Fairly Predicting Graft Failure in Liver Transplant for Organ Assigning

Area of Science:

• Medical Informatics
• Machine Learning
• Transplantation

Background:

• Liver transplantation is critical for end-stage liver disease but faces organ scarcity.
• Current organ allocation relies on the Model for End-stage Liver Disease (MELD) score, which overlooks post-transplant outcomes and donor/organ characteristics.
• Existing machine learning (ML) models for graft failure prediction may exhibit bias against certain demographic groups.

Purpose of the Study:

• To develop and evaluate a fair machine learning framework for predicting liver transplant graft failure.
• To address the limitations of existing organ allocation criteria and ML models by incorporating fairness considerations.
• To improve the accuracy and equity of organ distribution decisions in liver transplantation.

Main Methods:

• A novel fair machine learning framework was developed for graft failure prediction in liver transplantation.
• Knowledge distillation was employed to integrate dense and sparse features, leveraging the strengths of tree models and neural networks.
• A two-step debiasing methodology was specifically designed and implemented to enhance the fairness of the prediction model.

Main Results:

• The proposed framework demonstrated superior performance in both prediction accuracy and fairness compared to existing models.
• Experiments confirmed the presence of unfairness issues in current ML models used for liver transplant allocation.
• The knowledge distillation approach effectively combined diverse features for robust graft failure prediction.

Conclusions:

• The developed fair machine learning framework offers a promising solution for equitable organ allocation in liver transplantation.
• This approach can mitigate bias in ML models, leading to more just and accurate graft failure predictions.
• The findings highlight the importance of fairness in developing AI-driven healthcare decision-making tools, particularly in resource-limited scenarios like organ transplantation.