A transformer-based deep learning approach for fairly predicting post-liver transplant risk factors
Can Li1, Xiaoqian Jiang2, Kai Zhang2
1Department of Biostatistics and Data Science, School of Public Health, The University of Texas Health Science Center at Houston, Houston, TX, USA.
Journal of Biomedical Informatics
|November 22, 2023
Summary
This study introduces a deep learning model for predicting post-liver transplant risks, improving patient matching and ensuring equitable outcomes across diverse populations. The model balances multiple risk predictions effectively.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Medicine
- Transplantation Science
Background:
- Liver transplantation is crucial for end-stage liver disease but faces challenges in donor-patient matching and transplant equity.
- Current scoring systems like MELD focus on short-term mortality, neglecting crucial post-transplant risks.
- Accurate prediction of post-transplant complications is essential for improving patient outcomes and resource allocation.
Purpose of the Study:
- To develop advanced predictive models for multiple post-transplant risk factors in liver transplant recipients.
- To enhance donor-patient matching by incorporating post-transplant risk prediction.
- To ensure fairness and equity in transplant outcomes across diverse demographic subpopulations.
Main Methods:
- A deep neural network was employed using a multi-task learning approach to simultaneously predict five key post-transplant risks.
- Task-balancing techniques were utilized to optimize performance across all predicted risk factors.
- A novel fairness-achieving algorithm was developed and applied to mitigate prediction disparities among different demographic groups.
Main Results:
- The multi-task learning model successfully balanced task performance while maintaining high accuracy, reducing task discrepancy by 39%.
- The fairness-achieving algorithm significantly decreased prediction disparities across sensitive attributes including gender, age group, and race/ethnicity.
- The integrated approach demonstrated robust and equitable risk prediction capabilities for liver transplant patients.
Conclusions:
- Integrating multi-task learning with fairness-aware algorithms offers a powerful framework for improving liver transplant outcomes.
- The developed model enhances the accuracy and equity of post-transplant risk prediction, addressing critical challenges in transplantation.
- This approach holds significant potential for optimizing organ allocation and patient management in liver transplantation.
