Food liking-based diet quality indexes (DQI) generated by conceptual and machine learning explained variability in cardiometabolic risk factors in young adults

Abstract

The overall pattern of a diet (diet quality) is recognized as more important to health and chronic disease risk than single foods or food groups. Indexes of diet quality can be derived theoretically from evidence-based recommendations, empirically from existing datasets, or a combination of the two. We used these methods to derive diet quality indexes (DQI), generated from a novel dietary assessment, and to evaluate relationships with cardiometabolic risk factors in young adults with (n = 106) or without (n = 106) diagnosed depression (62% female, mean age = 21). Participants completed a liking survey (proxy for usual dietary consumption). Principle component analysis of plasma (insulin, glucose, lipids) and adiposity (BMI, Waist-to-Hip ratio) measures formed a continuous cardiometabolic risk factor score (CRFS). DQIs were created: theoretically (food/beverages grouped, weighted conceptually), empirically (grouping by factor analysis, weights empirically-derived by ridge regression analysis of CRFS), and hybrid (food/beverages conceptually-grouped, weights empirically-derived). The out-of-sample CRFS predictability for the DQI was assessed by two-fold and five-fold cross validations. While moderate consistencies between theoretically-and empirically-generated weights existed, the hybrid outperformed theoretical and empirical DQIs in cross validations (five-fold showed DQI explained 2.6% theoretical, 2.7% empirical, and 6.5% hybrid of CRFS variance). These pilot data support a liking survey that can generate reliable/valid DQIs that are significantly associated with cardiometabolic risk factors, especially theoretically-plus empirically-derived DQI.