Comparison of two statistical approaches to predict all-cause mortality by dietary patterns in German elderly subjects

Dietary patterns are comprehensive variables of dietary intake appropriate to model the complex exposure in nutritional research. The objectives of this study were to identify dietary patterns by applying two statistical methods, principal component analysis (PCA) and reduced rank regression (RRR), and to assess their ability to predict all-cause mortality. Motivated by previous studies we chose percentages of energy from different macronutrients as response variables in the RRR analysis. We used data from 9356 German elderly subject enrolled in the European Prospective Investigation into Cancer and Nutrition study. The first RRR pattern, subjects which explained 30.8% of variation in energy sources and especially much variation in intake of saturated fat, monounsaturated fat and carbohydrates was a significant predictor of all-cause mortality. The pattern score had high positive loadings in all types of meat, butter, sauces and eggs, and was inversely associated with bread and fruits. After adjustment for other known risk factors, the relative risks from the lowest to highest quintiles of the first RRR pattern score were 1.0, 1.01, 0.96, 1.32, 1.61 (P for trend: 0.0004). In contrast, the first two PCA patterns explaining 19.7% of food intake variation but only 7.0% of variation in energy sources were not related to mortality. These results suggest that variation in macronutrients is meaningful for mortality and that the RRR method is more appropriate than the classic PCA method to identify dietary patterns relevant to mortality.