Identification of dietary patterns using partial least square regression

Partial Least Square Regression (PLS-R) is a method of reducing the dimensionality of the data. PLS-R assumes that there is a common structure underlying the blocks of predictor (X) and response (Y) variables, and that this structure can be resumed by a few latent components, explaining the Y, calculated as a linear combination of the X. The components are obtained as to explain the maximum covariance between the Y and X variables. PLS-R is an iterative algorithm, and in any iteration Y-scores, X-weights, X-scores and Y-weights are sequentially calculated one as a function of the previous one. We applied PLS-R to an Italian case-control study of esophageal cancer in order to identify dietary patterns from 33 food groups and 6 nutrients as X and Y variables, respectively. We performed a logistic regression on PLS-R scores and found a strong direct association between a dietary pattern, characterized by a diet poor in fruit and vegetables, and rich in alcohol, red meat, bread, butter, unspecified seed oils, and the risk of esophageal cancer (odds ratio: 5.64 for the highest quintile compared with the lowest, confidence interval: 2.54-12.54).