Construction of a Risk Model and Deep Learning Network Based on Patients with Active Pulmonary Tuberculosis and Pulmonary Inflammation

Most patients with active pulmonary tuberculosis (TB) are difficult to be differentiated from pneumonia (PN), especially those with acid-fast bacillus smear-negative (AFB) and interferon-γ release assay-positive (IGRA) results. Thus, the aim of the present study was to develop a risk model of low-cost and rapid test for the diagnosis of AFB IGRA TB from PN. A total of 41 laboratory variables of 204 AFB IGRA TB and 156 PN participants were retrospectively analyzed. Candidate variables were identified by t-statistic test and univariate logistic model. The logistic regression analysis was used to construct the multivariate risk model and nomogram with internal and external validation. A total of 13 statistically differential variables were compared between AFB IGRA TB and PN by false discovery rate (FDR) and odds ratio (OR). By integrating five variables, including age, uric acid (UA), albumin (ALB), hemoglobin (Hb) and white blood cell counts (WBC), a multivariate risk model with a concordance index (C-index) of 0.7 (95% CI: 0.61, 0.8) was constructed. The nomogram showed that UA and Hb acted as protective factors with an OR <1, while age, WBC and ALB were risk factors for TB occurrence. Internal and external validation revealed that nomogram prediction was consistent with the actual observations. Collectively, it was revealed that an integration of five biomarkers (age, UA, ALB, Hb and WBC) may be used to quickly predict TB in AFB IGRA clinical samples from PN.