Measurement errors and dietary patterns in nutritional epidemiology. (Doctoral thesis)

Nutritional epidemiology is a science that aims to measure dietary exposure as well as assessing diet-disease relationships. In the last decades, the Food Frequency Questionnaire (FFQ) was the tool of choice for measuring self-reported dietary intakes of individuals in large-scale epidemiological studies. However, as with other dietary assessment methods, measuring usual individual’s diet with an FFQ is prone to measurement errors, which lead to a loss of statistical power as well as biased estimates. As a consequence, validation studies are often used to assess the degree of measurement errors and their potential effects on estimates.
Concerning modeling diet-disease relationships, the simple way of analyzing the effect of individual nutrient or foods has been complemented by the use of dietary patterns that consider how foods are consumed together. Dietary pattern analysis has many advantages and is believed to represent more closely the real world. Several statistical methods are available to compute dietary patterns. For example, dimension reduction techniques are meant to summarize individual’s dietary intakes by fewer new continuous variables, each representing different dietary patterns. These techniques are Principal Component Analysis (PCA) and Reduced Rank Regression (RRR). PCA aims to compute dietary patterns representing dietary behaviors adopted in a population whereas RRR computes dietary patterns that represent the most beneficial or detrimental patterns relative to health outcomes. On the other hand, cluster analysis and mixture models are two alternative methods which create groups of individuals in a way that individuals in a same group have similar dietary habits, whereas individuals in different groups tend to have different dietary habits. Nevertheless, all methods are not perfect and have several limitations.
The main objective of this doctoral thesis was to suggest methodological improvements of existing methods for computing dietary patterns. The secondary aim was to use these methods along with suggested ameliorations to describe dietary patterns of individuals living in the Greater Region, namely the Grand Duchy of Luxembourg, Wallonia (Belgium) and Lorraine (France). Associations with several characteristics, namely cardiovascular risk factors, socio-demographics as well as lifestyle factors, were also explored. For that purpose, we used data from the NESCaV survey (Nutrition, Environment and Cardiovascular Health), an interregional cross-sectional study based on individuals living in the Greater Region. However, since measurement of dietary intake was done with an FFQ, the present thesis also aimed to assess the validity of dietary exposure estimates obtained with the FFQ.
This thesis is structured in three major parts, including an introduction, five chapters each representing a scientific article, and a discussion.