Genome sequencing is critical for forecasting outcomes following congenital cardiac surgery

Abstract

While exome and whole genome sequencing have transformed medicine by elucidating the genetic underpinnings of both rare and common complex disorders, its utility to predict clinical outcomes remains understudied. Here, we use artificial intelligence (AI) technologies to explore the predictive value of whole exome sequencing in forecasting clinical outcomes following surgery for congenital heart defects (CHD). We report results for a prospective observational cohort study of 2,253 CHD patients from the Pediatric Cardiac Genomics Consortium with a broad range of complex heart defects, pre- and post-operative clinical variables and exome sequencing. Damaging genotypes in chromatin-modifying and cilia-related genes are associated with an elevated risk of adverse post-operative outcomes, including mortality, cardiac arrest and prolonged mechanical ventilation. The impact of damaging genotypes is further amplified in the context of specific CHD phenotypes, surgical complexity and extra-cardiac anomalies. The absence of a damaging genotype in chromatin-modifying and cilia-related genes is also informative, reducing the risk for some adverse postoperative outcomes. Thus, genome sequencing enriches the ability to forecast outcomes following congenital cardiac surgery.

Fig. 1. Damaging chromatin and cilia genotypes predict adverse post-operative outcomes in the context of CHD phenotypes and surgical complexity.

Bayesian networks display a best machine-learned relationship among genotypes, phenotypes, and outcomes for 2253 surgical patients with CHD. Each network node represents a present/absent variable. Damaging genotypes in chromatin-modifying genes (CHRMdGV) or cilia-related genes (CILIAdGV) were identified from the exomes of 2253 CHD patients by GEM. Phenotype classes were predicted from Fyler codes using XGBoost. Surgical outcomes for each patient were obtained from the Society of Thoracic Surgeons national database. Relative risks for selected adverse surgical outcomes were then estimated from each network using network-propagated probabilities. a An exact Bayesian network depicting the relationship among damaging de novo genetic variants in chromatin-modifying genes (green), phenotypes: LVO, HLHS, and ECAs (blue), surgical STAT4 or STAT5 category (yellow), and adverse surgical outcomes (orange). b An exact Bayesian network depicting the relationship among damaging recessive genetic variants in cilia-related genes (green), phenotypes: laterality defects (HTX) and extra cardiac anomalies (ECAs) (blue), surgical STAT4 category (yellow), and adverse surgical outcomes: long ventilation time, cardiac arrest, and mortality (orange). Directed acyclic graphs were moralized and displayed as non-directional networks. c Relative risk ratio estimates for adverse post-operative outcomes and CHD phenotypes or surgical complexity, comparing probands with and without damaging genotypes. Empirical ninety-five percent confidence intervals (CI 5, 95) are based on 1000 resampled network-based probability estimates. Because the resampling distribution estimates may be constrained by the Bayesian network structure, error bars may not be symmetric with respect to the median point estimate (see Methods). Abbreviations: CHRMdGV - de novo damaging genotypes in chromatin-modifying genes, LVO - left ventricular outflow tract obstruction, HLHS - hypoplastic left heart syndrome, CILIAdGV - biallelic damaging genotypes in cilia-related genes, ECA - extra cardiac anomaly, HTX - heterotaxy/laterality defects, MORT - mortality, STAT4 - surgical STAT4 category, STAT4-5 - surgical STAT 4 or STAT5 category, VENT - post-operative ventilation time >7 days. For Figs. 1c, 2, and 3, there were 8–35 patients in the conditional subsets and 1–5 patients in the target sets (see Supplementary Data 4).

Fig. 2. Damaging chromatin and cilia genotypes predict adverse post-operative outcomes in the context of extracardiac anomalies.

Relative risk ratios for adverse post-operative outcomes and extracardiac anomalies (ECAs), comparing probands with and without damaging genotypes in chromatin-modifying or cilia-related genes. Each risk estimate shows the point estimate of the network propagated relative risk. Empirical ninety-five percent confidence intervals (CI 5, 95) were generated by resampling the data matrix with replacement and re-estimating the network propagated risk 1000 times. Because the resampling distribution estimates may be constrained by the Bayesian network structure, error bars may not be symmetric with respect to the median point estimate. Target and conditional counts are listed in Supplementary Data 4. The \\ symbol represents a (CI 5, 95) that exceeds the x-axis range.

Fig. 3. Damaging genotypes in various gene categories/pathways are predictive of mortality for the most complex surgical procedures.

Relative risk ratios for adverse post-operative outcomes and surgical complexity compare probands with and without damaging genotypes in various gene pathways or categories. Gene lists are described in Supplementary Data 3 and have been previously published^,,. There is overlap between gene lists, with some genes represented in more than one gene pathway/category (Supplementary Fig. 4). Each estimate shows the point estimate of the network propagated relative risk. Empirical ninety-five percent confidence intervals (CI 5, 95) were generated by resampling the data matrix with replacement and re-estimating the network propagated risk 1,000 times. Because the resampling distribution estimates may be constrained by the Bayesian network structure, error bars may not be symmetric with respect to the median point estimate. Target and conditional counts are listed in Supplementary Data 4.