Chronic obstructive pulmonary disease (COPD), a manageable respiratory condition, is the third leading cause of death worldwide. Knowing which prescription medications are the most effective in improving health outcomes for people with COPD is essential to maximizing health outcomes.
To estimate the long-term benefits of combination long-acting β-agonists (LABAs) and inhaled corticosteroids compared with LABAs alone in a real-world setting.
Design, Setting, and Patients
Population-based, longitudinal cohort study conducted in Ontario, Canada, from 2003 to 2011. All individuals aged 66 years or older who met a validated case definition of COPD on the basis of health administrative data were included. After propensity score matching, there were 8712 new users of LABA–inhaled corticosteroid combination therapy and 3160 new users of LABAs alone who were followed up for median times of 2.7 years and 2.5 years, respectively.
Newly prescribed combination LABAs and inhaled corticosteroids or LABAs alone.
Main Outcomes and Measures
Composite outcome of death and COPD hospitalization.
The main outcome was observed among 5594 new users of LABAs and inhaled corticosteroids (3174 deaths [36.4%]; 2420 COPD hospitalizations [27.8%]) and 2129 new users of LABAs alone (1179 deaths [37.3%]; 950 COPD hospitalizations [30.1%]). New use of LABAs and inhaled corticosteroids was associated with a modestly reduced risk of death or COPD hospitalization compared with new use of LABAs alone (difference in composite outcome at 5 years, −3.7%; 95% CI, −5.7% to −1.7%; hazard ratio [HR], 0.92; 95% CI, 0.88-0.96). The greatest difference was among COPD patients with a codiagnosis of asthma (difference in composite at 5 years, −6.5%; 95% CI, −10.3% to −2.7%; HR, 0.84; 95% CI, 0.77-0.91) and those who were not receiving inhaled long-acting anticholinergic medication (difference in composite at 5 years, −8.4%; 95% CI, −11.9% to −4.9%; HR, 0.79; 95% CI, 0.73-0.86).
Conclusions and Relevance
Among older adults with COPD, particularly those with asthma and those not receiving a long-acting anticholinergic medication, newly prescribed LABA and inhaled corticosteroid combination therapy, compared with newly prescribed LABAs alone, was associated with a significantly lower risk of the composite outcome of death or COPD hospitalization.
Chronic obstructive pulmonary disease (COPD) affects 5% to 22% of adults older than 40 years, has a lifetime risk of more than 25%, and is the third leading cause of death worldwide.1-3
Medications are a mainstay of COPD management, and knowing which are most effective in real-world practice is essential. Combination therapy consisting of long-acting β-agonists (LABAs) and inhaled corticosteroids (ICSs) has been shown to decrease exacerbations and possibly decrease mortality compared with placebo.4-6 However, there are still gaps in what is known about its comparative effectiveness compared with LABAs alone.
A recent systematic review of randomized clinical trials (RCTs) found that exacerbation rates, but not hospitalizations or mortality, were lower in patients treated with LABA-ICS combination therapy compared with LABAs alone.7 However, it included RCTs that predominantly enrolled younger adult men with few comorbidities and it excluded people with asthma and who were taking long-acting anticholinergics (LAAs), a commonly used COPD medication. Thus, its results may not be generalizable to older, frailer COPD patients with asthma or other comorbidities taking other COPD medications in real-world settings.
There are also downsides to treating with LABAs and ICSs. In RCTs, patients receiving these medications have been shown to have an increased risk of pneumonia compared with patients receiving LABAs alone.7 A systematic review of RCTs has also shown a non–statistically significant increased fracture risk among LABA and ICS users compared with users of LABAs alone.8 Moreover, the cost of the combination product is usually greater. The large number of patients receiving LABAs and ICSs, at least in some regions, suggests that starting combination medication for people with COPD occurs often without a preceding trial of LABAs alone.9
Therefore, we conducted an observational study to examine the association of LABA-ICS combination therapy compared with LABAs alone and the composite outcome of mortality and COPD hospitalizations in older COPD patients with naturally occurring comorbidities, including asthma, in real-world conditions. Because confounding by indication is a potential limitation in observational studies, we performed a number of additional sensitivity analyses.
