Machine learning for post-liver transplant survival: Bridging the gap for long-term outcomes through temporal
Kiruthika Balakrishnan1, Sawyer Olson2, Gyorgy Simon3
1Department of Family, Community and Health Systems Science, University of Florida, Gainesville, FL, USA.
Computer Methods and Programs in Biomedicine
|October 5, 2024
Summary
New temporal variation features improve long-term liver transplant (LT) survival prediction. The Random Survival Forest (RSF) model incorporating these features outperforms the MELD score for better patient outcomes.
Area of Science:
- Medical Informatics
- Transplantation Medicine
- Survival Analysis
Background:
- Long-term survival of liver transplant (LT) recipients is crucial for organ allocation and mortality assessment.
- Current models like Model-for-End-Stage-Liver-Disease (MELD) inadequately predict post-LT survival.
- Accurate prediction of long-term post-LT survival is needed.
Purpose of the Study:
- To develop and evaluate predictive models for long-term post-liver transplant survival.
- To introduce novel temporal variation features for enhanced prediction accuracy.
- To compare the performance of different survival models using these features.
Main Methods:
- Utilized Cox Proportional-Hazards regression (CoxPH), Random Survival Forest (RSF), and Extreme Gradient Boosting (XGB) models.
- Incorporated patient demographics and waiting list duration.
- Analyzed data from 716 liver transplant patients (2011-2021).
Main Results:
- Temporal variation features combined with the RSF model achieved the highest predictive accuracy (C-index: 0.71, IBS: 0.151).
- This approach significantly outperformed the predictive capability of the MELD score (C-index <0.51).
Conclusions:
- Integrating temporal variation features with the RSF model improves long-term post-LT survival prediction.
- Enhanced predictions can aid clinical decision-making in organ allocation and patient management.
- This can lead to better overall outcomes for liver transplant recipients.
