点击显示 收起
1 From the Departments of Nutrition and Epidemiology, Harvard School of Public Health, Boston, and the Channing Laboratory, the Department of Medicine, Brigham and Women's Hospital, Boston.
2 Supported by research grants CA55075, HL35464, and HL60712 from the National Institutes of Health. FBH is a recipient of the Charles A King Trust Research Fellowship from The Medical Foundation, Boston. 3 Address reprint requests to FB Hu, Department of Nutrition, Harvard School of Public Health, 665 Huntington Ave, Boston, MA 02115. E-mail: frank.hu{at}channing.harvard.edu.
ABSTRACT
Background: Previous studies on diet and coronary heart disease (CHD) focused primarily on individual nutrients or foods.
Objective: We examined whether overall dietary patterns derived from a food-frequency questionnaire (FFQ) predict risk of CHD in men.
Design: This was a prospective cohort study of 44875 men aged 40–75 y without diagnosed cardiovascular disease or cancer at baseline in 1986.
Results: During 8 y of follow-up, we documented 1089 cases of CHD (nonfatal myocardial infarction and fatal CHD). Using factor analysis, we identified 2 major dietary patterns using dietary data collected through a 131-item FFQ. The first factor, which we labeled the "prudent pattern," was characterized by higher intake of vegetables, fruit, legumes, whole grains, fish, and poultry, whereas the second factor, the "Western pattern," was characterized by higher intake of red meat, processed meat, refined grains, sweets and dessert, French fries, and high-fat dairy products. After adjustment for age and CHD risk factors, the relative risks from the lowest to highest quintiles of the prudent pattern score were 1.0, 0.87, 0.79, 0.75, and 0.70 (95% CI: 0.56, 0.86; P for trend = 0.0009). In contrast, the relative risks across increasing quintiles of the Western pattern score were 1.0, 1.21, 1.36, 1.40, and 1.64 (95% CI: 1.24, 2.17; P for trend < 0.0001). These associations persisted in subgroup analyses according to cigarette smoking, body mass index, and parental history of myocardial infarction.
Conclusions: These data suggest that major dietary patterns derived from the FFQ predict risk of CHD, independent of other lifestyle variables.
Key Words: Health Professionals Follow-up Study • food-frequency questionnaire • coronary heart disease • dietary patterns • men
INTRODUCTION
Distinct eating patterns reflect different dietary traditions worldwide, and they may be related to rates of coronary heart disease (CHD) in different countries (1). Mediterranean and Asian populations have very low rates of CHD compared with Western populations. These low rates are attributed to high intakes of vegetables, fruit, whole-grain products, and fish and low intakes of red meat, high-fat dairy products, and other animal products in traditional Mediterranean and Asian diets. However, the differences in CHD rates among countries may also be due to differences in other CHD risk factors, including physical activity and obesity.
Many prospective cohort studies have examined the associations between intakes of individual nutrients or foods and risk of CHD (2), but few have evaluated the relation of overall dietary patterns to the risk. Conceptually, examination of overall dietary patterns would more closely parallel the real world, where people do not eat isolated nutrients but rather meals consisting of a variety of foods with complex combinations of nutrients that may be interactive or synergistic (3). Studies of individual foods or nutrients can be difficult to interpret because of strong correlations among them. In dietary pattern analyses, the collinearity of nutrients or foods can be used to advantage because patterns are characterized on the basis of habitual food use (4). In addition, dietary interventions may be easier to implement and may be more comprehensive when initiated as a change in the overall dietary pattern (5). In clinical studies, changes in dietary patterns appeared to be more effective in lowering blood pressure than was supplementation with single nutrients (6, 7). Finally, dietary pattern analysis is potentially useful in making dietary recommendations because overall dietary patterns might be easy for the public to interpret or translate into diets (8).
