Analysis of the nonlinear relationship between hypothermic machine perfusion parameters and delayed graft function and construction of an optimized predictive model based on sampling algorithms
10.12464/j.issn.1674-7445.2025032
- VernacularTitle:低温机械灌注参数与移植物功能延迟恢复的非线性关系分析及基于采样算法优化预测模型的构建
- Author:
Boqing DONG
1
;
Chongfeng WANG
1
;
Yuting ZHAO
2
;
Huanjing BI
1
;
Ying WANG
1
;
Jingwen WANG
1
;
Zuhan CHEN
1
;
Ruiyang MA
1
;
Wujun XUE
1
;
Yang LI
1
;
Xiaoming DING
1
Author Information
1. Department of Kidney Transplantation, the First Affiliated Hospital of Xi'an Jiaotong University, Institute of Organ Transplantation of Xi'an Jiaotong University, Xi'an 710061, China.
2. .
- Publication Type:OriginalArticle
- Keywords:
Kidney transplantation;
Delayed graft function;
Hypothermic machine perfusion;
Restricted cubic spline;
Nonlinear relationship;
Nomogram;
Balanced sampling;
Serum creatinine
- From:
Organ Transplantation
2025;16(4):582-590
- CountryChina
- Language:Chinese
-
Abstract:
Objective To analyze the nonlinear relationship between hypothermic machine perfusion (HMP) parameters and delayed graft function (DGF) and optimize the construction of a predictive model for DGF. Methods The data of 923 recipients who underwent kidney transplantation from deceased donors were retrospectively analyzed. According to the occurrence of DGF, the recipients were divided into DGF group (n=823) and non-DGF group (n=100). Donor data, HMP parameters and recipient data were analyzed for both groups. The nonlinear relationship between HMP parameters and the occurrence of DGF was explored based on restricted cubic splines (RCS). Over-sampling, under-sampling and balanced sampling were used to address the imbalance in the proportion of DGF to construct logistic regression predictive models. The area under the curve (AUC) of each model was compared in the validation set, and a nomogram model was constructed. Results Donor BMI, cold ischemia time of the donor kidney, and HMP parameters (initial and final pressures, resistance, and perfusion time) were significantly different between the DGF and non-DGF groups (all P<0.05). The RCS analysis revealed a threshold-like nonlinear relationship between HMP parameters and the risk of DGF. Among the models constructed using different sampling methods, the balanced sampling model had the highest AUC. Using this model, a nomogram was constructed to stratify recipients based on risk scores. Recipients in the high-risk group had higher serum creatinine levels at 1, 6, and 12 months after kidney transplantation compared to those in the low-risk group (all P<0.05). Conclusions There is a nonlinear relationship between HMP parameters and the risk of DGF, and the threshold is helpful for organ quality assessment and monitoring of graft function after transplantation. The predictive model for DGF constructed on the base of balanced sampling algorithms helps perioperative decision-making and postoperative graft function monitoring of kidney transplantation.
