Use of finite mixture models in dietary patterns analysis.

Background

In recent years, the dietary pattern approach has been used extensively to describe overall eating profiles in populations. In the literature, dietary patterns are often computed by cluster analysis and principal component analysis (PCA). However, PCA does not create distinct groups of individuals with different dietary habits; moreover the choice of the clustering method and of the number of clusters in cluster analysis remains difficult. On the other hand, finite mixture models (FMM) do not have those limitations and have many other advantages. However, they have been rarely used in dietary pattern analysis.

Objective
The objective of this study was to use FMM to compute dietary patterns based on data from the NESCaV survey (Nutrition, Environment and Cardiovascular Health), a large population-based study carried out between 2007 and 2011among the Greater Region population (N=2298 subjects).

Methods
A 134-food frequency questionnaire was used to assess dietary intakes. The most appropriate parameterization of the covariance matrix and number of clusters was chosen on the basis of the Bayesian information criterion (BIC).

Results

Four dietary patterns were determined. A ”non-prudent” and a “prudent” patterns were characterized respectively by non-healthy and healthy food choices. A “breakfast/low alcohol” pattern was characterized by high intakes of food items usually consumed at breakfast. Finally, a “vegetables/dairy products/low carbohydrate” pattern was characterized by low intakes of carbohydrates but high intakes of vegetables, pulses, fruits, animal protein and fat mostly from dairy products. The “non-prudent” pattern was the most prevalent with 34% of the population assigned to this cluster. The “prudent”, “breakfast/low alcohol” and “vegetables/dairy products/low carbohydrate” patterns accounted respectively for 25%, 29% and 19% of the population, respectively. Women, older people and non-smokers followed the “prudent” and “breakfast/low alcohol”, whereas the “non-prudent” and “vegetables/dairy products/low carbohydrate” were more adopted by men and smokers. In addition, the “non-prudent” pattern was associated with higher cardiovascular risk.

Conclusion

FMM should be considered more often as they do not have limitations encountered with other methods and are not restrictive on cluster geometry. Moreover, this study highlights the need for targeted health promotion campaigns focussing on the benefits of healthy dietary patterns.