Identifying sources of measurement error in assessing dietary intakes - Results of a multi-country ring-trial.

BACKGROUND AND AIMS: Epidemiological investigations include dietary intakes as primary exposures or potential confounders. To reduce bias, data collection protocols include the administration of questionnaires together with measurements of biomarkers. Some error, however, remains and needs to be considered in the analysis and interpretation of results. The European Food Safety Authority supported a ring-trial to compare the precision and reproducibility of dietary assessment methods applied in Europe. METHODS AND RESULTS: Software applications used to collect 24-hour recalls and food records in six countries (Estonia, Italy, Latvia, Portugal, Spain, and Sweden) were assessed. The intake of 256 foods was identically reported to each method. Experienced interviewers participated and were instructed to repeat national protocols closely. The error in recording quantities, compared with reference values, was variable but in about 60% of recorded quantities was in the range of ±20%. Errors were however unsystematic and independent of the food type or quantification method used - although food pictures performed better. The reproducibility of some tools was limited. The methods generally captured additional ingredients (usually flavoring agents), but not sweetening agents or fortification and failed to record packaging information in about 60% of the cases. CONCLUSION: In a design that eliminated respondent bias, this study indicates that softwares, supporting databases and interviewers generally introduce random error in dietary assessments. The inclusion of large sample sizes and food pictures to quantify portions, together with enhanced attention on interviewers' training, standardisation of procedures and regular tool upgrades are essential in assuring a study's quality and comparability.