Introduction

Non-tuberculous mycobacteria (NTM) are ubiquitous organisms distributed throughout the environment [1]. The isolation of NTM from humans has been increasing in South Korea and globally [2, 3]. However, despite the fact that Mycobacterium tuberculosis (TB) is an obligate pathogen in humans, isolation of NTM does not necessarily have clinical significance. The most common site of NTM infection is the lung; NTM can cause both asymptomatic infection and symptomatic disease [1].

According to the guidelines proposed by the American Thoracic Society (ATS), diagnosis of NTM lung disease (NTM-LD) requires fulfilment of specified microbiological, clinical and radiological criteria and exclusion of other respiratory diseases [1]. Similarly, the British Thoracic Society (BTS) guidelines propose that NTM-LD can be diagnosed when cultures of sputum samples obtained at least 7 days apart from a patient whose chest radiograph is suggestive of mycobacterial infection and who may or may not have positive symptoms and signs [4].

Because NTM-LD symptoms cannot be differentiated from those of other respiratory diseases and it takes 2–3 weeks for most NTM to grow [1, 5, 6], NTM-LD is usually initially suspected from radiological findings. Although several studies have identified typical radiological findings of NTM-LD [7, 8], the accuracy of disease diagnosis based on radiological findings has not yet been evaluated. Moreover, it is not easy to differentiate NTM-LD from bronchiectasis without NTM infection or from pulmonary TB. Furthermore, bronchiectasis and NTM infection often coexist, making causality difficult to determine [1], and NTM-LD-associated changes on chest radiographs are frequently indistinguishable from those caused by pulmonary TB [4]. Therefore, the aim of this study was to elucidate the accuracy and inter-observer agreement of NTM-LD diagnosis based on chest computed tomography (CT) findings.

Materials and methods

Study design

Two chest radiologists (C.H.L and H.L) and two pulmonologists (Y.A.K and J.H.L), all of whom have >10 years of clinical experience in a tertiary referral hospital, retrospectively interpreted chest CT images from patients with NTM-LD according to the diagnostic criteria of the ATS/Infectious Diseases Society of America (ATS/IDSA) and BTS [1, 4], and those of patients with non-cystic fibrosis bronchiectasis [9] or culture-confirmed pulmonary TB. The ATS/IDSA criteria for diagnosing NTM-LD include (a) pulmonary symptoms with radiological abnormalities suggestive of NTM-LD, (b) exclusion of other diagnoses as appropriate; and (c) positive culture results from at least two sputum samples or at least one bronchial wash or lavage, or lung biopsy showing histopathological features of mycobacterial infection and positive culture for NTM. According to the BTS criteria, NTM-LD can be diagnosed when positive cultures from sputum samples are obtained at least 7 days apart from a patient whose chest radiograph suggests mycobacterial infection with or without symptoms and signs.

The four observers were blinded to all clinical information, including the specific diagnosis, except for age and sex. The interpretations of the four observers were analysed. The study protocol was approved and the requirement for informed patient consent waived by the Institutional Review Board of Seoul National University Hospital. The protocol was registered with ClinicalTrials.gov (NCT02340897).

Selection of patients

The target sensitivity and specificity of chest CT for the diagnosis of NTM-LD was set at 80 %. Setting the correlation between the four observers at 1.0, half the width of the correlation at 0.10, the minimum number of patients with NTM-LD and other respiratory diseases required was calculated using the following formula:

$$ \mathrm{n}=\frac{Z_{\alpha /2}^2\mathrm{p}\left(1-\mathrm{p}\right)\left\{\frac{1}{\mathrm{m}}+\left(1-\frac{1}{\mathrm{m}}\right)\rho \right\}}{w^2} $$

Where p is the expected sensitivity (or specificity), m the number of observers, ρ the correlation between observers (0 ≤ ρ ≤ 1), and W half the width of the confidence interval.

