A correction for this article has been published in Respiratory Research 2008, 9:81

Research

Exacerbations of chronic obstructive pulmonary disease: when are antibiotics indicated? A systematic review

Milo A Puhan^*, Daniela Vollenweider, Tsogyal Latshang, Johann Steurer and Claudia Steurer-Stey

• * Corresponding author: Milo A Puhan milo.puhan@usz.ch

Author Affiliations

Horten Centre, University Hospital of Zurich, Postfach Nord, CH-8091 Zurich, Switzerland

For all author emails, please log on.

Respiratory Research 2007, 8:30 doi:10.1186/1465-9921-8-30

The electronic version of this article is the complete one and can be found online at: respiratory-research.com/content/8/1/30

Received:	19 December 2006
Accepted:	4 April 2007
Published:	4 April 2007

© 2007 Puhan et al; licensee BioMed Central Ltd.

This is an Open Access article distributed under the terms of the Creative Commons Attribution License (creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Background

For decades, there is an unresolved debate about adequate prescription of antibiotics for patients suffering from exacerbations of chronic obstructive pulmonary disease (COPD). The aim of this systematic review was to analyse randomised controlled trials investigating the clinical benefit of antibiotics for COPD exacerbations.

Methods

We conducted a systematic review of randomised, placebo-controlled trials assessing the effects of antibiotics on clinically relevant outcomes in patients with an exacerbation. We searched bibliographic databases, scrutinized reference lists and conference proceedings and asked the pharmaceutical industry for unpublished data. We used fixed-effects models to pool results. The primary outcome was treatment failure of COPD exacerbation treatment.

Results

We included 13 trials (1557 patients) of moderate to good quality. For the effects of antibiotics on treatment failure there was much heterogeneity across all trials (I² = 82%). Meta-regression revealed severity of exacerbation as significant explanation for this heterogeneity (p = 0.016): Antibiotics did not reduce treatment failures in outpatients with mild to moderate exacerbations (pooled odds ratio 1.09, 95% CI 0.75–1.59, I² = 18%). Inpatients with severe exacerbations had a substantial benefit on treatment failure rates (pooled odds ratio of 0.25, 95% CI 0.16–0.39, I² = 0%; number-needed to treat of 4, 95% CI 3–5) and on mortality (pooled odds ratio of 0.20, 95% CI 0.06–0.62, I² = 0%; number-needed to treat of 14, 95% CI 12–30).

Conclusion

Antibiotics effectively reduce treatment failure and mortality rates in COPD patients with severe exacerbations. For patients with mild to moderate exacerbations, antibiotics may not be generally indicated and further research is needed to guide antibiotic prescription in these patients.

Background

The use of antibiotics in exacerbations of chronic obstructive pulmonary disease (COPD) remains controversial [1,2]. It is unclear which patients should receive antibiotics. The uncertainty arises from a complex clinical situation where the cause of the exacerbation is often unidentifiable [3]. Around 40–50% of exacerbations may be attributed to bacteria while other causes include viral infections or environmental irritants [4-6]. Even if bacteria are identified, it is uncertain whether they actually caused the exacerbation or whether they were present as part of the flora before the exacerbation.

Diagnostic tests cannot reliably distinguish between bacterial, viral or other origins of exacerbations. As a consequence, many physicians decide to be on the "safe" side and prescribe antibiotics[7]. The uncertain role of antibiotics is reflected by current guidelines that insufficiently inform physicians about adequate prescription of antibiotics [3,8]. Guidelines suggest adding an antibiotic if sputum is purulent, if sputum volume is increased and/or if fever is present. However, evidence supporting this suggestion is not based on randomised trials. There are no randomised trials where prescription of antibiotics was guided by purulence of sputum or other criteria. In addition, the extent of symptom worsening is difficult to standardise and utility of sputum assessment is controversial [9,10].

