VIP:Visual-Valid Dietary Behavior Pattern Recognition for Local-National Trials - PROJECT SUMMARY
Chronic diseases and conditions such as obesity, diabetes and cardiovascular disease are among the
most common, costly, and preventable of all health problems in the United States. A healthy dietary pattern
is paramount in disease risk reduction. Since 2010, the dietary pattern approach has been recommended to
examine the relation of the totality of diet and health outcomes by U.S. Dietary Guidelines Advisory
Committees; meanwhile longitudinal dietary data have become increasingly available. Yet, methods are
underdeveloped for characterizing longitudinal diet-quality variations and even rudimentary for validating diet-
quality patterns that describe these dynamic variations, therefore, leading to unclear evidence for assessing
diet-health/disease relationships and formulating dietary guidelines. A noticeable gap exists between dietary
pattern literature and the fast-growing statistical learning field. We propose to develop an innovative statistical
learning tool for diet-quality trajectory pattern-recognition based on rich and highly-comparable longitudinal
dietary datasets from randomized controlled trials (RCT) and observational studies (OS) pertaining to a variety
of individuals, race/ethnicities, and geographical locations, and spanning up to 30 years, collected across 4 NIH-
funded RCTs in Massachusetts, and 2 large-scale multi-site national RCT and OS studies as well as
simulated dietary data based on these trials. Our project builds on PI Fang’s NIH-funded behavioral trajectory
pattern-recognition tool (Multiple-Imputation based Fuzzy Clustering, MIFuzzy) which processes longitudinal
trial data with missing and zero-inflated values, and identifies latent trajectory patterns that characterize
patients’ complex engagement and cognitive response variations during multi-component RCTs and better
explains the heterogeneity of treatment effects. This project will enhance and expand MIFuzzy to a Visual-
Valid Dietary Behavior Pattern Recognition tool (VIP), adapted to diet-quality trajectory pattern analyses and
chronic disease risk assessment. Our goal is to provide a new multi-view of diet-quality trajectory patterns and
associated outcomes from longitudinal studies. Based upon high-quality and comparable RCT and OS
longitudinal dietary data from NIDDK-, NHLBI-, and NIMH-funded studies, this VIP project will help grow more
valid evidence for developing dietary guidelines and clarify our understanding of diet-disease relationships for a
range of patient/individual types, potentially enabling better personalized, adaptive dietary strategies. Developing
this evidence-based VIP tool will also contribute to the infrastructure for diet-related studies, advance pattern-
recognition methods, help scientific communities and the lay public compare with local and national diet-quality
guidelines, and assess dietary health risks. In the long run, this VIP project will contribute to creating a data
management platform that support near-real-time pattern analyses and adaptive interventions.
