Related Experiment Videos
Predicting outcomes after liver transplantation. A connectionist approach
H R Doyle1, I Dvorchik, S Mitchell
1Section of Computational Medicine, University of Pittsburgh School of Medicine, Pennsylvania.
Annals of Surgery
|April 1, 1994
Summary
Artificial neural networks show promise for predicting early liver transplant outcomes, outperforming traditional models. These advanced models offer improved accuracy for patient survival and organ allocation decisions.
Area of Science:
- Transplantation research
- Artificial intelligence in medicine
- Machine learning for clinical prediction
Background:
- Predicting early outcomes after liver transplantation is crucial for organ allocation and patient survival.
- Traditional multivariate models lack the sensitivity and specificity for clinical use.
- Artificial neural networks (ANNs) offer a powerful alternative for complex pattern recognition in medical data.
Purpose of the Study:
- To train an artificial neural network (ANN) to predict early graft outcomes following orthotopic liver transplantation.
- To evaluate the performance of ANNs compared to traditional predictive models.
Main Methods:
- Ten feed-forward, back-propagation neural networks were trained using data from 155 adult liver transplants.
- Data included information available by the second postoperative day.
- Receiver operating characteristic (ROC) curve analysis was used to evaluate predictive accuracy.
Main Results:
- Four ANNs achieved perfect discrimination (Area Under Curve [Az] = 1.0).
- Two additional ANNs demonstrated excellent performance (Az = 0.95).
- Combined network sensitivity reached 77% with 96% specificity at a specific cutoff.
Conclusions:
- ANNs show encouraging results for predicting liver transplant outcomes.
- Their robustness with noisy, nonlinear data makes them suitable for clinical predictive modeling.
- ANNs represent a promising advancement over traditional multivariate models in transplantation.