Maximizing Survival in Pediatric Congenital Cardiac Surgery Using Machine Learning, Explainability, and Simulation Techniques

Background: Pediatric and congenital heart surgery (PCHS) is highly risky. Complications associated with this surgical procedure are mainly caused by the severity of the disease or the unnecessary, late, or premature execution of the procedure, which can be fatal. In this context, prognostic models are crucial to reduce the uncertainty of the decision to perform surgery; however, these models alone are insufficient to maximize the probability of success or to reverse a future scenario of patient death. Method: A new approach is proposed to reverse the prognosis of death in PCHS through the use of (1) machine learning (ML) models to predict the outcome of surgery; (2) an explainability technique (ET) to determine the impact of main risk factors; and (3) a simulation method to design health scenarios that potentially reverse a negative prognosis. Results: Accuracy levels of 96% in the prediction of mortality and survival were achieved using a dataset of 565 patients undergoing PCHS and assessing 10 risk factors. Three case studies confirmed that the ET known as LIME provides explanations that are consistent with the observed results, and the simulation of one real case managed to reverse the initial prognosis of death to one of survival. Conclusions: An innovative method that integrates ML models, ETs, and Simulation has been developed to reverse the prognosis of death in patients undergoing PCHS. The experimental results validate the relevance of this approach in medical decision-making, demonstrating its ability to reverse negative prognoses and provide a solid basis for more informed and personalized medical decisions.