Controlling for total energy intake in regression models for assessing macronutrient effects on disease

This paper illustrates the relationships between the (i) the standard multivariate method, (ii) the energy-adjusted method and (iii) the energy decomposition method for controlling for total energy intake. The setting is a general regression model which includes linear, logistic and Cox regressions. The discussion is confined to effects of macronutrients and to situations in which total energy intake is associated with disease. Pike, Bernstein & Peters [(1989) Am. J. Epidemiol. 129, 1312-1313] show how the three methods relate to each other and that the choice between them is a matter of taste. Here the relationships are illustrated using Venn-type diagrams. However, the rationale for all current controlling procedures relies on the assumption of external factors acting uniformly over different sources of energy. This assumption should be challenged. Furthermore, if individual macronutrients have independent, possibly even opposite effects on the risk for disease, then adjustment of any one of them will be relative to the effect of the others. This might not be satisfactory. Current methodology does not resolve these problems, and more elaborate data as well as further thought is needed.