基于归一化回归算法的多参数模型用于原发性肝癌微小血管侵犯预测的研究  

Multi-parameters model based on normalized regression algorithm in prediction of microvascular invasion of primary hepatocellular carcinoma

作　　者：王孜怡 肖潇 黄晨军 童林 高春芳 曹宏伟 WANG Ziyi;XIAO Xiao;HUANG Chenjun;TONG Lin;GAO Chunfang;CAO Hongwei(Department of Information,Changhai Hospital,Naval Military Medical University,Shanghai 200438,China;Clinical Laboratory Medicine Center,Yueyang Hospital of Integrated Traditional Chinese and Western Medicine,Shanghai University of Traditional Chinese Medicine,Shanghai 200437,China;Department of Clinical Laboratory,Shanghai Eastern Hepatobiliary Surgery Hospital,Shanghai 200438,China)

机构地区：[1]海军军医大学附属长海医院信息科,上海200438 [2]上海中医药大学附属岳阳中西医结合医院检验实验中心,上海200437 [3]上海东方肝胆外科医院检验科,上海200438

出　　处：《国际检验医学杂志》2022年第24期2973-2976,共4页International Journal of Laboratory Medicine

基　　金：上海市卫健委肿瘤分子医学协同创新集群(2019CXJQ03);上海市科委重点项目(17JC1404500)。

摘　　要：目的基于归一化的甲胎蛋白(AFP)、甲胎异质体(AFP-L3)、异常凝血酶原Ⅱ(PIVKAⅡ)检测数据建立逻辑回归(LR)模型用于预测原发性肝癌微小血管侵犯(MVI)。方法纳入手术后证实为原发性肝细胞癌(HCC)的患者1314例,使用Python进行数据集7∶3(建模组∶验证组)的划分,收集入组患者术前AFP、AFP-L3、PIVKAⅡ检测结果和术后MVI分级信息,检测数据在(-1,1)区间归一化,建立LR模型并在建模组和验证组评价模型预测MVI的效能。结果建模组区分HCC患者是否发生MVI的曲线下面积为0.647,虽然与单独采用AFP指标的相同,但在验证组中,采用LR模型诊断MVI的曲线下面积为0.720,高于AFP、AFP-L3、PIVKAⅡ单独使用的诊断效能。结论基于临床常用的AFP、AFP-L3、PIVKAⅡ3项肝癌标志物的数据归一化和LR建模,可辅助临床预测HCC是否发生MVI,对于HCC患者的精准施治和临床预后判断有积极意义。Objective To establish a logistic regression(LR)model based on the normalized detection results of AFP,AFP-L3 and PIVKAⅡfor the predicting of microvascular invasion(MVI)in primary hepatocellular carcinoma(HCC).Methods A total of 1314 cases of HCC verified by operation after entering the group used the Python to conduct the data set 7∶3(modeling group∶verification group)division.The detection results of AFP,AFP-L3 and PIVKAⅡbefore operation,and postoperative MVI classification information were collected from the patients entering the group.The detection data were normalized in the section of(-1,1).The LR model was constructed.The efficacy of the model for predicting MVI was evaluated in the modeling group and verification group.Results The area under curve(AUC)of the model in distinguishing whether MVI occurring was 0.647,although which was similar to that of independently adopting AFP indicator alone;but in the verification group,AUC of adopting LR model for diagnosing MVI was 0.720,which was higher than the diagnostic efficacy of AFP,AFP-L3 and PIVKAⅡused alone.Conclusion The normalization of data based on clinical common 3 liver cancer markers(AFP,AFP-L3 and PIVKAⅡ)and LR modeling could assist in clinical prediction whether MVI occurring in HCC,which has an important significance for the precision treatment and clinical prognosis in the patients with HCC.

关 键 词：微小血管侵犯 归一化 逻辑回归 模型 原发性肝癌 

分 类 号：R735.7[医药卫生—肿瘤]