Dietary patterns and cardiovascular mortality in the Melbourne Collaborative Cohort Study

¹ From the Department of Epidemiology and Preventive Medicine, Monash University, Melbourne, Australia (LRH, DRE, and AMT); the Cancer Epidemiology Centre, The Cancer Council Victoria, Melbourne Australia (LRH, DRE, GGG, and AMH); the Centre for MEGA Epidemiology, School of Population Health, The University of Melbourne, Melbourne, Australia (DRE and AMH); the Department of Public Health and Primary Care, Institute of Public Health, University of Cambridge, Cambridge, United Kingdom (JP); and the Department of Medicine (University of Melbourne), St Vincent's Hospital, Melbourne, Australia (LB and KO)

² Supported by VicHealth and The Cancer Council Victoria for cohort recruitment, by grants from the National Health and Medical Research Council (284476, 209057, 124317, and 251533) and VicHealth (supplement grant 2003-0759), and by The Cancer Council Victoria and Monash University for further infrastructure support.

³ Reprints not available. Address correspondence to LR Harriss, Department of Epidemiology and Preventive Medicine, Central and Eastern Clinical School, 3rd Floor, Burnet Tower, Commercial Road, Melbourne, VIC 3004, Australia. E-mail: linton.harriss{at}med.monash.edu.au.

ABSTRACT  
Background: Despite increased cardiovascular disease risk factors, migrants to Australia from Mediterranean countries have lower mortality than do native-born Australians. Dietary patterns may contribute to this.

Objective: The objective was to investigate the relation between dietary patterns and mortality from cardiovascular (CVD) and ischemic heart disease (IHD) in an ethnically diverse population.

Design: This was a prospective cohort study (mean follow-up: 10.4 y) of 40 653 volunteers (23 980 women) aged 40–69 y in the Melbourne Collaborative Cohort Study (1990–1994); 24% of the subjects were Mediterranean born.

Results: Four dietary factors were identified from a food-frequency questionnaire with the use of principal components analysis. They explained 69% of intake variance and reflected frequent intakes of Mediterranean foods, vegetables, meat, and fresh fruit. The Mediterranean factor was inversely associated with CVD and IHD mortality in models adjusting for diabetes, waist-to-hip ratio, body mass index, and hypertension. For IHD, the hazard ratio (HR) for the highest compared with the lowest quartile of consumption was 0.59 (95% CI: 0.39, 0.89; P for trend = 0.03). Associations persisted in analyses excluding people with prior CVD (HR: 0.51; 95% CI: 0.30, 0.88; P for trend = 0.03). Vegetable and fresh fruit factors were inversely associated with CVD mortality but only among those without prior CVD. HRs (highest compared with lowest quartile) were 0.66 (95% CI: 0.48, 0.92; P for trend = 0.02) for vegetables and 0.69 (95% CI: 0.52, 0.93; P for trend = 0.04) for fresh fruit. The meat factor was not associated with CVD or IHD mortality.

Conclusion: Our findings suggest that frequent consumption of traditional Mediterranean foods is associated with reduced cardiovascular mortality after controlling for important risk factors and country of birth.

Key Words: Diet • factor analysis • statistical analysis • food habits • cardiovascular disease • coronary heart disease • mortality • prospective studies • Mediterranean diet

INTRODUCTION  
Despite large declines in age-adjusted rates, cardiovascular disease (CVD) remains the single largest cause of death and disability in Australia for both men and women (1). Continuing declines are possible and can be achieved through modification of individual behavior. For example, cardiovascular risk factors, including diabetes, dyslipidemia, hypertension, and overweight, can be influenced by dietary habits. Although migrants to Australia from Mediterranean countries have high prevalence of these risk factors, they have lower CVD mortality than native-born Australians (2). It is not clearly understood why this paradox exists. A possible reason for this observation is the traditional Mediterranean diet which is associated with higher intakes of plant foods and fish, moderate intake of wine, and lower intake of animal products. As such, it is a rich source of bioactive phytochemicals, monounsaturated fats, n–3 fatty acids, and fiber and is relatively low in saturated fats. In combination, these characteristics are believed to have antioxidant, anti-inflammatory, and antithrombotic properties that may counterbalance adverse effects of CVD risk factors (3).