A major difficulty in studying the relation between dietary patterns and disease outcomes is that dietary patterns cannot be measured directly. One commonly used statistical method for quantifying dietary patterns is factor analysis. Factor analysis is a multivariate technique (9, 10) that, in a dietary context, uses information reported on food-frequency questionnaires (FFQs) (11–13) or in dietary records (14) to identify common underlying factors (patterns) of food consumption. Recently, Slattery et al (15) showed in a case-control study that eating patterns identified by factor analysis were significantly associated with colon cancer risk, independent of other lifestyle variables. In a previous study (16), we showed reasonable reproducibility and validity of major dietary patterns defined by factor analysis using food consumption data collected through an FFQ. In this study, we attempted to evaluate whether these dietary patterns predicted incidence of CHD during 8 y of follow-up in the Health Professionals Follow-up Study (HPFS).
SUBJECTS AND METHODS
Subjects
The HPFS began in 1986 when 51529 US health professionals (dentists, optometrists, pharmacists, podiatrists, and veterinarians) aged 40–75 y answered a detailed questionnaire that included a comprehensive diet survey and items on lifestyle practice and medical history. Follow-up questionnaires were sent in 1988, 1990, 1992, and 1994 to update information on potential risk factors and to identify new cases of cardiovascular and other diseases. We excluded from the analysis men who did not satisfy the a priori criteria of a reported daily energy intake between 3.3 and 17.6 MJ (800 and 4200 kcal) or who left blank >70 of a total of 131 food items on the diet questionnaire (n = 1595). We also excluded men with a prior diagnosis of myocardial infarction (MI), angina, coronary artery surgery, stroke, transient ischemic attack, or peripheral arterial disease at baseline (n = 5059). Thus, we followed 44875 men to determine the incidence of CHD over the subsequent 8 y.
Endpoints
The endpoints were incident CHD (including nonfatal MI and fatal CHD) occurring between the return of the baseline questionnaire and 31 January 1994. We inquired about the occurrence of cardiovascular endpoints on each biennial questionnaire. Participants reporting an incident MI were asked for their permission for us to review their medical records. Nonfatal MI was confirmed by using World Health Organization (WHO) criteria: symptoms plus either typical electrocardiographic changes or increased activities of cardiac enzymes (17). MIs that required hospital admission and for which confirmatory information was obtained by interview or letter, but for which no medical records were available, were designated as probable. We included all confirmed and probable cases in the analyses because results were the same after exclusion of probable cases. The follow-up rate for nonfatal events was 97% of the total potential person-years of follow-up.
Deaths were reported by next of kin, coworkers, or postal authorities or were obtained from the National Death Index (18). Using all sources combined, we estimated that follow-up for the deaths was >98% complete. Fatal CHD was confirmed by medical records or autopsy reports or if it was the underlying cause on the death certificate and a diagnosis of CHD was confirmed by other sources. Sudden death within 1 h of the onset of symptoms in men with no other plausible cause of death (other than CHD) was also included.
Semiquantitative food-frequency questionnaire
The reproducibility and validity of the FFQ used in this study was reported previously (19, 20). The FFQ includes 131 food items with specified serving sizes that are described by using natural portions (eg, 1 banana and 2 slices of pizza) or standard weight and volume measures of the servings commonly consumed in this study population. For each food item, participants indicated their average frequency of consumption over the past year in terms of the specified serving size by checking 1 of 9 frequency categories ranging from "almost never" to "6 times/d." The selected frequency category for each food item was converted to a daily intake. For example, a response of "2–4 servings/wk" was converted to 0.43 servings/d (3 servings/wk).
Assessment of dietary patterns
The procedure for deriving dietary patterns by using food consumption data from the FFQ was described in detail elsewhere (16). Briefly, we first classified 131 food items in the FFQ into 40 predefined food groups (Table 1) to minimize within-person variations in intakes of individual foods. Individual food items were preserved if they constituted a distinct item on their own (eg, eggs, butter, margarine, pizza, soup, coffee, and tea) or if they were thought to represent a particular dietary pattern (eg, garlic, liquor, wine, beer, and French fries). In secondary analyses, we used all 131 food items to derive dietary patterns.
