JMIR Publications

ABSTRACT

Background:

Prediction of post-transplant health outcomes for pediatric heart transplantation is critical for risk stratification and high-quality post-transplant care.

Objective:

The purpose of the current study is to examine the use of machine learning models to predict rejection and mortality for pediatric heart transplant recipients.

Methods:

Various ML models were used to predict rejection and mortality at 1-, 3-, and 5-years post-transplant in pediatric heart transplant recipients using the United Network for Organ Sharing data from 1987-2019. Variables utilized for predicting post-transplant outcomes included donor and recipient, medical and social factors. SHAP (SHapley Additive exPlanations) were calculated to estimate the importance of each variable for prediction.

Results:

Random forest and adaptive boosting (AdaBoost) models were the best performing algorithms for different prediction windows across outcomes. Random forest outperformed other ML algorithms in predicting five of the six outcomes (AUROC: 0.664 and 0.706 for 1-year and 3-year rejection, 0.697, 0.758 and 0.763 for 1-year, 3-year and 5-year mortality). AdaBoost achieved the best performance for prediction of 5-year rejection with AUROC 0.705.

Conclusions:

The current study demonstrates the comparative utility of ML approaches for modeling post-transplant health outcomes using registry data. ML approaches can identify unique risk factors and their complex relationship with outcomes, thereby identifying at-risk patients and informing the transplant community about the future potential of these innovative approaches to improve pediatric post-heart transplant care. Future studies are required to translate the information derived from prediction models to optimize counseling, clinical care, and decision-making within pediatric organ transplant centers.