Abstract
Background Acute exacerbation of idiopathic interstitial pneumonias (AE-IIPs) induces permanent pulmonary dysfunction and is potentially lethal. The unpredictable occurrence of AE-IIPs remains an important clinical issue in the management of IIPs.
Methods In this multicentre, retrospective, observational study, a predictive score for AE-IIPs was designed using clinical factors based on multivariate Fine–Gray analysis in patients with IIPs.
Results Based on multivariate Fine–Gray analysis in an exploratory cohort of 487 patients with IIPs, the predictive score for AE-IIPs was determined as follows: 1 point each was added for honeycombing on high-resolution computed tomography (H), age >75 years (A) and lactate dehydrogenase level >222 U·L−1 (L); the total score ranged from 0 to 3 (HAL score). The HAL score discriminated the risk of AE-IIPs with a C-index of 0.62 (95% CI 0.56–0.67); this discrimination was verified in a validation cohort of 402 patients with IIPs with a C-index of 0.67 (95% CI 0.60–0.73). In a combined cohort, the estimated cumulative risks for AE-IIPs at 1, 2, 3, 5 and 10 years were 1.9%, 3.5%, 5.1%, 7.7% and 12.9%, respectively, in the total score 0 group; 4.7%, 8.3%, 12.0%, 17.7% and 28.4%, respectively, in the total score 1 group; and 8.0%, 14.2%, 19.7%, 28.7% and 43.0%, respectively, in the total score ≥2 group. Subgroup analysis revealed that the HAL score was applicable to patients with and without idiopathic pulmonary fibrosis.
Conclusions The HAL score discriminated the risk of AE-IIPs and could aid in the management of IIPs.
Abstract
A predictive model using existence of radiographic honeycombing, age >75 years and serum lactate dehydrogenase level >222 U·L−1 discriminated the risk of acute exacerbation in patients with idiopathic interstitial pneumonias. https://bit.ly/3YvSkrc
Background
Acute exacerbation of idiopathic interstitial pneumonias (AE-IIPs) has been defined as acute respiratory deterioration (with a duration typically <1 month) accompanied by new widespread alveolar abnormalities (bilateral ground-glass opacity and/or consolidation), in the absence of an alternative explanation such as cardiac failure or fluid overload [1, 2]. The reported incidence of AE-IIPs in patients with idiopathic pulmonary fibrosis (IPF) is ∼10% per year [3–7]. Acute exacerbation also occurs in patients with non-IPF IIPs or secondary interstitial lung diseases (ILDs) [8–13]. Although the incidence and prognosis of acute exacerbations vary among IIPs, AE-IIPs is a potentially lethal event in any IIP. Even when patients avoid a fatal outcome, they often experience permanent pulmonary dysfunction because of pulmonary fibrosis caused by AE-IIPs.
The aetiology of AE-IIPs is unclear and its unpredictable occurrence remains an important clinical issue in the management of IIPs. Risk factors for AE-IIPs have been identified, including low forced vital capacity (FVC); low diffusing capacity of the lung for carbon monoxide (DLCO); poor baseline oxygenation; radiological honeycombing; and/or increased levels of blood lactate dehydrogenase (LDH), Krebs von Lungen (KL)-6 or surfactant protein (SP)-D [4, 6–9, 14–17]. However, the predictive abilities of these factors have varied between studies and/or IIP entities. There are no established predictors for AE-IIPs. Because IIPs are a group of heterogeneous disease entities, the accurate prediction of AE-IIPs cannot be achieved using a single factor.
In contrast, scoring models with multiple clinical factors can provide clinical utility in predicting the overall survival of patients with IIPs. For example, the GAP index, composed of gender (i.e. sex), age and physiology (% predicted FVC and DLCO), has demonstrated good discriminative accuracy for the prognosis of IPF [18–20]. The model weights the survival risk of clinical factors using multivariate analyses to eliminate potential confounding factors; this facilitates the accurate prediction of survival among patients with IPF. However, because the model was designed for the prediction of overall survival in patients with IPF, it is inappropriate for predicting AE-IIPs.
The present study was conducted to establish a scoring model for predicting AE-IIPs in patients with IIPs. We identified risk factors for AE-IIPs using clinical data, determined the optimal combination of risk factors using multivariate analyses and scored the selected factors according to their risk ratio. Furthermore, we verified the utility of the scoring model in another patient cohort.