We conducted a retrospective cohort study of Ontario residents aged 66 years or older using multiple linked population health care databases. Ontario has a diverse, multicultural population of approximately 2 million people aged 65 years or older. Virtually all residents of all ages receive universal access to physician and hospital services, which are recorded in administrative databases, and those aged 65 years or older receive prescription medications from publically funded insurance programs.
Ethics approval was obtained from the Research Ethics Board of Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada. A waiver of informed consent was obtained.
Patient demographic and health care records were linked using encrypted health card numbers, which served as unique and anonymous identifiers across databases. The Ontario Registered Persons Database contains basic demographic information including date of death.10 The Ontario Drug Benefits database contains prescription medication claims for all residents aged 65 years or older.11 The Canadian Institute of Health Information’s Discharge Abstract Database and National Ambulatory Care Reporting System contain detailed information for all admissions to hospitals and emergency departments. The Ontario Health Insurance Plan physician claims database contains information on all outpatient services provided by fee-for-service physicians and “shadow billings” for physicians paid under alternate payment plans.
Responses to 4 cycles of the Canadian Community Health Survey, which is a nationally representative cross-sectional survey designed by Statistics Canada,12 were used to obtain information such as individual-level smoking status.
All Ontario residents aged 66 years or older who met a validated case definition of physician-diagnosed COPD using health administrative data (eAppendix in the Supplement)13 and were new users of LABAs or LABA-ICS combination therapy between September 1, 2003, and March 31, 2011, were included. This COPD definition has been shown to have a positive predictive value of 81% and a negative predictive value of 88.5% in adults 35 years and older, and a positive predictive value of 86% in adults aged 65 years or older compared with clinical evaluation by a physician.13,14 New users were individuals who had not filled a prescription for either of these medications or for ICSs alone in the previous year and were studied to minimize bias due to improved outcomes among long-term users who were presumably benefitting from their medication and worse outcomes among patients who, due to problems, required a step up or down in therapy.14 LABA-ICS combination therapy and LABAs alone were the only medications considered. The initial prescription date was the index date. Individuals who received prescriptions for both medications on the index date, patients who were ineligible for health insurance, and those who had had lung volume reduction surgery or transplantation were excluded.
A wide variety of baseline characteristics were determined using the health administrative data (eTable 1 in the Supplement). Neighborhood income quintiles, shown to be a reasonable proxy for socioeconomic status,15 were determined by linking postal codes to census data. General measures of comorbidity based on diagnostic information from health services use in the 2 years prior to the index date were determined using the Johns Hopkins Adjusted Clinical Group Case-Mix System.16
The primary outcome was a composite of all-cause mortality and COPD hospitalization (International Statistical Classification of Diseases, Tenth Revision [ICD-10] codes J41-J43). We excluded COPD hospitalizations with acute lower respiratory tract infection (ICD-10 code J44.0) because we wanted to analyze pneumonia separately. A composite outcome was chosen because factors that lead to COPD hospitalization also lead to mortality; however, we also analyzed mortality and COPD hospitalizations as distinct outcomes.
Secondary outcomes included hospitalization for pneumonia (ICD-10 codes J10-J18 and J44.0) and hospitalization for fragility fractures of the spine, pelvis, forearm, or hip (ICD-10 codes S32, S52, and S72) likely to result from osteoporosis—both are suggested adverse effects of ICSs.7,8
Patients who did not experience an outcome of interest were censored on March 15, 2012.
Propensity score matching was used to compare patients with similar observed characteristics, all of whom were potential candidates for LABAs or LABAs and ICSs.17 Multivariable logistic regression was used to compute the propensity score for receiving LABAs alone using all the measured baseline characteristics as predictors. Individuals prescribed LABAs were then matched with up to 3 individuals prescribed LABAs and ICSs on the basis of age (±1 year), sex, codiagnosis of asthma, COPD duration, and propensity score (±0.2 SDs). An absolute standardized difference of less than 10% on all covariates was accepted as adequate balance.18
We used Cox regression analysis to compare outcomes between the groups while accounting for matching by using marginal proportional hazards models.19 For hospitalization outcomes, censoring on mortality would have been inappropriate because it is a more serious and common outcome. Therefore, we used subdistribution hazard models to account for the competing risk of mortality to avoid bias due to the high mortality rate.20
All statistical tests were 2-sided with P < .05 defined as the level of statistical significance and were performed using SAS, version 9.1 (SAS Institute Inc).