View this table:
TABLE 1.. Food groupings used in the dietary pattern analysis
We conducted factor analysis (principal components) to derive dietary patterns based on the 40 foods or food groups (10). The factors were rotated by an orthogonal transformation (resulting in uncorrelated factors) to achieve a simpler structure with greater interpretability. In determining the number of factors to retain, we considered components with an eigenvalue >1, the Scree test (9), and the interpretability of the factors. We did not use the percentage of variance explained by each factor because this criterion depends largely on the total number of variables included in the analyses (9). The factor score for each pattern was constructed by summing observed intakes of the component food items weighted by factor loadings (9). The analyses were conducted by using the FACTOR PROCEDURE in SAS (21).
In a subsample of men (n = 127) in the HPFS (16), the 131-item FFQ was administered twice with a 1-y interval, and two 1-wk diet records were collected during that year. The factor analysis generated 2 major eating patterns. The first factor (the prudent pattern) was characterized by a high intake of vegetables, fruit, legumes, whole grains, fish, and poultry, whereas the second factor (the Western pattern) was characterized by a high intake of processed meat, red meat, butter, high-fat dairy products, eggs, and refined grains. The reliability correlations for the factor scores between the 2 FFQs were 0.70 for the prudent pattern and 0.67 for the Western pattern. The correlations (corrected for week-to-week variations in diet records) between the FFQ and the diet records were 0.52 for the prudent pattern and 0.74 for the Western pattern.
Statistical analysis
Participants were followed from the time the 1986 questionnaire was returned until the occurrence of a nonfatal MI or CHD death, death, or 31 January 1994. Relative risks (RRs) were calculated by dividing the incidence of CHD among men in the highest quintile of dietary pattern scores by the incidence among men in the lowest quintile of dietary pattern scores, with adjustment for age (40–44, 45–49, 50–54, 55–59, 60–64, 65–69, and 70 y). To adjust for other risk factors, we used pooled logistic regression with 2-y intervals (22), which is approximately equivalent to Cox regression for time-dependent covariates when the event rate is rare. Multivariate models were adjusted for total energy intake (quintiles), smoking status [never, past, current (1–14, 15–24, and 25 cigarettes/d)], alcohol consumption (0–4, 5–9, 10–14, 15–29, and 30 g/d), history of hypertension, history of diabetes, history of high cholesterol, parental history of MI before age 60 y, body mass index (quintiles), current vitamin E supplement use, and physical activity in metabolic equivalents per week (quintiles) (23).
RESULTS
We entered food consumption data for the 40 predefined food groups into the factor analysis procedure. The Scree plot of eigenvalues indicated 2 major patterns and 1 minor pattern. Thus, we extracted 3 factors in the final model. Factor-loading matrixes for the 2 major factors are listed in Table 2. The larger the loading of a given food item or group to the factor, the greater the contribution of that food item or group to a specific factor. The first factor was loaded heavily with the following foods or food groups: vegetables, legumes, whole grains, fruit, fish, and poultry; the second factor was loaded heavily with red meat, processed meat, refined grains, sweets and dessert, French fries, and high-fat dairy products. The first factor explained 10.0% of the total variance and the second factor explained 7.4% of the total variance. As with our previous study (16), we labeled the first factor as the prudent pattern and the second factor as the Western pattern. The third (minor) factor, which explained 4.4% of the total variance, did not appear to have a clear pattern of factor loadings, with a positive loading for some foods (eg, 0.36 for coffee, 0.36 for wine, 0.35 for liquor, 0.34 for beer, and 0.37 for garlic) and a negative loading for others (eg, -0.47 for low-fat dairy products, -0.43 for cold breakfast cereal, -0.31 for fruit, and -0.31 for whole-grain products). Our analyses did not suggest a significant association between this pattern and risk of CHD (data not shown).