A systematic review of randomised, placebo-controlled trials could inform the debate about the role of antibiotics substantially. Eleven years ago, a meta-analysis suggested a small improvement of lung function by antibiotics in COPD patients with an exacerbation, but the review was limited by the restriction to articles in English and its focus on lung function [11]. A recent systematic review [12] considered patient-important outcomes but missed some studies and included a non-randomised trial[13]. Inclusion of all available trials is, however, crucial to avoid selection bias and to study factors modifying the effects of antibiotics such as severity of exacerbation. Therefore, our aim was to review all randomised placebo-controlled trials that assessed the effects of antibiotics on patient-important outcomes in COPD patients suffering from exacerbations.

Methods

Selection criteria

We included randomised controlled trials comparing any antibiotics with placebo or no antibiotics in COPD patients suffering from an acute exacerbation defined as a worsening of a previous stable situation with symptoms such as increased dyspnea, increased cough, increased sputum volume or change in sputum colour. We considered studies if >90% of patients had a clinical (physician-based) diagnosis of COPD or, ideally, spirometrically confirmed COPD. We excluded studies of patients with acute bronchitis, pneumonia, asthma or bronchiectasis.

We included trials evaluating any antibiotics that were administered orally or parenterally daily for a minimum period of at least three days. We chose three days because this is the minimum duration for which antibiotics are usually prescribed in clinical practice for COPD exacerbations.

The outcome measure of primary interest was treatment failure defined as (1) no resolution of symptoms and signs as reported by patients or physicians or as (2) need for further antibiotics. Outcome measures of secondary interest were duration of hospital admission, admission to an intensive care unit, health-related quality of life, symptoms, mortality, and any adverse events registered during the study period.

Search strategy

The search was carried out by information specialists (Bazian, London, UK) and included searches in the Cochrane Central Register of Controlled Trials (CENTRAL, 2005 issue 4), PREMEDLINE (1960 to 1965), MEDLINE (1966 to March 2006), EMBASE (1974 to March 2006), the Database of Abstracts of Reviews of Effectiveness (DARE, March 2006). We entered all included studies into the Pub-med "related articles" function and the science citation index. In addition, we scrutinised the reference lists of included studies and review articles as well the conference proceedings of the international congresses of the American Thoracic Society and the European Respiratory Society from 2000 to 2006 since these studies might not have been fully published yet. We also contacted representatives of the pharmaceutical industry for additional published or unpublished data (Novartis, GlaxoSmithKline, AstraZeneca, BoehringerIngelheim, Pfizer and MSD). Finally, we searched international data bases for trial registration to identify ongoing or recently completed trials [14-16].

Study selection

Two members of the review team independently assessed the titles and abstracts of all identified citations without imposing any language restrictions. The reviewers then evaluated the full text of articles that seemed potentially eligible by one of the reviewers. Final decisions on in- and exclusion were recorded in the Endnote file and agreement was assessed using chance-adjusted kappa statistics.

Data extraction

One reviewer recorded details about study design, interventions, patients, outcome measures and results in predefined Windows Excel forms and a second reviewer checked data extraction for correctness. We used a small sample of studies with high likelihood for inclusion to pilot test the data form. To obtain missing information, we tried to contact authors of primary studies at least three times by telephone or email.

We entered dichotomous data on into 2 × 2 tables. For continuous outcomes, we recorded summary estimates per group (means, medians) with measures of variability (SD) or precision (SEM, CI). In trials with two groups receiving different antibiotics, we treated these groups as one group if the effects of the two antibiotics did not differ statistically significantly or clinically importantly.

Quality assessment

Two reviewers independently evaluated the quality of included trials using a list of selected quality items assessing components of internal validity [17]. We recorded the initial degree of discordance between the reviewers and corrected discordant scores based on obvious errors. We resolved discordant scores based on real differences in interpretation through consensus or third party arbitration.

Statistical analysis

We expressed treatment effects as odds ratios with corresponding 95% confidence intervals (CI) and calculated, based on pooled odds ratios, numbers-needed-to-treat. We pooled data across studies only in absence of significant heterogeneity (p > 0.1 for χ²) using fixed effects models (inverse variance method). We analysed comparisons with events only in one group by adding 0·5 to "zero-cells".

