A combination of statistical methods for the analysis of the relative validation data of the quantitative food frequency questionnaire used in the THUSA study

Abstract

AbstractObjective:To apply structural equation modelling (SEM) and estimation of variance components to the relative validation data obtained from the quantitative food frequency questionnaire (QFFQ) used in the Transition, Health and Urbanisation in South Africa (THUSA) study.Design:A cross-sectional study.Setting:A community-based field study in an African population conducted during 1996.Subjects:Residents of the North West Province, South Africa, aged between 15 and 65 years.Methods:Relative validity of the QFFQ was tested against 7-day weighed food records, 24-hour urinary nitrogen (UN) excretion and estimated basal metabolic rate (BMR). SEM and estimation of variance components were applied to the log-transformed energy, protein, fat, calcium, iron, vitamin A and vitamin C intakes. UN excretion was used as a biomarker in the application of the SEM to protein and estimated BMR to energy intakes.Results:Constant bias (αQ) derived by the SEM varied from 0.85 (vitamin C) to 5.8 (energy). There was significant proportional bias for all nutrients except vitamin C. Validation coefficients (ρ(Q,T)) varied from 0.3 (fat, calcium, iron) to 0.7 (vitamin C). The inclusion of estimated BMR in the SEM for energy increased ρ(Q,T) from 0.38 to 0.42. The estimation of variance components gave slightly lower correlations for the relationship between intakes from the QFFQ and the unknown true intake.Conclusions:Robust statistical methods were successfully applied in a relative validation study for a QFFQ in an African population. Estimated BMR as a biomarker for energy intake produced more meaningful results than UN excretion as a biomarker for protein intake.