Machine Learning to Infer Neurocognitive Testing Scores Among Adolescents and Young Adults with Congenital Heart Disease

Congenital heart disease (CHD) affects approximately 1% of newborns and is associated with an increased risk for neurodevelopmental impairments. Identification and characterization of the factors affecting neurocognitive outcomes in adolescents and young adults (AYAs) with CHD remains an important area of research. We have integrated demographic, parental, socioeconomic, genetic, and brain magnetic resonance imaging (MRI) features into multivariate machine learning to infer neuropsychological testing scores with an enhanced forward inclusion backward elimination (FIBE) technique. Across 89 participants (aged 7–30 years), we included 15 neurocognitive assessments in 7 domains, achieving Pearson’s correlations 𝑟 = 0.25−0.65 between actual and inferred scores. General intelligence, working memory, and processing speed were most inferable, with various MRI features, the loss-of-functon(LoF) genetic variants in genes associated with neurodevelopmental disorders or that function as chromatin modifiers, sex, and father’s education level as key joint covariates. In contrast, oral language and perceptual organization were the least inferable. These findings highlight the combinatorial effects of genetic and environmental factors on the individual variability in neurocognitive functions among AYAs with CHD. Given the small sample size and data heterogeneity, further investigation in large cohorts is needed.