Consumption of meat, fish, dairy products, eggs and risk of ischemic heart disease: a prospective study of 7198 incident cases among 409,885 participants in the pan-European EPIC cohort

Background: There is uncertainty about the relevance of intake of animal foods to the etiology of ischemic heart disease (IHD). We examined the relationships of meat, fish, dairy products and eggs with risk for IHD in the pan-European EPIC cohort. Methods : A prospective study of 409,885 men and women in nine European countries. Dietary intakes were assessed using validated questionnaires, calibrated using 24-hour recalls. Lipids and blood pressure were measured in a subsample. During a mean 12.6 years follow up, 7198 participants had a myocardial infarction or died from IHD. The relationships of animal foods with risk were examined using Cox regression with adjustment for other animal foods and relevant covariates. Additional analyses excluded the initial 4 years of follow-up (to allow for possible reverse causation), and modelled substitutions of different animal foods for red and processed meat. Results: The hazard ratio (HR) for IHD was 1.19 (95% CI 1.06-1.33) for a 100 g/d increment in the intake of red and processed meat, and this remained significant after exclusion of the first 4 years of follow-up (HR 1.25 [1.09-1.42]). Risk was inversely associated with intakes of yogurt (HR 0.93 [0.89-0.98] per 100 g/d increment), cheese (HR 0.92 [0.86-0.98] per 30 g/d increment) and eggs (HR 0.93 [0.88-0.99] per 20 g/d increment); the associations with yogurt and eggs were attenuated and non-significant after excluding the initial four years of follow-up. Risk was not significantly associated with intakes of poultry, fish or milk. In analyses modelling dietary substitutions, replacement of 100 kcal/d from red and processed meat with 100 kcal/d from fatty fish, yogurt, cheese or eggs was associated with approximately 20% lower risk of IHD. Consumption of red and processed meat was positively associated with serum non-HDL cholesterol concentration and systolic blood pressure, and consumption of cheese was inversely associated with serum non-HDL cholesterol. Conclusions: The risk for IHD was moderately positively associated with consumption of red and processed meat, and modestly inversely associated with consumption of yogurt, cheese and eggs, although the associations with yogurt and eggs may be influenced by reverse causation bias. It is not clear whether the associations with red and processed meat and cheese reflect causality, but they were consistent with the associations of these foods with plasma non-HDL cholesterol, and for red and processed meat with systolic blood pressure, which could mediate such effects.


