Data source
We used the Diagnosis Procedure Combination database, a nationwide inpatient administrative claims and discharge abstract database in Japan. This database contains data on main diagnoses, primary diagnosis, and comorbidities at admission, recorded using International Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) codes and Japanese text data; age; sex; body height and weight; grade of activities of daily life on admission (Barthel Index) [6]; level of dyspnoea based on the Hugh-Jones classification [7]; ambulance use and discharge status. The database also includes the Age, Dehydration, Respiratory Failure, Orientation Disturbance and Blood Pressure (A-DROP) score for patients with community-acquired pneumonia, as well as data on whether C-reactive protein (CRP) was ≥ 20 mg/dL or infiltration covering at least two-thirds of one lung on chest radiography, mechanical ventilation during hospitalisation and hospitalisation costs. The sensitivity and specificity of the recorded ICD-10 codes and procedures in the Diagnosis Procedure Combination database were validated in a previous study [8].
The Hugh-Jones classification is a widely used dyspnoea scale with the following categories: I (the patient’s breathing is as good as that of other people of own age and build while working, walking and climbing hills or stairs), II (the patient is able to walk at the pace of normal people of the same age and build on level ground but is unable to keep up on hills or stairs), III (the patient is unable to keep up with normal people on level ground but is able to walk about a mile or more at their own pace), IV (the patient is unable to walk more than about 50 yards on level ground without resting), V (the patient is short of breath when talking or undressing or is unable to leave their home because of shortness of breath) and unspecified (the patient cannot be classified into any of the above grades because of bedridden status) [7].
The A-DROP score, established by the Japanese Respiratory Society, is a modified version of the CURB-65 (Confusion, Urea, Respiratory rate, Blood pressure-65) score [9]. The A-DROP score includes the following parameters: age (men: ≥ 70 years, women: ≥ 75 years), dehydration (blood urea nitrogen ≥ 21 mg/dL), respiratory failure (SaO2 ≤ 90% or PaO2 ≤ 60 mmHg), orientation disturbance (confusion) and low blood pressure (systolic blood pressure ≤ 90 mmHg).
The Institutional Review Board of The University of Tokyo approved this study and waived the requirement for patient informed consent because of the anonymous nature of the data.
Patient selection
We retrospectively collected data on patients admitted to hospitals for community-acquired pneumonia who were discharged from 1 April 2012 to 31 March 2014. Community-acquired pneumonia was defined according to the 2019 guidelines on the management of adults with community-acquired pneumonia published by the Infectious Diseases Society of America/American Thoracic Society [10].
We defined MRSA pneumonia patients as those who had both the ICD-10 code for MRSA pneumonia and records of the administration of anti-MRSA antibiotics (vancomycin, linezolid, teicoplanin or arbekacin) for more than 7 days.
Outcomes
The primary outcome of this study was all-cause in-hospital mortality. The secondary outcomes were 30-day in-hospital mortality, 90-day in-hospital mortality, length of stay and hospitalisation costs. The duration of antibiotic therapy was also evaluated.
Statistical analysis
The χ2 test was used to compare proportions between groups. The two-sample t-test was used to compare average values, and the Mann–Whitney test was used to compare the median values between groups.
Among the patients with pneumonia, we selected an MRSA pneumonia group and a non-MRSA pneumonia group with 1:4 matching: for each patient in the MRSA pneumonia group, we identified four non-MRSA patients of the same sex who were admitted to the same hospital in the same year and whose ages were within 5 years of the age of the MRSA patient. We used hospital identifiers for matching to cancel out site-specific effects such as physician practice patterns and treatment outcomes [11].
We performed multiple imputation for missing data on body mass index (BMI), Barthel Index, Hugh-Jones grade, A-DROP score, CRP ≥ 20 mg/mL or infiltration covering at least two-thirds of one lung on chest radiography, and hospitalisation costs. We replaced each missing value with a set of substituted plausible values by generating 20 complete datasets using the multivariate imputation by chained equations method. The following covariates were used to create these 20 complete datasets: MRSA pneumonia, age, sex, fiscal year, haemodialysis, mechanical ventilation at admission, ICU admission, arrival by ambulance, chronic obstructive pulmonary disease (COPD), interstitial lung disease, aspiration pneumonia, Pseudomonas aeruginosa pneumonia, cerebrovascular disease, Parkinson disease, diabetes, dementia, in-hospital death, 30-day in-hospital death and 90-day in-hospital death, with the assumption that data were missing at random [12, 13]. Estimates from these 20 imputed datasets were combined using Rubin’s rule to obtain combined imputation estimates and standard errors.
Then, using multivariable logistic regression analysis fitted with generalised estimating equations to account for the 1:4 matched-pair clustering, we examined the factors associated with all-cause in-hospital mortality. Multiple linear regression analysis fitted with generalised estimating equations was also used to assess hospitalisation costs. The following independent variables were included in the models: age, sex, MRSA pneumonia, BMI, Barthel Index, Hugh-Jones grade, A-DROP score, CRP ≥ 20 mg/mL or infiltration covering at least two-thirds of one lung on chest radiography, haemodialysis, mechanical ventilation at admission, ICU admission, arrival by ambulance, COPD, interstitial lung disease, aspiration pneumonia and Pseudomonas aeruginosa pneumonia. For sensitivity analyses, we added independent variables of chronic heart failure, chronic liver disease, sepsis, acute renal failure, leukopenia, immunosuppression and stroke to the multivariable regression models used in the main analyses. Statistical analyses were performed using SPSS, Version 22.0 (IBM SPSS, Armonk, NY, USA) and Stata, Version 16 (StataCorp, College Station, TX, USA).
Patient and public involvement
Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.
