Using machine learning to improve survival prediction after heart transplantation.
Brian Ayers1, Tuomas Sandholm2, Igor Gosev3
1Department of Surgery, The Massachusetts General Hospital, Boston, Massachusetts, USA.
Journal of Cardiac Surgery
|August 20, 2021
Summary
Modern machine learning (ML) models significantly enhance survival prediction for orthotopic heart transplantation (OHT) patients. These advanced techniques offer improved accuracy over traditional methods, aiding clinical decisions.
Area of Science:
- Cardiology
- Medical Informatics
- Machine Learning
Background:
- Orthotopic heart transplantation (OHT) outcomes require improved predictive models.
- Accurate prognostication is crucial for patient selection and management.
Purpose of the Study:
- To evaluate modern machine learning (ML) techniques for predicting survival after OHT.
- To compare the performance of ML models against traditional statistical methods.
Main Methods:
- Retrospective analysis of 33,657 adult OHT patients (2000-2019) from the UNOS registry.
- Development of an ensemble ML model combining multiple algorithms.
- Validation using area under the receiver-operating-characteristic curve (AUROC), net reclassification index (NRI), and decision curve analysis (DCA).
Main Results:
- The ensemble ML model achieved a superior AUROC of 0.764 (95% CI, 0.745-0.782), outperforming individual models and logistic regression.
- ML models demonstrated significant improvements in predictive performance (NRI: 72.9% ±3.8%) compared to logistic regression.
- Decision curve analysis confirmed the ensemble model's superior risk prediction across all risk levels.
Conclusions:
- Modern ML techniques offer enhanced risk prediction for OHT compared to traditional methods.
- Improved prediction can inform patient selection, program evaluation, and allocation policies.
- These findings support the integration of ML into clinical decision-making for OHT.