Stratification factors of interest were identified a priori. We stratified by codiagnosis of asthma21 because those with asthma might benefit more from LABA-ICS combination therapy and those without asthma are not normally considered to be corticosteroid responsive.22 We stratified by current or baseline use of LAAs, defined by receipt in the previous year, because they are commonly used in the management of COPD. In a post hoc analysis, we explored a potential 3-way interaction among the main treatment effect, receiving LAAs, and a codiagnosis of asthma. We also stratified by previous receipt of spirometry, the recommended diagnostic test for COPD,23 to determine whether more certainty about the diagnosis of COPD influenced the results.
Propensity score calibration, using supplementary data from the Canadian Community Health Survey, was used to incorporate data on variables not available in the health administrative data (eAppendix in the Supplement).24 We used an approach proposed by Schneeweiss,25 as we have done previously,14,26 to determine the effect on our findings of potential COPD misclassification due to the case definition having an imperfect positive predictive value (eAppendix in the Supplement). Although we realized that it might be informative censoring that creates bias, to assess whether low adherence was affecting the results we conducted 2 further analyses, one in which individuals were censored at the time that they stopped using their initially prescribed medication and another in which they were censored 2 months before stopping (eAppendix in the Supplement).
We identified 38 266 patients with physician-diagnosed COPD who newly started LABAs or LABAs and ICSs. Of those, 111 patients started both, 534 patients were ineligible for health insurance, and 74 patients had received lung reduction or transplantation and were excluded, leaving 3258 new users of LABAs and 34 289 new users of LABAs and ICSs. Compared with new users of LABAs and ICSs, prior to matching, new users of LABAs alone were older, more likely to receive other COPD medications, and more likely to have had a COPD or COPD-related hospitalization 6 months prior to the index date (Table 1 and eTable 1 in the Supplement). Prior to matching, new users of LABAs and ICSs had a significantly lower unadjusted risk of death or COPD hospitalization (hazard ratio [HR], 0.76; 95% CI, 0.72-0.79) compared with new users of LABAs alone (eTable 2 in the Supplement).
After matching, there were 3160 new users of LABAs and 8712 new users of LABAs and ICSs who were followed up for median times of 2.5 and 2.7 years, respectively. The standardized differences for all baseline characteristics were less than 10%, indicating acceptable balance.
The primary outcome was observed among 2129 new users of LABAs (1179 deaths [37.3%]; 950 COPD hospitalizations [30.1%]) and 5594 new users of LABAs and ICSs (3174 deaths [36.4%]; 2420 COPD hospitalizations [27.8%]). There was a modest but significantly lower risk of the composite outcome among new users of LABAs and ICSs compared with new users of LABAs alone (HR, 0.92; 95% CI, 0.88-0.96) (Table 2). Similarly lower risks were also observed for mortality and COPD hospitalizations analyzed separately. There was no significant difference in the risk of pneumonia or fracture hospitalization. Calibrating the propensity score using supplementary survey data on smoking, body mass index, and self-reported health had little effect (eTable 3 in the Supplement).
Differences in the association between LABA-ICS combination therapy vs LABAs alone and the primary outcome were observed in stratified analyses (Table 3). There was a statistically significant lower risk of the primary outcome among those with a codiagnosis of asthma (HR, 0.84; 95% CI, 0.77-0.91), those not receiving LAAs (HR, 0.79; 95% CI, 0.73-0.86), and those who had not received spirometry (HR, 0.87; 95% CI, 0.81-0.93). When a 3-way interaction among treatment, LAA use, and codiagnosis of asthma was explored, the interaction term was not significant (P = .31), but stratified analyses were conducted for exploratory purposes. Among individuals who did not have asthma and who were not receiving LAAs, new users of LABA-ICS combination therapy were at a significantly lower risk of the primary outcome compared with new users of LABAs alone (HR, 0.83; 95% CI, 0.75-0.92).
