Reproducibility of a diet history questionnaire in a case-control study of breast cancer

The reproducibility of a diet history questionnaire was assessed as part of a case-control breast cancer study in Caucasian and Japanese women in Hawaii. The method was designed to estimate the intakes of total and saturated fat, cholesterol, and animal protein during a usual week before symptoms or diagnosis of disease in cases, and during a usual recent week in controls. Except for the time period of reference, the dietary technique was identical for all cases and controls. A subsample of 61 Japanese women (19 cases and 42 controls) and 56 Caucasian women (23 cases and 33 controls) participated in the dietary reinterview conducted within 3 months of the first interview. The mean difference in intake between the two interviews was small and not statistically significant (p greater than 0.05) for all four nutrient items among Japanese cases and controls and Caucasian cases, whereas the mean difference for all four nutrients was substantial and statistically significant (p less than 0.05) among the Caucasian controls. The intraclass correlation and the product-moment correlation were moderately high for all four nutrient items among Japanese cases and controls and Caucasian cases, but not among Caucasian controls. These findings indicate that the diet history method is reasonably reproducible in three of the four groups of studied subjects. Possible reasons for the poor reproducibility among Caucasian controls are discussed.