(P41-15-26) Integrating Data-Driven Dietary Scores and Polygenic Risk Models to Enhance BMI Prediction in Multiethnic Populations

Objectives: To develop and validate data-driven dietary scores (DDS) as a primary modifiable risk factor for BMI and evaluate the complementary predictive value of trans-ancestry polygenic scores (PGS). We aimed to determine if integrating these distinct data layers improves metabolic risk stratification across diverse ancestries compared to single-predictor models.

Methods: We conducted a cross-sectional analysis using baseline data from the Multiethnic Cohort Study. DDS were developed using 520 variables from a quantitative food frequency questionnaire among 193,182 participants. A random forest algorithm was trained on the subset of participants without genetic data. The resulting models were then validated in the independent subset of participants with available genomic data (N = 82,720), where we integrated published trans-ancestry PGS models to assess the efficacy of a joint model across groups. Final models were adjusted for total energy intake, smoking, physical activity, education level, and the first 20 principal components.

Results: The joint model significantly outperformed single-predictor approaches, explaining 15.6%–22.9% of total BMI variation across groups (p < 0.001). Specifically, an integrative model combining both PGS and DDS significantly improved model fit; adding DDS to genetic models provided an additional 5.3%–8.5% in variance explained, while adding PGS to dietary models contributed 5.1%–12.0%. DDS alone explained 7.7%–14.3% of BMI variation, and trans-ancestry PGS alone achieved 7.4%–15.5%. We observed statistically significant but small interactions in Japanese American, Latino, and White populations (β = 0.01– 0.02, p < 0.05). In these groups, the results suggest that genetic risk for obesity may be dampened by healthy dietary patterns.<

Conclusions: Trans-ancestry PGS and DDS provide complementary information for BMI prediction across diverse populations. Our findings demonstrate that the systematic integration of modifiable environmental factors alongside genetic information significantly outperforms single-predictor approaches. These results highlight the importance of considering population-specific gene-diet interactions to improve metabolic risk stratification.

Funding Sources: National Institute of Health