Dietary Pattern Analysis

The statistical engine can be exactly principal component analysis, but dietary pattern analysis is defined by its substantive setup, not the algebra. The input is a food-group intake matrix spanning the whole diet, constructed through deliberate nutritional grouping of foods and usually energy-adjusted, and the output factors are interpreted and labeled as eating patterns such as 'prudent' or 'Western' and then linked to disease. Generic PCA applied to arbitrary variables is not dietary pattern analysis; the method's identity lies in the food-group input, the nutritional interpretation, and the diet-outcome focus.

A posteriori patterns are derived empirically from the data using factor, cluster, or reduced-rank methods, letting the observed correlations among foods define the patterns — this is what dietary pattern analysis in the data-driven sense refers to. A priori patterns are defined in advance from nutritional knowledge as diet-quality indices, such as the Healthy Eating Index or a Mediterranean diet score, scoring each person against fixed criteria. Hu's review highlighted both as complementary: a posteriori patterns reveal how a population actually eats, while a priori indices test adherence to a hypothesized ideal.

Reduced-rank regression (RRR) is a hybrid that derives dietary patterns to explain as much variation as possible in a chosen set of intermediate response variables — typically nutrients or biomarkers thought to lie on a disease pathway — rather than maximizing variation in the foods themselves. This orients the patterns toward a specific health outcome, combining the data-driven flexibility of factor analysis with the hypothesis-driven focus of a priori indices. RRR is used when investigators have a clear mechanistic target and want patterns that are predictive of it rather than merely descriptive of eating behavior.