We assessed heterogeneity using χ² statistic and expressed the proportion of variation due to heterogeneity as I² [18]. We explored sources of heterogeneity using meta-regression following a priori defined explanations, which included severity of exacerbation (defined as severe if requiring inpatient treatment and as mild to moderate requiring outpatient treatment according to the Operational Classification of Severity of the European Respiratory and American Thoracic Societies [19]), generation of antibiotics (before and after 1980), definition of outcomes, length of follow-up (≤ and > 10 days) and study quality. We assessed publication bias using the regression-based test of Egger [20].

We conducted all analyses with STATA for windows version 8.2, Stata Corp; College Station, TX)

Results

Identification of studies

Figure 1 summarises the process of identifying eligible clinical trials. We identified 765 citations from electronic databases and selected 35 of them for full text assessment. Together with 30 additional citations from hand-searching we studied 65 publications in detail. We included 13 trials with 1557 COPD patients in the analyses. We excluded most trials because they compared different antibiotics without having a placebo control group. From trial registers, we identified four randomised trials that are still ongoing [21-24]. The pharmaceutical industry did not provide any unpublished data.

Study characteristics

Table 1 shows the characteristics of the trials that were published between 1957 and 2001. In seven trials, patients suffered from mild to moderate exacerbations receiving outpatient treatment [25-31]. Six trials included patients admitted to the hospital because of severe exacerbations [32-37]. Nouira [34] included patients with very severe exacerbations, who needed mechanical ventilation. Severity of underlying COPD could not be compared across trials because lung function and other parameters were reported inconsistently between 1957 and 2001. In all trials, patients received co-interventions such as systemic corticosteroids, theophylline, β-mimetics, gastric ulcer prophylaxis or ventilation support with or without oxygen. But the proportion of patients receiving co-interventions was rarely specified and could not be considered as potential confounders in the analyses.

Ten trials used treatment failure as an outcome although definitions varied from patient reported failure of symptom resolution to the physicians' decision to prescribe additional treatment [25-28,30-32,34,36,37]. Four trials including patients with severe exacerbations assessed mortality [34-37] and three trials [32-34] the duration of hospital stay.

In one trial, analyses were based on the number of 116 patients with exacerbations as well as on the total number of exacerbations (n = 362) [26]. In our meta-analyses, we considered the analysis based on the number of patients only because the other trials also followed this approach. In addition, Anthonisen et al used a cross-over design for patients with more than one exacerbation. Thereby, patients with more than one exacerbation counted in the antibiotic and placebo group. In addition assessing antibiotics with a cross-over design may not fulfil the important requirement for cross-over studies that patients must return to their baseline state before starting the cross-over. COPD patients often do not fully recover from exacerbations and are, therefore, unlikely to return to their baseline state.

The quality of the trials was moderate to good (table 2). Ten trials described their method of randomisation. Concealment of random allocation was reported in eight trials and in nine trials, outcome assessors were blinded. Initial agreement for quality assessment among the two reviewers was high (88% for all items, chance-corrected kappa = 0.75, p < 0.001).

Effects of antibiotics

Median treatment failure rate was 0.12 for the antibiotic groups (range 0.00 to 0.47) and 0.34 for the placebo groups (range 0.10 to 0.80). Thus across all trials, one out of eight patients with antibiotics had a treatment failure whereas one out of three patients had a treatment failure with placebo.

Figure 2 shows that the effects of antibiotics were very heterogeneous across trials (I² = 82%). When we explored predefined sources of heterogeneity in meta-regression analyses we found that generation of antibiotic (p = 0.55), definition of outcomes (p = 0.20), length of follow-up (p = 0.38) and study quality (p = 0.92) did not explain heterogeneity. We could not assess severity of COPD as a source of heterogeneity because lung function parameters were not reported in earlier trials.