An integrated compositional approach to investigating the full effect of dietary intake on health

Preprint medRxiv: Abstract Nutritional epidemiology studies aim to investigate the impact of diet on disease risk, and their findings inform subsequent dietary recommendations. The issue of whether foods or nutrients should be the unit of exposure in such studies is a topic of debate. However, although nutrient-based analyses can elucidate biological mechanisms, they are technically problematic; furthermore, although foods have been considered the most suitable unit of dietary exposure because "we eat food, not nutrients" and the findings of food-based analyses can be easily translated into dietary advice, there is a greater risk of confounding and/or reverse causation because food selection and consumption are closely related to socio-economic and behavioural factors. This note describes an integrated approach that combines the positive aspects of nutrient-and food-based analyses. Individual dietary exposure is characterised by its overall macro-and micronutrient compositions and, in line with the compositional nature of dietary data, methods based on log-ratio transformations of nutrient compositions are used in diet/outcome association analyses that provide fully adjusted isocaloric estimates of the impact of specific nutrient balances on health outcomes. These estimates can then be combined with the nutrient compositions of selected foods to produce outcome-related scores which, as they are defined regardless of a subject's food selection/consumption, are not subject to confounding or reverse causation, and therefore have important implications for public health professionals. Data coming from an Italian population-based study were used to illustrate the approach in the context of diet/serum uric acid relationships, and some resulting new insights were briefly discussed.