Methods

SUBJECTS

The study is based on data from the Danish WHO MONICA surveys, previously described in detail.5 Briefly, random equal sized samples of 30 year, 40 year, 50 year, and 60 year old men and women living in the south western part of Copenhagen County, were drawn from the National Central Person Registry and invited to participate in surveys conducted in 1982–84 (DAN-MONICA I), 1987 (DAN MONICA II), and 1991–92 (DAN-MONICA III). The MONICA III survey also included men and women aged 70 years. The response rates were 79%, 75%, and 73%, respectively, and the total population comprised 3698 men (1940 +748 + 1010) and 3618 women (1845+756+1017) (table 1). After 10 years, 4130 subjects from the MONICA I sample, were still alive and lived in Denmark. They were invited to a follow up examination. Of these 2254 men and women attended both examinations.

All cohort members were followed up until 31 October 1998 for all cause mortality obtained from official death certificates at the National Board of Health.

DATA COLLECTION

Data on well established biological and behavioural cardiovascular risk factors were collected by clinical examinations and questionnaires, respectively. Comparable survey methods, which followed the WHO-MONICA protocol, were used in all three surveys. These methods are described in earlier reports.5 6 Data on the biological variables, smoking, drinking, and physical activity, were missing from less than 1% of the participants, whereas 19% of the participants had missing data for one or more of the variables in the questionnaire on dietary habits. Subjects with incomplete data were more often women, lower educated and from the MONICA II survey. All analyses were conducted on the subgroup of 2855 men and 2755 women, with complete information on all variables.

VARIABLES

In all three studies, an extensive questionnaire on sociodemographic variables, lifestyle, and health was completed before a general health examination. The subjects were questioned about theirpresent state of health and asked to describe their health as: (1) very good, (2) good, (3) average, (4) bad, or (5) very bad. Average, bad or very bad self rated health was considered to be poor health. The questionnaire also included a question on food intake in which respondents were asked: “How often do you, on average, eat the following food items ?” followed by a list with the 26 items specified in table 2, which gives the exact answer categories. The alternatives used in the frequency scale were as follows: never, once a month or less, twice a month, once a week, 2 to 3 times a week, once a day, 2 to 3 times a day, and 4 times or more daily. Participants were also asked about their average daily number of cups of tea or coffee. The food frequency questionnaire has been compared with a diet history interview as a part of this study.6 This cross comparison showed that the mean food intake measured by a diet history interview increased with increasing frequency category of the food frequency questionnaire indicating that this questionnaire was able to identify levels of food intake correctly. Data were aggregated using principal component analysis (factor analysis) described in detail elsewhere.2The frequency of intakes of the 28 food items aggregated in two factors (components) with eigenvalues over 2. The first component (termed “prudent food intake pattern”) that positively associated with frequent intakes of wholemeal breads (and inversely with other types ), pasta, rice, oatmeal products, fruit, vegetables, and fish (table 2). The second component (termed “western food intake pattern”) was positively associated with frequent intakes of meat, sausages, potatoes, butter and white bread (data not shown). The prudent food intake pattern was inversely associated with all cause mortality, while the western food intake pattern was not associated with mortality (2). The following known predictors of mortality were chosen as covariates for the analyses: Smoking: Questions on smoking concerned current habits (yes, occasionally, no/never), and the kinds and average daily quantities of tobacco consumed (cigarettes, cheroots, cigars, pipe). In the analyses of the data, only subjects who smoked daily were considered smokers. Participants were also asked about their weekly intake of beer (bottles), wine (glasses), or spirits (units). Alcohol consumption was assessed by the average number of beers, glasses of wine and units of spirits having been consumed per week during the last year. These data were classified according to weekly alcohol intake: <1 drink per week, 1–14 drinks per week for women and 1–21 drinks per week for men, and a consumption of more than 14 or 21 drinks per week, respectively. Data on leisure time physical activity were based on answers to a question for which the participants had to mark one of four alternatives: A, mostly sedentary; B, walking, bicycling or otherwise active at a corresponding level at least four hours per week; C, going jogging or demanding sports, or doing heavy activity during leisure for at least three hours per week; D, long distance running or competitive sports several times per week. We defined people in group A as being inactive the rest were considered active.Educational attainment was assessed with questions about whether participants had any vocational education. Anthropometric measures were collected by the same trained nurse in all studies. Body mass index (BMI) was calculated as weight in kilograms divided, by the square of height in metres and divided to four categories: under weight (BMI<20), normal weight (BMI 20–24.9), overweight (BMI 25–29.9) and obesity (BMI ⩾ 30).

STATISTICAL ANALYSIS

Bivariate associations between the prudent food intake pattern, self rated health and other covariates were evaluated by logistic regression. Association between risk factors and mortality was analysed using Cox's proportional hazards regression models with age as underlying time scale and delayed entry accordingly. A Cox model was developed that included the prudent dietary pattern (in quartiles), smoking status (0=non-smoker,1=smoker) , physical activity (0=active, 1=inactive), vocational education (0=yes, 1=no), BMI (in four categories), weekly alcohol intake (<1 drinks, 1–14 drinks for women and 1–21 drinks for men and over 14 or 21 drinks, respectively ), and self rated health (in five categories) as predictors of death. In the model we analysed for potential interactions between the prudent dietary pattern and self rated health. The proportional hazards assumption was evaluated for all variables by comparing estimated log-log surviour curves over the different categories of the variables being investigated and by a test based on the generalisation of Grambsch and Therneau.7 This showed that the assumption held for the models. Tests for the individual covariates indicated that the assumption was not valid for physical activity in women. Thus, the risk estimates must be interpreted with caution for this variable. The initial data analyses showed that self rated health could be collapsed into three categories: very good and good, average, bad and very bad, because the risks were very similar in the adjacent categories. Furthermore, moderate alcohol consumption was associated with good self rated health (data not shown), but alcohol did not predict mortality and was excluded from the analyses. In the MONICA I population the association between risk factors (including age) and impaired or improved self rated health were also analysed using multiple logistic regression. Impaired self rated health was defined as those with very good or good health at baseline, who reported their health as average, bad or very bad at the re-examination. Improved health was defined as those with average or bad health at baseline, who reported of very good or good health at the re-examination.