According to this formula, 62 patients with NTM-LD and 62 patients without NTM-LD were required. To achieve statistical power, an additional eight patients were added. Thus, we randomly selected 66 patients who had been diagnosed with NTM-LD between 1 July 2011 and 30 April 2014 from an NTM cohort that had participated in an earlier study [6] and comprised patients newly diagnosed with NTM-LD since 1 July 2011 at Seoul National University Hospital. In addition, we selected 33 age- and sex-matched patients with bronchiectasis that had been diagnosed between 1 January 2012 and 30 April 2014 from the non-cystic fibrosis bronchiectasis cohort of our previous study, which comprised patients newly diagnosed with bronchiectasis since 1 January 2012 at the same institution [6]. The diagnoses of these patients were based on low-dose CT findings. NTM infection had been excluded by two negative mycobacterial cultures of sputum at least 3 months apart. Lastly, 33 age- and sex-matched pulmonary patients diagnosed at our institution with TB between 1 July 2011 and 30 April 2014 were selected.

Image evaluation

All CT scans had been performed without injection of intravenous contrast media with one of the following four scanners (SOMATOM Sensation 16, Siemens, Forchheim, Germany; Brilliance 64, Philips Medical Systems, Eindhoven, The Netherlands; LightSpeed Ultra, GE Healthcare, Waukesha, WI, USA; Aquilion ONE, Toshiba Medical Systems, Otawara, Japan) before the initiation of specific treatment for each disease. CT scans were obtained with 1.0- to 2.5-mm collimation and images were reconstructed by using a medium sharp reconstruction algorithm with 3- to 5-mm thickness. Images were exported in DICOM format and forwarded to observers.

The observers evaluated: (a) the presence of bronchiectasis; (b) the severity of bronchiectasis; (c) the presence of cavitation; (d) the extent of cavitation (single or multiple cavities); (e) the presence of nodular and micro-nodular lesions; (f) tree-in-bud appearance; (g) consolidation; (h) ground-glass opacity; (i) atelectasis and (j) pleural effusion. The list above was modified from previous studies [10–12]. The severity of bronchiectasis was allocated a grade between 1 and 3 [13]. Micro-nodules were defined as opacities less than 3 mm in diameter [14]. Finally, the observers were requested to select a diagnosis of NTM-LD, bronchiectasis or pulmonary TB, and to allocate a degree of confidence for that specific diagnosis (definite, probable or possible).

Statistical analysis

The sensitivity, specificity, positive predictive value and negative predictive value for each disease were calculated using generalized estimating equations. The differences in these values between observers were measured using logistic regression. The degree of inter-observer agreement was assessed using Fleiss’ κ values as follows: κ < 0.0, poor; 0.0 < κ < 0.20, slight; 0.21 < κ < 0.40, fair; 0.41 < κ < 0.60, moderate; 0.61 < κ < 0.80, substantial and 0.81 < κ < 1.00, almost perfect [15]. Multiple logistic regression was performed to determine the CT features that led to the correct diagnosis of NTM-LD. A value of P < 0.05 was considered significant. All analyses were performed using SAS version 9.3 (Cary, NC, USA).

Results

Patient characteristics

Of the 66 patients with NTM-LD, 44 (66.7 %) were female and 22 (33.3 %) were male. Their median age was 64 years. The results of microbiological diagnosis were as follows: M. avium in 24 patients, Mycobacterium intracellulare in 21, M. abscessus sens stricto in nine, M. massiliense in five, M. fortuitum in three, M. kansasii in two, M. peregrinum in one and M. kyorinese in one. Of the 33 patients with bronchiectasis, 24 (72.7 %) were female. Their median age was 61 years. The patients have a previous history of measles (11 patients, 33.3 %), pulmonary TB (four patients, 12.1 %) and whooping cough (three patients, 9.1 %). Of the 33 patients with pulmonary TB, 20 (60.6 %) were female. The median age was 62 years. All of them had culture-confirmed pulmonary TB that was reported to be drug-susceptible (Table 1).

