Multivariate relative risk (RR) of dietary patterns and coronary heart disease, adjusted for age, period, smoking, body mass index, hormone replacement therapy, aspirin use, caloric intake, family history, history of hypertension, multivitamin and vitamin E use, and physical activity.
Fung TT, Willett WC, Stampfer MJ, Manson JE, Hu FB. Dietary Patterns and the Risk of Coronary Heart Disease in Women. Arch Intern Med. 2001;161(15):1857-1862. doi:10.1001/archinte.161.15.1857
Although substantial information on individual nutrients or foods and risk of coronary heart disease (CHD) is available, little is known about the role of overall eating pattern.
Using dietary information from a food frequency questionnaire in 1984 from the Nurses' Health Study, we conducted factor analysis and identified 2 major dietary patterns—"prudent" and "Western"—and calculated factor scores of each pattern for individuals in the cohort. We used logistic regression to examine prospectively the associations between dietary patterns and CHD risk among 69 017 women aged 38 to 63 years in 1984 without history of major chronic diseases.
The prudent pattern was characterized by higher intakes of fruits, vegetables, legumes, fish, poultry, and whole grains, while the Western pattern was characterized by higher intakes of red and processed meats, sweets and desserts, french fries, and refined grains. Between 1984 and 1996, we documented 821 CHD cases. After adjusting for coronary risk factors, the prudent diet score was associated with a relative risk (RR) of 0.76 (95% confidence interval (CI), 0.60-0.98; P for trend test, .03) comparing the highest with lowest quintile. Extreme quintile comparison yielded an RR of 1.46 (95% CI, 1.07-1.99; P for trend test, .02) for the Western pattern. Those who were jointly in the highest prudent diet quintile and lowest Western diet quintile had an RR of 0.64 (95% CI, 0.44-0.92) compared with those with the opposite pattern profile.
A diet high in fruits, vegetables, whole grains, legumes, poultry, and fish and low in refined grains, potatoes, and red and processed meats may lower risk of CHD.
TRADITIONALLY, nutritional epidemiology has largely focused on the effects of single nutrients or foods on disease outcomes.1- 3 However, because nutrients and foods are consumed in combination, their joint effects may be best investigated by considering the entire eating pattern. Analyzing food consumption in the form of dietary patterns offers a perspective different from the traditional single-nutrient focus and may provide a comprehensive approach to disease prevention or treatment, which has been used in several settings, including Dietary Approach to Stop Hypertension4 and the Lyon Diet Heart Study.5
In a previous study of men, we identified 2 major dietary patterns by factor analysis.6 One is labeled the "prudent pattern," characterized by a higher intake of fruits, vegetables, fish, whole grains, and legumes. The other pattern is labeled the "Western pattern," characterized by a higher intake of red and processed meat, high-fat dairy products, sweets, and desserts. These 2 patterns significantly predicted the risk of coronary heart disease (CHD) during 8 years of follow-up among 44 875 men in the Health Professionals' Follow-up Study.7 However, such associations have not been examined in women. Because eating patterns may differ between men and women, we therefore examined prospectively the associations between dietary patterns and the risk of CHD among women in the Nurses' Health Study.
The Nurses' Health Study began in 1976, when 121 700 female nurses aged 30 to 55 years in 11 US states responded to a questionnaire regarding medical, lifestyle, and other health-related information.8 Since then, questionnaires have been sent biennially to update this information. In 1980, the participants completed a 61-item food frequency questionnaire (FFQ). In 1984, the FFQ was expanded to 116 items. Similar FFQs were sent to the women in 1986, 1990, and 1994. We used the 1984 FFQ as baseline for this study because it was similar to the baseline (1986) FFQ used in the Health Professionals' Follow-up Study.