View this table:
TABLE 2.. Factor-loading matrix for the major factors (diet patterns) identified by using food consumption data from the food-frequency questionnaire used in the Health Professionals Follow-up Study in 19861
The 2 major patterns closely resembled those identified in a validation study using data from two 1-wk diet records in a subsample of the HPFS (16). When the whole cohort was randomly divided into 2 groups, the 2 major patterns were similar for the 2 groups and closely resembled those for the overall sample. Participants with a higher prudent pattern score were more likely to take multivitamin and vitamin E supplements and to engage in exercise, were less likely to smoke cigarettes, were slightly older and leaner, and were more likely to have hypercholesterolemia (Table 3). This cross-sectional relation probably reflects changes in diet after a diagnosis of high cholesterol concentrations. Men with a higher prudent pattern score also had higher intakes of dietary folate, fiber, protein, and carbohydrate and lower intakes of cholesterol, saturated fat, monounsaturated fat, and trans fat. As expected, consumption of vegetables, fruit, fish, poultry, low-fat dairy products, and whole-grain products was directly correlated with the prudent pattern score, whereas consumption of red meat, processed meat, and sweets and dessert was inversely correlated with the prudent pattern score.
View this table:
TABLE 3.. Age-standardized baseline characteristics according to quintiles of prudent pattern score in the Health Professionals Follow-up Study in 1986
In contrast, men with a higher Western pattern score were more likely to smoke and drink alcohol and less likely to take multivitamin or vitamin E supplements and to exercise (Table 4). These men also had higher intakes of cholesterol, saturated fat, and monounsaturated fat and lower intakes of folate, fiber, protein, and carbohydrate. Consumption of red meat, processed meat, eggs, butter, high-fat dairy products, refined grain, and sweets and dessert was positively correlated with the Western pattern score.
View this table:
TABLE 4.. Age-standardized baseline characteristics according to quintiles of Western diet score in the Health Professionals Follow-up Study in 1986
During 8 y of follow-up (311606 person-years), we documented 1089 incident cases of CHD (730 nonfatal MIs and 359 fatal cases of CHD). After adjustment for age, a higher prudent pattern score was strongly associated with a monotonic lower risk of CHD (Table 5). The age-adjusted RRs across increasing quintiles of prudent pattern score were 1.0, 0.84, 0.76, 0.71, and 0.66 (95% CI: 0.54, 0.80; P for trend < 0.0001). Further adjustment for smoking and other CHD risk factors did not change the results materially. This inverse association was similar for fatal CHD; multivariate RRs across quintiles of prudent pattern score were 1.0, 0.94, 0.78, 0.63, and 0.66 (95% CI: 0.46, 0.94; P for trend = 0.0003) and nonfatal MI. RRs after adjustment for smoking, BMI, and other CHD risk factors were 1.0, 0.83, 0.78, 0.81, and 0.70 (95% CI: 0.54, 0.91; P for trend = 0.03). The inverse association between the prudent pattern score and total CHD was persistent among subgroups of current smokers and nonsmokers, obese and nonobese individuals, and those with and without a parental history of MI (Figure 1).
View this table:
TABLE 5.. Relative risks (95% CIs) of coronary heart disease according to quintiles of prudent pattern score1
FIGURE 1. . Multivariate relative risks (RRs) of coronary heart disease according to quintiles (Q1–Q5) of prudent pattern score by smoking status, BMI, and parental history of myocardial infarction (MI). The models include all the nondietary covariates listed in Table 4 except the stratifying variable. Note that the scale of the y axis for the smoking subgroups is different from that of the other 2 panels.
In our analyses, intakes of fiber—especially cereal fiber (24)—and folate (25) were inversely associated with risk of CHD. To examine the degree to which the inverse association for the prudent pattern was mediated by these beneficial nutrients, we included dietary intakes of cereal fiber and folate in the multivariate model. Additional adjustment for these nutrients only slightly attenuated the inverse association for the prudent pattern score: the RR for the comparison of the highest with the lowest quintile was 0.75 (95% CI: 0.59, 0.95; P for trend = 0.02).