Table 1 Baseline characteristics of patients according to disease category
Full size table

CT findings for each disease

The four observers read a total of 132 chest CT scans and reported 528 CT findings (Table 2). Compared with patients with bronchiectasis, patients with NTM-LD tended to have more cavities, nodular lesions, micro-nodular lesions, tree-in-bud pattern and consolidation. The patients with NTM-LD had significantly less severe bronchiectasis than the bronchiectasis-only group (P < 0.001). Moreover, the most severe form of bronchiectasis (Grade 3) was less common in subjects with NTM-LD. The proportion of cases with atelectasis (P = 0.927) and ground-glass opacity (P = 0.106) were similar between these two groups.

Table 2 Presence of each finding on chest CTs reported by the four observers according to diagnosis
Full size table

The CT findings that favoured NTM-LD over pulmonary TB were the presence of bronchiectasis (P < 0.001) and atelectasis (P = 0.001). Cavities were less frequently observed in patients with NTM-LD than in those with pulmonary TB (P < 0.001). The distribution of micro-nodular lesion (P = 0.870), tree-in-bud pattern (P = 0.759), consolidation (P = 0.270), ground-glass opacity (P = 0.567) and pleural effusion (P = 0.663) was similar between NTM-LD and pulmonary TB (Fig. 1).

Fig. 1
figure 1

Chest CT images of patients with non-tuberculous mycobacterial (NTM) lung disease, pulmonary tuberculosis (TB) and non-NTM bronchiectasis showing the pattern of each disease. a Image of a 74-year-old woman with Mycobacterium intracellulare lung disease shows small nodules (arrow) and focal bronchiectasis (arrowhead). b Image of a 69-year-old woman with pulmonary TB shows cavity (arrow) and focal bronchiectasis (arrowhead). c Image of a 59-year-old woman with non-NTM bronchiectasis shows bilateral bronchiectasis (arrow)

Full size image

Accuracy of radiological diagnosis

The sensitivity of diagnosis of NTM-LD based on chest CT findings was 56.4 % (95 % confidence interval [CI] 47.9–64.7), whereas the sensitivity of diagnosis of pulmonary TB and bronchiectasis was 72.0 % (95 % CI 60.0–81.5) and 81.8 % (95 % CI 71.8–88.8), respectively. NTM-LD was misdiagnosed as bronchiectasis in 66 of the 264 readings by the four observers and as pulmonary TB in 49 of those readings. The specificity for NTM-LD was 80.3 % (95 % CI 73.1–86.0). Fifty-two readings misclassified pulmonary TB (42 readings) or bronchiectasis (ten readings) as NTM-LD (Table 3).

Table 3 Sensitivity, specificity, positive predictive value and negative predictive value of diagnosis based on chest CT for non-tuberculous mycobacterium lung disease (NTM-LD), pulmonary tuberculosis (TB) and bronchiectasis
Full size table

Sensitivity did not differ according to observer confidence: it was 57.0 % (95 % CI 43.6–69.5) for definite diagnoses, 56.1 % (45.6–66.1) for probable and 65.2 % (44.1–81.6) for possible (P = 0.706). In contrast, specificity increased with confidence: 95.2 % (87.2–98.2) for definite diagnoses, 77.4 % (67.4–85.0) for probable and 44.4 % (20.5–71.3) for possible (P < 0.001).

Inter-observer correlation

Observers 1 (pulmonologist), 2 (pulmonologist) and 3 (radiologist) demonstrated a similar degree of sensitivity (around 50 %) for an NTM-LD diagnosis. The specificities of these observers were also similar, ranging between 81.4 % and 89.5 %. The sensitivity of observer 4 (radiologist) (80.8 %) was higher but the specificity lower (49.2 %) than for the other observers. Consequently, the sensitivity (P < 0.001) and specificity (P < 0.001) of observer 4 differed significantly from those of the other observers. Moreover, the sensitivity for pulmonary TB (P = 0.014) and bronchiectasis (P = 0.004) and specificity for pulmonary TB (P = 0.007) and bronchiectasis (P = 0.002) between the four observers also differed significantly (Table 4).

