Accept/decline decision module for the liver simulated allocation model.

Area of Science:

• Medical Informatics
• Machine Learning
• Transplant Surgery

Background:

• Simulated allocation models (SAMs) are crucial for evaluating organ allocation policies.
• A key component of SAMs is predicting organ acceptance by potential recipients.

Purpose of the Study:

• To develop and evaluate machine learning classifiers for organ acceptance prediction.
• To improve the simulated allocation model for liver allocation.

Main Methods:

• Feature selection and imbalance correction methods were applied to organ transplant data.
• Classifiers were developed using logistic regression, support vector machines, boosting, classification and regression trees, and Random Forests.
• A novel evaluation tool, sample-path accuracy, was developed to assess predictor performance within simulations.

Main Results:

• The Random Forest method achieved the lowest overall error rate.
• Boosting techniques demonstrated higher accuracy when both sensitivity and specificity were critical.
• No single method outperformed all others across all performance metrics.

Conclusions:

• The logistic regression classifier offers interpretability, detailing feature contributions to acceptance probability.
• The Scientific Registry of Transplant Recipients selected the logistic regression model for their next-generation liver SAM.