After adjustment for age, a higher Western pattern score was associated with a monotonic higher risk of CHD (Table 6). The age-adjusted RRs across increasing quintiles of the Western pattern score were 1.0, 1.21, 1.24, 1.24, and 1.45 (95% CI: 1.19, 1.77; P for trend < 0.0001). Further adjustment for smoking, alcohol use, and other coronary risk factors slightly strengthened these associations. The positive association appeared to be stronger for fatal CHD: the RRs across quintiles of the score were 1.0, 1.63, 1.49, 1.50, and 2.15 (95% CI: 1.34, 3.46; P for trend = 0.009) than for nonfatal MI. Corresponding RRs of nonfatal MI were 1.0, 1.12, 1.30, 1.33, 1.46 (95% CI: 1.04, 2.05; P for trend = 0.0007). The positive association between the Western diet pattern score and total CHD persisted in subgroup analysis according to current smoking, obesity, and parental history of MI (Figure 2). Additional adjustment for intakes of nutrients that were directly correlated with this pattern, such as dietary cholesterol, saturated fat, and trans fat, only slightly attenuated these results: RRs across quintiles of Western pattern score were 1.0, 1.21, 1.27, 1.27, and 1.43 (95% CI: 1.01, 2.01; P for trend = 0.004).
View this table:
TABLE 6.. Relative risks (95% CIs) of coronary heart disease according to quintiles of Western pattern score1
FIGURE 2. . Multivariate relative risks (RRs) of coronary heart disease according to quintiles (Q1–Q5) of Western pattern score by smoking status, BMI, and parental history of myocardial infarction (MI). The models include all the nondietary covariates listed in Table 4 except the stratifying variable. Note that the scale of the y axis for the smoking subgroups is different from that of the other 2 panels.
In an additional analysis, we subdivided breakfast cereal into whole and refined cereals using the criteria developed by Jacobs et al (26). Specifically, breakfast cereals with 25% whole-grain or bran content by weight were classified as whole-grain and the rest were classified as refined-grain cereals. The associations between the 2 major patterns and risk of CHD obtained from this analysis were similar to those mentioned above. The multivariate RR comparing extreme quintiles of the prudent pattern score was 0.72 (95% CI: 0.58, 0.89) and the corresponding RR for the Western pattern was 1.62 (95% CI: 1.23, 2.15).
Although the 2 dietary pattern scores were statistically independent through the orthogonal transformation procedure, it is possible for one individual to have high or low scores on the 2 patterns at the same time. We therefore examined CHD risk according to joint classifications of Western pattern and prudent pattern scores. In a multivariate analysis, the observed associations for the 2 patterns appeared to be independent of each other. Compared with those with the lowest score for the prudent pattern and highest score for the Western pattern, the RR of CHD for men in the highest category of the prudent pattern compared with the lowest category of the Western pattern was 0.50 (95% CI: 0.34, 0.74).
We conducted several sensitivity analyses to examine the robustness of our findings. Use of the maximum likelihood rather than the principal component method to extract initial factors or an alternative algorithm [oblique rotation (producing correlated factors) rather than orthogonal rotation (producing uncorrelated factors)] to rotate factors yielded similar loadings for the 2 major diet patterns. When the final factor solutions were limited to 2 rather than to 3 factors, the factor loadings for the 2 factors did not change appreciably, and neither did the associations between the factors scores and the risk of CHD. The multivariate RRs across quintiles of the prudent pattern score were 1.0, 0.95, 0.78, 0.77, and 0.72 (95% CI: 0.58, 0.89; P for trend = 0.0009). Corresponding RRs across quintiles of the Western pattern score were 1.0, 1.15, 1.45, 1.30, and 1.61 (95% CI: 1.24, 2.10; P for trend < 0.0001). Finally, we derived dietary patterns based on food consumption data from all 131 food items in the FFQ rather than from the predefined food groups. The 2 major diet patterns identified from this analysis were qualitatively similar to those identified by using predefined food groups. Additionally, the associations between the 2 major diet patterns and risk of CHD did not materially change; the multivariate RRs across quintiles of the prudent pattern score were 1.0, 0.99, 0.85, 0.78, and 0.76 (95% CI: 0.62. 0.94; P for trend = 0.006). Corresponding RRs across quintiles of the Western pattern score were 1.0, 1.06, 1.20, 1.36, and 1.54 (95% CI: 1.19, 2.01; P for trend = 0.0003).