Misclassification of COPD diagnosis could explain the significant association between medication groups and the primary outcome. However, only in extreme and unlikely scenarios would the results in those with asthma or without asthma not receiving LAAs be negated (eTable 4 in the Supplement).
When individuals were censored when they stopped using their initially prescribed medication, the HR of the primary outcomes was 0.94 (95% CI, 0.88-1.00), and when they were censored 2 months prior to when they stopped using their initially prescribed medication, it was 0.96 (95% CI, 0.90-1.03).
We conducted a population-based study in adults aged 66 years or older with COPD and observed that new use of LABA-ICS combination therapy was associated with a modest but statistically significant lower risk of the composite outcome of death or COPD hospitalization compared with new use of LABAs alone and a more pronounced lower risk in subgroups of individuals with asthma and who were not receiving LAAs. To the best of our knowledge, this is the first large, real-world population study to compare LABAs and LABA-ICS combination therapy.
Our findings contribute new knowledge to the treatment of people with COPD with and without a codiagnosis of asthma. Patients with asthma-COPD overlap syndrome comprised 28% of our matched cohort, an amount consistent with other studies.28,29 Our finding of an association between LABAs and ICSs and outcomes helps clarify the management of patients with COPD and asthma, as many studies of COPD medications have excluded people with asthma and vice versa. In addition, practice guidelines for COPD recommend that LABAs be considered first-line treatment while asthma guidelines warn against use of LABAs without ICSs. Our findings also offer insight into the optimal treatment of COPD patients without asthma—those who would not be considered especially corticosteroid responsive. In this subgroup, those who did not have asthma who were not receiving LAAs and who received LABAs and ICSs had a lower risk of the composite outcome of a magnitude similar to people with asthma. People without asthma who received LAAs did not have an association with better outcome, perhaps because the gains conferred by LAAs reduced the incremental benefit of LABA-ICS therapy or because in general they had more severe disease, which made them less responsive to medications. Previous studies comparing triple therapy with LABAs, ICSs, and LAAs and dual therapy with LABAs and LAAs have not been of sufficient quality to draw conclusions.30
Our study is consistent with a recent systematic review and meta-analysis of RCTs that found that people taking combination LABA-ICS therapy had non–statistically significant lower hospitalization rates (rate ratio, 0.79; 95% CI, 0.55-1.13) and mortality (odds ratio, 0.92; 95% CI, 0.76-1.11) than people taking LABAs alone.7 Our study extends these findings by examining a population with more outcomes, thus allowing results to reach statistical significance. It also includes and is generalizable to patients excluded from many of the RCTs in the review. In addition, some RCTs included volunteers who could have been receiving at least 1 of the medications before and who, because they agreed to take it again, were more likely to tolerate it and have favorable outcomes.14
Our study is also consistent with a systematic review of RCTs that did not observe a higher occurrence of fractures among LABA-ICS users compared with users of LABAs alone.8 It is not consistent with a systematic review of RCTs that observed an increased risk of pneumonia among combination LABA-ICS users.7 Differences between observational studies and RCTs, notably our inclusion of frailer patients who were more likely to have risk factors for pneumonia hospitalization that overshadowed the risks of these medications, may explain this discrepancy. One observational study comparing ICSs with no ICSs that included frailer patients found pneumonia to be increased among the former.31 There are many differences in study design that might explain why its results differed from ours (eAppendix in the Supplement).
Our study had limitations. Unlike RCTs, observational study designs are susceptible to unmeasured confounding. Through the inclusion of multiple prognostic variables, and considering others through propensity score calibration, we strived to ensure that we accounted for all possible confounders. However, it is possible that there were residual differences between the groups.
Using health administrative data to identify people with COPD might have also led to misclassification of individuals with COPD. A plausible scenario that could have explained our overall results is that, due to an imperfect positive predictive value of our case definition, all the patients potentially misclassified as having COPD when they did not were in the LABA-ICS combination group and had a 20% decrease in risk of the primary outcome, perhaps because they truly had asthma. However, these patients would have had to have a 50% decreased risk of the primary outcome in those with asthma and a 40% decrease in those without asthma not receiving LAAs to explain the results—scenarios that we believe highly unlikely. It was also possible that due to an imperfect negative predictive value, there were people misclassified as not having COPD who were not included in our study when they should have been. Thus, while our findings likely apply to the majority of people with COPD, they might not apply to all.
