MD has received much attention as a diet associated with lower risk of coronary heart disease (Keys, 1980). Several reports, however, have indicated that this may extend to several forms of cancer (Trichopoulou et al, 2000b, 2003; Bosetti et al, 2003; Lagiou et al, 2006; Mitrou et al, 2007). We have examined the degree of adherence to traditional MD in relation to incidence of cancer overall in a prospective cohort investigation in Greece. Two advantages are relevant here: the cohort is based on the general population and large segments of it still adhere to the traditional MD (The Data Food Networking, 2008).
A total of 28 572 participants from all over Greece were recruited in 1994-99 in the Greek component of European Prospective Investigation into Cancer and nutrition (EPIC), the prospective cohort study investigating the role of biologic, dietary, lifestyle and environmental factors in the aetiology of cancer and other chronic diseases. It is conducted in 23 research centres in 10 European countries (Riboli et al, 2002). For Greece, the study protocol was approved by the ethics committees of the International Agency for Research on Cancer (IARC) and the University of Athens Medical School. All participants provided written informed consent and all procedures were in accordance with the Helsinki Declaration.
A validated, semi-quantitative, food-frequency questionnaire including 150 foods and beverages, as well as several complex recipes, was used to assess usual intake during the year preceding enrolment (Gnardellis et al, 1995; Katsouyanni et al, 1997). The questionnaire was administered in person by trained interviewers. For each item, participants were asked to report their frequency of consumption and portion size with the help of 76 photographs of portion sizes from which quantities consumed in grams per day were estimated. Nutrient and total energy intakes were calculated for each participant using a food-composition database modified to accommodate the particularities of the Greek diet (Trichopoulou and Georga, 2004). For this analysis, 11 all-inclusive foods groups as well as selected nutrients were considered: vegetables, legumes, fruits and nuts, dairy products, cereals, meat and meat products, fish and shellfish, potatoes, eggs, confectionery and non-alcoholic beverages, as well as ethanol, monounsaturated (mostly from olive oil), polyunsaturated and saturated lipids.
A 10-point scale was used to assess the degree of adherence to the traditional MD (Trichopoulou et al, 2003). A value of 0 or 1 was assigned for each of the nine components indicated below, using the sex-specific median as the cutoff value. For the presumed beneficial components (vegetables, legumes, fruits and nuts, cereals, and fish), participants were assigned a value of 0 for consumption below the median and of 1 for consumption at or above the median. For the presumed detrimental components (meat and meat products, and dairy products) the value of 0 was assigned for consumption at or above the median, while a value of 1 was assigned for below the median. For ethanol a value of 1 was assigned to men consuming 10 g to less than 50 g per day and to women consuming 5 g to less than 25 g per day. For lipid intake, a value of 1 was assigned to individuals with a monounsaturated to saturated lipids ratio at or above the median. Thus, the final MD score ranged from 0 (minimal adherence) to 9 (maximal).
A number of socio-demographic and lifestyle characteristics, including smoking, physical activity and use of dietary supplements were recorded at enrolment. Anthropometric measurements were also undertaken using standardised procedures and allowing for the calculation of body mass index (in kg/m). Finally, an overall metabolic equivalent (MET)-hour per day index, expressing the average daily energy expenditure level, was calculated from the frequency and duration of participation in occupational and leisure-time physical activities (Ainsworth et al, 1993; Trichopoulou et al, 2000a. 2003).
From the original cohort of 28 572 participants, 1696 (5.9%) were excluded because follow-up information was not available, 383 participants (1.3%) due to prevalent cancer at enrolment (excluding non-melanoma skin cancer) and another 870 (3.0%) due to missing values in one or more analysed variables. The final study population consisted of 25 623 individuals.
Active follow-up methods were implemented by specially trained health professionals through telephone interviews with the participants or, in case of death, their next of kin. A structured form was completed for each incident cancer and the diagnosis was verified through pathology reports, medical records, discharge diagnoses or death certificates, according to IARC guidelines for end point data in EPIC. Cancers were classified according to the International Classification of Diseases for Oncology (ICD-O-2) (World Health Organization, 1990). The outcome studied was the first cancer diagnosis (excluding non-melanoma skin cancer), or death from cancer whenever diagnosis was not reported.
Frequency distributions were used for descriptive purposes. Means and s.d. were used for dietary variables, including energy intake. Cox proportional hazards regression, with time to event as the primary time variable, was used to assess relationships with, alternatively, each dietary variables (per increment of intake calculated as a round number close to the s.d. of the relevant daily intake) or the MD score. The date of first diagnosis of cancer (or of death from incident cancer when diagnosis date was not reported) was used to calculate time to event. All models were adjusted for age, years of schooling, smoking status, body mass index, height, physical activity expressed in MET-hours per day, ethanol intake, supplement intake and total energy intake. In analyses of MD score and cancer, consumption of potatoes, eggs, confectionery and nonalcoholic beverages (which are not components of the score) were also controlled for (continuously), whereas ethanol (which is a component of the score) was not. Stratification by sex was also performed (except for the sex-specific models). In the Cox models, the assumption of proportionality (assessed through the Schoenfeld goodness of fit test) was met, no time-dependent covariates were used and there was no colinearity among the variables. All analyses were performed using STATA 8.0 statistical package (Stata Corporation, 2003).
