结核病与肺部健康杂志 ›› 2018, Vol. 7 ›› Issue (1): 23-28.doi: 10.3969/j.issn.2095-3755.2018.01.006

• 论著 • 上一篇    下一篇

基于大数据建立临床肺结核综合诊断数学模型的研究

温保江,周正华,梁志勇,李军,温文沛(),郭卉欣   

  1. 511500 广东省清远市慢性病防治医院结核病区(温保江、周正华、梁志勇、李军);广东省结核病控制中心门诊部(温文沛、郭卉欣)
  • 收稿日期:2018-03-10 出版日期:2018-03-30 发布日期:2018-07-24
  • 通信作者: 温文沛 E-mail:568323856@qq.com

A study on mathematical model of comprehensive clinicl diagnosis of tuberculosis based on big data

Bao-jiang WEN,Zheng-hua ZHOU,Zhi-yong LIANG,Jun LI,Wen-pei WEN(),Hui-xin. GUO   

  1. Department of Tuberculosis, Chronic Disease Prevention and Control Hospital of Qingyuan City, Guangdong Province,Qingyuan 511500,China
  • Received:2018-03-10 Online:2018-03-30 Published:2018-07-24
  • Contact: Wen-pei WEN E-mail:568323856@qq.com

摘要:

目的 基于大数据建立临床肺结核综合诊断数学模型,以提高临床诊断肺结核的质量。方法 从广东省清远市慢性病防治医院结核病区和呼吸病区每个月(2016年1月至2017年12月)按照分层随机抽样方法选取临床检查项目资料完整的患者15~20例,共计345例。诊断初治活动性肺结核患者198例(肺结核组),非肺结核患者147例(非肺结核组,含陈旧性肺结核)。收集所有患者临床资料,包括胸部X线摄影(胸片)表现(病灶范围、有无空洞)、结核菌素皮肤试验(PPD试验)、血白细胞(WBC)计数、血清白蛋白、痰涂片(萋尔-尼尔逊染色镜检法)、痰培养(改良罗氏液体培养)、痰GeneXpert MTB/RIF检测(简称“GeneXpert检测”)、结核感染T细胞斑点试验(T-SPOT.TB)检查结果,以及临床症状体征(咳嗽、咯血、体质量下降、纳差)和相关病史(结核病病史、吸烟史、糖尿病病史)等临床资料信息,应用logistic回归模型以及判别分析等统计学方法进行研究。方法 通过非条件logistic回归模型分析确定具有独立鉴别性的临床检查项目,包括性别[χ2=6.047,P=0.014;OR(95%CI)=0.410(0.201~0.834)]、结核病病史[χ2=29.273,P=0.000;OR(95%CI)=0.086(0.036~0.210)]、血WBC计数[χ2=10.266,P=0.001;OR(95%CI)=0.771(0.657~0.904)]、痰GeneXpert检测[χ2=18.633,P=0.000;OR(95%CI)=7.280(2.956~17.928)]、痰培养[χ2=10.400,P=0.001;OR(95%CI)=10.021(2.469~40.664)]、T-SPOT.TB检测[χ2=23.669,P=0.000;OR(95%CI)=5.769(2.848~11.688)]、PPD试验[χ2=6.462,P=0.011;OR(95%CI)=1.503(1.098~2.057)]、体质量下降[χ2=7.807,P=0.005;OR(95%CI)=3.335(1.433~7.761)];其中GeneXpert检测、痰培养、T-SPOT.TB检测、体质量下降等项目的OR值均>2,对诊断肺结核具有明显优势。建立的肺结核诊断函数模型,对非肺结核的判别准确率达83.67%(123/147)、对肺结核的判别准确率达79.29%(157/198)。 结论 本研究建立的肺结核的数学诊断模型判别准确率高,为临床医师临床诊断肺结核及降低误诊率提供了新的工具。

关键词: 结核, 肺, 计算机通信网络, 诊断, 鉴别, 模型, 统计学, 数据说明, 统计, 评价研究

Abstract:

Objective This study on mathematical model of comprehensive clinicl diagnosis of tuberculosis (TB) based on big data is aimed to improve the clinical diagnosis of tuberculosis. Methods A total of 345 patients from the Chronic Disease Prevention and Control Hospital of Qingyuan City between January 2016 and December 2017 were selected using stratified random sampling method.They were divided into TB group (n=198, initial therapy for active TB) and no-TB (n=147, including old TB).Results of X-ray,PPD, WBC, serum albumin, sputum smear, sputum culture, GeneXpert MTB/RIF test and T-SPOT.TB, clinical symptoms and medical history were analyzed using logistic regression model and discriminant analysis.Results After detecting sexual (χ2=6.047,P=0.014;OR (95%CI)=0.410 (0.201-0.834)), TB history (χ2=29.273,P=0.000; OR (95%CI)=0.086 (0.036-0.210)), WBC (χ2=10.266, P=0.001; OR (95%CI)=0.771 (0.657-0.904)), GeneXpert MTB/RIF (χ2=18.633, P=0.000; OR (95%CI)=7.280 (2.956-17.928)), sputum culture (χ2=10.400, P=0.001; OR (95%CI)=10.021 (2.469-40.664)), T-SPOT.TB (χ2=23.669, P=0.000; OR (95%CI)=5.769 (2.848-11.688)), PPD (χ2=6.462, P=0.011; OR (95%CI)=1.503 (1.098-2.057))and weight loss (χ2=7.807, P=0.005; OR (95%CI)=3.335 (1.433-7.761)), it was found that all ORs of GeneXpert MTB/RIF,sputum culture, T-SPOT.TB and weight loss were above 2, which was significantly help to diagnosis of TB. Function model of TB diagnosis is useful, the accuracy in differential diagnosis of no-TB was 83.67% (123/147), and the accuracy in TB diagnosis was 79.29% (157/198). Conclusion Function model of TB diagnosis in this study is of high accuracy in differential diagnosis, it help to clinical diagnosis of TB and could reduce the misdiagnosis rate.

Key words: Tuberculosis, pulmonary, Computer communication networks, Diagnosis, differential, Models, statistical, Data interpretation, statistical, Evaluation studies