Background
Long-Acting Beta-Agonists (LABAs) -salmeterol and formoterol- were introduced in the ‘90s when they demonstrated reducing symptoms and use of rescue medication [
1]. Concerns about their safety appeared in 1993 when Castle et al. reported a threefold mortality in a randomized clinical trial (RCT) comparing LABAs with SABAs [
2]. Post-marketing reports of adverse events showed an increased risk of death and serious asthma events [
3]. The Salmeterol Multicenter Asthma Research Trial was stopped in 2003 after an interim analysis showed a fourfold increased mortality amongst patients randomized to salmeterol vs. placebo [
4]. Similar concerns about formoterol were raised by a reanalysis of three RCTs. Meta-analyses of RCTs with LABAs as a monotherapy indicated an increased mortality risk [
5,
6].
Meta-analyses of RCTs examining the safety of LABAs in combination with inhaled corticosteroids (ICs) showed inconsistent results. Most of them found no significant differences in asthma-related hospitalizations and asthma-related mortality compared with patients treated with ICs alone [
7‐
11]. But a statistically significant increase of catastrophic asthma events for LABAs plus ICs was shown by the update of a meta-analysis [
12]. In 2010, the Federal Drug Administration (FDA) required label changes to indicate contraindication of use of LABAs without concomitant ICs, recommending only fixed-dose LABAs plus ICs combination, and calling for new studies to address this issue [
13].
Nevertheless, there is still a lack of knowledge regarding LABAs’ safety with concomitant ICs use, with both theoretical arguments and limited empirical evidence that ICs may mitigate LABA-associated risks [
14‐
16]. Most of the systematic reviews currently available are based on RCTs, which may present limitations to assess long-term and rare outcomes [
5,
8‐
11]. Moreover, RCTs may not reflect the actual patterns of use of these medications in asthma patients’ day-to-day regarding treatment duration and adherence. To our knowledge, there is only one systematic review of observational studies [
17]. Its meta-analysis showed that the combined treatment was associated with a lower risk of asthma-related hospitalizations and/or emergency room visits.
Since year 2008, end date of the above mentioned review, many non-randomized studies have been published, especially due to the FDA’s 2010 call for further evidence. The aim of this study was to assess the risk of serious adverse events in patients with asthma treated with LABAs and ICs in comparison to patients treated only with ICs, by synthesizing the available evidence from non-randomized studies through systematic review and meta-analysis.
Methods
The protocol registration number was CRD42012003387 (
http://www.crd.york.ac.uk/Prospero). We searched MEDLINE and EMBASE databases with a specific strategy (see Additional file
1) from 1990, when LABAs were commercialized, to January 20
th, 2013.
We looked for non-randomized studies in all languages (non-randomized controlled trials, controlled before-after studies, prospective or retrospective cohorts, case-control studies) on adults, adolescents or children with asthma diagnosis. Studies assessing treatment with LABAs plus ICs (either as two separate inhalers or as a single inhaler) compared with ICs monotherapy were considered, regardless of the dose (see Additional file
1). Co-therapy such as immunomodulators and leukotriene modifiers were not excluded. We defined ‘severe exacerbation’ following the American Thoracic Society/European Respiratory Society statement [
18] which was based on urgent health care utilization: asthma-related emergency department (ED) visits, hospitalizations, intubations, intensive care unit (ICU) admissions, and use of systemic corticosteroids were considered either specific or combined outcomes.
Two members of the study team, a physician (GH) and a pharmacist (MA), independently reviewed studies found in the literature search by examining titles, abstracts, and full text articles. A third reviewer (MF) resolved discrepancies. A pilot test was performed to homogenize criteria among reviewers. Finally, the selected articles’ reference lists were reviewed to identify other possible studies that could be included.
Data were extracted by agreement of two reviewers using a standardized, predefined data collection form, including: study and participants characteristics, interventions, comparator, outcomes, asthma severity, co-medication, and ethics consideration of each study. Authors were contacted if clarification was needed.
The risk of bias in the identified studies was assessed using a checklist developed by members of the Cochrane Non-Randomised Studies Methods group [
19]. We assessed 4 categories of potential biases: groups of comparison, reasons for allocation in groups, parts of the study that were prospective, and group comparability (Additional file
1).
Analytic strategy
Reported adjusted OR and 95% confidence intervals (95% CI) for the comparison of ICs plus LABAs versus ICs alone were considered. Where adjusted ORs were not reported, unadjusted ORs were held. To assess the risk of severe exacerbation in patients with asthma treated with LABAs plus ICs, compared to those treated only with ICs, meta-analyses were carried out for individual specific adverse events and combined outcomes. Subgroup analyses for children and administration mode were planned. The summary OR and 95% CI estimated in the meta-analyses, together with ORs from individual studies, were presented in forest plots.
