决策树分析在提高血清维生素D、CA153、CA125、CEA对乳腺癌辅助诊断作用中的应用分析  

Application of decision tree analysis in the improvement of serum vitamin D,CA153,CA125 and CEA in the assisted diagnosis of breast cancer

作　　者：尹志辉 周志伟[2] 刘继勇[2] 杨保昌[2] 王凌凌[3] YIN Zhihui;ZHOU Zhiwei;LIU Jiyong;YANG Baochang;WANG Lingling(Department of Clinical Laboratory,the Xingtai Third Hospital,Xingtai,Hebei 054000,China;Department of Clinical Laboratory;Department of Oncology,the Central Hospital of China National Petroleum Corporation,Langfang,Hebei 065000,China)

机构地区：[1]河北省邢台市第三医院检验科,河北邢台054000 [2]中国石油天然气集团公司中心医院检验科,河北廊坊065000 [3]中国石油天然气集团公司中心医院肿瘤科,河北廊坊065000

出　　处：《国际检验医学杂志》2019年第20期2482-2485,共4页International Journal of Laboratory Medicine

基　　金：廊坊市科技支撑计划项目(2016013025)

摘　　要：目的探讨血清指标建立决策树模型在乳腺癌患病诊断中的应用。方法选取2016年8月至2018年4月在中国石油天然气集团公司中心医院就诊的由组织病理学确诊的新发乳腺癌患者60例,作为乳腺癌组;选取同期体检非肿瘤、非生殖内分泌系统疾病的女性患者60例,作为非乳腺癌组;再选取本研究外首次就诊乳腺科患者60例组成验证组,其中乳腺癌30例,非乳腺癌30例。检测两组患者肿瘤标志物糖链抗原153(CA153)、CA125、癌胚抗原(CEA)和25-羟基维生素D,首先对两组患者检测结果进行统计学分析,然后对所有检测数据建立决策树模型,分析各指标在辅助诊断乳腺癌中作用,收集验证组患者相关指标应用受试者工作曲线(ROC曲线)对模型的临床诊断效能进行评价。结果(1)乳腺癌组未绝经患者维生素D水平低于对照组,乳腺癌组绝经患者CEA水平高于非乳腺癌组患者,差异具有统计学意义(P<0.05);其他指标差异无统计学意义(P>0.05);(2)进行自动建模分析得出,决策树100.0%预测能力最佳模型为CHAID决策树,该模型显示当维生素D水平小于或等于7.2 ng/mL时,对乳腺癌阳性预测达到100.0%;当维生素D水平大于7.2 ng/mL且小于或等于19.0 ng/mL时,加入是否绝经变量,对于未绝经患者阴性预测能力76.0%,绝经患者阳性预测能力65.7%;当维生素D水平大于19.0 ng/mL时,且CEA结果小于或等于2.48 U/mL时阴性预测能力87.9%,且CEA结果大于2.48 U/mL时阳性预测能力100.0%;通过ROC曲线分析得出决策树模型较多指标串联分析在筛查特异度方面提升明显。结论决策树模型分析可通过对检验指标递进分类,以树形式组成不同判断集合,增强了检测指标在乳腺癌中的辅助诊断价值。Objective To explore the application of decision tree model of serum indicators in the diagnosis of breast cancer.Methods From August 2016 to April 2018,60 newly diagnosed breast cancer patients diagnosed by histopathology were selected as breast cancer group,and 60 female patients with non-neoplastic and non-reproductive endocrine diseases were selected as non-breast cancer group.A total of 60 patients who first visited the department of mammary gland were included in the validation group,including 30 cases of breast cancer and 30 cases of non-breast cancer.Tumor markers(CA153,CA125,CEA)and 25-hydroxyvitamin D3 were detected in two groups.First,the results of two groups of patients were analyzed statistically.Then,a decision tree model was established for all detection data,and the role of each index in assistant diagnosis of breast cancer was analyzed.Relevant indicators of the validation group were collected and applied to the work of the subjects.The ROC curve was used to evaluate the clinical diagnostic efficacy of the model.Results(1)Vitamin D level of non-menopausal patients in breast cancer group was lower than that of control group,CEA level of menopausal patients in breast cancer group was higher than that of non-breast cancer group(P<0.05);there was no significant difference in other indicators(P>0.05);(2)Through automatic modeling analysis,100%prediction ability of decision tree was the best.The model is CHAID decision tree,which shows that when vitamin D≤7.2 ng/mL,the positive prediction of breast cancer reaches 100%;when 7.2 ng/mL<vitamin D≤19.0 ng/mL,whether to add menopausal variables,the negative predictive ability of premenopausal patients is 76.0%,and the positive predictive ability of menopausal patients is 65.7%;When the level of vitamin D>19.0 ng/mL,and the CEA≤2.48 U/mL,the negative predictive ability was 87.9%,and the positive predictive ability was 100%when the CEA>2.48 U/mL.Through the analysis of ROC curve,it was concluded that the multi-index tandem analysis of decision tree mode

关 键 词：决策树分析 乳腺癌 维生素D 诊断价值 

分 类 号：R737.9[医药卫生—肿瘤]