Epidemiologic studies addressing diet have focused predominantly on identifying individual dietary constituents that might predict disease (4). However, the exact mechanisms responsible for the effects of these individual nutrients remain poorly understood (5). Nutrients are not consumed in isolation, and bioavailability is often determined by the interaction and combination of foods consumed at any one time (6). The effects of single foods or nutrients may be too small to measure, but effects of combinations may be measurable in epidemiologic studies. For these reasons, research into dietary patterns provides a practical and complementary approach to study dietary effects on disease outcomes. Two reviews show the growing interest in such studies (7, 8). Since 1990, 5 studies have investigated the association between dietary patterns and total CVD or ischemic heart disease (IHD) mortality (9–13). In general, those studies support the contention that healthier dietary patterns (characterized by fruit, vegetables, and whole grains) are cardioprotective, whereas those deemed less healthy (characterized by meat products, high-fat and processed foods) are associated with elevated cardiovascular risk. The predominant feature of those studies, however, is their tendency to investigate only 2 dietary factors with the use of a relatively small number of food items or predefined food groups. Further, with the exception of one other study (13), cohort participants tend to be culturally similar, which may result in homogeneity of dietary intake. These attributes influence the composition of dietary factors and may therefore limit interpretation of results (14).

In this study, we use data from a food-frequency questionnaire (FFQ) containing 121 food items collected from an Australian cohort with a high proportion of middle-aged Mediterranean-born migrants. We used factor analysis to identify 4 main dietary patterns and prospectively investigated the relation between these dietary factors and total CVD and IHD mortality.

SUBJECTS AND METHODS  
Study population
The Melbourne Collaborative Cohort Study is a prospective cohort of 41 528 participants (24 479 women) aged 40–69 y at baseline. Full details of the design, recruitment, and study procedures were published previously (15). Recruitment occurred from 1990 to 1994 by the electoral roll, ethnic radio, clubs, and churches. Twenty-four percent of subjects were southern European migrants, deliberately oversampled to extend the range of dietary and lifestyle exposures and genetic variation. The Cancer Council Victoria Human Research Ethics Committee approved the study, and subjects gave their written informed consent to participate.

At baseline, face-to-face interviews were conducted in the language of choice, and questionnaires addressing lifestyle exposure, including dietary intake, were completed. Subjects missing data for dietary intake and any of the final covariates modeled were excluded from analyses. Those who had daily energy intakes considered implausibly high (>99th percentile) or low (<1st percentile) were also excluded. This left 40 653 subjects for final analysis (23 980 women, 16 673 men).

Assessment of dietary intake
Dietary information was collected with the use of a self-administered, 121-item FFQ designed to assess the diverse diet in the cohort, over the preceding year (16). Nine possible frequency responses were available for each food item, ranging from "never or less than once per month" to "6 or more times per day." This information was then used to calculate daily equivalent frequency. Vitamin supplements were not included as food items. Repeatability analyses for the FFQ have shown fair-to-moderate agreement for food items when completed 12 mo apart (17). A recent study that used the FFQ data found moderate-to-strong correlations between dietary intake of monounsaturated and polyunsaturated fatty acids and plasma phospholipid fatty acids (18). A similar study found that estimated dietary antioxidants (-carotene, ß-carotene, ß-cryptoxanthin, lutein or zeaxanthin, and lycopene) were useful when predicting plasma antioxidant concentrations (JA Simpson, K O'Dea, DR English, et al, unpublished observations, 2006).

Baseline demographic and risk factors
Potential confounders determined at baseline were used in multivariate regression analyses. These confounders included country of birth (Australia, United Kingdom, Italy, and Greece), smoking (never, current, and former), total daily energy intake (continuous variable as kJ/d), physical activity (4 categories that used a score combining frequency and intensity of exercise in the past 6 mo), highest education level obtained (primary, some secondary, secondary school, and tertiary qualification), history of CVD (angina, myocardial infarction, or stroke), family history of CVD (mother, father, sister, or brother), history of diabetes, history of hypertension, social isolation (men living alone or who have no social activities, men and women who have no friends they could visit without an invitation), waist-to-hip ratio (WHR), body mass index (BMI; in kg/m²), and sex.

