Diet-associated risks of disease and self-reported food consumption: how shall we treat partial nonresponse in a food frequency questionnaire?

Using data from a population-based case-control study on thyroid cancer, we studied two approaches of handling missing answers in a food frequency questionnaire when estimating food consumption and nutrient intakes. We analyzed the dietary reports of 165 cases and 248 control subjects. In the first approach, the omitted food items were considered as "null consumption." In the second approach, the missing answers were replaced with the median frequencies for subjects in the corresponding outcome category actually answering that specific food item. The results showed marginal differences between the two methods. In addition, the null consumption assumption was validated by means of complementary telephone interviews with a subsample of subjects who provided incomplete reports. Overall, the originally omitted answers actually corresponded to very rare consumption for 54.2% of cancer patients and for 54.7% of controls. However, this "true" proportion of null consumption varied greatly between different food items (range 0-96%) and between food groups (range 14-82%). To interpret the omitted self-reports of food consumption as indication of "zero consumption" is quite reasonable when the investigation does not focus on dietary items widely consumed in the source population.