Statistical modelling of usual intake

Within the EFSA Article 36 project “European Tool Usual Intake” (ETUI) a workshop was organised in May 2010 where the different available models to calculate usual intake were presented and discussed. This report integrates the workshop background document, the presentations given by experts, and the discussions during the workshop. The purpose of the workshop was to evaluate existing statistical methods for estimating usual intake with respect to a number of criteria, so that the performance of each method on each criterion will be well understood after the workshop. The outcome of the workshop allows choices to be made for a European Tool Usual Intake, to be implemented in the remainder of the project. A starting document was provided to the participants of the workshop with up-to-date information on methods, data and criteria, as a basis for discussion. It was apparent from the workshop that there is not one optimal model for all cases, rather a toolbox approach is suitable. The choice of the most appropriate model has to be fine-tuned case by case. Criteria to be considered are related to data availability and data pre-processing (e.g. compatibility of existing data formats, need to handle complicating factors like food code conversion, left-censored data, processing factors, brand loyalty, pooling over multiple datasets), the appropriateness of modelling assumptions for frequencies and amounts modelling (e.g. level of conservatism, additivity assumption and data transformations, the need to model intake as a function of covariates, correlations), the usefulness to combine survey and food frequency questionnaire data, the need to model single food intake or diet-aggregated intake, the need to evaluate uncertainties, and the usefulness of implementations. For the short term, case studies will be performed based on issues relevant for EFSA panels. Results from the workshop and possibilities of using models to calculate usual intake were recently discussed during the 5th meeting of the EFSA Expert group on food consumption data. Currently only few EU Member States use this kind of models but there is a lot of interest in gaining experience with the usual intake modelling. Several requests were received concerning the possibility of having more information about the ETUI project and even the organisation of a future training. In conclusion, this interim report is considered by EFSA to be a good review of the different models used and presents the main challenges related to the use of these techniques. Its publication could be useful to those who want to start using this kind of methodology