Material and methods
Study design
This multicentre, retrospective, observational study followed the ethical standards of the Declaration of Helsinki. The study protocol was approved by the institutional review boards of Hamamatsu University School of Medicine (Hamamatsu, Japan, approval 20-173), Seirei Hamamatsu General Hospital (Hamamatsu, Japan, approval 3472) and Seirei Mikatahara General Hospital (Hamamatsu, Japan, approval 20-45). The requirement for informed consent was waived because of the retrospective observational design.
Patients
The medical records of consecutive patients who were diagnosed with IIPs at the participating institutions between January 2001 and December 2020 were retrospectively analysed. The diagnosis of IIP was performed in accordance with American Thoracic Society/European Respiratory Society guidelines [21–23]. Only treatment-naïve patients were included, because prior use of medications for IIPs might have influenced some predictive factors. Patients were excluded if they had ILDs secondary to known causes (e.g. collagen vascular diseases, sarcoidosis or hypersensitivity pneumonia); if they received steroids, immunosuppressants or antifibrotic agents for the treatment of IIPs at the first visit; and/or if they exhibited acute exacerbation at the first visit. Patients in Seirei Hamamatsu General Hospital and Seirei Mikatahara General Hospital were regarded as the exploratory cohort; patients in Hamamatsu University Hospital were regarded as the validation cohort.
Data collection
The following clinical data were collected at the time of IIP diagnosis: age, sex, pack-year smoking history, history of dust exposure, laboratory data, pulmonary function and clinical and pathological diagnoses (pathological diagnosis was recorded only if performed). High-resolution computed tomography (HRCT) scans were obtained at three centres using various scanners at full inspiratory and spine position, with slice thicknesses from 1 to 3 mm and slice intervals from 2.5 to 10 mm. All images were evaluated at standard window settings for visualisation of the lung parenchyma (window level of −600 HU and window width of 1500 HU). Using HRCT images collected at the time of diagnosis, the presence of emphysema and honeycombing was evaluated by two experienced general radiologists (S. Ichikawa and N. Yoshizawa) who were blinded to all other patient data. The definitions of emphysema and honeycombing were established in accordance with the Fleischner Society guidelines [24]. Disagreements concerning the presence of emphysema and honeycombing were resolved by consensus decision in collaboration with a third radiologist (S. Goshima). Autoimmune features were recorded according to the diagnostic criteria for interstitial pneumonia with autoimmune features [25]. The definition of AE-IIPs was based on an international working group report [1]. Additional details regarding the data correction are provided in the supplementary methods.
Statistical analysis
The time to the first AE-IIPs and overall survival were measured from the time of IIP diagnosis. Gray's test was used to analyse the time to the first AE-IIPs. Fine–Gray analysis was performed using clinical data to identify risk factors for AE-IIPs. Death prior to acute exacerbation was treated as a competing risk event.
Variables with p<0.1 in univariate analyses were entered into multivariate analyses. When two variables exhibited strong correlations with each other (Pearson's correlation coefficient >0.7), only one of the two was selected for multivariate analysis to avoid multicollinearity. Stepwise selection of variables for the predictive model was performed using the Akaike information criterion, Bayesian information criterion and p-value-based methods. In Fine–Gray analysis, laboratory data (continuous variables) were converted into dichotomous categorical variables with cut-off values based on their upper normal limits.
To develop prediction scores for AE-IIPs, hazard ratios from the multivariate Fine–Gray analysis in the exploratory cohort were converted to logarithms, integral-multiplied and rounded to the nearest integer [26]. The Harrell C-index was used in the Fine–Gray model to assess prediction model performance; 95% confidence intervals were obtained from bootstrap resampling of 2000 replicates. Time-dependent positive predictive values and negative predictive values were calculated using the Fine–Gray model. The predicted and observed cumulative incidence of AE-IIPs and overall survival at 1, 2, 3, 5 and 10 years were calculated using the Fine–Gray model. The predictive accuracy was verified in the validation cohort.
All values were analysed using R 4.1.1 (R Foundation for Statistical Computing, Vienna, Austria). Additional details regarding the statistical analysis are provided in the supplementary methods.
