Objectives: Determining the clinical significance of nontuberculous mycobacterium (NTM) isolation from respiratory specimens remains challenging. No studies have yet investigated the clinical application of electronic nose (eNose) in this field.

Methods: Patients with at least two sets of Mycobacterium avium complex, M. abscessus complex or M. kansasii isolation from respiratory specimens were recruited and their breath samples were analyzed by eNose. Those whose NTM status (pulmonary disease [PD] or colonization [PC]) could be ascertained by the expert panel were classified into the NTM status-confirmed cohort. The others (at least one expert had different opinion) were classified into the NTM status-unconfirmed cohort. For them, re-assessment of clinical status was performed six months later. Three approaches, one support vector machine (SVM) and two convoluted neural network models (SimVGGNet-5 and SimResNet-5), were applied. Model trained on status-confirmed cohort (training: validation = 8:2) was then tested on the status-unconfirmed cohort.

Results: A total of 106 participants whose NTM status was confirmed and 27 whose NTM status was initially unconfirmed were included. The area under receiver-operating characteristic curve (AUC) of eNose on NTM status confirmed cohort of the three models was 0.87, 0.97, and 0.90, respectively.. For the validation cohort, the AUC of eNose of the three models was 0.68, 0.74, and 0.73, respectively. Applying to the NTM status-unconfirmed cohort, the eNose has an accuracy of 0.70, 0.59, and 0.56, respectively.

Conclusions: eNose can differentiate NTM clinical status and predict outcome among patients with clinical uncertainty. Breathomics could serve as a potential non-invasive clinical assessment modality in NTM management.