摘要
目的利用临床诊断决策树的方法建立结核性胸膜炎（tuberculouspleurisy，TBP）综合诊断策略，并探讨其对TBP的诊断价值。方法采用回顾性研究方法。根据自行制定的结核性胸膜炎及恶性胸腔积液的人组标准，收集首都医科大学附属北京胸科医院2014年1月至2015年12月符合入组标准的住院患者病历资料，共314例，分为TBP组（205例）和恶性胸腔积液组（109例）。采用随机数的生成方法将综合数据按照3：1比例，分为训练样本数据集及验证样本数据集，然后进行决策树算法（CART）分析，生成结核性胸膜炎的临床诊断决策树，最后将生成的决策树模型对验证样本数据集进行验证并计算出检测效应值，以完成对TBP综合诊断策略的验证。结果对25项用于构建临床诊断决策树的指标进了单因素统计分析显示，其中有16项指标在TBP组和恶性胸腔积液组差异有统计学意义。以独立构建的临床诊断决策树工作流程为基础，进行1000次模拟实验，全部循环构建的1000棵决策树平均利用的评判因素为（8．57±1．63）个。对实验结果进行相关评价指标的计算，结果显示临床诊断决策树用于TBP诊断的敏感度为98．14％，特异度为93．64％，符合率为95．01％。对决策树中各项指标的贡献得分排序显示，排名前9项指标依次是胸腔积液腺苷脱氨酶、血红细胞沉降率、发热、胸腔积液c反应蛋白、年龄、血结核抗体、血T细胞斑点试验B、性别、乏力等。结论临床诊断决策树方法是TBP与恶性胸腔积液有效鉴别诊断策略之一。
Objective To form comprehensive diagnosis strategy of tuberculous pleurisy （TBP） using the decision tree in the clinical, and to evaluate the value of decision tree in diagnosis of TBP. Methods Based on inclusion criteria of TBP and malignant pleural effusion, 314 patients from Beijing Chest Hospital affiliated to Capital Medical University between January 2014 to December 2015 were retrospectively studied. These patients were divided into TBP group （205 cases） and malignant pleural effusion group （109 cases）. And the comprehensive data were randomly divided into the training sample data set and validation sample data set according with the ratio of 3 .. 1 using random number statistics, and then a diagnosis tree for clinical diagnosis of TBP were builded, which was used to verify validation sample data and calculate detection effect value, in order to verify the comprehensive diagnosis strategy. Results A single factor statistical analysis was made on 25 indexes of constructing decision tree in the clinical and it was found that, 16 indexes were statistically significant between TBP group and malignant pleural effusion group. A total of 1000 simulated experiments were carried out based on the clinical decision tree, the average utilization of the 1000 decision trees constructed by the whole cycle is 8.57 ± 1.63. According to the algorithm of relevant evaluation indexes of the experimental results, the sensitivity for diagnosis of TBP by the constructed clinical decision tree was 98.14% and the specificity was 93.64%, the accuracy was 95.01%. According to the contribution of the indexes in the decision tree, the front 9 indexes were the pleural effusion, erythrocyte sedimentation rate, fever, pleural effusion C-reactive protein, age, blood tuberculous antibody, blood T lymphocyte spot test B, sex and fatigue, etc. Conclusion The clinical decision tree is one of the effective methods for differentiating TBP from malignant pleural effusion.
出处
《结核病与胸部肿瘤》
2016年第4期243-249,共7页
Tuberculosis and Thoracic Tumor
关键词
结核
胸膜
诊断
决策树
Tuberculosis, pleural
Diagnosis
Decision trees
