Dietary patterns obtained through principal components analysis: the effect of input variable quantification

Principal components analysis (PCA) is a popular method for deriving dietary patterns. A number of decisions must be made throughout the analytic process, including how to quantify the input variables of the PCA. The present study aims to compare the effect of using different input variables on the patterns extracted using PCA on 3-d diet diary data collected from 7473 children, aged 10 years, in the Avon Longitudinal Study of Parents and Children. Four options were examined: weight consumed of each food group (g/d), energy-adjusted weight, percentage contribution to energy of each food group and binary intake (consumed/not consumed). Four separate PCA were performed, one for each intake measurement. Three or four dietary patterns were obtained from each analysis, with at least one component that described 'more healthy' and 'less healthy' diets and one component that described a diet with high consumption of meat, potatoes and vegetables. There were no obvious differences between the patterns derived using percentage energy as a measurement and adjusting weight for total energy intake, compared to those derived using gram weights. Using binary input variables yielded a component that loaded positively on reduced fat and reduced sugar foods. The present results suggest that food intakes quantified by gram weights or as binary variables both resulted in meaningful dietary patterns and each method has distinct advantages: weight takes into account the amount of each food consumed and binary intake appears to describe general food preferences, which are potentially easier to modify and useful in public health settings.