Measurement error correction in nutritional epidemiology based on individual foods, with application to the relation of diet to breast cancer.

Nutrient intake is often measured with error by commonly used dietary instruments such as the food frequency questionnaire (FFQ) or 24-hour recall. More accurate assessments of true intake are obtained by using weighed diet records, in which subjects record what they eat on a real-time basis, but these records are expensive to administer. Validation studies are often performed to relate "gold standard" intake to intake according to surrogate instruments and to correct relative risk estimates obtained in the main study for measurement error. Most measurement error correction methods use validation study data at the nutrient level. However, subjects almost always report intake at the food rather than the nutrient level. In addition, the validity of measurement of different foods can vary considerably; it is relatively high for some foods (e.g., beverages) but relatively low for others (e.g., meats, vegetables). This differential validity could be incorporated into measurement error methods and potentially improve on nutrient-based measurement error methods. In this paper, the authors discuss correction methods for food-based measurement error and apply them to study the relation between FFQ intake in 1980 and incident breast cancer in 1980-1994 among approximately 89,000 women in the Nurses' Health Study, in whom approximately 3,000 incident breast cancers were observed.