DISCUSSION
With the use of dietary data from an FFQ, 2 major dietary patterns emerged by factor analysis. The first factor, the prudent pattern, was characterized by high intakes of vegetables, fruit, legumes, whole grains, fish, and poultry, whereas the second factor, the Western diet pattern, was characterized by high intakes of red meat, processed meat, refined grains, sweets and dessert, French fries, and high-fat dairy products. During 8 y of follow-up, we found that as prudent pattern score increased, the risk of CHD decreased, even after adjustment for potential beneficial nutrients such as folate and cereal fiber. In contrast, as Western pattern score increased, the risk of CHD increased, even after adjustment for potential deleterious nutrients such as saturated fat, trans fat, and cholesterol. These data suggest that the 2 major dietary patterns derived from the FFQ predict the risk of CHD, independent of the effects of several known beneficial or deleterious nutrients.
Our data are consistent with ecologic observations. Distinct eating patterns reflecting different dietary traditions have been related to disease rates in different countries (1). The Mediterranean and Asian diets have attracted considerable interest as alternatives to the Western diet because of the extremely low rates of CHD in Greece and Japan. Compared with the typical American diet, traditional Mediterranean and Asian diets contain substantially more grains, legumes, vegetables, fruit, and fish and less red meat, high-fat dairy products, and other animal products. However, because many other potential risk factors besides diet vary among countries, one cannot confidently attribute differences in CHD rates to distinct dietary patterns. Nevertheless, the overlap of the prudent diet pattern identified by the factor analysis with the Mediterranean and Asian diets increases our confidence that the factor analysis identified meaningful patterns.
Relating dietary patterns to disease outcomes within a population is relatively new in nutritional epidemiology. Limited data suggest that overall dietary patterns assessed by various methods are indeed associated with risk of diseases or mortality. Using dietary data collected by a diet-history questionnaire, Slattery et al (15) derived 2 major patterns that closely resembled ours: one was a Western-type diet pattern characterized by a high intake of red meat, processed meat, fast food, refined grains, and sugar-containing foods; the second was the prudent diet pattern, characterized by high intakes of vegetables and fruit, fish, and poultry. The authors found that the prudent diet pattern was associated with a lower risk of colon cancer, whereas the Western-type diet pattern was associated with an increased risk. In a study of elderly rural Greeks (27), greater adherence to the traditional Mediterranean dietary pattern (reflected by a composite score for intakes of vegetables, fruit, legumes, ethanol, dairy products, cereal, and monounsaturated rather than saturated fat) was significantly associated with a reduction in total mortality. Similarly, Huijbregts et al (28) found that a dietary indicator of a healthy dietary pattern, based on the World Health Organization's guidelines for the prevention of chronic diseases, was significantly associated with reduced overall mortality in the Seven Countries Study. Using cluster analysis, Farchi et al (29) identified 4 distinct dietary patterns in 2 Italian rural cohorts of the Seven Countries Study. After 20 y of follow-up, they found the lowest risk of CHD in the cluster with high vegetable oil consumption and the highest risk of stroke in the cluster with high alcohol consumption. Using data from the National Health and Nutrition Examination Survey Epidemiologic Follow-up Study, Kant et al found inverse associations between dietary diversity (a score counting the number of food groups consumed daily: dairy, meat, grain, fruit, and vegetable) and subsequent overall mortality (30) and cause-specific mortality (31). In addition, Slattery et al (32) found that diet diversity in plant intake, but not in meat intake, was associated with lower risk of colon cancer. However, the dietary diversity score is a crude measure of dietary patterns because it does not distinguish the effects of different types of foods.
