Enhancing prediction of post-nephrectomy renal function in living donors: Integration of ai-measured kidney volumes with machine learning models
Eun-Ah Jo1,2, Sangil Min2, Ahram Han2, Jongwon Ha2, Juhan Lee3, Sangwan Kim4, Seonggong Moon5, Jinsung Kim5, Yong Chul Kim6.
1Surgery, Chung-Ang University Hospital, Seoul, Korea; 2Surgery, Seoul National University Hospital, Seoul, Korea; 3Surgery, Yonsei University College of Medicine, Seoul, Korea; 4Institute of Health Policy and Management, Seoul National University Hospital, Seoul, Korea; 5Radiation Oncology, Yonsei University College of Medicine, Seoul, Korea; 6Nephrology, Seoul National University Hospital, Seoul, Korea
This study aimed to create a predictive model to estimate 1-year post-nephrectomy eGFR in living kidney donors, incorporating baseline characteristics, lab results, and AI-measured kidney volumes. Data was collected from a multicenter retrospective cohort of 1219 living kidney donors. Different machine learning models were juxtaposed to statistical models and compared using mean absolute error (MAE), R-squared (R²), and root mean square error (RMSE). The inclusion of remnant kidney volume (RKV) was a pivotal factor, significantly enhancing the predictive accuracy of the 1-year post-nephrectomy estimated glomerular filtration rate (eGFR). Notably, the Poisson regression model demonstrated the best performance, with mean absolute error (MAE) and root mean square error (RMSE) notably lower in the test (MAE 7.30, RMSE 9.48) and external validation sets (MAE 7.43, RMSE 9.49), compared to other models. Further validation and prospective testing of the model in diverse populations are warranted to consolidate its utility and broaden its applicability in clinical settings.
[1] kidney transplantation
[2] living donors
[3] nephrectomy
[4] chronic kidney disease
[5] kidney cortex
