Abstract
AbstractThe continuing controversy concerning the relationships between diet and serum cholesterol levels highlights the need for innovative analytical approaches to the question. It is now acknowledged that dietary data are compositional in nature, but it is less widely recognised that the same is also true of any variable expressed in concentration units (mg/mL), such as serum cholesterol. Compositional data are parts of a whole and convey essentially relative information, which means they need to be interpreted in terms of the ratios between the individual components of the composition. The various formulations of log-ratio transformations proposed for compositional data analysis provide new variables that can be included in standard regression models as dependent and explanatory variables. Using data from an Italian population-based study, we describe the use of such methods to evaluate the relationships between a two-part composition (non-HDL and HDL cholesterol and their total) and three dietary nutrient compositions, and define multivariate linear regression equations that have one cholesterol log-ratio and the composition total as dependent variables and some macro- and micronutrient log-ratios as explanatory terms. Two alternative models are fitted: one containing the nutrient log-ratios in the form of their simplified expression as orthogonal balances; the other estimating the impact of nutrient pivot balances, which express the relative dominance of each of the parts of the dietary compositions. This approach to investigating the relationship between diet and serum cholesterol allows the simultaneous examination of the effects of non-redundant dietary components on both the quantitative and qualitative aspects of serum cholesterol profiles, and provides insights into some matters concerning public health.
Publisher
Cold Spring Harbor Laboratory