Machine Learning With Ingredient-Level Food Trees Reveals Contributors to Systemic Inflammation Among Adults in the National Health and Nutrition Examination Survey, 2001-2010 and 2015-2018.

BACKGROUND

Methods for modeling the relationship between self-reported 24-hour dietary recalls and health outcomes are traditionally based on nutrients and/or dietary patterns. Machine learning (ML), combined with hierarchical representations of diet, may help improve estimates of health and identify specific foods associated with diet-induced inflammation.

OBJECTIVE

The aim of this study was to assess the accuracy of estimating systemic inflammation from hierarchically arranged ingredient-level diets in a large US cohort.

DESIGN

This was a cross-sectional analysis using data on US adults from the National Health and Nutrition Examination Survey 2001-2010 and 2015-2018 cycles.

PARTICIPANTS AND SETTING

The continuous National Health and Nutrition Examination Survey data representing adult men and women who completed 1 or more 24-hour dietary recalls with an energy intake between 500 and 4500 kcal/day had no active infection or acute phase response, and measurement of serum C-reactive protein (CRP) level (N = 19 460).

MAIN OUTCOME MEASURES

The main outcome measure was classification accuracy for predicting high and low inflammation based on the top and bottom tertiles of CRP level.

STATISTICAL ANALYSES PERFORMED

Mixed meal disaggregation was performed to generate an ingredient-level representation of diet that was further annotated to produce a hierarchical data structure, or food tree. Hierarchical feature engineering selected the most informative food tree features for predicting systemic inflammation (ie, CRP level). ML models were used to assess the accuracy of predicting CRP level from the food tree features compared with the Dietary Inflammatory Index (DII) score. Logistic regression was used to calculate the marginal effects of ingredients identified from ML models.

RESULTS

Representation of diet as an ingredient-level food tree reduced dietary features from 6412 unique foods to 566 unique ingredients. ML classifiers trained on food tree data predicted high vs low systemic inflammation (CRP level tertile) with similar accuracy (0.761) on never-seen data compared with models trained using DII scores (0.757) (McNemar test P = .5). Individual dietary components revealed contributions toward increased inflammation, including fruit punch, soda, and high-fat milk (marginal effects: 0.001 to 0.005; P < .05), and foods associated with decreased inflammation such as herbal tea, brewed espresso, decaf coffee, brown rice, and dry pasta (marginal effects: -0.08 to -0.001; P < .05).

CONCLUSIONS

Specific ingredients, selected from a food tree, performed as well as the DII in predicting systemic inflammation. Choice of common foods and beverages associated with inflammation varied in magnitude and direction, consistent with previous studies that have demonstrated pro- and anti-inflammatory responses to these dietary components.