Application of a dietary pattern approach in the Belgium context

Dietary pattern analysis has recently emerged as a new direction and a complementary approach to study the relationship between diet and morbidity or mortality. At present, two methods have been developed to construct dietary patterns: "a priori" method and "a posteriori" method. This paper presents the two methods and their application based on dietary data from the "Belgium Interuniversity on Nutrition and Health Study". A prospective study was conducted (1979-1984) in a sample of 5,225 males and 4,476 females from the Belgian population aged 25 to 74 years at the initial survey and followed for 10 years for all causes and specific mortality. Dietary data was collected by a 24-hour recall and for a sub-sample also by a 7 day-diet record. The "a priori" method was used by calculating an index based on the national dietary guidelines. We used the principal component analysis to identify dietary patterns a posteriori. We conducted a first principal component analysis using the data from the 24-hour recall and a second on the data collected by the 7 day-record. Both of the currently used approaches for extracting dietary patterns have advantages and limitations. We applied first the "a priori" approach by calculating an Index (IAR) which measures the adherence of the sample to the Belgian dietary guidelines. We obtained an index that ranged from 0 to 8, a higher score represented a "healthier diet". The index mean (sd) was 3.7 (+/- 1.2) for the entire sample with a significantly higher IAR for women. Using factor analysis, we identified 8 dietary patterns for men and for women. These were difficult to translate in terms of dietary intake profile. Inversely, with the factors identified with the 7 day record, we could find a "western" dietary profile and a "prudent" profile. Dietary pattern analysis offers the opportunity to evaluate the overall quality of the diet. The dietary profiles constructed by the two approaches should be related to morbidity or mortality in order to evaluate their predictive capacity.