Korean J Transplant.  2023 Nov;37(Suppl 1):S219. 10.4285/ATW2023.F-8097.

Prediction of recipient renal function in living donor kidney transplantation using baseline characteristics and donor renal cortex volume

• Ko M
• Kim S
• Kim YC
• Jo EA
• Kim J
• Moon S
• Han A
• Ha J
• Min S

• KMID: 2551839
• DOI: http://doi.org/10.4285/ATW2023.F-8097

Abstract

Background
The accurate anticipation of posttransplant renal function in living donor kidney transplantation (LDKT) recipients based on pretransplant variables is challenging. In this study, we aimed to develop a predictive model for recipient’s posttransplant renal function using baseline characteristics of the donor and recipient.
Methods
We analyzed 870 adult LDKT cases from 2010 to 2020. To measure donor’s total kidney volumes, we utilized a commercial software (Oncostudio, Oncosoft Inc.). To measure cortex volume of kidney, we developed a three-dimensional U-Net structure based full automated model. Recipients best estimated glomerular filtration rate (eGFR) within 2 weeks after transplantation was the primary outcome of interest. Linear regression was explored to assess the relationship between recipient eGFR and factors, including the cortical volume of the donated kidney. Moreover, we explored multiple statistical methods to establish a reliable predictive model, with 90% of cases for training and 10% for testing.
Results
The mean cortex volume of transplanted kidneys was 115.1±21.8 mL. For external validation of our automated renal cortex segmentation model, an independent dataset from a separate institution, yielding impressive concordance metrics such as a Dice similarity coefficient of 0.97 and a Hausdorff distance 95% of 0.77 mm for cortical volumetry. The best posttransplant eGFR of recipients was 81.1±23.8 mL/min/1.73 m². Recipient and donor factors, such as weight, height, initial eGFR, and cortex volume of donated kidney were associated with recipients best eGFR posttransplant. Among various predictive models examined, the generalized additive model (GAM) had the best performance, evidenced by a mean absolute error of 10.53.
Conclusions
Our study underlines the utility of preoperative computed tomography-derived donor kidney cortex volume as a predictive determinant for recipient eGFR subsequent to LDKT. The GAM model exhibited feasible accuracy, enabling comprehensive eGFR prediction through integration of recipient and donor factors, inclusive of the donated kidneys cortex volume. Further real-world applications and external validations will advance this predictive model.