Consumption of fried foods and risk of coronary heart disease: Spanish cohort of the European Prospective Investigation into Cancer and Nutrition study

Objective To assess the association between consumption of fried foods and risk of coronary heart disease.

Design Prospective cohort study.

Setting Spanish cohort of the European Prospective Investigation into Cancer and Nutrition.

Participants 40 757 adults aged 29-69 and free of coronary heart disease at baseline (1992-6), followed up until 2004.

Main outcome measures Coronary heart disease events and vital status identified by record linkage with hospital discharge registers, population based registers of myocardial infarction, and mortality registers.

Results During a median follow-up of 11 years, 606 coronary heart disease events and 1135 deaths from all causes occurred. Compared with being in the first (lowest) quarter of fried food consumption, the multivariate hazard ratio of coronary heart disease in the second quarter was 1.15 (95% confidence interval 0.91 to 1.45), in the third quarter was 1.07 (0.83 to 1.38), and in the fourth quarter was 1.08 (0.82 to 1.43; P for trend 0.74). The results did not vary between those who used olive oil for frying and those who used sunflower oil. Likewise, no association was observed between fried food consumption and all cause mortality: multivariate hazard ratio for the highest versus the lowest quarter of fried food consumption was 0.93 (95% confidence interval 0.77 to 1.14; P for trend 0.98).

Conclusion In Spain, a Mediterranean country where olive or sunflower oil is used for frying, the consumption of fried foods was not associated with coronary heart disease or with all cause mortality.

Frying is one of the most commonly used methods for cooking in Western countries. When food is fried its nutritional content changes—the food loses water and takes up fat, increasing its energy density.1 2 Frying modifies both the foods and the frying medium because oils deteriorate during frying, especially when reused, through the processes of oxidation and hydrogenation, leading to a loss of unsaturated fats and an increase in trans fats.1 Thus fried food absorbs degradation products of the frying oil.3 At the same time, frying may also improve palatability by making food crunchy.4 5

Fried foods have been associated with various cardiovascular risk factors in cross sectional studies. In Spain, the Pizarra study of 1226 adults showed that consumption of food fried with reused oils was associated with a higher prevalence of arterial hypertension.6 Furthermore, a cross sectional analysis of the Spanish cohort of the European Prospective Investigation into Cancer and Nutrition (EPIC) study concluded that consumption of fried foods was positively associated with general and central obesity.7 Similarly, in the SUN (Seguimiento University of Navarra) cohort in Spain, more frequent consumption of fried foods at baseline was associated with a higher risk of becoming overweight or obese during a six year follow-up.8 Lastly, in another cross sectional study in 2090 Italian adults, consumption of fried food was associated with lower high density lipoprotein cholesterol levels and a larger waist circumference.9

Only a few studies have evaluated the effect of fried foods on risk of cardiovascular disease. A case-control study in Costa Rica found no association between consumption of fried foods and risk of non-fatal acute myocardial infarction.10 Another case-control study, INTERHEART, observed a positive association between fried foods and acute myocardial infarction.11

Therefore, although interest in the associations of dietary patterns and specific cooking techniques with risk of disease is increasing,12 little is known about the health effects of fried foods. We evaluated prospectively the association between consumption of fried foods and the risk of coronary heart disease in the EPIC-Spain cohort.

The methods of the EPIC project have been reported elsewhere.7 13 14 15 16 For the present analysis we used the data from the Spanish cohort of EPIC, which included 41 438 healthy adults (15 632 men), aged 29-69. Study participants were recruited between 1992 and 1996 in five Spanish regions. Three of these regions are located in the north (Asturias, Gipuzkoa, and Navarra) and two in the south, including the Mediterranean shore (Granada and Murcia). Selection of these regions aimed to maximise variability in diet because food and nutrient intake has traditionally been different between the north and the south of Spain and between the inner regions and coastal areas. To increase variability in diet, participants were also recruited in urban and rural areas from various educational and social sectors. Study participants included mostly blood donors, civil servants, and the general population. Participation rates varied from 55% to 60% between regions. The exposures of interest were assessed at baseline, and participants were followed up until 31 December 2004. All participants gave written informed consent before enrolment.

Using a computerised dietary history questionnaire,17 previously validated in Spain18 19 20 and administered by trained interviewers, we obtained data on usual food consumption for each participant. We used household measures, standard units, and a collection of 35 sets of photographs of simple foods, mixed foods, and drinks to quantify the portion of each food (g/day) consumed. Each interview on dietary history lasted about 40-50 minutes.

