Are Complex Models In Nutritional Epidemiology Always Worth The Trouble?

Purpose
It has been repeatedly emphasized that diet could account for up to 40% among preventable causes of cancer, although the consensus around this estimate is not unanimous. Despite several decades of research, comparatively few nutrition-related factors have been established as playing a causal role in human cancer.
Methods
The evaluation of role of diet on the occurrence of cancer has entailed a number of methodological challenges. First, extensive focus was given to procedures designed to perform correction of risk parameters for random and systematic measurement errors in individuals’ dietary exposure estimates. Second, the evaluation of exposure/disease relationships in international multi-center study consortia motivated the need to exploit any level of etiological evidence, notably at the individual level (within-center) and at the aggregate level (between-center). Third, standard approaches have long focused on the relation between a limited list of foods or nutrients and the risk of cancer, which requires a relevant use of statistical assumptions when controlling for potential confounding by other dietary and lifestyle factors.
Results
Recognizing the multi-factorial nature of cancer and other chronic diseases, complementary holistic methodologies have been employed to address the notion of dietary patterns, a concept conceived to address the inherent inter-correlations between dietary variables. Strategies relying on a priori (evidence driven) or a posteriori (unsupervised or data driven) approaches have been proposed, thus contrasting analytical simplicity with computational sophistication. The merits and the pitfalls of each of the above points are illustrated and discussed.
Conclusions
In an effort to provide workable tools to understand the etiology and possibly prevent cancer and other chronic diseases, the day-to-day experience of nutritional epidemiologists is characterized by continuous concerns on the efficacy of cutting-edge statistical models to tackle biological complexity.