Agreement between a brief food frequency questionnaire and diet records using two statistical methods

Abstract

AbstractObjectiveTo compare intra- and inter-method reliability of a semi-quantitative food frequency questionnaire (FFQ) designed specifically to measure beta carotene (BC) and retinol intake, using two methods – the limits of agreement (LOA) and the correlation coefficient.DesignA cross-sectional study of dietary intake.SettingA randomized trial of vitamin A supplements in 2769 subjects with past asbestos exposure.SubjectsData from 57 men and 26 women, aged 28–72 years, living in Western Australia.MethodsThe FFQ was administered at baseline (FFQ1) and repeated 1 year later (FFQ2). Four 1-week diet records (DRs) were completed during the year.ResultsMean agreement between FFQ2 and FFQ1 was 120% for BC and 98% for retinol. LOA were 47–306% and 21–459%, respectively. Mean agreement between FFQ2 and the DR was 149% for BC and 63% for retinol; LOA were 50–447% and 11–349%, respectively. Mean agreement and LOA varied across energy intakes. Between the DR and FFQ2, correlation coefficients were 0.36 for BC and 0.51 for retinol. These varied considerably across age, gender and energy intakes and were not in accordance with limits of agreement findings.ConclusionAlthough correlation coefficients were positive and significant, there was less than ideal intra-method and inter-method reliability shown by the limits of agreement method. Bias was uneven across the range of intakes, the LOA were wide and, compared with the DR, the FFQ significantly over-estimated BC and under-estimated retinol. This shows the limitations of calculating correlation coefficients alone, for assessing reliability and validity.