Participants were asked about food consumed in a typical week during the previous 12 months. Foods consumed at least twice a month were recorded. The dietary history questionnaire recorded information on up to 662 different foods, including those consumed individually and those consumed as part of 203 recipes and regional dishes. Of these foods, 212 were fried. We calculated energy and nutrient intake using the EPIC food composition table.21

Information on cooking methods was collected as part of the dietary history questionnaire. The same food may be cooked by several methods as part of a single recipe. For example, in the preparation of paella (a typical Spanish recipe), rice is both fried and boiled. Thus, fried food was defined as food for which frying was the only cooking method used. Such food could be deep fried or pan fried and it could be battered, crumbed, or sautéed. To estimate the absorption of fat from frying, we used an absorption coefficient for each fried food.18 We also recorded the type of oil used for dressing, cooking, and frying.

Information on demographic variables, educational level, smoking, and physical activity was obtained through interview at recruitment. Physical activity was assessed with a validated questionnaire that included activity at work, at home, and during leisure time.22 Participants were also asked if they had diabetes mellitus, hypercholesterolaemia, hypertension, cancer, or angina, or had experienced myocardial infarction or stroke. In addition they were asked to report their menopausal status and any use of oral contraceptives or hormone replacement therapy. Trained observers used standardised procedures to measure weight, height, and waist circumference. The body mass index was calculated as weight (kg) divided by height squared (m2).

At baseline, prevalent coronary heart disease was self reported. During follow-up, we ascertained incident coronary heart disease with a telephone questionnaire (at three years after recruitment) and through record linkage with three sources of information: hospital discharge databases, population based myocardial infarction registries (available in Murcia, Navarra, and Gipuzkoa), and regional mortality registries and the national mortality database (managed by the National Statistical Institute, Madrid, Spain), which provided information on date and cause of death.

Coronary heart disease was classified according to the International Classification of Diseases, ninth revision (ICD-9 codes 410-414) and 10th revision (ICD-10 codes I20-I25). A team of trained doctors and nurses validated coronary heart disease events against hospital records and autopsy reports.

Coronary heart disease events were classified on the basis of symptoms, signs, biomarkers, and findings on electrocardiogram or at autopsy, according to standard criteria.23 A definite coronary heart disease event was defined as one meeting all relevant criteria for a fatal or non-fatal acute myocardial infarction, or angina requiring revascularisation (coronary artery bypass graft or percutaneous transluminal coronary angioplasty). We defined probable and possible acute coronary heart disease events as cases where electrocardiogram findings were non-specific or the biomarkers were equivocal or missing.13

Of the 41 438 participants in the cohort, we excluded the following from the analyses: 193 for having coronary heart disease at baseline; 167 for an implausibly high or low dietary consumption, defined as 3 standard deviations either way from the cohort mean (5710 kcal/day); and 321 for lacking data on important variables such as date of coronary heart disease event (n=12), smoking (n=22), diabetes mellitus (n=71), hypercholesterolaemia (n=197), and hypertension (n=60). Thus the analyses were carried out on 40 757 participants.

We used Cox regression to obtain hazard ratios for coronary heart disease according to sex specific quarters of fried food consumption. The quarter (first) with lowest consumption was used as reference. In the Cox models, age was the underlying time variable, with entry time defined as the participant’s age at recruitment and exit time as the age at the coronary heart disease event, death, or 31 December 2004, whichever came first. To reduce violations of the proportional hazards assumption we also stratified models by age at recruitment (five year groups), sex, and centre. We built three models, with progressive adjustment for potential confounders to determine their influence on the results. Model 1 was adjusted for energy intake and model 2 was additionally adjusted for educational level, smoking, physical activity (at work, at home, and during leisure time), diabetes mellitus, hyperlipidaemia, cancer, oral contraceptive use, menopause, hormone replacement therapy, ethanol intake, and consumption of non-fried foods (vegetables, fruits, nuts, dairy products, meat, and fish). Model 3 was additionally adjusted for possible mediators of the study association reported in the literature: body mass index, waist circumference, and hypertension. We categorised continuous variables into quarters, except for age, which was analysed as a continuous variable. The P for linear trend was calculated by modeling the quarters for fried food as a continuous variable. Moreover, we also calculated hazard ratios of coronary heart disease and of all cause mortality for a 100 g increase in fried food consumption analysed as a continuous variable.

New analyses were done after censoring the first two years of follow-up and after excluding those who reported a change in their diet during the previous year. To test whether the study association varied with type of oil used for frying (olive oil versus sunflower oil or other vegetable oils), we used the likelihood ratio test, which compared models with and without interaction terms (products of the quarters for fried food and type of oil). We also used the same procedure to test whether results varied by sex.

Statistical significance was set at P