Introduction

Dietary intake plays an important role in cardiovascular health [1•, 2, 3]. Observational studies have shown better cardiovascular health among adults consuming nutrients, such as n-3 fatty acids [4], and foods such as fruit and vegetables [5, 6], fish [7], and whole grains [5, 8]. Other foods, including red and processed meat, have been related to greater risk of cardiovascular disease (CVD) [9]. More recent studies have focused on the whole diet or dietary patterns, since we eat combinations of foods and beverages and not just one food or nutrient [10]. Numerous dietary patterns and scores have been created to generally characterize a population’s overall dietary intake and are reviewed in detail elsewhere [11, 12••]. Briefly, there are three general approaches to characterizing dietary patterns (Table 1). The “a priori” approach to characterize a pattern of dietary intake or diet quality is based on pre-defined criteria. For example, the Healthy Eating Index (HEI) created in 1995 [13] and recently updated in 2010 [14] reflects the 2010 U.S. Dietary Guidelines for Americans (DGA) [12••]. Another a priori diet quality score or diet pattern is the Dietary Approach to Stop Hypertension or DASH diet pattern, which was developed to reduce blood pressure [15]. A posteriori diet pattern is a data-driven approach where diet patterns are created using factor analysis, principal components analysis (PCA), or cluster analysis [11, 16]. Commonly labeled diet patterns are the “Western” and “Prudent” patterns. The Western pattern is characterized by higher intake of red and processed meat, fried foods, sweets, solid fats, and refined grains, while the Prudent pattern, in contrast, is usually characterized by higher intake of a variety of vegetables and fruit and whole grains. There are many dietary patterns, theoretically limitless combinations of foods and beverages habitually consumed, but common diet patterns are often based on food preferences such as a vegetarian or Mediterranean diet patterns [17, 18], religious or philosophical beliefs, and commercial diet plans including the popular diet plans such as Atkins, Pritikin, and South Beach [19].

Table 1 Types of diet patterns and methods of creating them
Full size table

Identifying healthy diets to prevent the development of CHD or unhealthy diets that increase CHD risk is key for developing and tailoring public health promotion strategies for specific populations. Recently, the Diet Pattern Methods Project (DPMP), a consortium to strengthen research evidence on dietary indices, reported consistency among five a priori diet quality scores/patterns as well as consistency among the associations for each of these diet scores/patterns with mortality from all-causes, CVD, and cancer [20•]. However, evidence from prospective studies is inconsistent for a posteriori derived diet patterns relative to CHD [21–32]. Although these study results were recently meta-analyzed [33], questions remain unanswered. The major limitation of a meta-analysis is the aggregation of studies with diverse study populations and methods; combining studies with stark methodologic differences is akin to comparing apples and oranges [34].

The purpose of this review was to explain the inconsistent findings among the studies. We compared and contrasted the methods from 12 published prospective studies that examined the relations of a posteriori derived dietary patterns with incident CHD.

Comparison of Study Methods

Studies included in this comprehensive review were restricted to studies included in a recent meta-analysis [33] plus 2 others published since 2000 [22, 24]. All studies had a prospective cohort design, dietary assessment at baseline, derivation of dietary patterns using factor analysis, principal components analysis, or cluster analysis, and follow-up for incident CHD [21–32]. Studies are described in Table 2.

Table 2 Characteristics of the 13 prospective studies of dietary patterns relative to coronary heart disease (CHD) according to geographic location
Full size table

Study Population: Differences in Underlying Populations

Of the 12 studies, 7 enrolled only white participants [21, 22, 25–29]; 1 study included 4 race/ethnic groups—White, Black, Hispanic, and Chinese [23]; 1 study with Black and White adults [24], and 3 studies enrolled Japanese and Chinese adults [30–32]. Participants in most studies were aged 40–75 years at baseline, although younger participants were enrolled in 3 studies [25, 28, 29]. Exclusion criteria varied somewhat across studies: (1) study participants with prevalent CHD were excluded from most studies except 2 [25, 31], and (2) participants with diabetes were also excluded in 3 studies [22, 23, 27].

