Andrew Smith Dietary patterns in the ALSPAC cohort: Cluster analysis EUCCONET International Workshop 18th October 2011
Andrew Smith Dietary patterns in the ALSPAC cohort: Cluster analysis Pauline Emmett P. K. Newby Kate Northstone World Cancer Research Fund
Dietary patterns Examine diet as a whole Start with many variables (e.g. FFQ, diet diary) Express as a small number of variables Image: Paul / FreeDigitalPhotos.net
Principal components analysis (PCA) Examine diet as a whole Start with many variables (e.g. FFQ, diet diary) Use correlations between foods Express as a small number of components Image: Paul / FreeDigitalPhotos.net
Cluster analysis Examine diet as a whole Start with many variables (e.g. FFQ, diet diary) Use similarities between people Express as a small number of clusters Image: Paul / FreeDigitalPhotos.net
Dietary patterns studies in ALSPAC SourcePCACluster Mothers FFQNorthstone and Emmett, 2008a Partners FFQNorthstone and Emmett, 2010 FFQ age 3North, Emmett et al FFQ age 4½Northstone, Emmett et al FFQ age 7Northstone, Emmett et al. 2005Smith et al FFQ age 9Northstone and Emmett, 2008b Diet diary age 10in submission FFQs age 13in submission
Case study: PCA of FFQ age 7 Northstone and Emmett, 2005 Image: Suat Eman, winnond / FreeDigitalPhotos.net Junk Traditional Health conscious
Dietary patterns studies in ALSPAC SourcePCACluster Mothers FFQNorthstone and Emmett, 2008a Partners FFQNorthstone and Emmett, 2010 FFQ age 3North, Emmett et al FFQ age 4½Northstone, Emmett et al FFQ age 7Northstone, Emmett et al. 2005Smith et al FFQ age 9Northstone and Emmett, 2008b Diet diary age 10in submission FFQs age 13in submission
Cluster analysis k-means is most widely-used method Must avoid pitfalls –Standardization –Algorithm –Reliability Image: Boaz Yiftach / FreeDigitalPhotos.net
Dietary patterns studies in ALSPAC SourcePCACluster Mothers FFQNorthstone and Emmett, 2008a Partners FFQNorthstone and Emmett, 2010 FFQ age 3North, Emmett et al FFQ age 4½Northstone, Emmett et al FFQ age 7Northstone, Emmett et al. 2005Smith et al FFQ age 9Northstone and Emmett, 2008b Diet diary age 10in submission FFQs age 13in submission
Case study: cluster analysis of FFQ age 7 FFQ administered to ALSPAC children at 81 months of age Input variables are the same as PCA (Northstone, Emmett et al. 2005) 8,279 children 3 clusters Smith et al. 2011
Case study: cluster analysis of FFQ age 7 Processed4177 children White bread Processed meat Fizzy drinks Ice cream Crisps Chips Squash Biscuits and sweets Plant- based 2065 children Brown/wholemeal bread Fruit and vegetables Vegetarian foods Pasta Rice Cereals Traditional British 2037 children Meat Potatoes Full-fat milk
Similarities with PCA
Cluster analysis of diet diary data 7473 children aged 10 Advantages –Similar 3-cluster solution (Processed, Plant-based, Traditional British) –Good separation between clusters Disadvantages –Not robust to under-reporting –PCA shows better associations with fat mass
Dietary patterns studies in ALSPAC SourcePCACluster Mothers FFQNorthstone and Emmett, 2008a Partners FFQNorthstone and Emmett, 2010 FFQ age 3North, Emmett et al FFQ age 4½Northstone, Emmett et al FFQ age 7Northstone, Emmett et al. 2005Smith et al FFQ age 9Northstone and Emmett, 2008b Diet diary age 10in submission FFQs age 13in submission
Dietary patterns studies in ALSPAC SourcePCACluster Mothers FFQNorthstone and Emmett, 2008a. Br J Nutr. 99 p Partners FFQNorthstone and Emmett, EJCN 64 p FFQ age 3North, Emmett and the ALSPAC Study Team, EJCN 54 p FFQ age 7Northstone, Emmett and the ALSPAC Study team, EJCN 59 p Smith, Emmett, Newby, Northstone EJCN 65 p FFQ age 9Northstone and Emmett, 2008b. Br J Nutr 100 p Diet diary age 10in submission FFQs age 13in submission