Commentary: An OPEN assessment of dietary measurement errors

epidemiological studies of chronic disease risk in relation to diet, a crucial question is whether assessments of dietary intake can accurately characterize an individual's habitual intake of foods and nutrients. Over the last two decades, this question has been addressed in numerous validation studies. The OPEN study 1 provides the best answer yet. It is a true landmark study both because of its size and the thoroughness of its design. The main innovation of the OPEN study is the use of doubly labelled water (i.e. water made from stable isotopes of both hydrogen and oxygen) to measure energy expenditure. Doubly labelled water is extremely expensive to produce. It is therefore unlikely that the OPEN study will be replicated in the foreseeable future. With this lack of reproducibility in mind, we would like to consider the extent to which the conclusions of the OPEN study can be generalized. Before considering the conclusions of OPEN, it is worth reviewing the evolution of dietary validation studies. Initially, dietary measurement validation studies were based on a comparison of two assessment methods, one of which (often based on a series of weighed food consumption records) was assumed to provide a perfectly valid intake measurement. This strong validity assumption was later relaxed by the use of statistical models for measurement error. These models impose certain constraints on the design of validation studies, where the nature of the constraint depends on the purpose of the study. If the aim is to correct for the 'attenuation' effect of measurement error on estimates of disease risk then two independent measurements are required, one of which must be unbiased. If the aim is to completely characterize the error properties of the dietary assessment methods then three independent measurements are required. 2,3 The difficulty is in finding three independent estimates of dietary intake. In practice, three main categories of dietary assessment can be distinguished: questionnaires for assessment of habitual, long-term intake; methods based on recording of actual food intake on one or more days (e.g. weighed food records, 24-hour diet recall interviews), and biomarkers of diet. It is clear that the measurement errors of instruments in the first two categories are correlated, not least because the same food tables must be used when converting foods to nutrients. This leaves only certain biomarkers as a possible alternative. One set of biomarkers of particular interest consists of those based on the urinary recovery …