Development of a Multibiomarker Panel of Healthy Eating Index in United States Adults: A Machine Learning Approach

Abstract

BackgroundDietary and nutritional biomarkers are objective dietary assessment tools that will enable a more accurate and precise determination of diet-disease relations. However, the lack of established biomarker panels for dietary patterns is concerning, as dietary patterns continue to be the focus of dietary guidelines. ObjectivesWe aimed to develop and validate a panel of objective biomarkers that reflects the Healthy Eating Index (HEI) by applying machine learning approaches to the National Health and Nutrition Examination Survey data. MethodsCross-sectional population-based data (eligible criteria: age ≥20 y, not pregnant, no reported supplement use of dedicated vitamin A, D, E, or fish oils; n = 3481) from the 2003 to 2004 cycle of the NHANES were used to develop 2 multibiomarker panels of the HEI, 1 with (primary panel) and 1 without (secondary panel) plasma FAs. Up to 46 blood-based dietary and nutritional biomarkers (24 FAs, 11 carotenoids, and 11 vitamins) were included for variable selection using the least absolute shrinkage and selection operator controlling for age, sex, ethnicity, and education. The explanatory impact of selected biomarker panels was assessed by comparing the regression models with and without the selected biomarkers. In addition, 5 comparative machine learning models were constructed to validate the biomarker selection. ResultsThe primary multibiomarker panel (8 FAs, 5 carotenoids, and 5 vitamins) significantly improved the explained variability of the HEI (adjusted R2 increased from 0.056 to 0.245). The secondary multibiomarker panel (8 vitamins and 10 carotenoids) had lesser predictive capabilities (adjusted R2 increased from 0.048 to 0.189). ConclusionsTwo multibiomarker panels were developed and validated to reflect a healthy dietary pattern consistent with the HEI. Future research should seek to test these multibiomarker panels in randomly assigned trials and identify whether they have broad application in healthy dietary pattern assessment.