Speaker: Natalie Gasca, Graduate Student, UW Biostatistics
Abstract: Nutritional scientists interested in diet-disease links have shifted from studying individual to multiple dietary components at once, but encounter difficulties when interpreting counterintuitive results due to correlated data and biologically complex systems. Additionally, most nutritional studies tend to characterize diets by indices or quantiles of food consumption, making it difficult to draw definitive conclusions about the size of dietary effects associated with minimized disease risk. We seek to fill this knowledge gap by expanding current statistical methods to better interpret links between cardiovascular disease risk and consumed foods. Specifically, our goal is to determine a reasonable way to use nutritional information to identify understandable heart-healthy diets that protect against heart disease. Although nutritional science has accorded much attention to outcome-independent approaches (e.g. a-priori or unsupervised), very few analyses have utilized supervised methods. We aim to broaden the use of supervised methods, when appropriate, to contribute to and complement the nutritional science literature. In particular, we focus on partial least squares extensions with reasonable features to build an interpretable heart disease risk score from the most relevant foods.