Modest validity and fair reproducibility of dietary patterns derived by cluster analysis.

Cluster analysis is widely used to analyze dietary patterns. We aimed to analyze the validity and reproducibility of the dietary patterns defined by cluster analysis derived from a food frequency questionnaire (FFQ). We hypothesized that the dietary patterns derived by cluster analysis have fair to modest reproducibility and validity. Dietary data were collected from 107 individuals from population-based survey, by an FFQ at baseline (FFQ1) and after 1 year (FFQ2), and by twelve 24-hour dietary recalls (24-HDR). Repeatability and validity were measured by comparing clusters obtained by the FFQ1 and FFQ2 and by the FFQ2 and 24-HDR (reference method), respectively. Cluster analysis identified a "fruits & vegetables" and a "meat" pattern in each dietary data source. Cluster membership was concordant for 66.7% of participants in FFQ1 and FFQ2 (reproducibility), and for 67.0% in FFQ2 and 24-HDR (validity). Spearman correlation analysis showed reasonable reproducibility, especially in the "fruits & vegetables" pattern, and lower validity also especially in the "fruits & vegetables" pattern. κ statistic revealed a fair validity and reproducibility of clusters. Our findings indicate a reasonable reproducibility and fair to modest validity of dietary patterns derived by cluster analysis.