基于多检验变量和机器学习算法的结肠癌诊断模型建立及价值评估  

Establishment and value assessment of colon cancer diagnostic models based on multiple variables and different machine learning algorithms

作　　者：梁永媛[1] 蔡培飞[2] 郑桂喜[1] LIANG Yongyuan;CAI Peifei;ZHENG Guixi(Department of Clinical Laboratory,Qilu Hospital of Shandong University,Jinan 250012,Shandong,China;Department of Blood Transfusion,Affiliated Hospital of Jining Medical College,Jining 272000,Shandong,China)

机构地区：[1]山东大学齐鲁医院检验科,山东济南250012 [2]济宁医学院附属医院输血科,山东济宁272000

出　　处：《山东大学学报（医学版）》2024年第2期51-59,共9页Journal of Shandong University:Health Sciences

基　　金：国家自然科学基金(82172347);山东省科技创新重大专项(2021CXGC010603)。

摘　　要：目的 采用不同机器学习算法,建立基于多检验变量的结肠癌诊断模型,并评估其临床应用价值。方法 收集119例结肠癌患者(结肠癌组)和125例健康对照(健康对照组)的血清样本,提取血清外泌体,采用RT-qPCR方法测定miR-214-3p分子在两组中的表达水平,进而绘制受试者工作特征(receiver operating characteristic, ROC)曲线,评估其对结肠癌的诊断效能。同时,收集结肠癌组和健康对照组的常规检验项目结果。将以上指标均纳入研究筛选出特征性变量,采用11种不同算法结合ROC曲线和机器学习曲线综合评价筛选出最优算法,建立结肠癌诊断模型。结果 结肠癌组血清外泌体中miR-214-3p的表达水平明显高于健康对照组(P<0.001),其诊断结肠癌的ROC曲线下面积(area under curve, AUC)为0.820,具有较好的诊断效能。将结肠癌组和健康对照组的血清外泌体miR-214-3p及30种常规检验指标纳入后,筛选出尿素、癌胚抗原、单核细胞计数、外泌体miR-214-3p共4个特征性变量,且逻辑回归算法是建立机器学习模型的最优算法,其AUC为0.93,且学习曲线呈现很好的拟合状态。结论 血清外泌体miR-214-3p是结肠癌的潜在标志物,基于4个特征性变量和逻辑回归算法建立的机器学习模型对结肠癌有良好的诊断效能。Objective To establish a colon cancer diagnostic model based on multiple variables using various machine learning algorithms and to assess its clinical application value.Methods Serum samples from 119 colon cancer patients and 125 healthy controls were collected.Serum exosome was extracted,and miRNA 214-3p(miR-214-3p)level was measured using RT-qPCR.Receiver operating characteristic(ROC)curve was plotted to evaluate the diagnostic efficiency of colon cancer.Additionally,30 routine laboratory items of colon cancer patients and healthy controls were collected.Characteristic variables were screened,and 11 algorithms were used to establish the diagnostic model.The optimal model was selected with ROC and machine learning curves.Results The expression level of miR-214-3p in colon cancer patients was significantly higher than that in healthy controls(P<0.001),with the area under the ROC curve(AUC)being 0.820,indicating good diagnostic performance.After the expression level of miR-214-3p and other 30 routine laboratory items were enrolled,4 characteristic variables were screened to establish the diagnostic model,including UREA,carcinoembryonic antigen,monocyte and miR-214-3p.The Logistic regression algorithm was identified as the optimal one(AUC=0.93).Conclusion Serum exosome miR-214-3p is a potential biomarker of colon cancer.The model based on 4 characteristic variables and Logistic regression algorithm has an excellent diagnostic performance for diagnosing colon cancer.

关 键 词：结肠癌 miR-214-3p 外泌体 机器学习 

分 类 号：R783[医药卫生—口腔医学]