In addition, as reflective of real life, adherence to medication was at the same low level as found in previous studies14,32 (eTable 5 in the Supplement). We considered but rejected conducting a time-on-treatment sensitivity analysis because people receiving LABAs often stepped up to LABAs and ICSs but not vice versa. As a result, people in the LABA-ICS group would be more likely to have advanced disease and to do worse. By using only new users, we were able to create comparable groups and follow them up in an intention-to-treat manner, taking medication changes that commonly occur in the community into consideration. More people in the LABA group were receiving LABAs alone—at least until they stepped up to LABAs and ICSs—and more people in the LABA-ICS group were receiving LABAs and ICSs (eTable 5 in the Supplement). Thus, real-world conditions, specifically crossover, attenuated the groups’ differences, producing conservative results that likely caused us to underestimate the incremental benefit of LABAs and ICSs over LABAs alone. When we censored people at or before the point when they stopped using the initially prescribed medication, the results were similar.
The strengths of our study include its population, real-world relevance, and power to examine the relative effects of LABAs compared with LABAs and ICSs in various subgroups.
Among older adults with COPD, particularly those with asthma and those not receiving an LAA, newly prescribed LABA-ICS combination therapy, compared with newly prescribed LABAs alone, was associated with a significantly lower risk of the composite outcome of death or COPD hospitalization. These findings should be confirmed in RCTs.
Corresponding Author: Andrea S. Gershon, MD, MSc, Institute for Clinical Evaluative Sciences, G1 06, 2075 Bayview Ave, Toronto, ON M4N 3M5, Canada (firstname.lastname@example.org).
Author Contributions: Dr Gershon and Mr Campitelli had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. These data sets were held securely in a linked, deidentified form and analyzed at the Institute for Clinical Evaluative Sciences.
Study concept and design: All authors.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Gershon, Campitelli, Croxford.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Campitelli, Croxford, Stukel.
Obtained funding: Gershon.
Administrative, technical, or material support: Campitelli, Croxford.
Study supervision: Gershon.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: Dr Gershon is supported by a Fellowship for Translational Health Research from the Physicians’ Services Incorporated Foundation, Toronto, Ontario, and was supported by a New Investigator Award funded by team grant OTG-88591 from the Canadian Institutes of Health Research Institute of Nutrition, Metabolism and Diabetes while working on this study. Dr To is supported by the Dales Award in Medical Research from the University of Toronto, Toronto, Ontario, Canada. Funding for this project was made available through a Drug Innovation Fund Grant from the Government of Ontario. This study was also supported by the Institute for Clinical Evaluative Sciences, which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care.
Role of the Funders/Sponsors: The study funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: The opinions, results, and conclusions reported in this article are those of the authors and are independent from the funding sources. No endorsement by the Institute for Clinical Evaluative Sciences or the Ontario Ministry of Health and Long-Term Care is intended or should be inferred.
Additional Contributions: We thank Brogan Inc, Ottawa, Ontario, Canada, for the use of their Drug Product and Therapeutic Class Database.
