Search

Recent large-scale sequencing efforts have shed light on the genetic contribution to the etiology of congenital heart defects (CHD); however, the relative impact of genetics on clinical outcomes remains less understood. Outcomes analyses using genetic data are complicated by the intrinsic severity of the CHD lesion and by interactions with conditionally dependent clinical variables. Here we apply Bayesian Networks, an explainable Artificial Intelligence solution, to describe the intertwined relationships between clinical variables, demography, and genetics in a cohort of children with single ventricle CHD. As isolated variables, a damaging genetic variant in a gene related to abnormal heart morphology and prolonged ventilator support following stage I palliative surgery increased the probability of having a low Mental Developmental Index (MDI) score at 14 months of age by 1.9- and 5.8-fold, respectively. However, in combination, these variables acted synergistically to further increase the probability of a low MDI score by 10-fold. Likewise, genetic information was predictive of a favorable neurodevelopmental outcome. For example, the absence of a damaging variant in a known syndromic CHD gene and a shorter post-operative ventilator support increased the probability of a normal MDI score 1.7- and 2.4-fold, respectively, but in combination increased the probability of a good outcome by 59-fold. Our analyses suggest a modest genetic contribution to neurodevelopmental and growth outcomes as isolated variables, similar to known clinical predictors. By contrast, genetic, demographic, and clinical variables interact synergistically to markedly impact clinical outcomes. These findings underscore the importance of capturing and quantifying the impact of damaging genomic variants in the context of multiple, conditionally dependent variables, such as pre- and post-operative factors, and demography.