Construction of etiological diagnosis model for pathogen-negative pulmonary tuberculosis using tuberculosis scores of GBP5, DUSP3, and TBP genes combined with inflammatory factors
10.3760/cma.j.cn112150-20250501-00382
- VernacularTitle:GBP5和DUSP3及TBP三基因TB评分联合炎症因子构建病原学阴性肺结核鉴别诊断模型
- Author:
Miaomiao ZHAO
1
;
Yanyang ZHOU
;
Qiuxiang HU
;
Hui CHEN
;
Tingting CHEN
;
Yingqi CHEN
;
Ping XU
Author Information
1. 苏州大学苏州医学院，苏州 215123
- Publication Type:Journal Article
- Keywords:
Pathogen-negative pulmonary tuberculosis;
Three-gene tuberculosis score;
Inflammatory factors;
Discriminative diagnostic model
- From:
Chinese Journal of Preventive Medicine
2025;59(11):1965-1971
- CountryChina
- Language:Chinese
-
Abstract:
To evaluate the diagnostic performance of a three-gene (GBP5, DUSP3, and TBP) tuberculosis (TB) score in bacteriologically-negative pulmonary tuberculosis, and to develop and validate a discriminative diagnostic model by integrating inflammatory cytokines (IL-2, IL-5, IL-17, and IFN-γ). A prospective cohort study was conducted, a total of 238 patients admitted to the Affiliated Infectious Disease Hospital of Soochow University from May 2023 to May 2024 were enrolled, including 119 pathogen-negative pulmonary tuberculosis patients and 119 patients with other pulmonary diseases (OPD). The GeneXpert MTB-HR kit was used to detect the three-gene TB scores from residual blood routine samples. The diagnostic performance was assessed using receiver operating characteristic (ROC) curve analysis. Concurrent data on 12 inflammatory cytokines were collected from patients. Potential biomarkers were screened using univariate analysis and multivariate logistic regression, and selected features were incorporated into the construction of four machine learning models: logistic regression, support vector machine (SVM), K-nearest neighbors (KNN), and adaptive boosting (AdaBoost). The samples were randomly split into a training set (85%) and a test set (15%). The models were trained on the training set, and their diagnostic performance was validated using the test set. The predictive ability of each model was evaluated based on ROC curve parameters. The results showed that the three-gene TB score alone yielded an AUC of 0.539 (sensitivity: 50.94%, specificity: 60.50%) in distinguishing pathogen-negative pulmonary tuberculosis from OPD. Four non-col-linear inflammatory factors (IL-2, IL-5, IL-17, and IFN-γ) were selected and combined with the three-gene TB score to construct machine learning models. The AdaBoost model demonstrated the best performance, achieving an AUC of 0.893 (sensitivity: 85.4%, specificity: 73.0%) in the training set and an AUC of 0.873 (sensitivity: 88.2%, specificity: 72.2%) in the test set. In conclusion,the AdaBoost diagnostic model integrating the three-gene TB score with inflammatory factors (IL-2, IL-5, IL-17, and IFN-γ) exhibits superior discriminating performance for pathogen-negative pulmonary tuberculosis compared to OPD, significantly outperforming the three-gene TB score alone.
