Machine learning-based analysis on factors influencing blood heavy metal concentrations in the Korean CHildren's ENvironmental health Study (Ko-CHENS)

Abstract

Heavy metal concentration in pregnant women affects neurocognitive and behavioral development of their infants and children. The majority of existing research focusing on pregnant women's heavy metal concentration has considered individual environmental factor. In this study, we aim to comprehensively consider lifestyle, food, and environmental factors to determine the most influential factor affecting heavy metal concentration in pregnant women. The Ko-CHENS (Korean CHildren health and ENvironmental Study) is a nationwide prospective birth cohort study in South Korea enrolling pregnant women from 2015 to 2020. A total of 5458 eligible pregnant women were included in this study, and 897 variables were included in questionnaire comprising: maternal general information, indoor and living environment, dietary habits, health behavior, exposure to chemicals. Lead, cadmium and mercury concentration on blood were measured in early, late pregnancy and in cord blood at birth. Variables that might be related to heavy metal concentrations were included in machine learning models. Random forest and XGBoost machine learning models were conducted for predictions. Both models had similar but better performance than multiple linear regression. Kimchi (β = 1.55), seaweed (β = 0.40), fatty fish (β = 1.55), intakes respectively affected lead, cadmium, and mercury exposure through early, late pregnancy and cord blood.