(P38-19-26) Bias-Free Approach Tests Links of Nutrition With Physiology and Aging Processes: Method Demonstration in the Multiethnic Cohort – Adiposity Phenotype Study (MEC-APS)

Objectives: An important challenge is to improve understanding of the role(s) of specific foods/patterns, nutrients, etc. in human health. We have developed, theoretically-grounded, and validated a domain- and distribution-independent method termed DIRAC (DIversity of Ranks and ACcuracy) for objectively determining if combining individual variables or mathematical models adds independent predictive information. DIRAC does not require assumptions, additional data, or secondary validation and resists over-fitting. We evaluated DIRAC’s utility for studying diet’s role(s) in adiposity and metabolism. We used data from the MEC-APS, a complex test-case, since, while ectopic fat is obesity-associated, its prediction by total adiposity, including proxies (e.g., weight, BMI), is not strong and differs across racial/ethnic groups.

Methods: The MEC has followed 215,000 Hawaii and Los Angeles residents of African, Japanese, European, Latino, and Native Hawaiian ancestry since 1993-1996. In the MEC-APS (~20 years later), we measured visceral, liver, and subcutaneous fat (VFAT, LFAT, SFAT) using abdominal MRI and total body fatness by DXA in a cross-sectional subset of 1781 healthy women and men aged 60-77 years with BMIs of 17-46 kg/m². Statistics: DIRAC.

Results: Exploratory analyses found that prediction of VFAT, LFAT, and the VFAT/SFAT ratio improved when AHEI, HEI, and DASH scores from both baseline and 20-year follow-up were included. In contrast, prediction of HOMA-IR and HOMA-β improved when the same dietary pattern scores were included, but only the concurrent dietary measures contributed. Examination of the DASH dietary pattern component scores revealed insights into mathematically-independent influence (or lack thereof) of specific diet components and outcomes (e.g., dairy and legumes added no information to VFAT models). We further identified mathematically-dominant effects of other diet components (e.g., processed foods and increased VFAT, LFAT). The data revealed both sex- and ethnicity-dependent and independent relationships.

Conclusions: The DIRAC framework can provide objective, unbiased, assumption-free information about dietary interactions and their relationships with outcomes that can complement standard epidemiological analyses.

Funding Sources: USDA, NIH