Ascertainment of deaths
Deaths were included in the analysis if they occurred between the recruitment date and 31 December 2003. By this date, 51 (0.13%) participants (29 Australian, 10 English, 2 Italian, and 10 Greek) had moved from Australia. Mortality definitions were based on primary cause of death with the use of the codes for CVD (390–459 and I00–I99) and IHD (410–414 and I20–I25), respectively, from the 9th and 10th revisions of the International Classification of Diseases (19, 20). The Victorian Cancer Registry assigned codes for underlying cause of death for deaths that occurred in 1991 and 1992. All other deaths were coded by the Australian Bureau of Statistics.

Statistical analysis
Analyses were performed with the use of STATA 8.2 (Stata Corp, College Station, TX). Factor analysis was used to determine the main dietary factors from the 121 food items. Six additional food items were calculated from separate questions. These items included intake of garlic, olive oil, vegetable oil, and alcohol as beer, wine, and spirits. Garlic intake was measured on a transformed scale in which a score of 1 indicated daily use, and 0.01 indicated no intake or less than once per month. Intake of olive and vegetable oils was calculated from total household intake divided by the number of people in the household to give mL/wk. Consumption of alcoholic beverages was measured as grams of alcohol per day.

Dietary factors were calculated by using data from all 40 653 eligible subjects. Orthogonal (varimax) rotation was then used to generate uncorrelated factors for greater interpretability. We determined the number of main dietary factors by choosing those accounting for the greatest variation of dietary intake and by graphical assessment of the Scree plot. Four factors with eigenvalues > 2 were retained. These accounted for 69% of total intake variance. Factor loadings with absolute values 0.20 were used to interpret the factors. Subjects received a factor score for each dietary factor, and these scores were divided into quartile groups for analyses.

Cox proportional hazards models were constructed with age as the time axis (21). Follow-up began at baseline and ended at the time of death, date left Australia, or 31 December 2003, whichever came first. Diabetes, WHR, BMI, and hypertension may be mediators of any relation between diet and CVD. We therefore performed 2 sets of analyses, first excluding and then including these covariates. Adjusted hazard ratios (HRs) were calculated for each dietary factor with the lowest quartile as the reference group. Confidence limits were set at the 95% level and two-sided P values are presented. A P value < 0.05 was regarded as significant. The proportional hazards assumption was tested for all variables in the model. Violation of the test occurred for variables representing history of CVD, family history of CVD, and sex. Final analyses were therefore stratified by these covariates. Tests of linear trend were performed across dietary factor categories by constructing a continuous variable assigning median values from each category. Tests for interaction were performed for country of birth (as a dichotomous variable using Australia and United Kingdom compared with Italy and Greece), sex, diabetes, and history of CVD.

Several sensitivity analyses were completed to determine the robustness of our results. First, the statistical analyses were repeated after removing those with a baseline history of CVD (n = 2449). Second, we performed factor and regression analyses for non–Mediterranean and Mediterranean-born subjects separately. Four factors were created by using each of the non-Mediterranean and Mediterranean subcohorts. All 4 factors from the non-Mediterranean subjects, and 3 factors from the Mediterranean subjects were similar and highly correlated with factors created by using all subjects. The fourth factor from the Mediterranean-born subjects was not correlated with any other factor. This factor was characterized by frequent intake of items, including salami, hard cheeses, sweet biscuits, ham, bacon, pasta, white bread, sausage or frankfurters, vegetable or chicken dishes, chocolate, corn or potato crisps, ice cream, soft drinks, tinned fish, peanut butter or peanuts, margarine, butter, and other confectionery. Adjusted HRs (CVD and IHD) for the highest compared with the lowest quartile of consumption for each of the 3 similar factors were consistent for each subgroup compared with those that used all subjects. None of the results reached statistical significance, and CIs were considerably wider, especially for the Mediterranean-born group. For these reasons, we opted to use data from all subjects for the main analyses.