Introduction
Ischemic heart disease (IHD) is the commonest disease and cause of death in Europe. 1 The Lipids were measured in stored plasma samples as part of the EPIC-CVD case-cohort study, 23 which is nested within EPIC. 11 The sub-cohort was randomly selected from participants with 24 a stored blood sample, with selection stratified by the 23 EPIC recruitment centres. Details of 1 methods are in the Supplementary material. 2 3 Ascertainment and verification of cases of ischemic heart disease 4 5 The outcome was IHD, defined as the composite of first non-fatal myocardial infarction (MI: 6 ICD-10 I21) or death from IHD (ICD-10 I20-25). Incident non-fatal MIs were ascertained in 7 each EPIC centre using a combination of record linkage to morbidity or hospital registries, 8 and self-reports followed by confirmation with medical records. 11 Information on vital status 9 was collected from mortality registries at the regional or national level in most centres except 10 in Greece where vital status was ascertained by active follow-up of study participants and 11 next of kin. Centres in Denmark, Greece, Italy, Norway and Spain validated all suspected 12 cases of MI, whereas centres in France, the Netherlands, Sweden and the UK validated a 13 subset of the suspected cases to assess the accuracy of the overall ascertainment process. A 14 range of methods was used to confirm the diagnosis of IHD and included retrieving and 15 assessing medical records or hospital discharge notes, contact with medical professionals, unknown history of MI or stroke at baseline (n=11,308), 23 cases whose date of diagnosis 1 was after the end of follow-up for each centre, and 23 participants with no follow-up data. 2 These exclusions left a total of 490,311 participants, and further restricting the dataset to 3 EPIC centres with known values for all of the animal foods (which meant excluding 4 Heidelberg, Potsdam, Naples and Umeå) left a total of 409,885 participants, including 7198 5 incident cases of non-fatal MI (n=5392) or fatal IHD (n=1806). 6 7 Follow-up was measured from recruitment until the date of first non-fatal MI or fatal IHD 8 event, or censoring at the date of death from other causes, non-fatal non-MI IHD, the date at 9 which follow-up for IHD events was considered complete, or emigration or other loss to 10 follow-up (1.3%). Relative risks as hazard ratios (HRs) and their 95% confidence intervals 11 (95% CIs) were estimated using Cox regression models. All analyses were stratified by sex 12 and EPIC centre and adjusted for exact age at recruitment (continuous), smoking, self- 13 reported histories of diabetes, hypertension, and hyperlipidemia, physical activity, 14 employment status, level of education, BMI (these latter eight covariates were all categorical 15 variables, with 'unknown' categories added), current alcohol consumption (categorical), and 16 intakes of energy, fruit and vegetables, dietary fibre from cereals, and percent energy from 17 sugars (each continuous). In the main analyses of calibrated food intakes, the results for each 18 animal source food were also adjusted simultaneously for the other animal source foods. 19 20 Participants were divided into fifths of self-reported intake for each animal food based on the 21 recruitment questionnaire (for any foods with more than 20% zero values the categories were 22 approximate fifths), with the quintiles calculated for all included participants, and a trend test 23 performed by scoring the categorical fifths of intake 1 to 5 and treating this as a continuous 24 variable. To test for whether the data were compatible with a linear trend, we also fitted models with the fifths of intake treated as a categorical variable; there were no significant 1 improvements in fit when comparing the categorical intake model with the continuous (trend 2 test) intake model, suggesting that any associations between food intake and risk were 3 approximately linear. Then, to improve the comparability of dietary data across participating 4 centres and to correct for measurement error in relative risk estimates, the dietary data from 5 the subset of participants with 24-hour recalls were used to provide statistically calibrated 6 estimates of dietary intakes for all included participants. HRs were calculated for increments 7 in observed and calibrated intake of each food. Observed food intakes were calibrated using a 8 fixed-effect linear model in which centre and sex specific 24-hour recall data from an 8% 9 random sample of the cohort were regressed on the observed intakes, generating a calibrated 10 intake corresponding to each observed intake. 12 15 The sizes of the increments were chosen to 11 approximate the difference in mean 24 hour recall intake between participants in the lowest 12 and highest fifths of observed intake, and with reference to the increments used in previous 13 publications such the World Health Organization's review of the carcinogenicity of red and 14 processed meat. 16

16
Using the results from the mutually-adjusted risks model, the effects of substituting 100 17 kcal/d of each other animal food for 100 kcal/day of red and processed meat were estimated 18 from the ratios of the risk (as measured by the hazard ratio) for each food in turn and the risk 19 for red and processed meat. 17 For example, if P and R represent the hazard ratios per 100 20 kcal/day yogurt and per 100 kcal/day red and processed meat in the mutually-adjusted risks 21 model, the effect of substituting 100 kcal/day yogurt for 100 kcal/day red and processed meat 22 is estimated by the ratio P/R; the difference in covariance was used to estimate the 95% 23 confidence interval. 24 To examine whether the overall results might be influenced by reverse causality, we repeated 1 the analyses after excluding the first 4 years of follow-up (i.e. with follow-up for all 2 participants commencing 4 years after the date of recruitment). To examine whether 3 associations between the animal foods and IHD risk were consistent across sub-groups of 4 other risk factors, we also conducted separate analyses for subsets of sex, smoking status 5 (never, former and current), prior disease status (participants with or without a history of Southern Europe: Greece, Italy, Spain, Provence, SW France), and countries with partial 10 (France, Netherlands, Sweden, UK) or complete (Denmark, Greece, Italy, Norway, Spain) 11 validation of cases. Tests for heterogeneity of trend between sub-groups were obtained by 12 comparing the risk coefficients for each sub-group using inverse variance weighting, testing 13 for statistical significance using a chi-square test on k-1 degrees of freedom where k is the 14 number of sub-groups. 15 16 To examine whether dietary risk factors might act through major established physiological 17 IHD risk factors, we examined the associations of food intakes with non-HDL cholesterol and 18 systolic blood pressure, calculating mean levels of these biomarkers in each category of 19 animal food intake (using linear regression to estimate least-squares means), with adjustment 20 for age, sex and EPIC centre. 21 22 All analyses were performed using Stata version 15.1 (Stata Corporation, College Station, 23 TX, USA), and a P-value less than 0.05 was considered statistically significant.

