Machine Learning for Investigation on Endocrine-Disrupting Chemicals with Gestational Age and Delivery Time in a Longitudinal Cohort

Abstract

Endocrine-disrupting chemicals (EDCs) are widespread environmental chemicals that are often considered as risk factors with weak activity on the hormone-dependent process of pregnancy. However, the adverse effects of EDCs in the body of pregnant women were underestimated. The interaction between dynamic concentration of EDCs and endogenous hormones (EHs) on gestational age and delivery time remains unclear. To define a temporal interaction between the EDCs and EHs during pregnancy, comprehensive, unbiased, and quantitative analyses of 33 EDCs and 14 EHs were performed for a longitudinal cohort with 2317 pregnant women. We developed a machine learning model with the dynamic concentration information of EDCs and EHs to predict gestational age with high accuracy in the longitudinal cohort of pregnant women. The optimal combination of EHs and EDCs can identify when labor occurs (time to delivery within two and four weeks, AUROC of 0.82). Our results revealed that the bisphenols and phthalates are more potent than partial EHs for gestational age or delivery time. This study represents the use of machine learning methods for quantitative analysis of pregnancy-related EDCs and EHs for understanding the EDCs mixture effect on pregnancy with potential clinical utilities.