food groups the quintile is the individual’s score for that group, I.E., those who consume the most are in quintile five and would have a score of five (the highest score), and for the three discouraged groups this is inverted so that the greatest consumers have a score of one. The quintile scores for all eight groups are summed to produce a measure of DASH accordance that has a potential range of 8–40, with higher scores indicating that the individual’s diet is in greater accordance with the DASH diet pattern. This score is then divided into five quintiles (range in this study: 9–19, 20–23, 24–25, 26–29, 30–38), so that Quintile 5 contained those people with the most DASH-accordant diets. This method for assessing accordance with the DASH diet pattern has been used previously in the EPIC-Norfolk cohort and details of the foods contributing to each group are reported in a study by Monsivais et al. [31].

The EPIC study collects data on incident events and mortality through linkage to hospital admissions and death certificates, which has been described in detail by Lachman et al. [32]. Follow-up for incident events continued until 31st March 2009 and until 31st December 2013 for mortality.

The outcomes of interest were incident CVD and CVD mortality, looking at the following specific outcomes [International Classification of Disease (ICD-10) codes in brackets where appropriate]: (1) incident Ischaemic heart disease (IHD) (I20–25), (2) incident cerebrovascular disease (stroke) (I60–69), (3) total incident CVD (I20–25, I60–69), (4) total CVD mortality (I20–25, I60–69), (5) non-CVD mortality, and (6) all-cause mortality. All-cause and non-CVD mortality were examined because they can reveal whether any association between DASH and health outcomes is specific to CVD or is associated with other causes of death, which may indicate residual confounding.

In order to address potential confounding, we used multivariable models to adjust for total dietary energy (continuous), sex (male, female), age in years (continuous), smoking (current smoker, former smoker, never smoked), body mass index (continuous), physical activity (inactive, moderately inactive, moderately active, active), alcohol consumption (mean units per day), SES (based on occupation, according to the Registrar General’s classificaiton scheme: non-skilled, semi-skilled, skilled—manual, skilled—non-manual, manager, professional), marital status (single, married, widowed, separated, divorced), diabetes (yes, no), presence of antihypertensive medication (yes, no), presence of lipid-lowering medication (yes, no), and previous history of cardiovascular disease (yes, no) (although in models where an incident case of that disease was the outcome these people were excluded). Although presumed to be associated with DASH accordance and CVD outcomes, blood pressure and blood lipids were not included in the main models because they lie on the proposed causal pathway between DASH accordance and CVD, although they were included in a sensitivity analysis. Physical activity was measured using the EPAQ2 questionnaire, which aimed to capture physical activity across multiple domains to give a measure of overall activity [26]. This questionnaire has been validated against objectively measured physical activity and found to be acceptable for this use [33].

Cox proportional hazards models were fitted to calculate the hazard ratio (HR) and a 95% confidence interval for the relationship between quintile of DASH accordance and the outcomes studied, using the quintile of least accordance (quintile 1) was the reference category. For each outcome of interest two models were produced: (1) a model adjusted for age, sex and total dietary energy; (2) as Model 1 but further adjusted for smoking status, physical activity, alcohol intake, diabetes, BMI, SES, presence of antihypertensive medication, presence of lipid-lowering medication, and previous history of cardiovascular disease. We also calculated the absolute rate for each disease to enable comparisons with studies conducted in other populations. Time at risk began from the date of the first health check and ended at date of death or date of diagnosis, as appropriate, or was right-censored by the end of follow-up.

Two sets of sensitivity analyses were also conducted. One of these was as Model 2, above, but also included total serum cholesterol and systolic blood pressure. The other used a continuous rather than categorical measure of DASH accordance.

All analyses were conducted in Stata SE 13.1 [34].

The association observed here between accordance to the DASH diet and incident CVD is in line with expectations, with a meta-analysis showing a significant association for this relationship (RR 0.80 95% CI 0.74, 0.86), in samples from the USA, Italy and Sweden [17]. The estimate of a 12% lower risk found here is smaller than the 20% reduction found in that meta-analysis but the majority of studies used in the meta-analysis only contained women, and several studies were based on occupational cohorts, which is likely to reduce the absolute risk of CVD in the sample when compared to a population-based sample of older men and women like the one used here. The dominant study in the meta-analysis was completed by Fung et al. based on an occupational cohort of women from the USA. The absolute risk of incident IHD observed in the lowest quintile of DASH accordance in that cohort was 165 cases per 100,000 person-years, in contrast to 1262 cases per 100,000 person-years in the present study [19]. The greater overall risk in this cohort may result in an apparent weaker association between CVD and dietary risk factors. Rothman’s Sufficient-Component Cause model [35] stated that a given disease outcome may have multiple causes that are each sufficient to cause disease, with only some being necessary. Since dietary risk factors are not a necessary cause for CVD events, a high level of CVD risk in this sample may reflect a lower relative contribution of diet compared to other risk factors (particularly age-related factors), making the association between this measure of diet and CVD less detectable. Given this, the difference in sample characteristics may explain the difference in the magnitude of the association between that found by Salehi-Abargouei et al. and the present study.