For the present analysis, women were included if they completed the 1984 semiquantitative FFQ with less than 70 missing items and a total caloric range (as calculated from the FFQ) between 500 and 3500 kcal/d. We excluded women with a history of myocardial infarction (MI) (n = 484), angina (n = 1604), coronary artery bypass surgery (n = 40), hypercholesterolemia (n = 5669), diabetes (n = 2127), and stroke (n = 178). We excluded subjects with diabetes and high cholesterol level because diagnoses of these conditions at baseline may lead to change in diet. We then included in this analysis 69 017 women with follow-up for up to 12 years, from 1984 to 1996.
Dietary intake information was collected by FFQs. The questionnaire was designed to assess average food intake during the previous year, and standard portion size was given for each food item. Cohort members were asked to choose from 9 possible frequency responses, ranging from "never" to "more than 6 times a day" for each food. Total caloric intake was calculated by summing up intakes from all foods. For this analysis, we used information from the FFQ administered in 1984, which had 116 items. Foods from the FFQ were classified into 38 food groups based on nutrient profiles or culinary usage in a manner similar to that in a previous study in men (food group classifications are available from the authors).7 Foods that did not fit into any of the groups or that may represent distinctive dietary behaviors were left as individual categories (eg, pizza, tea, beer). Previous validation studies among members of the NHS cohort showed good correlations between nutrients assessed by the FFQ and multiple weeks of food records completed during the previous year.9
Our end point included fatal CHD and nonfatal MI that occurred between the return of the 1984 questionnaire and June 1, 1996. We sought permission to review medical records of all self-reports of incident MI for confirmation of diagnosis. These records were reviewed by physicians with no knowledge of the participants' risk factor status. Myocardial infarctions were confirmed by World Health Organization criteria based on symptoms and changes in electrocardiogram or cardiac enzymes.10 Infarctions that required hospitalization were classified as probable when confirmatory information was obtained by letter or interview, but medical records were not available. Probable cases were included in our analysis (approximately 17%).
Deaths were reported by family members, by the postal service, or through searches in the National Death Index,11 and ascertainment was 98% complete. Fatal MI was confirmed by medical records, autopsy reports, or death certificate if a diagnosis of CHD was also identified from other sources. We also designated the cause of death as presumed CHD if it was indicated on death certificates but additional medical records were not available. In addition, deaths that occurred within 1 hour after the onset of symptoms with no other plausible non–CHD-related cause of death were also included as fatal CHD cases (about 12% of fatal CHD cases).
Dietary patterns were generated by factor analysis (principal components) on the basis of 38 food groups by means of the orthogonal rotation procedure.12 This results in uncorrelated factors, which are easier to interpret. We determined the number of factors to retain by eigenvalue (greater than 1), Scree test, and factor interpretability. The factor score for each pattern was found by summing intakes of food groups weighted by factor loading,13 and each individual received a factor score for each identified pattern. Good reproducibility with this method has been obtained in a parallel cohort of men.6 Factor analysis was conducted with SAS PROC FACTOR (SAS Institute Inc, Cary, NC).14
To quantify association between dietary patterns and CHD risk, we used person-time analysis. Study participants began contributing follow-up time from the date of return of the 1984 FFQ. Accumulation of follow-up time ceased on first diagnosis of MI, death, the last date of questionnaire return, or June 1, 1996. The incidence rate for each dietary pattern was calculated by dividing the number of CHD cases in each quintile of pattern scores by the person-years of follow-up.
Multivariate relative risks (RRs) of each of the dietary patterns in 1984 for CHD were calculated by means of pooled logistic regression with 2-year increments. For rare events, this method approximates the Cox proportional hazard model with time-dependent covariates.15,16 We adjusted for age (5-year categories), smoking (never, past, and 1-14, 15-24, or ≥25 cigarettes per day), energy intake, body mass index (calculated as weight in kilograms divided by the square of height in meters; 5 categories), multivitamin and vitamin E supplement use, physical activity (5 categories in hours per week), hormone replacement therapy (premenopause, never, past, or current), parental history of MI, and history of hypertension. We also tested for interaction between the 2 dietary patterns by creating categorical interaction terms of the 2 patterns and comparing the log likelihood of this model with the model that contained the main effects only. We also considered dietary pattern scores as cumulative average to reduce within-person variation and to represent long-term intake.17 For example, the 1984 dietary patterns were used in relation to incidence from 1984 to 1986, the average of 1984 and 1986 dietary pattern scores were used to predict incidence in 1986 and 1988, and so forth. We stopped updating dietary information if an individual developed hypertension or an intermediate cardiac end point such as angina.
