Observations on the reliability and validity of the design and diet history method in the Melbourne Colorectal Cancer Study

The paper is a description of, and a practical guide to, the questions, problems, and pitfalls that can arise in the planning and execution of a large-scale epidemiological study of diet and disease; this study was an eight-year project in which diet histories were collected from over 2,000 people. Qualified dietitians obtained a quantitative diet history, which covered all foods eaten in Australia, and assessed as the most representative of the previous 20 years. Because over half of both cases and controls changed their diets in the previous 20 years, the period of recall nominated varied, but this was not a major bias. There was no effect of seasonality on recall. The average duration of the interview was one hour for both cases and controls. The response rate was 84% for cases and 94% for controls, suggesting that the time for interview of the diet history method is not a deterrent for a high response rate. In general, between-interviewer variation was small, although the intakes of total vegetables, cruciferous vegetables, and leafy green vegetables were the least repeatable. Between-interviewer differences were not found to affect the relative risks. Reproducibility of placing a dietary factor into a particular quintile of consumption was good for most foods, but it was lowest for cruciferous and leafy green vegetables. Interviewer bias regarding the dietary causes of colorectal cancer was not a major bias. Analysis of within-interviewer variation over time and analysis of recall bias were not major biases, as far as could be assessed in this study. Indirect validation of the nutritional data in the study (by comparison with other studies, by national per capita consumption, and by a comparison of energy intake with estimated energy requirements based on height, weight, and activity levels) suggested that the data are not an overestimate of intake. Cases and controls were both overestimating and underestimating their dietary intakes to a similar degree. For epidemiological studies of diet and disease, the quantitative diet history method as used here is recommended, particularly if the main objective of the study is to confirm hypotheses.