Diet Assessment: Differences in the Capture of Dietary Intake

Dietary intake was assessed using a food frequency questionnaire (FFQ) in 11 studies, while a diet history questionnaire was used in 1 study [28]. The number of foods listed in each FFQ ranged from 26 to 127 food items. Among study participants responding to the diet history questionnaire, 662 foods were reported. Further, there was considerable variation in the food lists on the FFQs between studies, since each study’s FFQ was developed for its respective population. For example, the list of vegetables on the Asian FFQs is similar within this culture, but quite different than those listed on the US and European FFQs, including seaweeds, yard-long beans, wild rice stems, Chinese cabbage and greens, wax gourd, Hyacinth beans and snow peas, white turnips, lotus roots, bamboo shoots, edible wild plants, green tea, oolong tea, and others [30–32].

Dietary Pattern Construction: Differences in Pattern Constituents and Analytic Technique

Many of the procedures and decisions made to create the diet patterns were not standardized among the 12 studies. For example, the number of foods/food groups that made up each of the diet patterns varied considerably; in 5 studies, individual food items were categorized into a smaller number of food groups (range 31–47 food groups) [21–23, 28, 29]; while in the other 7 studies, individual food items were included in the factor or cluster analysis model (range 40–127 individual food items) [24–27, 30–32]. Gender-specific diet patterns were created in 4 studies [21, 22, 26, 31], while the diet patterns in the other 7 studies were based on both genders [23–25, 27–30]. The a posteriori Prudent diet pattern (or similar ‘healthy’ pattern) was created in all 12 studies, while the Western diet pattern (or similar ‘unhealthy’ pattern) was created in 11 studies, but not in the study by Akesson et al. [26]. Further, in 4 of the studies, up to 3 additional diet patterns were created [23, 24, 27, 31]. Nettleton et al. [23], Shikany et al. [24], and Cai et al. [31] retained 4, 5, and 3 factors, respectively, in their PCA models based on subjective evaluation of eigenvalues, congruence of the solution among sex, race, and geographic region, or interpretation of the factor solutions. Brunner et al. reported 4 patterns identified in cluster analysis [27].

Definition of CHD—Differences in the Outcome

In US and European studies, incident CHD was defined as (1) confirmed and probable incident non-fatal myocardial infarction (MI) and (2) fatal CHD. Incident non-fatal MI was generally defined the same in US [21–24] and European studies [25–29], but the Asian studies [30–32] limited the CHD outcome definition to fatal MI. CHD mortality was assessed similarly in all studies.

Ascertainment of CHD

In US studies, medical records were reviewed for incident MI [21–24]. Fatal CHD was confirmed by medical records, autopsy reports, underlying cause of death on the death certificate, or confirmation by other sources. In European studies, CHD diagnosis was obtained from national hospital registries or population-based MI registries [25, 26, 28, 29] or through study exam, doctor diagnosis, and medical records [27]. Death information was obtained from cause of death registries [25–29]. CHD mortality in Asian countries was identified by review of death certificates obtained from regional Public Health Centers or death registries [30–32].

Comparison of Study Results

Study participants in the 12 studies were aged 20 to 84 years at baseline and were followed for incident CHD or CHD death over 4.6 to 15 years (Table 2). For the Western diet pattern, there was a 37 to 64 % higher risk of incident CHD [21, 22, 24] in 3 US studies; however, results were not reported for the “Fat, processed meat” (or Western) diet pattern relative to CHD in the Nettleton et al. study [23]. In contrast, null results were observed in all 4 European and 3 Asian studies. For the Prudent diet pattern, the risk of incident CHD was lower by 13 to 65 % in 3 US and 4 European studies, while no relation was found in 1 US study [24], 1 European study [25], nor in the 3 Asian studies [30–32].

A recent meta-analysis by Rodriguez-Monforte et al. reported findings for factor analysis and cluster analysis derived diet patterns relative to incident CVD, CHD, and stroke [33]. In this meta-analysis, the Prudent diet pattern was related to a 17 % lower risk of CHD, but there was no relation with the Western diet pattern [33]. Further, stratification by geographic location (USA + Europe; Asia) showed similar findings.

