Establishment and Validation of a Predictive Model for Non-tuberculous Mycobacterial Infections in Acid-fast Bacilli Smear-positive Patients

Abstract

Background: To identify non-tuberculous mycobacteria (NTM) infections in initially acid-fast bacilli (AFB) smear-positive patients more quickly and effectively, we attempted to establish a predictive model and validate it in clinical practice.Methods: A total of 125 AFB smear-positive patients with mycobacteriosis diagnosed in the Respiratory Department of Shanghai Pulmonary Hospital were retrospectively analyzed，including 64 cases of NTM and 61 cases of pulmonary tuberculosis (PTB), to identify the clinical features distinguishing NTM from PTB patients. A bivariate regression was then set up to determine the independent predictive risk factors to NTM infections. A receiver operating characteristic (ROC) curve was used to determine the model’s predictive discrimination. Finally, the model was verified both internally and externally for its predictive ability.Results: Compared with PTB, female patient, the symptom of hemoptysis, the exists of bronchiectasis, and negative test for QuantiFERON tuberculosis (QFT) were more common in NTM patients (78.1% vs. 34.4%, 37.5% vs. 11.5%, 76.6% vs. 19.7%, 82.8% vs. 8.2%, respectively, P≤0.001). The lesions on chest high-resolution computed tomography (HRCT) in NTM patients were more commonly involved in the right middle lobe (87.5% vs. 52.5%, P<0.001) and left lingular lobe (71.9% vs. 50.8%, P=0.016), and cystic change was more commonly morphological alteration (32.8% vs. 8.2%, P=0.001). Binary regression analysis showed that female patient, the exists of bronchiectasis, negative test for QFT and right middle lobe lesion were independent risk factors for NTM in AFB smear-positive patients (P<0.05). A ROC curve combining the four risk factors showed a sensitivity and specificity of 85.9% and 93.4%, respectively, and the area under the curve (AUC) was 0.963. Moreover, internal validation and external validation of the later clinical data both confirmed the effectiveness of the model.Conclusions: The predictive model would be useful for early differential diagnosis of NTM in initially AFB smear-positive patients.