Machine Learning Improves Kidney Discard Predictability
1IEMS Department, Northwestern University, Evanston, IL, 2IEMS Department, MEAS, Northwestern University, Evanston, IL
Meeting: 2020 American Transplant Congress
Abstract number: C-020
Keywords: Donors, marginal, Kidney transplantation
Session Information
Session Name: Poster Session C: Kidney Deceased Donor Allocation
Session Type: Poster Session
Date: Saturday, May 30, 2020
Session Time: 3:15pm-4:00pm
Presentation Time: 3:30pm-4:00pm
Location: Virtual
*Purpose: Despite the shortage of kidneys supply, 18%-20% of deceased donor kidneys are discarded annually.
*Methods: The cohort consisted of adult deceased donor kidneys donated between 12/3/2014-03/01/2019. The used classification models, along with performance metrics of interest, are shown in Table 1. Half of the data records are used for training the models. The models are then tested on the other half of the data records.
*Results: All models’ classification performance metrics are shown in Table 1. Overall, Random Forest (RF) outperforms other models. Of 6,707 discarded kidneys in the test dataset, RF correctly classifies 4,160 kidneys, whereas Logistic Regression (LR) correctly classifies 3,401 kidneys. Considering a subset of kidneys with KDPI ≥ 75% (16,274), RF significantly outperforms LR in classifying true discards and true transplants (RF: 14,406 (accuracy=0.885); LR: 12,248 (accuracy=0.750)). Also, when considering 3,300 kidneys with KDPI ≥ 75% discarded as “No recipient located – list exhausted,” RF classifies 2,900 true discards, compared to LR with 2,385 true discard classifications. It means 515 more kidneys are correctly identified as potential for discard using RF. Overall, the classification performance of Adaptive Boosting (AdaBoost) is comparable to that of RF.
*Conclusions: RF and AdaBoost significantly perform better than classical logistic regression-based models in classifying deceased donor kidney discards.
| Model | Confusion Matrix (Predicted/ Observed) | Discarded (6,707) | Transplanted (26,385) | Recall Average (SD) | Precision Average (SD) | F-Measure Average (SD) | Accuracy Average (SD) |
| Random Forest | Discarded Transplanted |
4,160 2,547 | 1,377 25,008 |
0.620 (0.006) | 0.751 (0.005) | 0.680 (0.004) | 0.881 (0.001) |
| Adaptive Boosting | Discarded Transplanted |
4,081 2,626 | 1,363 25,022 |
0.609 (0.010) | 0.750 (0.005) | 0.672 (0.006) | 0.879 (0.002) |
| Support Vector Machines | Discarded Transplanted |
3,206 3,501 | 940 25,445 |
0.478 (0.006) | 0.773 (0.006) | 0.591 (0.005) | 0.866 (0.001) |
| Neural Networks | Discarded Transplanted |
3,553 3,154 | 1,378 25,007 |
0.530 (0.023) | 0.772 (0.20) | 0.610 (0.010) | 0.863 (0.003) |
| K-Nearest Neighbors | Discarded Transplanted |
2,334 4,373 | 1,722 24,663 |
0.348 (0.005) | 0.576 (0.006) | 0.434 (0.005) | 0.816 (0.001) |
| Naive Bayes | Discarded Transplanted |
4,416 2,291 | 4,118 22,267 |
0.658 (0.008) | 0.518 (0.006) | 0.579 (0.004) | 0.806 (0.003) |
| Logistic Regression | Discarded Transplanted |
3,401 3,306 | 1,189 25,196 |
0.507 (0.005) | 0.741 (0.005) | 0.602 (0.004) | 0.864 (0.001) |
To cite this abstract in AMA style:
Barah M, Mehrotra S. Machine Learning Improves Kidney Discard Predictability [abstract]. Am J Transplant. 2020; 20 (suppl 3). https://atcmeetingabstracts.com/abstract/machine-learning-improves-kidney-discard-predictability/. Accessed May 16, 2025.« Back to 2020 American Transplant Congress