a1 BIPS – Institute for Epidemiology and Prevention Research, Achterstrasse 30, 28359 Bremen, Germany
a2 Department of Public Health, Ghent University, Ghent, Belgium
a3 International Agency for Research on Cancer (IARC), Dietary Exposure Assessment Group (DEX), Lyon, France
a4 Research Foundation–Flanders (FWO), Brussels, Belgium
a5 Medical Faculty, Department of Pediatrics, University of Pécs, Pécs, Hungary
a6 GENUD (Growth, Exercise, Nutrition and Development) Research Group, University of Zaragoza, Zaragoza, Spain
a7 Fondazione IRCSS, Istituto Nazionale dei Tumori, Department of Preventive and Predictive Medicine, Nutritional Epidemiology Unit, Milan, Italy
a8 Department of Surveillance and Evaluation, National Institute for Health Development, Tallinn, Estonia
a9 Research and Education Foundation of Child Health, Paphos, Cyprus
a10 Institute of Food Sciences, CNR, Avellino, Italy
a11 Department of Public Health and Community Medicine, University of Gothenburg, Gothenburg, Sweden
Objective Measurement errors in dietary data lead to attenuated estimates of associations between dietary exposures and health outcomes. The present study aimed to compare and evaluate different approaches of handling implausible reports by exemplary analysis of the association between dietary intakes (total energy, soft drinks, fruits/vegetables) and overweight/obesity in children.
Design Cross-sectional multicentre study.
Setting Kindergartens/schools from eight European countries participating in the IDEFICS Study.
Subjects Children (n 5357) aged 2–9 years who provided one 24 h dietary recall and complete covariate information.
Results The 24 h recalls were classified into three reporting groups according to adapted Goldberg cut-offs: under-report, plausible report or over-report. In the basic logistic multilevel model (adjusted for age and sex, including study centre as random effect), the dietary exposures showed no significant association with overweight/obesity (energy intake: OR=0·996 (95 % CI 0·983, 1·010); soft drinks: OR = 0·999 (95 % CI 0·986, 1·013)) and revealed even a positive association for fruits/vegetables (OR = 1·009 (95 % CI 1·001, 1·018)). When adding the reporting group (dummy variables) and a propensity score for misreporting as adjustment terms, associations became significant for energy intake as well as soft drinks (energy: OR = 1·074 (95 % CI 1·053, 1·096); soft drinks: OR = 1·015 (95 % CI 1·000, 1·031)) and the association between fruits/vegetables and overweight/obesity pointed to the reverse direction compared with the basic model (OR = 0·993 (95 % CI 0·984, 1·002)).
Conclusions Associations between dietary exposures and health outcomes are strongly affected or even masked by measurement errors. In the present analysis consideration of the reporting group and inclusion of a propensity score for misreporting turned out to be useful tools to counteract attenuation of effect estimates.
(Received February 23 2012)
(Revised August 08 2012)
(Accepted August 24 2012)
(Online publication October 09 2012)