Latent class analysis to define radiological subgroups in pulmonary nontuberculous mycobacterial disease

Pulmonary non-tuberculous mycobacterial disease is a challenging infection which is associated with a high mortality [1‐3]. Treatment is frequently poorly tolerated, expensive, and in many cases response rates are poor [4]. Radiology is essential for the diagnosis of disease and has important implications in guiding treatment. NTM-pulmonary disease has been observed to commonly fall into two clinico-radiological patterns of disease, fibrocavitary and nodular-bronchiectatic disease, and this classification forms the basis of treatment recommendations in guidelines [5] and has important prognostic implications [6‐8]. Whilst being of great clinical utility this division may not capture the spectrum of radiological appearances, some of which may not fit clearly into either group [3, 9‐11].

The aim of this study was to characterise the radiological features of NTM-pulmonary disease using an NTM-specific scoring system, to use latent class analysis (LCA) as an unbiased method to identify subgroups sharing common patterns of radiological features, and to examine the clinical characteristics associated with any such patterns.

Individuals with NTM-pulmonary disease were recruited prospectively from the outpatient department of the Royal Brompton Hospital and Chelsea and Westminster Hospital between September 2012 and November 2013. All met American Thoracic Society 2007 disease criteria [5]. Subjects were excluded if they had a diagnosis of cystic fibrosis, HIV infection or other primary or secondary immunodeficiency or if they were receiving any immunosuppressant medication other than oral prednisolone. A HRCT scan of the thorax was performed in subjects who had not undergone such a scan in the previous 6 months. In those who declined additional imaging as part of the study their most recent clinical HRCT was obtained. Subjects also underwent a full clinical assessment, blood and sputum sampling, St George’s Respiratory Questionnaire (SGRQ) and lung function testing. Written consent was gained from all participants and the study was approved by the local Research Ethics Committee (reference 12/LO/1034). Further details are provided in the supplementary materials.

CT scoring was performed as previously described [12] by a specialist radiologist with 5 years of experience in thoracic imaging, blinded to clinical details. For bronchiectasis extent, bronchiectasis severity, tree-in-bud opacification, nodules and consolidation, tertiles of the maximum possible score for each feature were used to categorise the features as low, medium or high. Cavitating nodules, severe cavitation and aspergilloma scores were dichotomised as present or absent. As a measure of the overall severity of an individual’s radiological disease burden, composite CT scores were calculated for each participant by expressing the individual scores for bronchiectasis extent, bronchiectasis severity, tree-in-bud opacification, nodules, consolidation, cavitating nodules, severe cavitation and aspergilloma as a percentage of the maximum possible score for that feature, then summing the scores together. The scoring proforma is detailed in Additional file 1: Table S1 of the supplementary material.

Statistical analysis was performed in the R environment version 3.4.0 [13]. Latent class analysis (LCA) was performed on the matrix of CT scores using the fpc and flexmix packages [14, 15]. LCA is a statistical method which takes a set of multivariate data and uses this to identify groups of related subjects (‘latent classes’) within the data which share similar characteristics. The Aikake information criteria (AIC) was used to identify the optimum number of classes in the model as the number giving the lowest value of AIC [16]. Composite CT scores were compared between groups using analysis of variance (ANOVA). Clinical characteristics were compared between groups using ANOVA, Tukey’s Honestly Significant Difference test and Fisher’s exact test. Survival curves were generated using the survival package [17] and compared using the Log-rank test. Correction for multiple testing was performed using the Benjamini-Hochberg method.

To examine the reproducibility of the groups identified by LCA and their radiological characteristics, the analysis was repeated on an independent cohort from another tertiary centre. Subjects with NTM-pulmonary disease at Papworth Hospital who underwent CT scanning at as part of their clinical care were identified retrospectively, their anonymised imaging and clinical data were retrieved and data analysed using the exact methodology used for the original cohort.

A total of 85 subjects were recruited, their clinical and radiological characteristics are shown in Additional file 1: Tables S2 and S3 of the supplementary material. Twelve subjects declined further imaging but had undergone HRCT for clinical purposes which were obtained for analysis. The median interval between imaging and recruitment was 23 days. In 80 (94%) subjects a HRCT was obtained within twelve months of recruitment.

In this study, LCA identified three groups with distinct radiological characteristics in two independent cohorts of subjects with NTM-pulmonary disease. The finding of compatible radiological changes is a cornerstone of the diagnosis of NTM-pulmonary disease [5]. Historically the radiological pattern associated with NTM-pulmonary disease was one of cavitation similar to tuberculosis, typically seen in older males with underlying lung disease such as emphysema [18]. In the later twentieth century the first reports were published of a pattern of disease characterised by nodules and bronchiectasis which in contrast occurred in those with no underlying lung disease and was more common in women [19, 20]. The differentiation between the former ‘fibrocavitary’ disease and latter ‘nodular-bronchiectatic’ disease is of major clinical importance, as several studies have consistently demonstrated an association between fibrocavitary disease and mortality [6‐8], disease progression [7] and treatment failure [21, 22]. This has been reflected in successive guidelines which recommend more aggressive treatment of fibrocavitary disease [5]. Other patterns of disease have been reported, in particular consolidation and infiltrative patterns have been associated with poor outcomes [9‐11] however with no agreed consensus regarding definitions for these patterns, the findings remain unique to the individual studies. Even when an additional ‘consolidative’ category was used, one study found that 27% of subjects still did not clearly fit any category [3] and another was still unable to classify the radiological appearances of 229/481 subjects [9].

