Dietary variability and its impact on nutritional epidemiology

Dietary intakes are subject to measurement errors and to day-to-day variation, which have contributed to obscuring the suspected relation between dietary lipids and ischemic heart disease. The effect of measurement error on the correlation between dietary intakes and serum cholesterol levels has been studied by others. In this paper we study the effects of errors on the categorization of subjects according to the quantities of their intakes, and the effects of this misclassification on the observed relation between observed dietary intakes and disease. Our model is based on a bivariate normal joint distribution of true and observed intakes, from which various conditional probabilities can be calculated. Tables are given to simplify many of these computations. We conclude that the usual period of collection of dietary records, 1 week, is usually adequate. The model developed is applicable to any measurement recorded with error, and two examples of its application to the classification of subjects as normotensive or hypertensive are given. The model does depend on a large number of assumptions, some of which are clearly not met. Hence the actual numerical values obtained should be treated with some scepticism. If, however, the assumptions are approximately met, then the results should be reasonable approximations to the truth.