Data source
Inpatient data were extracted from the Japanese Diagnosis Procedure Combination database, the details of which have been reported elsewhere [16]. More than 1000 hospitals voluntarily contributed to the database, representing approximately 50% of all discharges from acute care hospitals in Japan. We collected data including those of sex and age; hospitalisation and discharge dates; weight and height; severity of dyspnoea based on the Hugh–Jones dyspnoea scale [17]; level of consciousness upon admission; smoking index; activities of daily living; frequency of hospitalisation; intensive care unit (ICU) admission during hospitalisation; main diagnoses, pre-existing comorbidities upon admission and complications after admission as recoded by the attending physicians based on the International Classification of Diseases, 10th revision (ICD-10) codes accompanied by text in Japanese; procedures and their dates; dates and doses of drugs administered during hospitalisation; and discharge status.
This study was approved by the institutional review board of The University of Tokyo. The need for informed consent was waived, because anonymised data were used.
Patient selection
This study used data collected from 1 January 2014 to 31 March 2018. The inclusion criteria were patients aged ≥ 15 years, those diagnosed with interstitial pneumonia (ICD-10 codes J84.1, J84.8 and J84.9), those who underwent computed tomography scan within 1 day after admission and those who received treatment with intravenous mPSL at a dose of 500–1000 mg/day for 3 days, which was started within 4 days after admission [18, 19]. Patients with IPF were selected as follows. Initially, patients with idiopathic interstitial pneumonias other than IPF, such as idiopathic nonspecific interstitial pneumonia, respiratory bronchiolitis-associated interstitial lung disease, cryptogenic organizing pneumonia, acute interstitial pneumonia, desquamative interstitial pneumonia, lymphoid interstitial pneumonia, idiopathic pleuroparenchymal fibroelastosis and unclassifiable idiopathic interstitial pneumonia, were excluded using the diagnoses in Japanese. Moreover, we did not include patients with secondary interstitial lung diseases identified using ICD-10 codes (hypersensitivity pneumonitis [J67], connective tissue disease associated with interstitial lung disease [M05, M06 and M30–35], sarcoidosis [D86], amyloidosis [E85], drug-induced lung disease [J70], radiation pneumonitis [J70], Pneumocystis jirovecii pneumonia [B59], pneumoconiosis [J60–65], pulmonary alveolar proteinosis [J84.0] eosinophilic pneumonia [J82], Langerhans cell histiocytosis [C96] and lymphangioleiomyomatosis [D21.9]; those receiving medications including furosemide, azosemide, carperitide, landiolol hydrochloride, digoxin, deslanoside and tolvaptan for acute heart failure within 1 day after admission; and those who received intra-aortic balloon pump therapy during hospitalisation [18, 19]. The remaining patients were assumed to have AE-IPF. Next, we excluded patients with missing data about treatment year, those without mechanical ventilation and those who died within 6 days after admission to prevent immortal time bias. Patients were divided into two groups: those who received high-dose mPSL alone (mPSL alone group) and those who received rTM for at least 3 days, which was started within 4 days after admission, combined with high-dose mPSL (rTM group).
Characteristics of patients
The characteristics of patients evaluated in this study were sex, age, treatment year, body mass index, Hugh–Jones dyspnoea scale scores upon admission, level of consciousness upon admission, comorbidities, Charlson Comorbidity Index, smoking index, activities of daily living scale (Barthel Index) upon admission, history of previous hospitalisation (0, 1–2, or ≥ 3), type of hospital (academic or non-academic hospital) and ICU admission. Furthermore, we examined data about procedures and treatments, including mechanical ventilation, continuous renal replacement therapy, high-flow nasal cannula oxygen therapy, transfusion and use of medications for IPF within 3 days after admission. Level of consciousness upon admission was evaluated using the Japan Coma Scale, [20, 21] which is widely used in Japan and is well correlated with the Glasgow Coma Scale score [22]. The following comorbidities were identified using ICD-10 codes (Additional file 1: Table S1): bronchial asthma, pulmonary emphysema, pneumonia, mycotic infection, pulmonary embolism, bronchiectasis, pneumothorax, cor pulmonale, lung and other types of cancer, sepsis, chronic heart failure, tachycardia, acute coronary syndrome, diabetes mellitus, stroke, dementia, renal failure, liver dysfunction and gastroesophageal reflux disease. The Charlson Comorbidity Index scores were classified into four categories (0, 1, 2 and ≥ 3).
Outcome
The primary outcome was all-cause in-hospital mortality. The secondary outcomes were 14- and 28-day mortality, post-hospitalisation bleeding events and length of hospital stay. The following bleeding events were assessed using ICD-10 codes: epistaxis (R040), hemoptysis (R042), pulmonary hemorrhage (R048), subcutaneous hemorrhage (R233), purpura (D692), muscle hemorrhage (T146), hematuria (R31), gastrointestinal bleeding (K228, K922), bloody stool (K921) and intracranial hemorrhage (I61, I629) [15].
Statistical analysis
Dichotomous and categorical variables were presented as numbers with percentages and continuous variables as median and interquartile range (IQR).
To account for differences in baseline characteristics, including comorbidities and treatments, stabilised inverse probability of treatment weighting (IPTW) analyses using propensity scores were performed to compare outcomes between the two groups. Stabilised IPTW uses propensity scores and adjusts for measured potential confounders while preserving sample size [23]. To control covariate imbalance, the specific stabilised weights were generated using propensity scores, which can predict the probability of receiving rTM combined with high-dose mPSL therapy. To estimate the propensity score, a logistic regression model for receiving high-dose mPSL alone therapy was used with the following independent variables: sex, age, treatment year, body mass index, Hugh–Jones dyspnoea scale score, level of consciousness upon admission, Charlson Comorbidity Index, smoking index, Barthel Index upon admission, frequency of hospitalisation, ICU hospitalisation within 3 days after admission, comorbidities and procedures (hemodialysis, high-flow nasal cannula oxygen therapy, fresh frozen plasma transfusion and concentrated platelet transfusion) and drugs for AE-IPF and disseminated intravascular coagulation (DIC) (noradrenaline, azithromycin, cyclophosphamide, cyclosporin, tacrolimus, azathioprine, pirfenidone, nintedanib, sivelestat sodium hydrate, heparin calcium, dalteparin and tranexamic acid). Variables included in the logistic regression model were those that were considered as potential confounders with reference to previous studies [18, 19]. Covariate balance was assessed using a standardised mean difference. A value of < 0.20 indicated an acceptable balancing of covariates between the two groups. Stabilised IPTW analyses can preserve sample size and appropriately estimate average treatment effects over the marginal distribution of measured covariates in a study cohort.
We used generalised linear models with cluster-robust standard errors treating each hospital as a cluster to compare the primary and secondary outcomes. Logistic regression analyses of in-hospital mortality, 14- and 28-day mortality and post-hospitalisation bleeding events were conducted. Then, odds ratios and their 95% confidence intervals (CIs) were calculated. The lengths of hospital stay between the two groups were compared via Poisson regression analysis, and the incidence rate ratios and their 95% CIs were calculated. To address competing outcomes, secondary outcomes were evaluated among the survivors alone and all patients.
A two-tailed significance level of 0.05 was used in all statistical analyses. All tests were performed using STATA/MP version 16 software (STATA Corp., College Station, TX, USA).