3
After a mean follow-up of 12.6 years there were 7198 incident cases of MI or death from 4 IHD. Table 1 shows participant characteristics by sex for all cohort participants and also for 5 incident cases. On average, cases were 6-10 years older than average for the cohort, with 6 higher mean BMI and lower mean alcohol intake. Cases were more likely to smoke, be 7 inactive, unemployed, diabetic, have elevated blood pressure or proatherogenic lipids, lower 8 mean observed intakes of fruit and vegetables, and moderate differences in intakes of animal 9 foods.  Table 2 shows the HRs and 95% CIs for IHD in each fifth of observed intake of animal 12 foods, relative to the bottom fifth of intake, and P values for tests of trend based on the 13 observed intakes. HRs in the top fifth of intake compared with the bottom fifth of intake were  Figure 1 shows the associations of IHD risk with statistically calibrated increments in intake 20 of eight mutually-exclusive animal foods (including red and processed meat combined, but 21 not red meat and processed meat separately), with mutual adjustment of risks for the animal 22 foods (see Supplementary material online Table S1 for HRs for uncalibrated and calibrated 23 increments, without mutual adjustment). For red and processed meat combined, the HR (95% 24 CI) was 1.19 (1.06-1.33) for a 100 g/day increment in calibrated intake. The HRs for calibrated intakes of yogurt (100 g/d), cheese (30 g/d) and eggs (20 g/d) were 0.93 (0.89-1 0.98), 0.92 (0.86-0.98) and 0.93 (0.88-0.99), respectively. 2 3 In analyses excluding the first 4 years of follow-up the association of risk with intake of red 4 and processed meat was marginally stronger (HR per 100 g/day increment 1.25 (1.09-1.42), 5 P=0.001), whereas the associations with calibrated intakes of yogurt and eggs were attenuated 6 and neither these associations, nor the association with cheese, were statistically significant 7 ( Table 3).  Table 4 shows the HRs for modelled substitution of 100 kcal/day of calibrated intake of red 12 and processed meat by 100 kcal/d of each of the other animal foods. Fatty fish, yogurt, 13 cheese and eggs were associated with significantly lower risks for IHD than red and 14 processed meat (15% to 24% reductions in risk per 100 kcal substituted per day). 15 16 Sub-group analyses 17 18 In analyses subdivided by history of diabetes, previous hypertension or hyperlipidemia, there 19 was no appreciable heterogeneity in the associations of animal foods with IHD risk except for 20 white fish, but this was not significantly associated with risk in either sub-group (see 21 Supplementary material online, Table S2). In analyses subdivided by smoking status, there 22 was no appreciable heterogeneity in the associations of animal foods with IHD risk except for 23 yogurt, which was inversely associated with risk in current smokers but not in never smokers 24 or former smokers (Supplementary Table S3). In analyses subdivided by age, there was no appreciable heterogeneity in the associations of animal foods with IHD risk except for red 1 and processed meat, which was strongly positively associated with risk in participants 2 recruited before age 55, but not in older people (Supplementary Table S4). In analyses 3 subdivided by sex, there was no appreciable heterogeneity in the associations of animal foods 4 with IHD risk except for eggs, which were inversely associated with risk in men but not in 5 women (Supplementary Table S5). There was no appreciable heterogeneity in the  Table S8); for red and processed meat, there was a large 10 and highly significant association with risk in the countries with complete case verification, 11 but not in the other countries. For milk there was a small positive association with risk in the 12 countries with complete verification, but not in the other countries. 13 14 Associations of foods with plasma lipids and blood pressure 15 16 Comparing participants in the highest fifth of intake of red and processed meat with those in 17 the lowest fifth of such intake, non-HDL cholesterol was higher by 0.19 mmol/l (4.3%), and 18 systolic blood pressure was higher by 3.3 mm Hg (2.5%); for processed meat, the difference 19 in systolic blood pressure among these groups of participants was 3.7 mm Hg (2.8%). 20 Comparing participants in the highest fifth of intake of cheese with those in the lowest fifth of 21 such intake, non-HDL cholesterol was lower by 0.10 mmol/l, whereas the intake of cheese 22 was unrelated to systolic blood pressure (see Supplementary material online, Tables S9 and 23 S10). 1 2 In this large European cohort we observed a positive association between red and processed 3 meat intake and risk of IHD, with a 19% (95% CI 6%-33%) higher risk per 100 g/day 4 increment in calibrated intake. Red and processed meat showed separate (albeit borderline 5 significant) associations with risk, which were each of similar magnitude. The association of 6 risk with red and processed meat was observed after excluding the first 4 years of follow up 7 and in participants without diabetes, or elevated blood pressure or proatherogenic lipids. 8 These additional results, therefore, reduce the likelihood of reverse causation or residual 9 confounding. By comparison, a previous meta-analysis of meat and risk of IHD reported that 10 unprocessed red meat consumption was not associated with risk of IHD, whereas processed 11 meat was, with a 42% higher risk per 50 g/d increment in intake. 5 However, that previous 12 review included only 769 events from four studies for unprocessed red meat, including one 13 case-control study; for processed meat it included 21,308 events from five studies, but most 14 cases derived from one study for which the endpoint was total cardiovascular mortality rather 15 than incident MI and fatal IHD. A subsequent meta-analysis also concluded that processed 16 meat but not unprocessed red meat was associated with IHD mortality, based on up to 1370 17 deaths from IHD. 6 Hence, further work is needed to understand potential reasons for the 18 differences with the current study's results, which were based on over 7000 IHD events. 19 20 We observed no clear association of IHD risk with consumption of either white fish or fatty 21 fish (although there was a borderline significant inverse association for fatty fish, and a 22 significant inverse association for fatty fish in the substitution analyses; see below). The 23 possible protective role of fish in IHD has been investigated for more than 30 years. A 24 previous analysis of fish consumption and mortality in EPIC found no evidence that higher intakes of total, white or fatty fish were associated with mortality from IHD. 18 By contrast, a 1 meta-analysis of 4472 deaths in 17 cohort studies suggested that there was an overall 2 significant inverse association between fish intake and IHD mortality, but the association was 3 not linear and the relative risk in the highest category of fish intake was not significantly 4 lower than that in the lowest intake. 9

