OBJECTIVES To elucidate the accuracy and inter-observer agreement of non-tuberculous mycobacterial lung disease (NTM-LD) diagnosis based on chest CT findings.

METHODS Two chest radiologists and two pulmonologists interpreted chest CTs of 66 patients with NTM-LD, 33 with pulmonary tuberculosis and 33 with non-cystic fibrosis bronchiectasis. These observers selected one of these diagnoses for each case without knowing any clinical information except age and sex. Sensitivity and specificity were calculated according to degree of observer confidence. Inter-observer agreement was assessed using Fleiss' κ values. Multiple logistic regression was performed to elucidate which radiological features led to the correct diagnosis.

RESULTS The sensitivity of NTM-LD diagnosis was 56.4 % (95 % CI 47.9-64.7) and specificity 80.3 % (73.1-86.0). The specificity of NTM-LD diagnosis increased with confidence: 44.4 % (20.5-71.3) for possible, 77.4 % (67.4-85.0) for probable, 95.2 % (87.2-98.2) for definite (P < 0.001) diagnoses. Inter-observer agreement for NTM-LD diagnosis was moderate (κ = 0.453). Tree-in-bud pattern (adjusted odds ratio [aOR] 6.24, P < 0.001), consolidation (aOR 1.92, P = 0.036) and atelectasis (aOR 3.73, P < 0.001) were associated with correct NTM-LD diagnoses, whereas presence of pleural effusion (aOR 0.05, P < 0.001) led to false diagnoses.

CONCLUSIONS NTM-LD diagnosis based on chest CT findings is specific but not sensitive.

KEY POINTS • Diagnosis of NTM-LD based on radiological findings showed high specificity. • Sensitivity of NTM-LD diagnosis was around 50 %. • Inter- observer agreement was moderate. • Identification of tree-in-bud pattern, consolidation and atelectasis led to correct diagnoses.