Table 4 Each observer’s sensitivity and specificity for correct diagnosis based on chest CT according to actual diagnosis
Full size table

Agreements for each radiological finding between the four observers were as follows (κ values); atelectasis, 0.230; tree-in-bud pattern, 0.411; consolidation, 0.437; nodular lesion, 0.460; micro-nodular lesion, 0.612; cavity, 0.744; bronchiectasis, 0.857; and pleural effusion, 0.947. Finally, overall agreement for the final diagnosis between the four observers was moderate with a κ value of 0.527. Agreement was weaker for an NTM-LD diagnosis (κ value of 0.453) and stronger for pulmonary TB (κ value of 0.529) and bronchiectasis diagnoses (κ value of 0.607).

Agreement for final diagnosis between radiologists (κ value of 0.518) and pulmonologists (0.582) was moderate. Agreement between observer 1 (pulmonologist) and observer 4 (radiologist) was the weakest (κ value of 0.419), while the strongest agreement was found between observer 2 (pulmonologist) and observer 4 (radiologist) (κ value of 0.598).

Radiological characteristics associated with a correct diagnosis of NTM lung disease

Identification of tree-in-bud pattern (adjusted odds ratio [aOR] 6.24, 95 % CI 3.35–11.65, P < 0.001), consolidation (aOR 1.92, 95 % CI 1.04–3.53, P = 0.036) and atelectasis (aOR 3.73 95 % CI 1.81–7.66, P < 0.001) on chest CT were associated with correct NTM-LD diagnoses (Fig. 2). Furthermore, the severity of bronchiectasis had an impact on NTM-LD diagnosis, although the presence or absence of bronchiectasis did not alter the diagnosis. However, once the presence of bronchiectasis had been identified, less severe forms favoured correct diagnoses of NTM-LD as compared with the most severe form of bronchiectasis (Grade 1 vs. Grade 3, aOR 3.77, 95 % CI 3.23–7.71, P < 0.001; Grade 2 vs. Grade 3; aOR 3.58, 95 % CI 1.74–7.37, P < 0.001).

Fig. 2
figure 2

Chest CT images of patients with non-tuberculous mycobacterial (NTM) lung disease. a Image of 49-year-old woman with Mycobacterium intracellulare lung disease. All observers identified focal bronchiectasis with atelectasis (arrow) and tree-in-bud pattern (arrowhead) and correctly diagnosed NTM lung disease. b Chest CT image of 54-year-old man with M. avium lung disease. All observers identified bronchiectasis (arrow) and pleural effusion (arrowhead) and misdiagnosed it as pulmonary TB (one observer) or non-NTM bronchiectasis (three observers)

Full size image

Conversely, when observers identified pleural effusions, the possibility of misdiagnosing NTM-LD as bronchiectasis or pulmonary TB increased significantly (aOR 0.05, 95 % CI 0.01–0.24, P < 0.001; Table 5).

Table 5 Radiological features on chest CT that led to correct diagnoses of non-tuberculous mycobacterium lung disease
Full size table

Discussion

Differentiating NTM-LD, which is generally not infectious, from pulmonary TB is crucial, especially in South Korea where the annual incidence of TB was 97 per 100,000 people in 2013 [16] and the identification of NTM-LD is rapidly increasing [3, 17]. In addition, differentiation of NTM-LD from bronchiectasis not associated with NTM-LD is important but challenging. The prevalence of NTM-LD among subjects with bronchiectasis is known to range from 10 % to 34 % [18, 19].

Previous studies have identified the radiological features of NTM-LD. Compared with pulmonary TB, NTM-LD tends to involve the middle and/or lower lung zones and bilateral lung more frequently [20], but forms cavities less frequently [21]. Compared with NTM-negative bronchiectasis, NTM-LD more frequently shows mucus plugging, cavities and bilateral involvement [18, 19].

