Dietary patterns and confounders
Most studies obviously try to account for the greatest possible number of confounders when assessing the role of a given nutrient or group of nutrients. Multivariate approaches represent the most widely used statistical tools in this respect. Another interesting and probably complementary way of addressing this issue is the use of clustering techniques, which may reveal that, despite being individually ‘‘independently’’ related to a given medical condition in multivariate analysis, some dietary habits belong to common patterns. This topic has been reviewed recently by DEVLIN et al. [28], who highlighted important methodological issues and reviewed dietary patterns identified in healthy population and several diseases. Altogether, two distinct patterns emerged almost constantly, i.e. the ‘‘healthy pattern’’ (consumption of
fruit and vegetables) and the ‘‘unhealthy’’, ‘‘Western’’ or ‘‘traditional’’ pattern (see later). These data underline that the ‘‘independence’’ of factors associated with a given condition should be taken as a relative notion, which may be highly influenced by assessment methods. Conversely, dietary patterns identified by clustering techniques are rather rough in that they may not capture very specific profiles and thus do not allow mechanisms of action to be explored.