Assessment of past diet in epidemiologic studies.

The reproducibility of recall of diet was examined for 44 men in Toronto, Ontario, Canada, by comparing estimates of consumption obtained from a dietary questionnaire in 1982 with estimates of consumption made by recalling the original diet at an interview conducted one year later in 1983. Estimates of average consumption obtained by recall were significantly lower than those originally reported for most foods and nutrients, but the magnitude of the differences was never greater than 20% of the original estimate. Correlations between individuals' levels of consumption were greater than 0.7 for nine of the 13 foods and nutrients studied. Current diet, assessed from two-day food records, was also associated with consumption originally reported for some nutrients. Fecal levels of hemicellulose were associated with fiber consumption originally reported and with current fiber consumption, and urine levels of 3-methylhistidine were associated with past meat consumption. The best prediction of past consumption of fiber and fat, however, was obtained from the recalled diet. No significant additional contribution to the prediction was made from estimates of current consumption or from biochemical measures.