This meta-analysis of Salehi-Abargouei et al. did not report the same distinction between IHD and stroke as shown here, with the pooled results showing a significant reduction in risk for both incident Stroke and incident IHD. That meta-analysis was pooled from three studies, of which only one found a significant association and this was conducted by Fung et al. [19]. The Fung et al. study was much larger (n = 88,517) than either the present study or the two others included in the meta-analysis, which may have been the cause of reduced uncertainty around the estimate and the significant result they report for that outcome. However, the lower risk of IHD associated with DASH observed here was only 10%, less than the 21% observed in the meta-analysis, so both the scale of the reduction and lack of significance is at odds with that observed there.

The finding that increased accordance to the DASH diet is associated with CVD mortality is in accordance with the findings of previous studies, all of which used samples from the USA [21, 36, 37]. Given the concordance between these studies and the present findings, both in direction and magnitude of the association observed, it is probable that the results presented here are true findings. Furthermore, the specificity of the association between DASH accordance and CVD mortality (but not non-CVD mortality) suggests biological plausibility and reduces our concern that these associations were the result of residual confounding by unmeasured social, behavioural or clinical risk factors.

The association observed between the third quintile of DASH accordance and both incident IHD and non-CVD mortality, yet not any other quintile of DASH accordance with these outcomes, is harder to interpret. It is possible that these are chance findings or that there are unobserved factors associated with these outcomes and the third quintile of DASH accordance to the exclusion of the other quintiles. By comparison, the associations between DASH and stroke and CVD-related mortality followed an apparent dose–response pattern, with largest associations observed in the contrast between the least and most DASH-accordant quintiles, these can reasonably be interpreted as being likely to be due to the composition of the DASH diet.

The main limitation of this study stems from the use of self-reported dietary data. This method is subjective and vulnerable to both random and systematic error, with the potential to lead to misclassification bias [38]. This could result from random error, which would bias the results towards the null and attenuate any real association, or it could be driven by social desirability bias, where intakes of healthier foods are overstated while intakes of less-healthy foods are under-reported [39]. However, with no knowledge of how social desirability is distributed across the sample, it is not possible to meaningfully comment on the direction or magnitude that this may have on the results reported here. It is plausible that obese participants were more likely to selectively underreport dietary intakes, as this has been shown in previous studies [40], which could have led to differential misclassification of DASH accordance based on obesity. However, it should be noted that limitations of this nature affect all subjective methods of dietary assessment, so this should not limit the comparability of these results to those of other studies. We excluded approximately 7% of the sample, mainly due to lack of dietary or covariate information. The excluded participants had characteristics associated with elevated CVD risk: older, more likely to smoke, and with higher blood pressure. This exclusion likely biased our results in a way that attenuated the associations we observed, since these individuals would have likely consumed less DASH-accordant diets and would have been at higher risk for CVD. Although the cohort design allows the observation of a temporal relationship between the exposure and outcome, it is still an observational design and is therefore susceptible to residual confounding, although the direction or magnitude of any such effect is not known.

This study has a number of strengths, including its large sample size, prospective design, basis in the general population, independent assessment of outcome events, and a broad range of available factors to include as covariates. The ascertainment of dietary exposures and covariate data at the start of the study eliminates potential recall bias that would affect retrospective study designs. The use of administrative records to identify outcomes minimizes loss to follow-up and ascertainment bias. To our knowledge, this is the first prospective study to examine the association between DASH accordance and CVD in the UK.

The generalisability of these results may be affected by the age of the study sample, which is older than the general UK population. However, CVD events are more likely to occur in people aged over 40 years, meaning that this sample represents an important population in which to assess the effectiveness of a diet intended to reduce the risk of CVD. Adjustment for possible confounding factors means that the results presented here can be compared to those of cohorts with a different distribution of these factors, allowing for its inclusion in future meta-analyses. As stated previously, at the time the EPIC-Norfolk study began, this cohort was similar to the wider UK population in terms of key risk factors, including height, weight, blood pressure and total cholesterol, which lends support to the external validity of the results presented here.

The epidemiological evidence presented here adds to a well-established literature on the importance of diet for prevention of chronic disease. In both the UK and USA, dietary recommendations are part of government strategy for population-level health promotion. However the uptake of dietary guidance and the adoption of healthier eating habits in the population may be hindered by a number of behavioral and social factors. Studies have suggested that healthy eating depends on food literacy and nutrition knowledge [41], cooking [42, 43], and the availability of time to spend in food preparation [44], and the eating habits of family, friends and peers [45, 46]. The cost of food and neighborhood food access may also be important in particular for the DASH diet. Research in the UK and USA indicate that DASH accordant diets tend to cost more than less-healthy diets [31, 47, 48] and studies from the UK and Ireland have found that living in closer proximity to supermarkets may enable dietary habits that are congruent with the DASH pattern [48, 49]. Given the balance of evidence, the promotion of DASH and other cardioprotective diets needs to be matched with both population- and individual-level interventions to support the uptake of healthy eating.