LAUSR

Abstract Objective: In the field of nutritional epidemiology, principal component analysis (PCA) has been used extensively in identifying dietary patterns. Recently, compositional data analysis (CoDA) has emerged as an alternative approach for obtaining dietary patterns. We aimed to directly compare and evaluate the ability of PCA and principal balances analysis (PBA), a data-driven method in CoDA, in identifying dietary patterns and their associations with the risk of hypertension. Design: Cohort study. A 24-h dietary recall questionnaire was used to collect dietary data. Multivariate logistic regression analysis was used to analyse the association between dietary patterns and hypertension. Setting: 2004 and 2009 China Health and Nutrition Survey. Participants: A total of 3892 study participants aged 18–60 years were included as the subjects. Results: PCA and PBA identified five patterns each. PCA patterns comprised a linear combination of all food groups, whereas PBA patterns included several food groups with zero loadings. The coarse cereals pattern identified by PBA was inversely associated with hypertension risk (highest quintile: OR = 0·74 (95 % CI 0·57, 0·95); P for trend = 0·037). None of the five PCA patterns was associated with hypertension. Compared with the PCA patterns, the PBA patterns were clearly interpretable and accounted for a higher percentage of variance in food intake. Conclusions: Findings showed that PBA might be an appropriate and promising approach in dietary pattern analysis. Higher adherence to the coarse cereals dietary pattern was associated with a lower risk of hypertension. Nevertheless, the advantages of PBA over PCA should be confirmed in future studies.