Heterogeneity among studies was evaluated using Galbraith plot and I
2 statistic categorized as follows: <30% not important; 30%-50% moderate; 50%-75% substantial; and 75%-100% considerable [
19]. If significant heterogeneity was identified among studies, further examination of the individual studies was conducted, and random effects models (Dersimonian-Laird Method) were used to obtain the summary OR estimates. Otherwise, fixed effects models were used (Mantel-Haentzel Method). Publication bias was assessed by Egger regression asymmetry test and funnel plots. The meta-analytic software program used was STATA.12.
Discussion
To date, less than 10% of all systematic reviews have adverse events’ assessment as a primary objective [
39]. Our findings support the relevance and suitability of performing systematic reviews of harms to provide valuable information on these risks. This systematic review identified 19 studies which met the inclusion criteria: 16 retrospective cohorts, 1 prospective cohort, 1 case-control, and 1 before-after study (1,165,342 participants). The meta-analyses performed (69,939-624,303 participants according to the outcome considered) showed that the LABAs/ICs combined treatment was not associated to a higher risk of adverse events, when compared with ICs alone. The OR ranged from 0.75 to 1.02 for the different outcomes explored, which is congruent with findings from meta-analyses of RCTs assessing asthma-related serious adverse events for salmeterol (0.95; 95% CI 0.52-1.73) [
8] and formoterol (0.53; 95% CI 0.28-1.0) [
10].
This consistency between our results and those from meta-analyses of RCTs reinforces the evidence available on this topic. It is well known that RCTs are the gold standard in evaluating efficacy and safety of emerging therapies. However, their poor external validity [
40] is a particular concern for long term chronic conditions that affect large and heterogeneous patient populations, such as asthmatics. In fact, it has been estimated that only 1.2% [
41] or 5% [
42] of the usual care asthma population could have been eligible for a typical asthma RCT. In this context, despite potential issues regarding observational studies’ internal validity, they are gaining widespread recognition [
43,
44] providing valuable information on treatment effectiveness and safety, especially in long-term outcomes.
To our knowledge, this is the first systematic review of non-randomized studies including children and adults with asthma to assess adverse events of LABAs, as the only systematic review of observational studies previously published was limited to adult patients (asthma-related hospitalizations OR = 0.85; 96%CI 0.74-0.97) [
17]. It included mainly unpublished studies identified from a pharmaceutical company’s research register. Since then, the publication of 12 observational studies permitted the inclusion of a larger number of patients. In comparison with systematic reviews of salmeterol and formoterol RCTs (with 15,309 and 13,366 patients, respectively), synthesis of non-randomized studies provides results from larger representative asthma samples, more accurate reflection of the usual clinical practice, and longer follow-up periods. The follow-up periods of studies included in our systematic review ranged from 3 months to 5 years, being in most cases 1 year (12 studies), an adequate frame of time for the assessment of adverse events [
19].
We identified several limitations on our review process. First, four retrospective cohorts could not be included in any of the meta-analyses performed due mainly to the lack of the specific estimator needed, but their results were consistent with our findings [
21,
22,
37,
38]. Second, outcomes of these retrospective cohorts varied substantially, from systemic corticosteroids use to deaths. Related to this wide range of clinical outcomes, there was a limitation for synthesizing them by meta-analysis – mortality, a primary outcome of interest, was reported only by one study [
28]. Furthermore, the use of composite endpoints could give misleading conclusions because the components have different relevance [
45]. However, not only the composite endpoints, but also the individual adverse events which compose them have been considered in the meta-analyses. Third, internal validity of the summary provided by a meta-analysis depends on the quality of primary studies. Confounding and selection bias could distort the findings from observational studies and therefore meta-analyses including them would produce biased estimates also. In our systematic review, sensitivity analysis by quality assessment was not performed as risk of bias was homogeneous among studies. Quality assessment was considered moderate for most of them because the studies were mainly comparative, allocation was based on treatment decisions, and adjusted by potential confounders. In fact, only few unadjusted estimators were included in the meta-analysis, and the sensitivity analyses carried out to assess the impact of excluding them, showed similar summary ORs: 0.89 (95% CI 0.69-1.15) for asthma-related hospitalization and 0.75 (95% CI 0.67-0.85) for asthma-related ED visits.
Most of the retrospective cohort studies identified in this systematic review obtained data from administrative medical claims and electronic health records, with definitions based on medication prescriptions and ICD-9 diagnosis (i.e. asthma codes for inclusion and other respiratory conditions for exclusion). The main limitations derived from designs of this nature include: a) presence of a prescription claim does not necessarily indicate that the medication was taken; and b) asthma severity criteria were not applied in most studies, and in those that did, severity definitions were based on medication use instead of spirometry or clinical parameters. To balance treatment groups, most of the studies made adjustments on baseline risk factors and socioeconomic variables by using regression models and propensity score matching. Nevertheless, possible confounding factors such as severity and adherence could still remain.