AS. Global burden of COPD: risk factors, prevalence, and future trends. Lancet
. 2007;370(9589):765-773.PubMedGoogle ScholarCrossref
T. Lifetime risk of developing chronic obstructive pulmonary disease: a longitudinal population study. Lancet
. 2011;378(9795):991-996.PubMedGoogle ScholarCrossref
et al. Global and regional mortality from 235 causes of death for 20 age groups in 1990 and 2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet
. 2012;380(9859):2095-2128.PubMedGoogle ScholarCrossref
et al; Trial of Inhaled Steroids and Long-Acting β2 Agonists Study Group. Combined salmeterol and fluticasone in the treatment of chronic obstructive pulmonary disease: a randomised controlled trial. Lancet
. 2003;361(9356):449-456.PubMedGoogle ScholarCrossref
et al; TORCH Investigators. Salmeterol and fluticasone propionate and survival in chronic obstructive pulmonary disease. N Engl J Med
. 2007;356(8):775-789.PubMedGoogle ScholarCrossref
KB. Risk for death associated with medications for recently diagnosed chronic obstructive pulmonary disease. Ann Intern Med
. 2008;149(6):380-390.PubMedGoogle ScholarCrossref
P. Combined corticosteroid and long-acting beta(2)-agonist in one inhaler versus long-acting beta(2)-agonists for chronic obstructive pulmonary disease. Cochrane Database Syst Rev
. 2012;9:CD006829.PubMedGoogle Scholar
S. Risk of fractures with inhaled corticosteroids in COPD: systematic review and meta-analysis of randomised controlled trials and observational studies. Thorax
. 2011;66(8):699-708.PubMedGoogle ScholarCrossref
JD. A cross-sectional study of the identification of prevalent asthma and chronic obstructive pulmonary disease among initiators of long-acting β-agonists in health insurance claims data. BMC Pulm Med
. 2014;14:47.PubMedGoogle ScholarCrossref
DG. Living and Dying in Ontario: An Opportunity for Improved Health Information. Toronto, ON: Institute for Clinical Evaluative Sciences; 2008. ICES Investigative Report.
D. Coding accuracy of administrative drug claims in the Ontario Drug Benefit database. Can J Clin Pharmacol
. 2003;10(2):67-71.PubMedGoogle Scholar
B. Combining cycles of the Canadian Community Health Survey. Health Rep
. 2009;20(1):53-58.PubMedGoogle Scholar
T. Identifying individuals with physician diagnosed COPD in health administrative databases. COPD
. 2009;6(5):388-394.PubMedGoogle ScholarCrossref
et al. Comparison of inhaled long-acting β-agonist and anticholinergic effectiveness in older patients with chronic obstructive pulmonary disease: a cohort study. Ann Intern Med
. 2011;154(9):583-592.PubMedGoogle ScholarCrossref
MM. Ecologic proxies for household income: how well do they work for the analysis of health and health care utilization? Can J Public Health
. 2004;95(2):90-94.PubMedGoogle Scholar
M. Performance of the ACG case-mix system in 2 Canadian provinces. Med Care
. 2001;39(1):86-99.PubMedGoogle ScholarCrossref
Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med
. 1998;17(19):2265-2281.PubMedGoogle ScholarCrossref
GM. A comparison of the ability of different propensity score models to balance measured variables between treated and untreated subjects: a Monte Carlo study. Stat Med
. 2007;26(4):734-753.PubMedGoogle ScholarCrossref
J. A proportional hazards regression model for the subdistribution with right-censored and left-truncated competing risks data. Stat Med
. 2011;30(16):1933-1951.PubMedGoogle ScholarCrossref
T. Identifying patients with physician-diagnosed asthma in health administrative databases. Can Respir J
. 2009;16(6):183-188.PubMedGoogle Scholar
GL. Inhaled corticosteroids or long-acting beta-agonists alone or in fixed-dose combinations in asthma treatment: a systematic review of fluticasone/budesonide and formoterol/salmeterol. Clin Ther
. 2009;31(12):2779-2803.PubMedGoogle ScholarCrossref
RJ. Adjusting effect estimates for unmeasured confounding with validation data using propensity score calibration. Am J Epidemiol
. 2005;162(3):279-289.PubMedGoogle ScholarCrossref
S. Sensitivity analysis and external adjustment for unmeasured confounders in epidemiologic database studies of therapeutics. Pharmacoepidemiol Drug Saf
. 2006;15(5):291-303.PubMedGoogle ScholarCrossref
et al. Cardiovascular safety of inhaled long-acting bronchodilators in individuals with chronic obstructive pulmonary disease. JAMA Intern Med
. 2013;173(13):1175-1185.PubMedGoogle ScholarCrossref
CJ. Combination inhaled steroid and long-acting β2-agonist in addition to tiotropium vs tiotropium or combination alone for chronic obstructive pulmonary disease. Cochrane Database Syst Rev
. 2011;(3):CD008532.PubMedGoogle Scholar
P. Inhaled corticosteroids in COPD and the risk of serious pneumonia. Thorax
. 2013;68(11):1029-1036.PubMedGoogle ScholarCrossref