In contrast with the traditional analytic approach used in nutritional epidemiology, dietary pattern analysis considers overall diet rather than individual nutrients or foods. This approach would more closely parallel the real world, where dietary intakes consist of nutrients that occur together in common foods. It can take into account complicated interactions among nutrients and nonnutrient substances in studies of free-living people. On the other hand, because there are many potential differences in nutrients between dietary patterns, this approach cannot be specific about the particular nutrients responsible for the observed differences in disease risk, and thus it may not be very informative about biological relations between dietary components and disease risk. However, our findings are consistent with associations between intakes of nutrients and foods identified in previous epidemiologic studies. In particular, higher consumption of fruit and vegetable (33), whole-grain products (26, 34, 35), and fish (36–38) has been associated with a reduced risk of CHD. Intakes of nutrients that are directly correlated with the prudent diet pattern, including folate (25), vitamin E (39–41), and fiber (24, 42, 43) are protective against CHD. On the other hand, red meat consumption was associated with increased CHD risk (44, 45). Intake of heme iron, primarily from red meat, was significantly related to an increased risk of MI (46). Whole milk rather than skim milk consumption has been associated with increased risk of CHD (34). Carbohydrate-containing foods with a high glycemic index, such as white bread and mashed potatoes, have been associated with an increased risk of type 2 diabetes (47, 48) and CHD (49).
Statistical methods used to define dietary patterns such as factor analysis are somewhat subjective (50) because decisions on the number of factors are usually based on empirical guidelines rather than on an exact quantitative solution (9, 10). In determining the number of factors, we adhered to established empirical guidelines as closely as possible. Meanwhile, multiple sensitivity analyses suggested the robustness of our findings on both the derived major patterns and their relations with CHD. Additionally, in our previous study (16), we showed reasonable reproducibility over time and comparability between the FFQs and diet records in characterizing dietary patterns in a subsample of the HPFS. The same procedures for defining food groups and deriving dietary patterns were used in this study.
A previous study (51) found that patients with hypercholesterolemia consuming a National Cholesterol Education Program Step I diet containing primarily lean red meats or lean white meats (poultry and fish) had similar reductions in LDL-cholesterol cholesterol. It is possible that lean red meats and fat meats may belong to different eating patterns and have different associations with risk of CHD. We were unable to separate lean red meats from fat meats in the dietary pattern analysis because our FFQ only asked about overall meat consumption. Nevertheless, in our previous analysis of the Nurses' Health Study (52), overall red meat consumption was associated with a modest increase in CHD risk, whereas white meat consumption was associated with a lower risk. Substitution of white meat for red meat was associated with significantly reduced risk of CHD.
Dietary patterns are likely to vary by sex, socioeconomic status, ethnic group, and culture. Thus, it is necessary to replicate the results of our study in other populations. Additionally, because of changes in food preferences and food availability, the meaning of dietary pattern could change over time. Finally, the 2 major diet patterns derived from our data explained <20% of total variance, suggesting the potential existence of other patterns. In a previous study (15), the Western and prudent diet patterns explained 19% of the variance in men and 15% of the variance in women. However, the percentage of variance explained by the factors should be interpreted with caution because it depends heavily on the total number of variables used in the factor analysis (9).
In conclusion, our data suggest that 2 dietary patterns derived from food consumption data assessed by the FFQ significantly predict the incidence of CHD, independent of other lifestyle variables. This study provided strong evidence that a diet high in vegetables, fruit, legumes, whole grains, fish, and poultry and low in red meat, processed meat, high-fat dairy products, and refined grains may reduce the risk of CHD. Our findings may provide practical guidance for nutrition intervention and education because overall patterns of dietary intake might be easy for the public to interpret or translate into diets.
REFERENCES