Results
Patient characteristics
In the exploratory and validation cohorts, 549 and 453 patients with IIPs, respectively, were screened for inclusion in the study. 62 and 51 patients, respectively, were excluded because of unavailable data (no HRCT imaging data n=25 and n=18; no spirometric data n=24 and n=22; no laboratory data n=2 and n=2) and the use of medications for IIPs at their first visit (n=11 and n=9); thus, 487 and 402 patients, respectively, were included in the analysis (figure 1). The clinical characteristics of the study cohorts are presented in table 1. Both cohorts consisted mainly of men; more than half of the patients were aged >70 years; and ∼30% of the patients had a history of smoking. Both cohorts generally had comparable demographic characteristics; however, the validation cohort had a significantly greater proportion of patients with a history of dust exposure (p<0.001) and significantly lower percentage FVC, percent predicted forced expiratory volume in 1 s and percentage DLCO values, compared with the values of the exploratory cohort (p<0.001, p<0.001 and p=0.031, respectively). In the exploratory and validation cohorts, 42.9% and 52.7% of the patients, respectively, had FVC <80%. Slightly more than 30% of patients had emphysema and slightly more than 40% of patients had honeycombing on chest HRCT images; these proportions were comparable between the two cohorts.
The proportions of types of IIPs were generally comparable between the two cohorts, except the validation cohort had a significantly greater proportion of patients with cryptogenic organising pneumonia (p=0.008). In the combined cohort, 317 (35.7%), 55 (6.2%), 55 (6.2%), 39 (4.4%) and six (0.7%) patients were diagnosed with IPF, nonspecific interstitial pneumonia, pleuroparenchymal fibroelastosis, cryptogenic organising pneumonia and desquamative interstitial pneumonia/respiratory bronchiolitis-associated interstitial lung disease, respectively. The remaining 417 (46.9%) patients had unclassifiable IIPs and slightly fewer than 10% of the patients had autoimmune features.
No patients had received steroids, immunosuppressants or antifibrotic agents for the treatment of IIPs before the start of the study. During the observation period, 237 (26.4%), 349 (39.3%) and 118 (13.3%) patients in the combined cohort received antifibrotic agents, steroids and immunosuppressants, respectively. Among the 237 patients who received antifibrotics, 99 (41.8%), 94 (39.7%) and 44 (18.6%) patients received pirfenidone, nintedanib and sequential use of both, respectively. The median observation time was 39.0 months (range 1.0–238.9 months) in the combined cohort. The interobserver reproducibilities (κ statistics) for emphysema and honeycombing on HRCT were 0.74 (95% CI 0.69–0.78) and 0.82 (95% CI 0.78–0.86), respectively.
Predictive factors for AE-IIPs
During the study period, 103 and 56 patients developed AE-IIPs in the exploratory and validation cohorts, respectively. Univariate Fine–Gray analysis identified the following predictive factors for acute exacerbation: older age; lower %DLCO; increased levels of LDH, KL-6 or SP-D; and presence of emphysema or honeycombing (table 2). In multivariate Fine–Gray analysis, stepwise selection according to p-value identified the following independent predictive factors: presence of honeycombing, age >75 years and LDH level >222 U·L−1 (table 2). Stepwise selection using the Bayesian information criterion identified high honeycombing and LDH as independent predictive factors; however, these factors were not employed for the final prediction model because the C-index of 0.59 was lower than the C-index of 0.62 obtained from the p-value selection. Stepwise selection using the Akaike information criterion did not identify a combination that consisted of significant factors. The results of the other candidate models in multivariate analyses are shown in supplementary table S1.
Predictive scores for AE-IIPs
Based on the risk ratio provided in table 2, the predictive score for AE-IIPs was determined as follows: 1 point each was added for honeycombing (H), age >75 years (A) and LDH level >222 U·L−1 (L); the total score ranged from 0 to 3 (HAL score) (figure 2a). A small proportion of patients had a total score of 3, and patients with total scores of 2 and 3 had comparable risks of AE-IIPs (supplementary figure S1). Therefore, the total scores of 2 and 3 were merged, and the patients were categorised into three groups: total score of 0, 1 and ≥2 (figure 2b–d and table 3). The HAL score discriminated the risk of AE-IIPs with a C-index of 0.62 (95% CI 0.56–0.67); it demonstrated close agreement between the observed and predicted occurrence of AE-IIPs at each score level (figure 2b, table 3). In the validation cohort, the HAL score discriminated the risk of AE-IIPs with a C-index of 0.67 (95% CI 0.60–0.73) (figure 2c and table 3). In the combined cohort, the estimated cumulative risks for AE-IIPs at 1, 2, 3, 5 and 10 years were 1.9%, 3.5%, 5.1%, 7.7% and 12.9%, respectively, in the total score 0 group; 4.7%, 8.3%, 12.0%, 17.7% and 28.4%, respectively, in the total score 1 group; and 8.0%, 14.2%, 19.7%, 28.7% and 43.0%, respectively, in the total score ≥2 group (figure 2d and table 3). In the combined cohort, the respective time-dependent positive predictive values of the HAL score at 1, 2, 3, 5 and 10 years were 6.4%, 10.8%, 15.6%, 22.9% and 33.4% in the total score ≥1 group (versus <1), whereas they were 7.6%, 13.7%, 20.8%, 30.2% and 45.1% in the total score ≥2 group (versus <2). The respective time-dependent negative predictive values at 1, 2, 3, 5 and 10 years were 98.9%, 96.6%, 94.8%, 93.1% and 82.7% in the total score ≥1 group (versus <1), whereas they were 96.0%, 93.2%, 90.6%, 86.3% and 76.3% in the total score ≥2 group (versus <2) (table 4).
