There are currently 121,678 people waiting for lifesaving organ transplants in the U.S. and of these, 100,791 await kidney transplants (as of 1/11/16). Over 3,000 new patients are added to the kidney waiting list each month, and 13 people die each day awaiting a kidney transplant. Transplant success rates are generally good at 1 year post surgery but fall off at 5 years to 32 percent, putting those recipients back on the organ waiting list. Given this background, it is of critical importance to properly match donors to recipients to generate better long-term outcomes.
To address the above, Navy researchers have developed a machine-learning tool to generate a minimized Bayesian network that better predicts organ failure at 1 and 3 years after transplantation, based solely on preoperative variables. In this model, donor age, recipient BMI, and gender were among the pre-transplant variables with strongest association to outcome – variables which are not currently incorporated in allocation schemes. Prediction accuracy is greater than 80 percent at both the 1 year and 3 year endpoints. Currently, kidneys are allocated based on human leukocyte antigen matching and time on list. This new screening model may be used to supplement those decision criteria for better patient outcomes. As evidence-based medicine is becoming the standard of care, clinicians look toward these prognostic tools to assist in decision making.
This application is related to US application number 20110289035; US application number 20110295782; and US patent 8,510,245.
- Model was developed from a training set of over 7,000 patients, with 793 pre- and post-transplant variables selected from the database eventually narrowed to 52
- Model could be deployed in an XML format such that transplant centers could enter known data into a web interface and compare the risks of failure
- US patent 9,561,006 available for license
- Potential for collaboration with Navy researchers