Categorisation of input variables for deriving dietary patterns.

We would like to congratulate Smith et al. for their recent paper on the relevant issue of input variable quantification when deriving dietary patterns by principal component analysis (PCA). Indeed, previous studies have compared different methods of deriving empirical patterns such as clusterv. factor-based analysis, while some authors have compared exploratory factor methods, such as PCA, with confirmatory methods, such as factor analysis. But as Smith et al. make clear, few studies have specifically dealt with the issue of how the choice of input variables influences the derived dietary patterns. In their paper, Smith et al. compared dietary patterns derived from PCA of four types of input variables, of which three were interval variables (weight in g/d, mean weight for total energy intake and percentage contribution to total energy intake), the fourth one being binary yes/no variables, coding whether the food groups had been consumed or not. From our understanding, their main conclusion was that there were no obvious differences between the patterns derived using any of the three interval variables, while the patterns derived using the binary variable appear somewhat different, reflecting more general food preferences. This result seems meaningful, but we would like to make a few comments that we think could be useful regarding this matter of input variable quantification for deriving dietary patterns.