Abstract:

Objective To explore the risk factors related to delayed graft function （DGF） after kidney transplantation using machine learning algorithms and to establish a predictive model. Methods Clinical data of kidney transplant donors and recipients and pathological data of donor kidney biopsy from January 2018 to December 2020 at the Institute of Organ Transplantation of Tongji Hospital Affiliated to Tongji Medical College， Huazhong University of Science and Technology were collected. The contribution of factors related to DGF were calculated through greedy algorithm，and logistic regression predict model were fitted using the reduced data set. Performance of the models is assessed by accuracy and area under the receiver operating characteristic curve （AUROC）. Results The observed incidence of DGF was 21.9%. Risk factors that were highly correlated with DGF include donor body type， blood urea nitrogen， cold ischemia time， extent of artery lesions， as well as tubular atrophy score（ct） and interstitial fibrosisscores （ci）. The prediction model was established using the above tested factors，the AUROC of the predict model was 0.71， and the prediction accuracy was 0.73. Conclusion Machine learning algorithms can be used to analyze the risk factors of DGF occurrence and establish predictive models. 

Key words:

 , Kidney transplantation； Logistic regression； Delayed graft function； Predict model

摘要：

目的 利用机器学习算法探究与移植肾功能延迟恢复（delayed graft function，DGF）发生相关的风险因素并建立预测模型。方法 收集 2018 年 1 月至 2020 年 3 月华中科技大学附属同济医院器官移植研究所实施的公民逝世后捐献供肾和肾移植受者的临床资料以及供肾活检病理资料，通过贪心算法筛选与 DGF发生相关的因素的贡献度，再利用逻辑回归算法建立预测模型并利用模型精确度，受试者工作特性曲线下面积（area under the receiver operating characteristic curve，AUROC）对模型效果进行评估。结果 术后 DGF 的发生率为 21.9%。与术后 DGF 发生相关性较高的因素包括供者体型、末次尿素氮、冷缺血时间、供器官小动脉病变范围、慢性肾小管萎缩评分（ct）和慢性间质纤维化评分（ci）。使用上述因素建立预测模型，模型的AUROC 约为 0.71，预测准确率约为 0.73。结论 利用机器学习算法可以分析 DGF 发生的风险因素并建立预测模型，以供临床预测 DGF 的发生风险。 

关键词:

肾移植 , 逻辑回归 , 移植肾功能延迟恢复 , 预测模型