Evaluation of a food frequency questionnaire to determine dietary intake in a large cohort: The SAPALDIA study

Introduction: Valid and scientifically based assessments of dietary intake at the population level are imperative for further investigations on the impact of diet on health and disease. Food frequency questionnaires (FFQs) are the most common tools used in large epidemiologic studies. Since complete and coherently answered FFQs are necessary to determine dietary intake, guidelines for error analysis and handling of missing information are very important, especially in self-administered paper questionnaires.

Aims: First, to establish guidelines for error analysis and handling of missing information in a FFQ developed at the ZHAW in Wädenswil (Zurich, Switzerland) for the SAPALDIA (Swiss Cohort Study on Air Pollution and Lung and Heart Diseases in Adults) study part three, which includes nutrition behaviour assessments. Second, to estimate the use and possible limitations of the FFQ.

Methods: Standardised procedures for data cleaning, data management and data control were performed. Implausible combinations of frequency, portion size, number of portions, double or multiple entries of food items and missing information on several food items were the main examined errors. Error analysis was carried out at the subject level, considering gender and age, and at the food item level. Descriptive analyses were performed to examine the nature of errors and scope of missing information. Error corrections and rules for imputation of missing values were proposed and applied.

Results: Collected data (n=1965) showed a big variation of number of overall errors per subject and per food item. In general, more errors were found in the data from female participants and participants aged over 60 years than in the data from males and younger subjects. Missing values of portion size and number of portions was the most frequent error followed by missing values in only the number of portions. The analysis per food item showed that fewer errors occurred for more frequently consumed foods such as pasta, rice and potatoes than for rarely consumed foods (e.g. tofu, soy milk, tomato juice).

Conclusions: The present analysis focused on potential errors and missing values occurring in the FFQ paper form in the SAPALDIA cohort. At the food group level, the FFQ appears to be a useful tool for assessing usual food intake. However, further investigations are needed to assess its use in specific population groups.