6
Dairy products are a major source of dietary saturated fatty acids, but prospective 7 observational studies have generally not shown a higher risk of IHD with a higher intake of 8 foods such as milk, yogurt and cheese. 19 20 We observed no association of milk with risk of 9 IHD, which is consistent with a meta-analysis of 4391 incident IHD cases in six prospective 10 studies. 21 We observed that yogurt consumption was inversely associated with risk of IHD. 11 However, this association did not persist after exclusion of the initial 4 years of follow-up, 12 and it showed heterogeneity by smoking status, with no association in never smokers 13 (suggesting, therefore, that the observed association may partly be explained by changes in 14 diet due to preclinical disease and/or residual confounding by smoking). Yogurt consumption 15 is associated with healthy dietary patterns, behaviors and lifestyle factors 22 , yet a meta- 16 analysis of 5 prospective studies (number of cases unclear) reported no association between 17 yogurt consumption and risk of IHD. 23 We also observed that cheese consumption was 18 inversely associated with risk of IHD; again, this inverse association was not significant after 19 excluding the first four years of follow-up, although the estimate was only slightly attenuated. 20 A meta-analysis of 8 prospective studies with 7425 incident cases showed a lower risk for 21 IHD in participants with a relatively high intake of cheese. 24 It has been suggested that cheese 22 has constituents which might act to reduce the risk of IHD, for example that the calcium in 23 cheese forms insoluble soaps with fatty acids thus reducing absorption of saturated fatty acids, and that the calcium also binds to bile acids, reducing their enterohepatic circulation 1 and possibly leading to a cholesterol lowering effect. 19  Egg consumption was inversely associated with IHD risk overall, but this association was no 4 longer evident after excluding the first 4 years of follow up, perhaps due to limited power, or 5 because people with preclinical disease may have reduced their egg consumption. A recent 6 meta-analysis of six prospective studies including 5847 incident cases reported no association 7 of egg consumption with risk of coronary heart disease. 8 , whereas a recent large prospective 8 study in China including 31,169 incident cases of IHD reported that egg intake was inversely 9 associated with risk; 26 it is possible that the risk associations found in the observational 10 studies due to the a dietary pattern often accompanying high egg intake and/or the cluster of 11 other risk factors in people with high egg consumption. 27

