Using machine learning to investigate the influence of the prenatal chemical exposome on neurodevelopment of young children

Abstract

Research investigating the prenatal chemical exposome and child neurodevelopment has typically focused on a limited number of chemical exposures and controlled for sociodemographic factors and maternal mental health. Emerging machine learning approaches may facilitate more comprehensive examinations of the contributions of chemical exposures, sociodemographic factors, and maternal mental health to child neurodevelopment. A machine learning pipeline that utilized feature selection and ranking was applied to investigate which common prenatal chemical exposures and sociodemographic factors best predict neurodevelopmental outcomes in young children. Data from 406 maternal-child pairs enrolled in the APrON study were used. Maternal concentrations of 32 environmental chemical exposures (i.e., phthalates, bisphenols, per- and polyfluoroalkyl substances (PFAS), metals, trace elements) measured during pregnancy and 11 sociodemographic factors, as well as measures of maternal mental health and urinary creatinine were entered into the machine learning pipeline. The pipeline, which consisted of a RReliefF variable selection algorithm and support vector machine regression model, was used to identify and rank the best subset of variables predictive of cognitive, language, and motor development outcomes on the Bayley Scales of Infant Development-Third Edition (Bayley-III) at 2 years of age. Bayley-III cognitive scores were best predicted using 29 variables, resulting in a correlation coefficient of r = 0.27 (R²=0.07). For language outcomes, 45 variables led to the best result (r = 0.30; R²=0.09), whereas for motor outcomes 33 variables led to the best result (r = 0.28, R²=0.09). Environmental chemicals, sociodemographic factors, and maternal mental health were found to be highly ranked predictors of cognitive, language, and motor development in young children. Our findings demonstrate the potential of machine learning approaches to identify and determine the relative importance of different predictors of child neurodevelopmental outcomes. Future developmental neurotoxicology research should consider the prenatal chemical exposome as well as sample characteristics such as sociodemographic factors and maternal mental health as important predictors of child neurodevelopment.

Keywords: APrON; Developmental neurotoxicology; Epidemiology; Infant and child neurotoxicity studies; Neurobehavioural Testing.