Abstract
Background: Diet research focuses on the characteristics of "dietary patterns" regardless of the statistical methods used to derive them. However, the solutions to these methods are both conceptually and statistically different. Methods: We compared factor analysis (FA) and latent class analysis (LCA) methods to identify the dietary patterns of participants in the Chinese Wuxi Exposure and Breast Cancer Study, a population-based case-control study that included 818 patients and 935 healthy controls. We examined the association between dietary patterns and plasma lipid markers and the breast cancer risk. Results: Factor analysis grouped correlated food items into five factors, while LCA classified the subjects into four mutually exclusive classes. For FA, we found that the Prudent-factor was associated with a lower risk of breast cancer [4th vs. 1st quartile: odds ratio (OR) for 0.70, 95% CI = 0.52, 0.95], whereas the Picky-factor was associated with a higher risk (4th vs. 1st quartile: OR for 1.35, 95% CI = 1.00, 1.81). For LCA, using the Prudent-class as the reference, the Picky-class has a positive association with the risk of breast cancer (OR for 1.42, 95% CI = 1.06, 1.90). The multivariate-adjusted model containing all of the factors was better than that containing all of the classes in predicting HDL cholesterol (p = 0.04), triacylglycerols (p = 0.03), blood glucose (p = 0.04), apolipoprotein A1 (p = 0.02), and high-sensitivity C-reactive protein (p = 0.02), but was weaker than that in predicting the breast cancer risk (p = 0.03). Conclusion: Factor analysis is useful for understanding which foods are consumed in combination and for studying the associations with biomarkers, while LCA is useful for classifying individuals into mutually exclusive subgroups and compares the disease risk between the groups.