The reproducibility of a food frequency questionnaire among controls participating in a case-control study on cancer.

This study was designed to test the reproducibility of a food frequency questionnaire used in a population-based case-control study on diet and pancreatic cancer. Repeat questionnaires covering the same time period were obtained using 63 male and female population controls, 35-79 years of age. For selected food items included in the case-control study, the attenuation of the odds ratios due to random error was estimated. Using 54 male and female population controls, 35-79 years of age, we conducted a second study to examine the agreement between original and repeat interviews when the time interval between interview and the period of interest was constant. In the first study, the median correlation coefficient was 0.72 for foods (ranging from 0.36 and 0.59 for subgroups of vegetables to 0.96 for alcoholic beverages) and 0.77 for nutrients (ranging from 0.62 for beta-carotene to 0.85 for energy and 0.91 for ethanol). In the second study, the median correlation coefficient was 0.68 for foods (ranging from 0.28 for eggs to 0.87 for alcoholic beverages) and 0.75 for nutrients (ranging from 0.48 for beta-carotene to 0.76 for energy). We conclude that for most items the agreement between original and repeat estimates was moderate (r > 0.50) to high (r > 0.70). Moderate agreement was found for 28 of 33 food items (85%) and for all 21 nutrient items (100%) and high agreement for 19 of 33 of food items (56%) and 15 of 21 nutrient items (71%). In the second study, agreement was somewhat lower but closely paralleled the results of the first study. On average, random error presumably attenuated most of the observed diet-cancer relationships only moderately; i.e., an observed odds ratio of 1.5 and a correlation coefficient of 0.70 yield an unattenuated odds ratio of 2.1.