a1 MRC Epidemiology Resource Centre, University of Southampton, Southampton General Hospital, Southampton SO16 6YD, UK
There is growing interest in the use of dietary patterns as measures of exposure in studies of diet–disease relationships. However, relatively little is known about the impact of the type of dietary assessment method on the patterns observed. Using FFQ and food diary data collected from 585 women in early pregnancy we used principal component analysis to define dietary patterns. The first pattern was very similar in both datasets and was termed the ‘prudent’ diet. The second pattern, whilst comparable for the FFQ and food diaries, showed greater variation in coefficients than the prudent pattern; it was termed the ‘Western’ diet. Differences between the FFQ and diary scores were calculated for each woman for both the prudent and Western diet patterns. Of the differences in the prudent diet score, 95 % lay within ± 1·58 sd of the mean, and 95 % of the differences in the Western diet scores lay within ± 2·22 sd of the mean. Pearson's correlation coefficients were 0·67 (P < 0·001) for the prudent diet score and 0·35 (P < 0·001) for the Western diet score. The agreement between the FFQ and diary scores was lowest amongst respondents who were younger, had lower educational attainment and whose diaries were coded as ‘poor, probably incomplete’, although these effects were small. The first two dietary patterns identified in this cohort of pregnant women appear to be defined similarly by both FFQ and diary data, suggesting that FFQ data provide useful information on dietary patterns.
(Received December 08 2006)
(Revised July 31 2007)
(Accepted August 06 2007)
Abbreviations: PCA, principal component analysis