Identification of Dietary Patterns Using Two Statistical Approaches and Their Association With Breast Cancer Risk: A Case-Control Study in Southern France

Abstract

The aim of this case-control study was to compare two different statistical methods in the identification of dietary patterns by use of principal component analysis (PCA) and variable clustering (VC) and to examine their association with the risk of breast cancer (BC). A dose-response association was then performed by the use of an adaptation of free knot spline function in logistic models. A "Western" pattern was revealed by PCA and VC and was then shown to be associated with a nonsignificant increase of BC risk. Only PCA identified a "meat/alcohol" pattern. Above the spline threshold, BC risk increased significantly (OR ≥ s vs. < s = 2.56, 95% CI 1.54-4.27). When we used PCA, a "Mediterranean" pattern was identified, but no association between BC risk and this pattern was shown. VC split the "Mediterranean" dietary pattern in two: "raw vegetables and olive oil" and "fish." Above the spline threshold, the "fish" pattern tended to protect against BC risk (OR ≥ s vs. < s = 0.77, 95% CI 0.58-1.01), whereas an excess of raw vegetables and olive oil increased BC risk (OR 1 se = 1.22, 95% CI = 1.06-1.32). Some results from the PCA and the VC methods were similar, whereas others were different but gave complementary results.