Subgroup analysis
Next, patients in the combined cohort were divided into two subgroups of IIPs: IPF and non-IPF. In the IPF group, the HAL score discriminated the risk of AE-IIPs (C-index 0.59, 95% CI 0.54–0.64) (figure 3a). The estimated cumulative risks at 1, 2, 3, 5 and 10 years were not estimable, 3.6%, 7.3%, 10.5% and 16.2%, respectively, in the total score 0 group; 5.0%, 11.6%, 19.3%, 27.4% and 39.0%, respectively, in the total score 1 group; and 6.3%, 14.4%, 23.2%, 32.5% and 46.5%, respectively, in the total score ≥2 group (supplementary table S2). Moreover, in the non-IPF group, the HAL score discriminated the risk of AE-IIPs (C-index 0.63, 95% CI 0.57–0.69) (figure 3b). The estimated cumulative risks at 1, 2, 3, 5 and 10 years were 2.3%, 3.5%, 4.3%, 6.6% and 11.5%, respectively, in the total score 0 group; 4.8%, 7.0%, 8.6%, 13.1% and 23.5%, respectively, in the total score 1 group; and 8.6%, 13.2%, 15.7%, 23.3% and 36.2%, respectively, in the total score ≥2 group (supplementary table S2).
In the combined cohort, patients who did and did not receive antifibrotic therapy before the occurrence of AE-IIPs (antifibrotic and non-antifibrotic groups, respectively) were analysed separately. 16 patients who received antifibrotics after the occurrence of AE-IIPs were included in the non-antifibrotic group. In the antifibrotic group (n=221) and non-antifibrotic group (n=668), 37 (16.7%) and 122 (18.3%) patients, respectively, experienced AE-IIPs (p=0.686). In both groups, the HAL score discriminated the risk of AE-IIPs (figure 3c–d). In a separate evaluation of patients in the IPF groups who did and did not receive antifibrotics, there was a tendency toward an increased risk of AE-IIPs as the HAL score increased, regardless of antifibrotic treatment; however, the HAL score did not discriminate the risk of AE-IIPs, particularly when comparing between the scores of 1 and ≥2 (supplementary figure S2). Among patients in the non-IPF group who did not receive antifibrotics, the HAL score discriminated the risk of AE-IIPs; however, only 10 AE-IIPs occurred among patients in the non-IPF group who received antifibrotic treatment, which was insufficient to confirm the utility of the HAL score. Nevertheless, those patients also exhibited a tendency for an increased risk of AE-IIPs as the HAL score increased (supplementary figure S2). When patients were stratified according to %FVC of 70% or 80%, the HAL score discriminated the risk of AE-IIPs in all groups (supplementary figure S3).
113 (12.7%) patients in the combined cohort (34 and 79 in the exploratory and validation cohorts, respectively) had CT images with a slice thickness from >1.5 to 3 mm. When we evaluated the HAL score in the remaining 776 patients who had CT images with a slice thickness from 1.0 to ≤1.5 mm, the HAL score discriminated the risk of AE-IIPs (supplementary figure S4).
Model prediction of overall survival
When the HAL score was applied to overall survival, the model discriminated the risk of death with a C-index of 0.61 (95% CI 0.58–0.64) (figure 4 and supplementary table S3). The estimated survival rates at 1, 2, 3, 5 and 10 years were 97.0%, 92.3%, 88.0%, 84.1% and 54.0%, respectively, in the total score 0 group; 94.4%, 86.4%, 79.1%, 65.2% and 32.3%, respectively, in the total score 1 group; and 90.8%, 78.4%, 67.9%, 49.2% and 15.4%, respectively, in the total score ≥2 group (supplementary table S3).
