342 - Beyond Sugar: Nutrient Patterns and Dental Caries Risk in US Children Using Explainable Machine Learning

Abstract:

Background: In this study, we looked at the connection between dental caries in US children
ages 2 to 11 and dietary patterns rather than specific nutrients using NHANES data from 2011 to
2018.

Methods: We employed a supervised machine learning classifier (XGBoost) with
unsupervised nutritional pattern extraction (principal component analysis [PCA] and non-
negative matrix factorization [NMF]) to predict any caries experience against caries-free status.
Using SHAP and model-based dose-response curves, we assessed the relative importance and
type of connections for sugar, calcium, phosphorus, vitamin D, and other nutrients. Our major
hypothesis was that dietary patterns rich in preventative micronutrients (such Ca/P/vitamin D)
would be associated with a lower risk of dental caries even after adjusting for sugar and
sociodemographic factors.

Results: Using nationally representative NHANES data (2011-2018), this study found that dental caries are common among U.S. children aged 2-11, with higher rates in older children, those from lower socioeconomic status, and those with higher sugar intake and lower-quality diets overall. While higher intakes of calcium, phosphorus, and vitamin D were linked to lower
predicted caries risk, machine learning models that integrated nutrient intakes, dietary patterns,
and sociodemographic factors showed good predictive performance and highlighted high sugar
consumption as a major risk factor.

Conclusion: While patterns richer in calcium, phosphorus, vitamin D, and other micronutrients
seem protective, our explainable machine learning analysis of NHANES 2011–2018 shows that
dietary nutrient patterns combining high sugar and low mineral/vitamin intakes are linked to an
increased caries risk in US children.