基于机器学习的肾移植受者术后抑郁的预测模型建立  

Development of a depression prediction model for renal transplant recipients based on machine learning

作　　者：李斐雯 李萍[2] 白春花[1] LI Fei-wen;LI Ping;BAI Chun-hua(不详;Department of Surgery,Nanjing Drum Tower Hospital,the Affiliated Hospital of Nanjing University Medical School,Nanjing,Jiangsu 210008,China)

机构地区：[1]南京大学医学院附属鼓楼医院外科,江苏南京210008 [2]南京大学医学院附属鼓楼医院泌尿外科,江苏南京210008

出　　处：《中国临床研究》2022年第8期1168-1172,共5页Chinese Journal of Clinical Research

基　　金：国家自然科学基金(82172639)。

摘　　要：目的基于机器学习算法,构建终末期肾病患者肾移植术后抑郁症的预测模型,并验证该模型的预测效果。方法选取2015年2月至2019年3月南京某三甲医院泌尿外科行肾移植手术的189例患者为模型建立对象,根据术后1年的随访结果将其分为抑郁组(n=74)和非抑郁组(n=139)。使用过采样技术对数据不平衡处理后,以0.75∶0.25的比例将数据集划分为训练集和测试集。选用6种常见机器学习算法在训练集上进行10倍交叉验证,根据AUC值选出表现最好的算法在完整训练集上训练,于测试集上评估,并筛选出预测相关重要特征。结果经10倍交叉验证后随机森林算法被用作最终模型的建立。在测试集上,随机森林模型准确率、灵敏度和特异性分别为80.6%、83.3%和78.1%。贡献最大的前五个特征为肌酐、护理模式、收入、总蛋白和白蛋白。结论本研究建立的随机森林模型预测效果良好,可为患者肾移植术后抑郁症的预测提供借鉴。此外,护理模式在模型中较高的贡献也间接表明护理工作在改善此类患者心理健康方面的重要性。Objective Based on machine learning algorithm,to develop a depression predict model in patients with endstage renal disease after renal transplantation and verify the prediction effect of the model.Methods A total of 189patients undergoing renal transplantation in the Urology Department of a tertiary A-level hospital in Nanjing from February 2015 to March 2019 were selected for the model development.According to the one-year follow-up results after operation,they were divided into depression group(n=74)and non-depression group(n=139).After oversampling of the raw imbalanced data,the dataset was randomly stratified to training set and testing set in the ratio of 0.75∶0.25.Using six commonly used machine learning algorithms,10-fold cross-validation on the training set was performed to test out the most suitable algorithm based on the AUC value and to evaluate them on the test set,screening out the important features related to prediction.Results After 10-fold cross-validation,the random forest algorithm was selected for the development of the final model.On the testing set,the accuracy,sensitivity,and specificity of the final model were80.6%,83.3%and 78.1%respectively.The top five features with the highest contribution of prediction model were creatinine,care mode,income,total protein and albumin.Conclusions The random forest model for depression prediction among the renal transplant recipients shows good effect and provides valuable information for specialists to refer to.Additionally,collaborative nursing mode is important in improving the mental health of such patients.

关 键 词：肾移植 抑郁 机器学习 随机森林 协同护理模式 

分 类 号：R473.6[医药卫生—护理学]