Development and validation of a blood-based diagnostic model for pulmonary tuberculosis combining GBP5 expression and routine laboratory indicators.
Zhao Miaomiao M, Hu Qiuxiang Q, Wang Qing Q, Cha Xinlang X et al.
WHO reports that only 54% of tuberculosis (TB) patients received rapid diagnostic tests at their initial presentation. Conventional laboratory methods in TB detection have high specificity (>90%) but low sensitivity (<50%). This means a large number of tuberculosis patients are missed. A better diagnostic approach is essential to decrease the TB burden. Studies on multi-indicator blood-based models for rapid TB diagnosis remain limited. We developed a rapid, non-sputum-based model to improve the efficiency of TB diagnosis. This was a retrospective case-control study including 301 patients with active tuberculosis (ATB) and 191 patients with other pulmonary diseases (OPD) who were evaluated at the Department of Pulmonary Medicine of the Affiliated Infectious Diseases Hospital of Soochow University between May 2023 and May 2024. A composite clinical diagnosis served as the gold standard for ATB diagnosis, including either a clinical diagnosis or bacteriological confirmation. The diagnostic outcome of ATB (ATB vs OPD) was defined as the dependent variable, while guanylate-binding protein 5 (GBP5) expression levels and routine laboratory indicators (including blood cell counts and plasma protein measurements) were defined as independent variables. Univariate and multivariate logistic regression analyses were performed to identify key predictors, and an AdaBoost algorithm was used to construct a TB diagnostic model. The performance of the model was compared with traditional laboratory-based TB tests. DeLong's Test was used to evaluate the statistical difference of AUC between AdaBoost and traditional methods. Through univariate and multivariate logistic regression analyses, the GBP5 gene, white blood cell (WBC) count, platelet (PLT) count, and prothrombin time (PT) were selected to construct an ATB diagnosis model using the AdaBoost algorithm. The AdaBoost model achieved an AUC of 0.808 in the training set and 0.805 in the test set, while the AUC of smear microscopy, MTB culture, Xpert MTB/RIF, and IGRA were 0.67, 0.59, 0.63 and 0.75 respectively. Therefore, our model demonstrated better diagnostic performance than conventional methods. This study preliminarily demonstrates that our AdaBoost diagnostic model based on GBP5 gene expression and routine blood indicators could serve as a potential tool for the clinical diagnosis of ATB. However, to be applied to the clinic, large samples are needed to validate the performance of the model.