The published factor loadings or correlations of individual foods/food groups for each of the Western and Prudent diet patterns in 10 studies are shown in Tables 3 and 4, respectively. Factor loadings were not reported in the Akesson et al. or Brunner et al. papers [26, 27]. Loading scores for absolute values <0.30 were not reported in the Hu et al. study, and for the other studies, loading scores <0.15 were not reported. Positive factor loadings for a specific item indicate higher intake of that item characterizes the corresponding diet pattern and negative values indicate relatively lower intake of the item characterizes the pattern. For the Western diet pattern (Table 3), several studies, but not all, had positive loading scores for red and processed meat, French fries or fried food, eggs, high-fat dairy products, refined grain, and added fats (butter, margarine, oils, and olive oil) and negative loading scores for whole grains. For the Prudent diet pattern in all studies, positive loading scores were observed for fish, fruit, and fruit juice (Table 4). Further, factor loading scores for many of the US and European studies were positive for poultry, low-fat dairy, nuts/seeds, and whole grains. For US and Asian studies, loading scores were positive for a number of vegetables and legumes (miso, soybean, tofu). For studies conducted in Europe and Asia, positive loading scores were observed for sweets (candy, desserts).

Table 3 Food group factor loading scores for the Western (or similar) diet pattern of study participants from 10 cohorts
Full size table
Table 4 Food group factor loading scores for the Prudent (or similar) diet pattern in study participants from 10 cohorts
Full size table

The average number of servings consumed per day for major food groups for the Western and Prudent diet patterns are shown by lowest and highest quantile of intake in Tables 5 and 6, respectively, for several studies. This information was not published in 6 of the 12 studies [23, 25, 27–29, 31]. For the Western pattern, Americans [21, 22, 24] generally consumed more servings of total meat (red, processed, and poultry), dairy products, fruit and vegetables, and whole grains than Asians [30, 32]; however, Asians reported consuming more servings of fish and rice (refined grain) than Americans (Table 5). Servings of food intake were not reported for the Western diet pattern in European studies. For the Prudent pattern (Table 6), Americans and Europeans consumed more servings of total meat, dairy products, fruit, and vegetables than Asians, but Asians reported consuming more fish and rice. Alcohol consumption was not consistently reported across studies.

Table 5 Number of average daily servings of major food groups consumed in the lowest and highest quantile for the Western (or similar) diet pattern in 5 prospective studies
Full size table
Table 6 Number of average daily servings of major food groups consumed in the lowest and highest quantile for the Prudent (or similar) diet pattern in 6 prospective studies
Full size table

Discussion and Conclusions

Meta-analysis results showed a 17 % lower risk of incident CHD for the a posteriori Prudent diet pattern [33]; however, lower CHD risk ranged from 13 to 65 % as reported in 7 of 12 individual studies [21–23, 26–29]. The Prudent diet pattern was characterized by higher intake of fish, poultry, low-fat dairy, fruit, fruit juice, nuts/seeds, and/or whole grains. The Western diet pattern was generally characterized by higher intakes of red and processed meat, French fries or fried food, eggs, high-fat dairy products, refined grain, and added fats, and while the Western pattern was not associated with CHD in meta-analysis [33], the risk of CHD was significantly higher in 3 of 4 US studies [21, 22, 24].

We propose several explanations for the inconsistency of findings for the diet pattern-CHD associations. First, a posteriori or data-driven derived diet patterns are based on subjective methods and require decisions that have consequences for the analysis and interpretation of the study results. Importantly, the methods to construct diet patterns in the 12 studies were not standardized. These methodological differences likely contribute to the inconsistency of results in this body of evidence. In constructing the diet pattern, the choice and treatment of food and beverage variables included in the model may influence study results. For example, the decision to include individual food items or food groups varied among the 12 studies; half of the studies collapsed individual food items into a smaller number of food groups (ranging from 30 to 50 food groups) [21–23, 28, 29], while the others input all individual food items in the factor analysis model. Subjective decision-making has also been involved with food grouping. For example, should red meat and processed meat be grouped together or kept separate? Or should all vegetables be grouped into 1 group or grouped into botanically similar groups or grouped according to another scheme. These decisions influence the factor loadings and study results. Another aspect to consider in diet pattern construction is whether to use absolute (i.e., servings or grams of food groups per day) or density amounts (i.e., mean weight of food adjusted for total energy intake). However, Smith et al. determined no differences between diet patterns derived using absolute and density amounts in a children’s diet study [35]. Finally, there is evidence suggesting that dietary intake differs between men and women; not only amount of food and beverage intake but choices of food and beverages vary [36]. Half of the studies created gender-specific diet pattern scores, while the others created diet patterns inputting data from both genders into the factor analysis/PCA model.

