病原学阴性初治肺结核患者诊断模型的建立及初步评价

目的 建立病原学阴性初治肺结核患者的诊断模型, 以提高临床诊断的规范性及准确性。方法 收集 2017 年 8 月至 2019 年7 月湖南省胸科医院收治的年龄≥15岁、结核分枝杆菌病原学检查阴性, 且从未因结核病应用过抗结核药物治疗或不规范化疗未满1个月已确诊的200例病原学阴性初治肺结核患者和212例非结核性肺部疾病患者的临床资料。采用简单随机数字表法将两组患者分为两类, 一类为训练样本, 一类为考核样本, 两类样本比例为3∶1。通过对训练样本临床观察指标进行单因素和多因素非条件logistic回归分析, 筛选出独立预测因子并建立诊断模型, 并以训练样本的临床诊断结果为金标准, 采用受试者工作特征(ROC)曲线评估模型的诊断能力, 再使用考核样本对建立的诊断模型进行前瞻性考核。结果 对训练样本临床观察指标的多因素非条件logistic回归分析显示, 血清白蛋白/球蛋白比值(A/G)> 1.2、血清糖类抗原-125(CA-125)> 35kU/L、全血结核分枝杆菌γ-干扰素释放试验(IGRA)阳性、咳嗽≥2周、有咯血或者痰中带血、全身结核中毒症状、并发其他疾病、病灶发生于单侧或两侧的肺上叶尖后段和(或)下叶背段和空洞形成是病原学阴性初治肺结核与非结核性肺部疾病鉴别的相关独立因素Wald χ ²= 7.264, P= 0.007, OR(95%CI)= 3.433(1.400~8.417); Wald χ ²= 17.114, P= 0.000, OR(95%CI)= 6.980(2.780~17.524); Wald χ ²= 63.643, P= 0.000, OR(95%CI)= 19.283(9.320~39.894); Wald χ ²= 5.557, P= 0.018, OR(95%CI)= 0.434(0.217~0.869); Wald χ ²= 6.237, P= 0.013, OR(95%CI)= 0.255(0.087~0.745); Wald χ ²= 24.930, P= 0.000, OR(95%CI)= 0.126(0.056~0.284); Wald χ ²= 12.062, P= 0.001, OR(95%CI)= 10.139(2.743~37.746); Wald χ ²= 16.224, P= 0.000, OR(95%CI)= 4.428(2.147~9.135); Wald χ ²=16.228, P=0.000, OR(95%CI)=14.437(3.939~52.919)]。ROC曲线分析模型的诊断效能为曲线下面积(AUC)=0.881(95%CI:0.839~0.915), 最佳临界值为0.42, 诊断病原学阴性初治肺结核的敏感度和特异度分别为86.67%和76.10%。经考核样本检验, 建立的诊断模型对两组患者总的准确判别率为79.61%(82/103)。结论 本研究建立的模型敏感度和准确率均较高, 可以作为临床医师诊断病原学阴性初治肺结核的辅助参考工具。

Establishment and preliminary evaluation of a diagnostic model for the new patients with pathogen-negative pulmonary tuberculosis

Objective To establish a diagnostic model for new patients with pathogen-negative pulmonary tuberculosis and improve the standardization and accuracy of clinical diagnosis. Methods From August 2017 to July 2019, the case information of 200 new patients with pathogen-negative pulmonary tuberculosis and 212 patients with non-tuberculous pulmonary disease who had been confirmed in Hunan Chest Hospital were collected. Inclusion criteria of patients were as follows: (1) age ≥15 years old; (2) pathogen-negative Mycobacterium tuberculosis; (3) never been treated with antituberculosis drugs or nonstandard chemotherapy for less than one month. These patients were divided into training and verification groups by a simple random number table, with the proportion of 3∶1. Univariate and multivariate unconditional logistic regressions were used to analyze main clinical observation indicators of training samples, and then the independent predictors were screened out and the diagnostic model was established. The receiver operating characteristic (ROC) curve analysis was used to evaluate the diagnostic performance of the model, taking the clinical diagnosis results of the training samples as the gold standard. A prospective assessment of the established diagnostic model was performed using the validation samples. Results Multivariate unconditional logistic regression analysis showed that serum albumin/globulin ratio (A/G)>1.2 (Wald χ ²=7.264, P=0.007, OR (95%CI)=3.433 (1.400-8.417)), serum carbohydrate antigen-125 (CA-125)>35 kU/L (Wald χ ²=17.114, P=0.000, OR (95%CI)=6.980 (2.780-17.524)), positive interferon-gamma release assay (IGRA) for whole-blood Mycobacterium tuberculosis (Wald χ ²=63.643, P=0.000, OR (95%CI)=19.283 (9.320-39.894)), coughing time ≥2 weeks (Wald χ ²=5.557, P=0.018, OR (95%CI)=0.434 (0.217-0.869)), hemoptysis or bloody phlegm (Wald χ ²=6.237, P=0.013, OR (95%CI)=0.255 (0.087-0.745)), poisoning symptoms of tuberculosis (Wald χ ²=24.930, P=0.000, OR (95%CI)=0.126 (0.056-0.284)), complicating with other diseases (Wald χ ²=12.062, P=0.001, OR (95%CI)=10.139 (2.743-37.746)), lesion site in the posterior segment of the upper lobe tip and/or the dorsal portion of the lower lobe on one or both sides of the lung (Wald χ ²=16.224, P=0.000, OR (95%CI)=4.428 (2.147-9.135)), and cavity formation (Wald χ ²=16.228, P=0.000, OR (95%CI)=14.437 (3.939-52.919)) were independent factors in the identification of pathogen-negative pulmonary tuberculosis and non-tuberculous pulmonary disease. The ROC curve analysis showed that area under the ROC curve (AUC) was 0.881 (95%CI: 0.839-0.915), the optimal cut-off value was 0.42, and the sensitivity and specificity for the diagnosis of pathogen-negative pulmonary tuberculosis were 86.67% and 76.10%, respectively. The total accurate discrimination rate of the diagnostic model in the validation group was 79.61% (82/103). Conclusion The diagnostic model shows high sensitivity and accuracy, and can be used as a reference tool for clinicians to diagnose new patients with pathogen-negative pulmonary tuberculosis.