Factors affecting reproducibility of dietary reports using food frequency questionnaires

Department of Medical Epidemiology, Karolinska Institute, Stockholm, SwedenObjective: To study the reproducibility of a food frequency questionnaire used to assess past dietary habits.Design: Repeated retrospective assessment of dietary habits of a population sample.Setting: Uppsala-O¨rebro Health Care Region of Sweden.Subjects: One-hundred and four subjects participating as control subjects in a case–control study on thyroidcancer.Interventions: Between spring 1993 and spring 1994 a case–control study was conducted in the study area, inorder to investigate risk factors for thyroid cancer—diet among others. The study subjects reported theirconsumption of selected foods with reference 1–5y before, as well as dietary changes that occurred fromadolescence to adult age. One year later 197 control persons were invited to provide a second report of the samefood items. Of the invited subjects 104 agreed to participate.Results: The median Spearman rank correlation coefficient between the first and second assessment was 0.58 forfood consumption and 0.56 for nutrients intake. Correlation between the two assessments varied greatly betweenfood items, ranging from y0.03 to 0.83. The correlation was positively related to the frequency and negativelyrelated to the skewness of consumption. The recall of dietary changes from adolescence had a median correlationcoefficient of 0.38 (range 0.19–0.69). There were hints of a higher reproducibility of dietary reports among menand among highly educated subjects (more than 11y of education), but these differences were statisticallysignificant only for recall of adolescent diet. Reproducibility of nutrients intake was also significantly higheramong men than among women, as well as among subjects older than 45y.Conclusions: The reproducibility of dietary reports was satisfactory. Among factors affecting reproducibility,frequency and homogeneity of consumption in the source population are presumably the most important. Age,sex and education modify the inter-subject reproducibility of past diet. Based on reproducibility of a semi-quantitative index, the recall of adolescent diet probably has a low validity.Sponsorships: The study was partially funded by the Swedish Cancer Society, grant no. 3136-B92-02XBB, theSwedish Society of Endocrinology and the Swedish Medical Research Council, grant no. 102.Descriptors: food frequency questionnaire; reproducibility; past dietary habitsEuropean Journal of Clinical Nutrition (2000) 54, 658–664