Our current study used a predefined scoring system to reduce subjectivity in assessing NTM related radiological changes, and latent class analysis as an unbiased method to identify groups sharing radiological characteristic without presupposing the existence of any specific patterns of disease.

The findings support the existence of fibrocavitary disease as a distinct pattern, identifying a group characterised by severe cavitation associated with markers of disease severity and a high mortality. There was a high prevalence of aspergilloma within this group, which has been associated with increased mortality in NTM-pulmonary disease [12].

Within the remaining majority of subjects, LCA identified two distinct subgroups, split evenly into those with very few radiological changes apart from nodules, and those with marked bronchiectasis. As may be expected, significantly more subjects in the latter Bronchiectatic group has pre-existing bronchiectasis, although this was also the case in over a third of those with ‘nodular’ disease. Interestingly despite having the best-preserved lung function and lowest symptom scores, there was no significant difference in mortality between the nodular and bronchiectatic groups. The duration of disease was significantly shorter in the nodular group, raising the possibility that this represents early disease which may progress to Cavitary or Bronchiectatic forms. A pattern of nodules and bronchiectasis have been associated with M. avium complex infection in a single study [23] however there were no differences in species prevalence between groups.

The validation cohort confirms the presence of three subgroups whose radiological characteristics shared many similarities with the Brompton cohort. There was a smaller group characterised by high composite CT scores with severe cavitation, the common presence of aspergillomas and severe and extensive bronchiectasis, corresponding to the Cavitary group. The other subjects were divided into those dominated by severe and extensive bronchiectasis without nodules, cavitation or aspergilloma (corresponding to the Bronchiectatic group) and those with low composite CT scores and low scores for most CT features (corresponding to the Nodular group). A higher proportion of subjects fell into the Bronchiectatic group, likely due to the higher overall prevalence of underlying bronchiectasis in this cohort.

The only major difference to the Brompton cohort was the higher prevalence of tree-in-bud changes, which were moderate or severe in 76% compared with 17%. The prevalence was higher in all three disease subgroups and does not appear to be due to the increased prevalence of underlying bronchiectasis. As the scoring of the validation cohort was performed separately to the Brompton cohort observer bias may be responsible for this difference, although this was performed by the same individual, alternatively it may be that differences in treatment, coinfection, or other clinical factors between centres is responsible. There was also more severe cavitation and aspergilloma in the nodular Papworth group compared with the Brompton, however only four subjects accounted for this difference.

In contrast to other studies mentioned previously, the presence of a separate consolidative pattern of NTM-pulmonary disease was not seen in this study. The vast majority of subjects in both cohorts had only mild consolidation, although this feature was slightly more prominent in the Cavitary group. Only one subject had consolidation and no other feature of NTM disease. Our findings may in part be explained by the absence of immunocompromised or critically ill subjects from our cohorts. When the analysis was restricted to subjects with known underlying bronchiectasis only two subgroups were identified, with the Bronchiectatic group being split in half dividing the cohort into groups more consistent with the conventional ‘fibrocavitary’ and ‘nodular-bronchiectatic’ patterns. However, in the validation cohort the three subgroups were still clearly identified.

The principle limitation of the study is the heterogeneous nature of the cohort, comprising subjects with differing underlying respiratory diseases, NTM species, antibiotic treatments and infecting co-pathogens. Nevertheless the findings were replicable in an independent cohort suggesting they are not merely related to the specific case mix at a single centre. Furthermore, such a mix of subjects is representative of real-life clinical practice, although as both cohorts come from tertiary centres the proportion of subjects with severe bronchiectasis and aspergillomas may be higher than the wider population of individuals with NTM-pulmonary disease. A number of subjects had previously undergone HRCT and declined further imaging, therefore in some subjects the collected clinical data may not reflect the point in time when the HRCT was performed and in 10% of cases more than a year had elapsed from imaging to study enrolment. A single radiologist was responsible for performing CT scoring and despite the use of an objective scoring system this remains a potential source of bias, although the scoring system was based on the Bhalla score [24] which has been shown to have low interobserver variability [25].

This study provides validation of the existence of cavitary disease as a distinct phenotype of NTM-pulmonary disease associated with a poor prognosis. In addition they suggest that ‘nodular-bronchiectatic’ disease is formed of two separate groups with important radiological and clinical differences, which may possibly reflect differences in underlying lung disease or duration of infection. These data underline that fact that NTM-pulmonary disease is a heterogeneous condition and more precise phenotyping will be valuable in clinical decision making and stratification in clinical trials.