Second, dietary assessment varied between studies, including the instrument to assess dietary intake (FFQ vs. diet history), differences in the FFQ food list (food and beverage items queried), the number of food and beverages listed on the FFQ varied between studies (ranging from 26 to 131 items), interviewer- vs. self-administration of the FFQ, and food and beverage intake not listed on the FFQ (alcohol or other beverages, high-calorie snack foods, or other foods to capture total diet). Osler et al. used a short item FFQ (26 items) which did not capture total dietary intake, which may have contributed to the null relation between both food patterns and CHD [25].

Third, ascertainment of CHD cases differed by country. It is possible that the number of cases captured in US studies was underestimated since national hospital charge registries were not available as in Europe. In addition, non-fatal CHD was not captured in the studies conducted in Japan and China [30–32]. Diverse capture of CHD cases, including different CHD definition in Asian studies, may potentially attenuate the diet-CHD point estimate, thus providing another potential explanation for differing study results.

Furthermore, geographic and/or cultural differences influence eating habits, including availability of food and type of food consumed, food preparation and cooking method (such as roasting, steaming, food preservation using salt), and amount of food consumed. Amount of food intake varied by geographic location; Asians consumed more fish and less meat for both lower (Q1) and higher (Q4 or Q5) Prudent and Western pattern scores compared to American quantile values. In the individual Asian cohorts, no relation was observed between the Prudent diet pattern and CHD, which was somewhat surprising given their higher fish and legume intake, and lower meat intake; although the study samples were relatively young and only CHD mortality was ascertained over 5–12 years of follow-up. In meta-analysis, results showed a statistically significant 18 % lower risk of CHD mortality associated with the Prudent pattern among Asian cohorts, a result likely driven by the large pooled sample size. Although sodium intake was not queried, sodium intake is high among Chinese and Japanese populations and may have contributed to CHD mortality [37, 38]. The risk of CHD was higher among adults consuming a Western diet pattern in US cohorts, but not in Europe or Asia; however, the meta-analysis results did not stratify between U.S. and Europe [33]. Since the European studies did not report average number of servings from each food group for the Western diet pattern, we were unable to compare food intake between studies from the two geographic areas. For comparison between studies, it is important to be transparent and report methodology related to constructing the diet patterns as well as the results of pattern analysis, including the factor loading scores and average food intake for all quantiles of diet pattern score. Further, more research is needed to understand the differences in diet pattern-disease associations by geographic location and culture as well as to conduct collaborative studies for development of standardized methods to reduce heterogeneity to better compare diet patterns among the numerous cohort studies.

In summary, the recommendations of the 2015–20 Dietary Guidelines for Americans [12••] and the AHA/ACC guidelines on lifestyle management [1•] that a diet pattern characterized by higher intake of fish, vegetables, legumes, fruit, whole grain, nuts, and lower intake of red and processed meat is related to lower risk of CHD. Other studies of the health benefits of this type of healthy diet pattern have been demonstrated in the PREDIMED Study [2]; the Lyon Heart Study [3], the 7th Day Adventist study [17]; the DASH trial, [15], and others. The Western dietary pattern has been inconsistently and positively associated with CHD risk and the inconsistency in results across studies is likely explained by the shifting constituents of the pattern from study to study in addition to the other methodological variations discussed. Although studies of diet patterns do not identify mechanisms contributing to lower risk, we may speculate that higher intake of omega3 fatty acids [4, 7], bioactive fruit and vegetables [6], whole grains [5, 8], nuts [39], and olive oil [2, 3] may work synergistically [10] to promote cardiovascular health.