Although guidelines stress the importance of radiological findings for the diagnosis of NTM-LD, the results of our study suggest that chest CT findings are not completely reliable for differentiating NTM-LD from pulmonary TB or bronchiectasis. The sensitivity for NTM-LD diagnoses (56.4 %) was lower than for pulmonary TB (72.0 %) or bronchiectasis (81.8 %), whereas the specificity for diagnosing these three disease entities was similar. In addition, inter-observer agreement was lower for an NTM-LD diagnosis (κ value = 0.453) than for pulmonary TB (κ value = 0.529) or bronchiectasis (κ value = 0.607) diagnoses. These observations suggest that diagnosing NTM based on chest CT findings is trickier than diagnosing pulmonary TB or bronchiectasis.

Our study suggests several predictors for correctly diagnosing NTM-LD based on chest CT findings. If pleural effusion is identified, the possibility of a correct NTM-LD diagnosis decreases significantly (aOR 0.05, 95 % CI 0.01–0.24, P < 0.001). The four observers reported the presence of pleural effusion in 15 (5.7 %) of 256 readings of 66 chest CTs from patients with NTM-LD and misdiagnosed seven of them as pulmonary TB and six as bronchiectasis. It has previously been believed that pleural effusion is rarely present in NTM-LD [20]. However, a recent study reported identifying pleural effusion in 13.3 % of patients with NTM-LD [22]; it is thus more common than previously believed [20]. Observers’ misconceptions regarding the rarity of pleural effusion among patients with NTM-LD may partially explain the unsatisfactory sensitivity of NTM-LD diagnosis.

Conversely, when observers identified tree-in-bud (centrilobular branching) appearance, consolidation and atelectasis, they tended to make a correct NTM-LD diagnosis. Centrilobular branching opacities and lobular consolidation are reportedly more prevalent in NTM-LD than in NTM-negative bronchiectasis [18]. Accordingly, the presence of tree-in-bud pattern and consolidation may help observers to differentiate NTM-LD from NTM-negative bronchiectasis. It is interesting that the presence of atelectasis was associated with correct NTM-LD diagnoses in our study because the presence of atelectasis has been reported to be similar among NTM-LD, pulmonary TB and bronchiectasis cases [18, 23]. While the presence of bronchiectasis did not have an impact on NTM-LD diagnosis, a less severe degree of bronchiectasis was associated with a correct diagnosis once bronchiectasis had been identified. Indeed, the most severe bronchiectasis (Grade 3) was less common in NTM-LD (19.9 %) than in NTM-negative bronchiectasis (42.3 %).

Despite the unsatisfactory sensitivity for the diagnosis of NTM-LD, overall specificity was 80.3 %. This varied with likelihood of diagnosis of NTM-LD based on chest CT findings, increasing with observer confidence as follows: 44.4 % for possible diagnoses, 77.4 % for probable and 95.2 % for definite. Diagnosis of NTM-LD requires fulfilment of radiological criteria [1]. Given that a confirmatory test must have high specificity [24], the present results indicate that radiological diagnosis based on the features on chest CT that we have identified could be used to confirm a diagnosis of NTM-LD, especially when the observer’s confidence is high.

Overall agreement for the final diagnosis between the four observers was moderate (κ value = 0.527), with observer 4 showing major discrepancy. Because the kappa values for radiological findings ranged from 0.230 to 0.947, the discrepancy of observer 4 is attributable to missing some radiographic findings or different interpretation of the findings that were identified.

In this study, we did not classify NTM-LD into upper lobe cavitary disease and nodular/bronchiectatic disease in the enrolment or in the analysis. This could be recognized as a limitation of the study. Our rationale is that a clear distinction between the two types of disease is not always possible. For example, cavities frequently exist in NTM-LD with bronchiectasis [18].

In conclusion, diagnoses of NTM-LD based on chest CT findings were inaccurate. With moderate inter-observer agreement, the sensitivity for the diagnoses of NTM-LD was only around 50 % and the specificity was similar to that of pulmonary TB or non-cystic fibrosis bronchiectasis. However, the stronger an observer’s confidence was, the higher the specificity of the observer’s diagnosis of NTM-LD.