The planned subgroup analysis for children and administration mode (as single or two separate inhalers) could not be conducted, and merits further comments. Stanford et al. [
25] performed an analysis stratified by age groups with similar results for adults and children aged 4-18 years: OR was 0.917 (95% CI 0.85-0.98) for ED visit and 0.88 (95% CI 0.7-1.11) for hospitalization. The case-control study by Jacobs et al. [
23] showed that paediatric LABA use in combination with ICs did not increase the likelihood of intensive care unit admission among hospitalized children, compared to ICs alone. Regarding administration mode, the little available evidence is controversial. The largest retrospective cohort identified in this review [
21] is remarkable because it showed higher risk for single inhalers compared with separate inhalers on a combined outcome composed of asthma-related hospitalizations, intubations or asthma-related ED visits: OR of 1.13 (95% CI 1.09-1.16) among newly diagnosed patients, and OR of 1.12 (95% CI 1.10-1.12) among those with pre-existing asthma. On the contrary, in the study by O’Connor et al. [
38] patients receiving LABAs plus ICs in a single inhaler were less likely to have an ED visit or to be hospitalized, compared with patients receiving the same treatment in separate inhalers (OR 0.69, 95% CI 0.51-.95). Delea et al. [
34] showed similar results in both administration modes.
Heterogeneity was substantial (66.5% and 70.5%) in two of the four meta-analyses reported. In those conducted with asthma-related hospitalization risk, the only study out of the confidence limits in the Galbraith plot was deVries et al. [
28]. This study differs from the other ones in having a follow-up period of 5 years, but many other possible reasons could explain such heterogeneity. In the meta-analysis conducted with the combined outcome, the only estimator that fell outside the Galbraith plot limits was the group with high dose of corticosteroids and salmeterol in Campbell et al. [
32], with an OR higher than 1. This might reflect that despite the adjustments, patients taking high corticosteroid doses represented a more severe group.
Almost two thirds of the studies were performed by Glaxo Smith Kline Beecham, while others received industry support without describing the extent of involvement of their sponsors. Usually, publication bias refers to the journals’ rejection of studies with negative results. Yet safety studies sponsored by the pharmaceutical industry could suffer from publication bias in the opposite direction, as it is more likely to publish negative results and to select the most favourable outcomes. We have found no evidence of publication bias in the meta-analyses reported, but Egger’s test has limited power when the number of studies is low, and funnel plots may have subjective interpretation.
Acknowledgments
We would like to thank A. Martin (IMIM Hospital del Mar Medical Research Institute) for the English review and editing of the manuscript.
Members of the ASTRO-LAB group are: Marijn De Bruin, Alexandra Dima (ASCoR, University of Amsterdam, The Netherlands); Eric Van Ganse, Laurent Laforest, Sandrine Herbage, Manon Belhassen, Marine Ginoux (LBBE, University Claude Bernard Lyon 1, France); Stéphane Schück, Nathalie Texier, Sandy Leproust, Hélène Le Cloarec (Kappa Santé, France); Richard Hubbard (University of Nottingham, England); Alison Bourke, Mary Thompson , Delphine Vial, David Ansell (Cegedim Strategic Data, England); Javier Olaiz, Ana Valcarcel Orti (Lyon Ingénierie Projet, France) ; and Montse Ferrer, Olatz Garin, Gimena Hernández (IMIM - Hospital del Mar Medical Research Institute, Spain).
Fundings
Financial support for this study was provided by the Health Research Fund (European Union FP7, ASTROLAB project EC HEALTH-F5-2011-282593); the Agency for Management of University and Research Grants AGAUR (2012FI_B1 00177); and a Fundación Carolina Fellowship. The funding agreement ensures the authors’ independence in designing the study, interpreting the data, and writing and publishing the report. None of these organizations had any role in the design or conduction of the study, nor in the data collection, management or interpretation, nor in the manuscript writing, reviewing, or approval.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.
The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
GH contributed to the conception and design of the article, conceptualized and oversaw analyses, contributed to the interpretation of data, and wrote the article. MA contributed to the reviewing and web search of included and excluded articles. AP contributed to the analysis and gave statistical support. OG, JA, CC, LL oversaw all aspects and reviewed the article for important intellectual content. MF oversaw all aspects, contributed to the conception and design of the article, contributed to the statistical analyses, carried out the interpretation of data, and contributed to the writing of the article. All the co-authors critically revised the manuscript and approved the final draft before submission.