Discussion
In the present study, univariate analyses revealed the following predictive factors for AE-IIPs: age, lower %DLCO, increased levels of serum KL-6, SP-D and LDH and presence of emphysema and honeycombing on HRCT images. Among them, multivariate analyses selected the scoring model using honeycombing, age >75 years and LDH level >222 U·L−1, which discriminated the risk of AE-IIPs. The utility of the HAL score was also verified in the validation cohort. Subgroup analysis revealed that the HAL score could be applied to patients with and without IPF. Furthermore, the HAL score discriminated the overall survival of patients with IIPs. The clinical factors employed in the HAL score were readily available in clinical practice; this simple model could be used for the stratification of AE-IIPs risk in the management of patients with IIPs.
Both honeycombing and LDH have strong predictive value among well-known risk factors, such as lower %DLCO and increased levels of KL-6 and SP-D. Honeycombing and LDH were also selected as predictive factors in other candidate models selected by different stepwise methods (supplementary table). The presence of honeycombing is a hallmark of fibrotic changes in pulmonary tissue and an important finding for the radiological diagnosis of IPF [22]. There are two possible explanations for the predictive ability of the presence of honeycombing. First, the presence of honeycombing suggested a diagnosis of IPF, which has the highest risk of acute exacerbation among IIPs [3–7, 10, 22, 27, 28]. Second, radiological honeycombing itself is reportedly a risk factor for AE-IIPs in patients with and without IPF [8, 10, 16, 29–32]. Honeycombing is the result of progressive pulmonary fibrosis and is associated with disease severity in patients with IIPs. It is reasonable that the presence of honeycombing was a predictor of AE-IIPs across various IIP phenotypes.
An increased level of LDH, another strongly predictive factor selected in the present study, is used to monitor disease activity in clinical practice and has been identified as a risk factor for AE-IIPs [14, 29, 33, 34]. LDH, an enzyme found in every organ, can indicate the occurrence of lung damage; therefore, it is used for the assessment of various IIPs. The disease-nonspecific utility of LDH presumably contributed to its predictive ability for acute exacerbations in various types of IIPs. Another possible explanation is that LDH had less confounding influence from other predictive factors, which allowed it to serve as a more effective independent predictor.
However, other factors significant in univariate analysis (e.g. lower %DLCO and/or increased levels of KL-6 and SP-D) were not included in the optimal model according to multivariate analysis. Although they are controversial, these factors have been identified as predictors of AE-IIPs [4, 6–10, 14–17]. Lower %DLCO and/or increased levels of KL-6 and SP-D are associated with IIP disease severity and/or activity. Therefore, these three factors might have confounding effects on honeycombing and LDH, which are strongly associated with AE-IIPs; these effects resulted in the exclusion of the three factors from the final predictive model. Notably, %DLCO, KL-6 and SP-D had weak but significant correlations with each other, as well as honeycombing and LDH.
The HAL score could discriminate the risk of AE-IIPs regardless of antifibrotic treatment during the study period. Antifibrotic agents have been shown to reduce the risk of AE-IIPs, which might affect the natural occurrence of AE-IIPs [35]. However, the utility of the HAL score was maintained even in patients who received antifibrotics. This finding suggests that, for antifibrotic-naïve patients, the HAL score could provide supplemental information regarding the potential use of antifibrotic agents based on current clinical guidelines. Furthermore, after patients receive antifibrotics, the HAL score could continue to be used for stratification of AE-IIP risk.
The advantage of the HAL score is that it requires only simple clinical factors without precise classification of IIPs. The risks for AE-IIPs differ among IIP phenotypes; however, the phenotypic classification of IIPs is not simple in clinical practice. First, it is difficult to accurately diagnose the IIP phenotype in the initial examination. For example, some patients with IPF lack typical honeycombing on HRCT images at the early stage, although they eventually develop honeycombing with disease progression. Additionally, patients with advanced non-IPF IIPs sometimes develop radiological and pathological honeycombing that mimics IPF [22]. Second, surgical lung biopsy for the diagnosis of IIPs cannot be performed in some patients because of its invasive procedure [36]. Third, pathological findings can vary among lung sites in patients with IIPs. If surgical lung biopsy is not performed for representative lesions, it is difficult to ensure an accurate diagnosis [37]. Furthermore, discordant diagnosis is frequent even among specialists [38]. The current model could provide cross-phenotypic utility for the prediction of AE-IIPs in patients with various IIPs.
