Abstract: SA-PO1079
Time-to-Graft Loss Prediction in Kidney Transplant Recipients
Session Information
- Transplantation: Clinical - II
November 04, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Transplantation
- 2102 Transplantation: Clinical
Authors
- Szili-Torok, Tamas, University Medical Center Groningen, Department of Internal Medicine, Groningen, Netherlands
- Verbeek, Max J., University Medical Center Groningen, Department of Internal Medicine, Groningen, Netherlands
- Kremer, Daan, University Medical Center Groningen, Department of Internal Medicine, Groningen, Netherlands
- Tietge, Uwe Jf, Division of Clinical Chemistry, Department of Laboratory Medicine, Karolinska Institutet, Stockholm, Sweden
- Bakker, Stephan J.L., University Medical Center Groningen, Department of Internal Medicine, Groningen, Netherlands
- De Borst, Martin H., University Medical Center Groningen, Department of Internal Medicine, Groningen, Netherlands
Background
Several predictions models have been proposed which estimate graft failure risk in kidney transplant recipients (KTR). However, the risk resulting from the model is challenging to interpret. Therefore, this study aims to propose a machine learning model which predicts time to kidney graft loss.
Methods
Data were derived from the transplant registry of the University Medical Center Groningen, including KTR transplanted after 1-1-1995. Data for model training was obtained 1 year after transplantation. KTR with missing data <5 years after transplantation were excluded. Graft loss was defined as death or graft failure. To predict the time to graft loss, extreme gradient boosting (XGBoost) with the accelerated failure time objective was utilized. The model was trained using a bootstrap aggregating (bagging) approach. Variables included in the model were selected by feature selection. The performance of the machine learning model was internally evaluated using the C-index on 20% of the dataset which was not used for training.
Results
From the 2195 included patients (age 52 [41-61] years, 42% female), 1225 (56%) developed graft loss during median follow-up of 8.1 [5.1-13.0] years. The final model was trained using 45 commonly determined clinical variables, including eGFR, hemoglobin, age, and several feature-engineered variables. Feature engineered variables are new variables created from existing ones to improve the performance of the model. Hematocrit*eGFR, recipient age and gamma-glutamyltransferase were identified as the most important variables driving time to kidney graft loss. Internal validation showed a C-index of 0.74.
Conclusion
The machine learning model developed predicts time to graft loss with a reasonable accuracy.