RESULTS  
Dietary factors
Four main factors emerged with the use of the factor analysis procedure. Food items that had rotated factor loadings with absolute values of 0.2 or greater are given in Table 1. The first factor (Mediterranean) was characterized by frequent intake of items, including garlic, cucumber, olive oil, salad greens, capsicum, cooked dried legumes, legume soups, feta and ricotta cheeses, olives, steamed fish, and boiled chicken. This factor was also negatively associated with consumption of tea, margarine, sweet biscuits, and cake. Factor 2 (vegetables) was characterized by frequent intake of cauliflower, broccoli, carrot, cabbage or Brussels sprouts, pumpkin, green beans or peas, leafy greens, celery or fennel, potato cooked without fat, beetroot, zucchini or squash or eggplant, coleslaw, salad greens, cucumber, and capsicum. Factor 3 (meat) was associated with frequent intake of beef rissoles, roast beef or veal, fried potato, beef or veal schnitzel, savory pastries, mixed dishes with lamb, fried eggs, beef steaks, fried fish, and bacon. Factor 4 (fresh fruit) was characterized by frequent intake of apricots, peaches or nectarines, plums, cantaloupe or honeydew, grapes, watermelon, pears, strawberries, oranges or mandarins, figs, apples, and pineapple.

View this table:
TABLE 1. Rotated factor loadings for food items that constitute the 4 main dietary factors identified with the use of food-frequency questionnaire data from the Melbourne Collaborative Cohort Study (1990–1994)¹

 
Country of birth explained 36% of the variation in the Mediterranean factor, with 69% of those in the highest quartile born in Italy or Greece (Table 2). Education explained 20% of the variation in the Mediterranean factor, with 51% of those in the highest quartile having completed primary school only. This is because the Mediterranean-born subjects had lower levels of formal education than did Australian-born subjects. Country of birth explained 11% of variation in the vegetable factor, with 90% of those in the top 2 quartiles born in Australia or the United Kingdom. Thirty-three percent of the variation in daily energy intake was explained by the meat factor, with those in the highest quartile having the greatest mean total daily energy intake.

View this table:
TABLE 2. Characteristics of 40 653 Australian men and women aged 40–69 y at baseline (1990–1994) who participated in the Melbourne Collaborative Cohort Study, according to dietary factor quartiles¹

 
Mortality
After a mean follow-up of 10.4 y, 697 CVD deaths, including 407 from IHD, had occurred (Table 3). In model 1 (without adjustment for diabetes, WHR, BMI, and hypertension) none of the factors showed significant associations with CVD or IHD mortality. After further adjustment for other covariates (model 2), the Mediterranean factor was inversely associated with both CVD and IHD death. The fresh fruit factor was weakly inversely associated with CVD death, although the linear trend was not significant.

View this table:
TABLE 3. Hazard ratios (95% CIs) for the association between dietary factor quartiles (Qs) and death from all cardiovascular diseases (total CVD) and ischemic heart disease (IHD), during mean follow-up of 10.4 y, for 40 653 Australian men and women aged 40–69 y¹

 
Tests of interaction between dietary factors and mortality for country of birth, sex, and history of CVD were not significant. However, history of diabetes modified the relation between IHD mortality and the Mediterranean and fresh fruit factors (P for interaction = 0.03 and 0.04, respectively). A borderline interaction was observed for history of diabetes and the vegetable factor (P interaction = 0.06). IHD data were then reanalyzed to investigate the association between these 3 factors and persons with diabetes. Analyses for the Mediterranean and vegetable factors gave notably different results among persons with diabetes than did the remainder of the cohort without diabetes. Compared with quartile 1, HRs for the Mediterranean factor were 0.42 (95% CI: 0.18, 0.97) for quartile 2, 0.40 (95% CI: 0.17, 0.97) for quartile 3, and 0.21 (95% C: 0.09, 0.47) for quartile 4. As a result of this finding, the overall association observed among persons without diabetes was slightly weakened for the Mediterranean factor when compared with the model controlling for the presence of diabetes (HR for the highest compared with the lowest quartile of intake: 0.71; 95% CI: 0.47, 1.08). For the vegetable factor, HRs for people with diabetes were 2.32 (95% CI: 1.02, 5.26) for quartile 2, 2.50 (95% CI: 1.07, 5.84) for quartile 3, and 1.74 (95% CI: 0.71, 4.27) for quartile 4 when compared with quartile 1. For the fresh fruit factor, no association for IHD was observed in either stratum of the diabetes variable.

Among participants without a baseline history of CVD (Table 4), 473 CVD deaths, including 242 from IHD, occurred since baseline. In model 1 (without diabetes, WHR, BMI, and hypertension) both the vegetable and fresh fruit factors were inversely associated with CVD mortality. After further adjustment for other covariates (model 2), these associations remained virtually identical, and the Mediterranean factor was inversely associated with both CVD and IHD death.

