State-of-the-art machine learning algorithms for the prediction of outcomes after contemporary heart transplantation: Results from the UNOS database.
Kampaktsis, P. N., et al.Clinical Transplantation 2021 [record in progress].
Aims
The aim of this study was to create machine learning (ML) models for predicting mortality following heart transplantation (HT).
Interventions
The study dataset was randomly divided into the training cohort and the validation cohort.
Participants
18,625 heart transplant recipients
Outcomes
Predictive performance of ML models for mortality following HT.
Follow-up
N/A
CET Conclusions
This study used UNOS data to develop machine learning (ML) models to predict mortality following heart transplant. The authors randomly split a dataset of 18,625 transplants between 2010 and 2018 to training and validation cohorts and developed a number of models to predict 1-, 3- and 5-year survival. Predictive ability of the best ML model was moderate (AUC 0.689 for 1-year survival), but superior to regression models and the IMPACT score. A nice aspect of this study is that the authors attempted to create explainable models using local interpretable model-agnostic explanations (LIME) to identify the salient features in predictions. The clinical utility of the models is not discussed and use in allocation or pre-transplant decision making is limited by the inclusion of transplant variables such as operative time and ischaemic time, which would not be known at the time of listing or organ offering.
Trial registration
N/A