We identified 2 major dietary patterns with the factor analysis procedure. Table 1 shows the factor loading matrix for the 2 patterns. A positive loading indicates positive association with the factor, and vice versa for a negative loading. The magnitude of the loading indicates the degree of contribution of the particular food group to the factor. The first factor was heavily contributed to by fruits, vegetables, whole grains, legumes, poultry, and fish. The second factor was heavily contributed to by refined grains, processed and red meats, desserts, high-fat dairy products, and french fries. As with our previous studies, we labeled the first factor the "prudent" pattern and the second factor the "Western" pattern.6
Individuals with high prudent-pattern scores tended to smoke less; use more vitamin supplements; drink more alcohol; consume more folate, fiber, and protein; and consume less saturated and monounsaturated fats (Table 2). As expected, they also had higher intakes of fruits, vegetables, whole grains, and low-fat dairy products. On the other hand, those with high "Western" pattern scores were more likely to be current smokers, use fewer vitamin supplements, consume more fat, and consume less folate and fiber (Table 3). Their diets also contained more red and processed meats, eggs, butter, and refined grains.
Between 1984 and 1996, we accumulated 801 075 person-years of follow-up and identified 821 incident cases of CHD. After adjustment for age, a higher prudent-pattern score was associated with a lower risk of total CHD (P for trend, <.001) (Table 4). The RR comparing the highest prudent-pattern quintile to the lowest was 0.61 (95% confidence interval [CI], 0.49-0.76). After adjusting for body mass index, smoking, caloric intake, supplemental vitamin use, hormone replacement therapy, and other coronary risk factors, the prudent pattern remained significantly and inversely associated with the risk of CHD (P for trend, .03) (Table 4). The RR for the top quintile was 0.76 (95% CI, 0.60-0.98) compared with the lowest quintile of prudent-pattern score. In contrast, a higher Western-pattern score was associated with a higher risk of total MI after adjusting for age (P for trend, <.001) (Table 4). The RR for the top quintile vs the lowest quintile was 1.44 (95% CI, 1.16-1.78). The significant positive association remained after multivariate adjustment (P for trend = .02). The RR for the top quintile vs the lowest quintile was 1.46 (95% CI, 1.07-1.99). Separate analyses of fatal CHD and nonfatal MI produced similar results (data not shown). The inverse association between the prudent pattern and the risk of CHD was not appreciably different between smokers and nonsmokers, lean and overweight individuals, and those with and without a family history of CHD (data not shown). Also, the positive associations between the Western pattern and CHD persisted in all subgroups. Results using average dietary patterns between 1984 and 1994 were also similar to those using the 1984 dietary information only. The RRs comparing extreme quintiles were 0.76 (95% CI, 0.60-0.95) for the prudent pattern and 1.30 (95% CI, 0.98-1.73) for the Western pattern.
In an additional analysis, we examined the association between joint classifications of both patterns and risk of CHD. As expected, the risk of CHD tended to decrease with increasing prudent-pattern score at any level of Western pattern score (Figure 1). Similarly, the risk of CHD tended to increase with increasing Western-pattern score. The lowest risk was at the highest level of prudent score and lowest level of Western score (RR, 0.64; 95% CI, 0.44-0.92). The test for interaction between the 2 dietary patterns was not statistically significant (P = .26 for interaction).