View this table:
TABLE 4. Hazard ratios (95% CIs) for the association between dietary factor quartiles (Qs) and death from all cardiovascular diseases (total CVD) and ischemic heart disease (IHD), among subjects without prior history of CVD, during a mean follow-up of 10.4 y for 38 204 Australian men and women aged 40–69 y¹

 

DISCUSSION  
We identified 4 main dietary factors with the use of factor analysis, which appeared to represent intakes of Mediterranean foods, vegetables, meat, and fresh fruit. After adjusting for established risk factors, the Mediterranean factor was inversely associated with CVD and IHD mortality in models adjusting for CVD history, diabetes, WHR, BMI, and hypertension. The Mediterranean, vegetable, and fresh fruit factors were inversely associated with CVD mortality among those without previous history of CVD. Frequent intake of meat was not associated with CVD or IHD mortality. Among persons with diabetes, the Mediterranean factor was strongly associated with reduced IHD mortality risk, whereas those with high scores for vegetable intake had increased risk.

This is a large prospective study with an average of 10.4 y follow-up and minimal loss to follow-up. The ethnically diverse cohort ensured heterogeneity of dietary intake and, importantly, resulted in 4 distinct dietary factors: Mediterranean foods, vegetables, meat, and fresh fruit. Although baseline measurement of established cardiovascular risk factors was thorough, unknown potential confounders may also have influenced our findings. Reliability of memory is recognized as the core problem in dietary measurement (22); however, other factors may be involved such as interpretation of portion size, integrating intake over time to fit frequency responses, and interpreting the FFQ items in relation to the foods actually eaten. Sex, culture, health status, education, and questionnaire design may also have an effect. Consequently, the level of exposure misclassification likely to have been associated with reliance on a FFQ will have tended to bias point estimates toward the null. Furthermore, our FFQ was designed to measure dietary intake for the 12 mo before baseline, which may not have been representative of intake over the causative relevant period. In addition, repeatability analyses for the FFQ showed only fair-to-moderate agreement for food items when completed 12 mo apart (17). Finally, misclassification of cause of death could have affected the associations observed for dietary factors.

Aside from dietary measurement error, the subjective nature of factor analysis itself may influence the importance of dietary factors produced (14). Of particular importance are the dietary items included for analysis, the number of factors chosen to be extracted, and the value of the factor loading chosen to describe the factor. Furthermore, dietary factors derived from factor analysis depend on the population being studied, making direct comparisons between studies difficult (4). Most studies focusing on dietary patterns and CVD have used only 2 factors for analysis, "prudent" and "Western," with the latter label apparently not encompassing components of the Mediterranean diet. These studies also pregrouped their FFQ items before doing factor analysis, which requires the researcher to make some judgment about how individual food items go together. We allowed the data to be grouped without making any assumptions about how items were to be combined.

In Denmark, a study analyzing 2994 men and 2877 women separately (10) found that a prudent diet (characterized by frequent intakes of whole-meal bread, porridge, grains, and oats; pasta; rice; raw vegetables; boiled vegetables; fruit; juices; jam and honey; cakes and cookies; candy and chocolate; milk and yogurt; fish; low-fat margarine; and tea) was associated with reduced CVD death in women but not men. A Western diet factor (characterized by frequent intakes of white bread, potatoes, juices, jam and honey, cakes and cookies, candy and chocolate, ice cream and soft drinks, milk and yogurt, eggs, meat, sausages, liver paste and cold meat, butter, lard, and margarine) was not associated with CVD mortality in either men or women. We chose not to do separate analyses for men and women because of power considerations and also because there was no evidence of effect modification by sex.

The Nurses' Health Study (11, 23) used factor analysis to identify 2 dietary factors. The prudent pattern (characterized by frequent intake of vegetables, fruit, fish, poultry, and whole grains) was associated with reduced risk of IHD and stroke. The Western pattern (characterized by frequent intake of refined grains, red and processed meat, desserts, sweets, and full-fat dairy products) was associated with increased risk of IHD and stroke. The Health Professional's Study of 44 875 US men produced 2 similar factors (12). Again, the prudent pattern was associated with reduced IHD risk, and the Western pattern was associated with increased risk.