Nevertheless, the C-index and positive predictive value of the HAL score were not sufficiently high for use as a prediction model. These findings may be related to the existence of other potential risk factors for AE-IIPs that were not included in the HAL score. Low FVC, low DLCO and high serum levels of KL-6 and SP-D have been identified as risk factors for acute exacerbation, although those factors were not included in the model based on multivariate analysis. Additionally, there may be unknown risk factors, such as genetic factors, detailed chest CT findings, the extent and localisation of CT abnormalities, bronchoalveolar lavage findings or levels of serum biomarkers other than KL-6 and SP-D. Alternative predictive models might have been selected if we had included more detailed data concerning IIPs. Additionally, the positive predictive value of the HAL score was not sufficiently high to predict the incidence of AE-IIPs, possibly because of the low prevalence of AE-IIPs in the study cohort. Notably, the positive predictive value increased over time as the incidence of AE-IIPs increased. Nevertheless, the high negative predictive value indicates that the HAL score may be useful in the identification of patients with a low risk of AE-IIPs. The HAL score does not completely predict the incidence of acute exacerbation with high accuracy, but it can be used to determine the approximate risk of AE-IIPs (based on three simple factors) in clinical practice.
The present study had three main limitations. First, the detection of honeycombing varied among HRCT scanners and observers. HRCT images were obtained at specific intervals (the interval ranged from 2.5 to 10 mm among the centres and scanners). Furthermore, ∼13% of the study patients had a slice thickness from >1.5 to 3 mm. Mild honeycombing might have been missed in the thick and interspaced HRCT images. Although there was considerable interobserver reproducibility in terms of honeycombing detection, we could not guarantee complete consistency among observers. In fact, a certain level of disagreement regarding the detection of honeycombing among observers has been reported [39]. There is a need to understand the limit of honeycombing detection when interpreting the HAL score in clinical practice. Second, the present study only evaluated Japanese patients with IIPs. There are ethnic differences in the prevalence of ILDs and the risk of AE-IIPs [1, 5, 16, 40, 41]. Therefore, the optimal models for AE-IIPs may differ according to ethnicity. Third, differences in predictive utility among IIP phenotypes are unknown. Approximately half of the study patients had unclassifiable IIPs. Additionally, 159 instances of AE-IIPs were observed in the combined cohort, which was insufficient for the identification of individual predictive models among IIP phenotypes. Further large-scale studies are needed to establish optimal predictive models for AE-IIPs that are tailored for different ethnicities and/or IIP phenotypes.
Conclusion
The HAL score using radiographic honeycombing, age, and LDH discriminated the risk of AE-IIPs. This simple model may be helpful for the stratification of AE-IIPs risk in the clinical management of patients with IIPs.
Supplementary material
Supplementary Material
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Supplementary methods, tables and figures ERJ-01634-2022.SUPPLEMENT
Shareable PDF
Supplementary Material
This one-page PDF can be shared freely online.
Shareable PDF ERJ-01634-2022.Shareable
Acknowledgements
We thank Ryan Chastain-Gross, from Edanz (https://jp.edanz.com/ac) for editing a draft of this manuscript.
Footnotes
Data availability: The datasets used and/or analysed in this study are available from the corresponding author upon reasonable request.
This article has an editorial commentary: https://doi.org/10.1183/13993003.00459-2023
Author contributions: M. Karayama: conceptualisation, methodology, interpretation of data, writing – review and editing, and supervision; Y. Aoshima, T. Suzuki: acquisition of data, interpretation of data, and writing – original draft; K. Mori: analysis and interpretation of data; N. Yoshizawa, S. Ichikawa: evaluation of chest imaging; S. Kato, K. Yokomura, M. Kono, D. Hashimoto, Y. Inoue, H. Yasui, H. Hozumi, Y. Suzuki, K. Furuhashi: conceptualisation and supervision; S. Goshima: evaluation of chest imaging, conceptualisation, supervision; T. Fujisawa, N. Enomoto, N. Inui: conceptualisation, supervision, and writing – review and editing; T. Suda: conceptualisation, methodology, writing – review and editing, and supervision.
Conflict of interest: All authors have nothing to disclose.
- Received August 20, 2022.
- Accepted February 6, 2023.
- Copyright ©The authors 2023.
This version is distributed under the terms of the Creative Commons Attribution Non-Commercial Licence 4.0. For commercial reproduction rights and permissions contact permissions{at}ersnet.org