Using Machine-Learning Techniques to Predict Postdonation Kidney Function in Living Kidney Donors

Background: Predicting kidney function after kidney donation is critical to properly selecting donors for living kidney donation. We aimed to predict postdonation kidney function after a living kidney donation using machine learning techniques. Methods: This retrospective cohort study included 823 living kidney donors from 2009 to 2020. The dataset was randomly divided into training (80%) and test (20%) sets. The main outcome was the postdonation estimated glomerular filtration rate (eGFR) at 12 months after kidney donation. We compared the performance of various machine learning techniques, traditional regression models as well as model from previous study. The best-performing model was selected using the mean absolute error (MAE) and root mean square error (RMSE). Results: The mean age was 45.2 ± 12.3 years, and 48.4% were males. The mean predonation and postdonation eGFRs were 101.3 and 68.8 ± 12.7 mL/min/1.73 m2, respectively. The XGBoost model showed the best performance with an MAE of 6.23 and RMSE of 8.06 with feature importance, including eGFR, age, serum creatinine, 24-hour urine creatinine, 24-hour urine sodium, creatinine clearance, cystatin C, cystatin C-based eGFR, computed tomography volume of the remaining kidney/body weight, normalized GFR of the remaining kidney measured through a diethylenetriaminepentaacetic acid (DTPA) scan, and sex. An easy-to-use web application titled Kidney Donation with Nephrologic Intelligence (KDNI) was developed. Conclusion: The prediction model using XGBoost accurately predicted the postdonation eGFR after living kidney donation. This model can be applied in clinical practice using KDNI, the developed web application. Funding: Government Support – Non-U.S.