Session Time: 3:15pm-4:45pm
Presentation Time: 3:27pm-3:39pm
*Purpose: There is a shortage of transplantable kidneys in the US. Yet, the discard rate among older donors (55+) is approx. 38% per year. The longterm graft function is predicted by the 1 yr estimated glomerular filtration rate (eGFR). KDPI, the current measure of kidney quality, does not consider recipient characteristics. A tool to assess transplant outcomes accounting for the specific traits of donor and recipient could prove beneficial information. Using 1 yr eGFR goals predicting function described by Kasiske (AJKD 57:466 2011) we developed a personalized score, called Transplant Risk and Benefit (TRB) score for donor-recipient pairs. The hypothesis is that there is significant potential for transplant benefit in kidneys not currently used.
*Methods: We used the 2000-2017 National UNOS STAR file dataset, containing information on donors, waitlisted patients, transplants and follow-ups. Outcomes of adult recipients of older-donor kidneys were analyzed. 23,045 transplants and 603 variables, from which 131 influential variables were extracted was used to calculate likelihood of obtaining certain 1 yr eGFR thresholds.
Using advanced machine-learning methods (e.g. SVM, Random Forest, Neural Networks, Logistic Regression, and Ensemble Methods) a TRB score was developed. This score depends on two thresholds, b and r, which indicates the level of benefit (eGFR) and risk of not realizing that b, respectively. The center and patient can choose the degree of benefit and likelihood of achieving that benefit for each donor kidney. Formally, for each donor i and recipient j
For example, if b=45 ml/min and r=0.2, then kidneys would be identified with at least an 80% probability that 1 year eGFR will be above 45. We applied the TRB score to 4,190 older-donor kidneys that were discarded due to “donor characteristics” to determine how many of these could have resulted in sufficiently positive graft outcome.
*Results: Depending on the threshold of 1 yr eGFR desired, significant numbers of waitlisted people could have received benefit from non-used organs.
|b (eGFR)||1-r||# non-used kidneys with TRB > 1-r||KDPI mean, std. dev.|
*Conclusions: The TRB score is able to identify the probability that a donor kidney will provide a predictable level of function in a particular recipient. A tool that provides a personal threshold of transplant benefit and the risk of failure achieve it would allow recipients and centers more ability to assess the right kidney for the right patient. Predictability from machine learning will improve with more data.
To cite this abstract in AMA style:Pruett T, Martin P, Gupta D. A Machine-Learning Approach for Estimating Likelihood of Transplant Benefit from Older-Donor Kidneys [abstract]. Am J Transplant. 2020; 20 (suppl 3). https://atcmeetingabstracts.com/abstract/a-machine-learning-approach-for-estimating-likelihood-of-transplant-benefit-from-older-donor-kidneys/. Accessed April 20, 2021.
« Back to 2020 American Transplant Congress