In this large cohort of women, we identified 2 major dietary patterns by means of factor analysis. The prudent pattern was characterized by a higher intake of fruits, vegetables, whole grains, legumes, poultry, and fish. On the other hand, the Western pattern was characterized by a higher intake of red and processed meats, high-fat dairy products, french fries, refined grains, and desserts and sweets. We found a significant inverse association between the prudent pattern and the risk of total CHD and a positive association between the Western pattern and the risk of CHD.
The 2 major patterns were similar to those observed among men in the Health Professionals' Follow-up Study.7 Also, our observed association between dietary patterns and MI were consistent with that in our previous study in men, in which the RR of CHD comparing extreme quintiles was 0.70 (95% CI, 0.56-0.86) for the prudent pattern and 1.64 (1.24-2.17) for the Western pattern.7 Therefore, the major dietary patterns derived through factor analysis appear to be applicable to both men and women. The relationship between eating patterns and CHD risk may act through biochemical risk factors for CHD. In a previous study of 466 men, we found that the prudent pattern was associated with lower levels of insulin, and the Western pattern was associated with higher levels of tissue plasminogen activator, fasting insulin, leptin, and homocysteine.18 Women with a higher prudent-pattern score in our study had a higher intake of fiber, protein, and folate. Thus, it is likely that the inverse association between prudent pattern and CHD was mediated by these nutrients.
The 2 major patterns were also similar to those observed in other studies. In a study of diet and colon cancer from members of the Kaiser Permanente Medical Care Program, several patterns were identified, and the 2 major patterns were similar to those we have identified.19 In a group of Canadians, 3 food patterns—high-energy, traditional, and health-conscious—were identified.20 The health-conscious pattern was similar to the prudent diet in our study. A major challenge in studying dietary patterns and disease risk is to establish a quantitative method to identify eating patterns, unless a specific pattern (eg, Mediterranean) has been specified. The factor analysis approach involves several arbitrary but important decisions, including the consolidation of food items into food groups, and the number of factors to extract. Nevertheless, the major patterns identified through factor analyses are consistent with a priori expectation of patterns. The lower rate of CHD in Asia and the Mediterranean region may in part result from their respective dietary patterns.21 Diets in Japan and Crete, for example, are in some ways similar to the prudent pattern in this study, with low amounts of animal products and higher amounts of vegetables and whole grains.22,23 Intervention studies have demonstrated the effectiveness of modifying multiple aspects of the diet in prevention or treatment of diseases. The Dietary Approach to Stop Hypertension trial emphasized fruits, vegetables, and low-fat dairy products and successfully reduced both diastolic and systolic blood pressure in hypertensive subjects.4 Also, in the Lyon Diet Heart Study, patients with a history of MI who followed the Mediterranean diet with a high amount of α-linolenic acid had significantly fewer subsequent cardiac events and mortality than those who followed a regular low-fat diet.5
The present study used prospectively collected data with minimal loss to follow-up and therefore rendered information and selection bias highly improbable. We also extensively controlled for variables that could possibly confound the association, such as physical activity, vitamin supplements, smoking, obesity, and use of postmenopausal hormones. However, because different ethnic groups or populations at different regions may have different dietary patterns, our results need to be verified in different populations.
In conclusion, major dietary patterns identified by means of factor analysis significantly predicted the risk of CHD in this cohort of women. This study indicates that a diet high in fruits, vegetables, legumes, whole grains, poultry, and fish and low in red and processed meats and refined grain products may lower risk of CHD in women.
Accepted for publication January 11, 2001.
This study was supported by grants HL60712, CA40356, HL34594, and HL24074 from the National Institutes of Health, Bethesda, Md.
Presented as a poster at the Annual Meeting of the Society of Epidemiological Research, Seattle, Wash, June 16, 2000.
Corresponding author and reprints: Teresa T. Fung, ScD, RD, Programs in Nutrition, Simmons College, 300 The Fenway, Boston, MA 02115 (e-mail: email@example.com).