Our Mediterranean factor, which was inversely associated with CVD and IHD mortality, was most similar to the prudent pattern described earlier. If these inverse associations indicate cause and effect, there are several candidate mechanistic hypotheses that could account for this. These include the role of antioxidants which are abundant in these foods. Phytochemicals such as bioflavonoids and carotenoids (a group of fat-soluble plant pigments such as -carotene, ß-carotene, and lycopene) and vitamins such as E and C were shown to be inversely associated with CVD (6). The null results from supplementation trials with antioxidant vitamins indicate that the health benefits of plant food intake cannot be attributed to individual micronutrients with antioxidant activity (24, 25). Many qualities of a diet rich in plant foods could contribute to the health benefits. For example, vegetable protein was shown to reduce serum cholesterol concentrations (26), and certain plant nutrients, including potassium, calcium, magnesium, and fiber, can reduce blood pressure (27–32). Folate, which is found in leafy green vegetables, citrus fruits, dried beans and peas, is important for lowering homocysteine concentrations which is a possible risk factor for IHD (33), although results from clinical trials testing folate have been null (34). Our Mediterranean factor was also characterized by the infrequent intake of certain foods as cream, sour cream, ice cream, chocolate, sausages, jams, honey, cake, and sweet biscuits. These foods are low in fiber and high in saturated fats, refined carbohydrates, and salt. Diets with these attributes were associated with increased cardiovascular risk (6). Therefore, the avoidance of these foods in the Mediterranean factor may partially explain the cardiovascular benefits observed in our study, especially among persons with diabetes.

We found no association between the frequent intake of meat and CVD or IHD mortality. This is similar to the Danish study (10) but unlike the 2 US studies (11, 12) in which the Western pattern was clearly associated with increased IHD mortality. If our results are correct, a possible explanation may be that cardioprotection is mediated primarily through plants. Diets high in meat products may therefore displace the consumption of more beneficial plant foods, thus reducing the cardioprotective potential of the overall diet (35).

We found that the associations were stronger for persons without a baseline history of CVD. Persons with previous disease are more likely to have modified their dietary habits as a consequence of illness and medical advice. For example, in our study, those with a history of CVD reported significantly lower intake of saturated fats than did the rest of the cohort, as well as higher intakes of fiber, polyunsaturated fats, and fish, and their reported diets might not be representative of those at the causative relevant time. It is also possible that other nondietary risk factors (not measured) operate after disease is established. However, if dietary risk factors for initial disease events are different from those for subsequent disease progression, previous disease would be an effect modifier, and this was not found in our study. However, tests for effect modification are often underpowered. We believe there is real plausibility to expect that history of CVD does modify the relation between diet and mortality in these analyses.

Our results suggest that the more frequent consumption of foods from the Mediterranean pattern may reduce CVD and IHD risks and may be most beneficial for persons with diabetes. Frequent intake of foods from the vegetable and fresh fruit patterns may reduce CVD risk for those without a prior history of CVD. In addition, frequent intake of meat does not increase risk after controlling for established risk factors. These findings are consistent with current recommendations to increase daily intakes of fruit and vegetables (6). Future research should explore the relation between these dietary factors and concentrations of biomarkers, including lipids, fatty acids, and inflammatory markers, to establish a better understanding of potential causal mechanisms.

ACKNOWLEDGMENTS  
This study was made possible by the contribution of many people, including the original investigators and the diligent team who recruited the participants and who continue working on follow-up. We thank the many thousands of Melbourne residents who continue to participate in the study.

The author's responsibilities were as follows—LRH: design of analysis, statistical analysis, interpretation of results, tables, writing of manuscript, and funding; DRE: funding, interpretation of results, and commented on manuscript drafts; JP: design of original Melbourne Collaborative Cohort Study(MCCS), interpretation of results, and commented on manuscript drafts; GGG: design and conduct of original MCCS and funding; AMT: funding, interpretation of results, and commented on manuscript drafts; AMH: design of analysis and commented on manuscript drafts; LB: design of analysis and commented on manuscript drafts; KO: design of original MCCS, funding, interpretation of results, and commented on manuscript drafts. None of the authors had a conflict of interest.

REFERENCES