13
The positive association we observed between red and processed meat and risk of IHD might 14 be related to the saturated fat content of these foods. However, although dairy products are 15 also relatively rich in saturated fats, intake of dairy products was not positively related to IHD 16 risk in this study; in fact there was a suggestion of an inverse association between cheese 17 intake and future risk of IHD. This finding might suggest that different food sources of 18 saturated fat, and/or different proportions of individual saturated fatty acids contained within 19 meat and dairy foods, may differ in their impact on risk of IHD, which would affect the 20 interpretation of previous studies of total dietary saturated fatty acids and risk. 28 It is also 21 possible that plant sources of protein may be associated with a lower risk of IHD than animal 22 foods, 29 and this should be considered in future analyses. 1 2 Our analyses showed that red and processed meat were positively associated with risk for 3 IHD, whereas the other animal foods were not associated or inversely associated with risk. 4 We therefore conducted analyses modelling isocaloric dietary substitutions, which showed 5 that fatty fish, yogurt, cheese and eggs were associated with significantly lower risks for IHD 6 when substituting for red and processed meat (15% to 24% lower risk per 100 kcal 7 substituted per day). Plant foods might also be associated with a lower risk of cardiovascular 8 disease than animal foods 27 and may be considered in future analyses. 9 Possible roles of plasma lipids and blood pressure 11 12 The positive associations of red and processed meat and the inverse association of cheese 13 consumption with the risk of IHD might be explained through the associations of these foods 14 with well-established risk factors for IHD, such as cholesterol fractions and systolic blood 15 pressure. Compared to participants in the lowest fifth of intake of red and processed meat, 16 those in the top fifth had a higher non-HDL cholesterol by 0.19 mmol/l and a higher systolic 17 blood pressure by 3.3 mm Hg; the difference in systolic blood pressure was larger for 18 processed meat than for red meat (3.7 and 2.2 mm Hg, respectively), consistent with previous  Collaboration 31 32 , these differences would be expected to be associated with higher IHD risks 23 of 8% and 12%, respectively. Such modelling suggests that the observed (uncalibrated) 13% 24 higher risk in the top fifth of intake of red and processed meat could be readily explained by the differences in blood lipids and blood pressure. Other mechanisms might also be involved, 1 for example, higher intakes of red and processed meat might increase the risk of IHD through 2 the conversion of carnitine in meat into trimethylamine oxide. 33   Strengths and limitations 11 12 Strengths of this study are the large number of cases, the prospective design, the wide range 13 of diets across Europe, the calibration of the dietary data using 24-hour recalls, and the ability 14 to adjust for major risk factors for IHD and to estimate the impacts of associations with 15 circulating lipids and blood pressure. 16 17 As with all observational studies, a potential limitation is that the associations may be 18 influenced by confounding by other risk factors. We have adjusted our results for major risk 19 factors for IHD, including smoking and BMI as well as socio-economic factors. However, as 20 the magnitudes of the associations we observed were relatively modest, we cannot discount 21 that the results have been influenced by residual confounding by adiposity, socio-economic 22 factors or other unmeasured factors. Another potential limitation is that, due to the multi- 23 centre design of the cohort, there were some variations in the ascertainment and validation of 24 the endpoint; the positive association of red and processed meat with risk for IHD was strong in the countries with complete validation of cases. It is also possible that associations of 1 specific foods with risk may vary between populations due to differences in associations with 2 other aspects of diet. This large prospective study in Europe shows a moderate positive association between 7 consumption of red and processed meat and risk of IHD, and it suggests a modest inverse

Substitution of other animal foods for red and processed meat
8 association between consumption of cheese and IHD risk. It is not clear whether these 9 associations reflect causality, but they were consistent with the associations of these foods 10 with plasma non-HDL cholesterol, and for red and processed meat with systolic blood 11 pressure, which could mediate such effects.

Contributors
The study was conceived and designed by TJK, PNA, KEB, AB, ER, and JD. The data were analysed by PNA. The first draft of the manuscript was prepared by TJK, PNA and KEB, and edited with input from the writing team (IJ, TK, MS, EW, MW and AMLW). All other authors provided the data and revised the manuscript critically for important intellectual content. All authors gave final approval of the version to be published and have contributed to the manuscript. TJK is the guarantor.

Data sharing
For information on how to submit an application for gaining access to EPIC data and/or biospecimens, please follow the instructions at http://epic.iarc.fr/access/index.php     , current alcohol consumption (non-drinkers and sex-specific fifths of intake among drinkers), and calibrated intakes of energy, fruit and vegetables combined, sugars (as % energy), fibre from cereals, and each other food (each continuous), and stratified by sex and EPIC centre. # Tests of trend were performed using the calibrated intake (continuous).

Table 4
Hazard ratios* (95% confidence intervals) for first non-fatal MI or fatal IHD for substitution of 100 kcal/day increment in calibrated energy intake from each food for 100 kcal/day increment in calibrated energy intake from red and processed meat