*Purpose: Real-time communication regarding organ placement is critical. For this purpose, UNOS launched DonorNet®, a computerized allocation algorithm, to aid and facilitate organ placement. However, by increasing the probability of an organ matching, this led to a remarkable increase in the volume of unwanted offers as a side effect. There is still an area for improvement to DonorNet® with novel computational techniques such as machine-learning to decrease the time spent reviewing organ offers that will not translate to an organ transplant based on the hospitals’ or surgeons’ preference. Hence, we aimed to develop a series of predictive machine-learning (ML) algorithms to aid in the incoming kidney donor offers through DonorNet®, taking special care in not impacting negatively in the organ allocation process.

*Methods: We analyzed ~45,000 historical kidney offers from our institution between 2003-2017 and built a series of ML predictive models to predict whether a given offer was going to be accepted in our center (Comparison A) or whether it was going to occur in a different hospital (Comparison B) (thereby addressing institutional preferences). Also, for Outcome A, we tested the predictiveness of our models to allocate the organ to the first patient on the list or any subsequent case thereafter. All our models were built in R v.3.5.1 using the CARET package.

*Results: Our most successful model, the bagged CART (Tree bagging), is an optimal classifier of incoming kidney organ offers with sensitivities, specificities, and accuracies ≥95% for both comparisons (Table). Other model families such as random forests or extreme gradient boosting yielded 100% in all model parameters for outcome B. A detailed model performance is shown in the Table.

*Conclusions: In a single-center experience, ML techniques could be useful for improving the increased number of unwanted organ offers thereby providing transplant surgeons with a manageable number of offers tailored to their preferences. We propose that a combination of techniques specifically, bagged CART and Extreme Gradient Boosting, could be the most optimal strategy to address the increased untailored kidney organ offers.

« Back to 2019 American Transplant Congress