Analysis of patterns of food intake in nutritional epidemiology: food classification in principal components analysis and the subsequent impact on estimates for endometrial cancer.

OBJECTIVE To assess the effect of different methods of classifying food use on principal components analysis (PCA)-derived dietary patterns, and the subsequent impact on estimation of cancer risk associated with the different patterns. METHODS Dietary data were obtained from 232 endometrial cancer cases and 639 controls (Western New York Diet Study) using a 190-item semi-quantitative food-frequency questionnaire. Dietary patterns were generated using PCA and three methods of classifying food use: 168 single foods and beverages; 56 detailed food groups, foods and beverages; and 36 less-detailed groups and single food items. RESULTS Classification method affected neither the number nor character of the patterns identified. However, total variance explained in food use increased as the detail included in the PCA decreased (approximately 8%, 168 items to approximately 17%, 36 items). Conversely, reduced detail in PCA tended to attenuate the odds ratio (OR) associated with the healthy patterns (OR 0.55, 95% confidence interval (CI) 0.35-0.84 and OR 0.77, 95% CI 0.49-1.20, 168 and 36 items, respectively) but not the high-fat patterns (OR 0.95, 95% CI 0.57-1.58 and OR 0.85, 0.51-1.40, 168 and 36 items, respectively). CONCLUSIONS Greater detail in food-use information may be desirable in determination of dietary patterns for more precise estimates of disease risk.