Related Experiment Video
Updated: Jul 26, 2025
10:32
Murine Cervical Heart Transplantation Model Using a Modified Cuff Technique
Published on: October 12, 2014
14.2K
Survival analysis for pediatric heart transplant patients using a novel machine learning algorithm: A UNOS analysis
Awais Ashfaq1, Geoffrey M Gray2, Jennifer Carapelluci3
1From the Cardiovascular Surgery, Heart Institute, Johns Hopkins All Children's Hospital, St. Petersburg, Florida.
Summary
Machine learning models effectively predict 1-year mortality after pediatric heart transplantation. Random forest models showed the best performance, identifying key risk factors like bilirubin levels and BMI for improved survival prediction.
Area of Science:
- Cardiology
- Pediatric Medicine
- Transplantation Science
- Data Science in Healthcare
Background:
- Pretransplantation risk factors for 1-year mortality after heart transplantation are not well understood.
- Predictive models for pediatric heart transplant outcomes are needed.
Purpose of the Study:
- To identify clinically relevant pretransplantation risk factors for 1-year mortality in pediatric heart transplant recipients.
- To develop and evaluate machine learning models for predicting 1-year mortality post-pediatric heart transplantation.
Main Methods:
- Utilized data from the United Network for Organ Sharing Database (2010-2020) for pediatric patients (0-17 years) undergoing their first heart transplant.
- Employed machine learning algorithms (Scikit-Learn, Scikit-Survival, Tensorflow) with expert and literature-based feature selection.
- Validated models using N-repeated k-fold cross-validation and assessed performance with the concordance index (C-index).
Main Results:
- Machine learning models outperformed traditional Cox proportional hazards models in predicting 1-year mortality.
- Random forest model achieved the highest C-index (0.68), indicating superior predictive accuracy.
- Key predictors identified included serum total bilirubin, travel distance, body mass index, donor SGPT/ALT, and donor PCO2.
Conclusions:
- A combined machine learning and expert-driven approach effectively predicts 1- and 3-year survival outcomes in pediatric heart transplantation.
- SHapley Additive exPlanations (SHAP) can be valuable for understanding complex interactions in survival prediction models.
