Original article Urban/rural differences in diet and atopy in South Africa It has been known since at least the 1970s that atopy manifests far less in rural than in urban areas of poorer countries (1–3). This is not due to lack of allergic sensitization in rural areas (4), and the explanation is not yet clear. Lower levels of atopy have been linked to infections in early childhood [the so-called Ôhygiene hypothesisÕ (5)] as well as parasite infestation (6), but there is no evidence that parasites are less common in urban areas of poor, tropical countries than in rural areas, or that conditions in rural areas are Ôless hygienicÕ. In a randomized controlled trial carried out in schools in Ecuador, treatment against geohelminth parasites had no effect on atopy (7). There is accumulating evidence that diet affects the prevalence of asthma and possibly of atopic disease in general. Areas of specific interest have been dietary antioxidants (8), dietary lipid (9) and dietary electrolytes (10), as well as the consumption of probiotics (11). Interpretation of the evidence is difficult, however, and many of the findings described in observational studies have not been confirmed in trials, possibly because of the wide potential for confounding and effect modification in nutritional studies (12). Methods of data analysis are available for dealing with questions of confounding and interaction of dietary exposures, but the sheer number of features of diet that can be measured will often defeat a comprehensive investigation of this kind. Over the last decade, researchers have moved away from the analysis of individual foods and nutrients, and towards the extraction of a small number of dietary ÔpatternsÕ using data analytic methods – most popularly principal com- ponents analysis (PCA) (13). PCA of a multivariate dietary assessment, such as a food frequency question- naire (FFQ), proceeds by standardizing food weights to have the same variance, and forming the weighted linear combination of these standardized food weights with the greatest variance, that is the dimension of diet along which there is the most variation. This is the first principal component. Each subsequent principal compo- nent is derived as the dimension of diet which shows the most variation subject to being uncorrelated with all previous components (14). Several studies have used PCA to research relation- ships between diet and disease, including coronary heart Background: There are large differences in the prevalence of atopic disease between urban and rural areas of developing countries, without good explana- tion. Diet has been associated with atopic disease, but studies of specific nutri- ents are contradictory, cross-sectional studies often being unsupported by trials. We investigated diet as an explanation for the difference in the prevalence of atopy between urban and rural areas in South Africa. Methods: A cross-sectional analysis of food frequency questionnaires and allergen skin tests from 698 children aged 8–13 years, recruited from 24 schools in Cape Province, South Africa, who were taking part in a case–control study of exercise-induced bronchospasm. Food frequency data were analysed with a principal components analysis (PCA). Results: The first two principal components of diet explained 25% of the vari- ance, and discriminated almost perfectly between urban and rural subjects. The ÔurbanÕ component of diet was strongly associated with positive skin tests even after adjusting for urban residence. There were no significant associations between individual foods or nutrients and positive skin tests, allowing for multiple testing. Conclusions: Diet explains part of the difference in prevalence between urban and rural areas. The ability to demonstrate this using PCA, but not by exhaustive analysis of all foods, reflects the value of reducing the number of dietary dimensions. The number of foods and nutrients which can be assessed, and the possibility of confounding and effect modification, make it difficult to identify the features of diet most directly implicated in disease. This may explain inconsistencies in dietary studies. R. Hooper 1 , J. Calvert 2 , R. L. Thompson 3 , M. E. Deetlefs 4 , P. Burney 1 1 National Heart & Lung Institute, Imperial College London, London, UK; 2 North Bristol NHS Trust, Bristol, UK; 3 Institute of Human Nutrition, University of Southampton, Southampton, UK; 4 Oxford Radcliffe NHS Trust, Oxford, UK Key words: children; dietary patterns; food frequency questionnaire; skin tests; South Africa. Dr R. Hooper Respiratory Epidemiology & Public Health Group National Heart & Lung Institute Imperial College London Emmanuel Kaye Building Manresa Road London SW3 6LR UK Accepted for publication 25 November 2007 Allergy 2008: 63: 425–431 Ó 2008 The Authors Journal compilation Ó 2008 Blackwell Munksgaard DOI: 10.1111/j.1398-9995.2008.01627.x 425