Korean J Transplant.  2022 Nov;36(Supple 1):S43. 10.4285/ATW2022.F-1573.

Machine learning based prediction model for renal adaptation in living kidney donors

  • 1Department of Nephrology, Samsung Medical Center, Seoul, Korea
  • 2Department of Digital Health, Samsung Advanced Institute for Health Sciences and Technology, Sungkyunkwan University, Seoul, Korea
  • 3Department of Smart Health Lab, Research institute of Future Medicine, Samsung Medical Center, Seoul, Korea
  • 4Department of Emergency Medicine, Samsung Medical Center, Seoul, Korea


Development of chronic kidney disease or end-stage kidney disease after donation is a critical issue in living kidney donation. It is imperative to predict postdonation kidney function for deciding eligibility because of long life expectancy in young donors and comorbidities affecting kidney function in elderly donors. We aimed to develop a machine learning based prediction model for renal adaptation after donation in living kidney donors.
This retrospective cohort study included a total of 823 living kidney donors from 2009 to 2020. We developed a pre-diction model using AutoScore, a machine learning-based clinical score generator. Two main outcomes were analyzed: fair renal adaptation and good renal adaptation, defined as estimated glomerular filtration rate (eGFR) 60 mL/min/1.73 m2 or above and 65% or above of predonation eGFR between postdonation 6 and 12 months, respectively. A web-application tool was devel-oped using 'shiny' R package.
Mean age was 45.2 years and 51.6% was female. Among predonation variables, predonation eGFR, age, sex, body mass index, remaining kidney CT volume percentage, and remaining kidney CT volume/weight, GFR of remaining kidney measured with DTPA were selected as significant factors for prediction models. Additionally, cystatin C based eGFR was selected for fair renal adaptation, while creatinine clearance and serum creatinine were selected for good renal adaptation. Areas under the receiver operating characteristic were 0.847 (95% confidence interval, 0.769–0.924) and 0.632 (0.546–0.718), and areas under the precision-recall curve were 0.967 (0.946–0.979) and 0.708 (0.656–0.781) for fair and good renal adaptation, respectively. An interactive web-application for clinical decision support system was entitled as "Renal Adaptation Prediction Tool prior to Op-eration (RAPTO)".
We developed a novel prediction model for renal adaptation after donation. The RAPTO may help clinical decision for eligibility of living kidney donation and selection of high-risk donors requiring more intense follow-up after donation.

Full Text Links
  • KJT
export Copy
  • Twitter
  • Facebook
Similar articles
Copyright © 2023 by Korean Association of Medical Journal Editors. All rights reserved.     E-mail: koreamed@kamje.or.kr