Types of studies

Randomised controlled trials (RCTs) comparing short‐duration systemic corticosteroid treatment (seven or fewer days) versus conventional, longer‐duration treatment (longer than seven days).

Types of participants

Adults with an acute exacerbation of COPD. The definition of an acute exacerbation could include any combination of an increase in breathlessness, sputum volume, sputum purulence, cough or wheeze. We excluded studies that included patients with asthma and other lung diseases (e.g. interstitial lung disease, bronchiectasis), unless separate data on participants with COPD alone were available.

Types of interventions

Systemic corticosteroid (SCS) given for a period of seven or fewer days (SCS ≤ 7) versus systemic corticosteroid given for longer than seven days (SCS > 7). Co‐interventions were required to be standardised across groups. We excluded studies in which participants received assisted ventilation (invasive or non‐invasive).

Types of outcome measures

Electronic searches

We identified trials from the Cochrane Airways Trials Register, which is maintained by the Information Specialist for the Group. The Cochrane Airways Trials Register contains studies identified from several sources:

1. Monthly searches of the Cochrane Central Register of Controlled Trials (CENTRAL), through the Cochrane Register of Studies Online (crso.cochrane.org);

2. Weekly searches of MEDLINE Ovid SP 1946 to date;

3. Weekly searches of Embase Ovid SP 1974 to date;

4. Monthly searches of PsycINFO Ovid SP 1967 to date;

5. Monthly searches of CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature) 1937 to date;

6. Monthly searches of AMED EBSCO (Allied and Complementary Medicine);

7. Handsearches of the proceedings of major respiratory conferences.

Studies contained in the Trials Register are identified through search strategies based on the scope of Cochrane Airways. Details of these strategies, as well as a list of handsearched conference proceedings are in Appendix 1.

We conducted additional supplementary searches of MEDLINE (PubMed platform, dates covered 1950 to March 2017) and CENTRAL (2017, Issue 3). EMBASE (Embase.com platform, dates covered 1947 to July 2014) searches were conducted in previous versions of this review.. Search strategies used in the Cochrane Airways Register of Trials and other databases are shown in Appendix 2.

For this review update, searches were conducted in the Cochrane Airways Group Register of Trials, and in the other databases in March 2017.

Searching other resources

Bibliographies of RCTs and review articles identified for inclusion were searched for additional papers. Registers of ongoing trials at Clinicaltrials.gov and the WHO trial portal were searched in March 2017.

Selection of studies

All potentially relevant trials were assessed for relevance by at least two review authors (DT, CW), who screened the full text to independently select trials for inclusion and to identify and record reasons for exclusion of ineligible studies. We resolved disagreements through discussion or, if required, we consulted a third person (JW). We identified and excluded duplicates and collated multiple reports of the same study, so that each study (rather than each report) was the unit of interest in the review. We recorded the selection process as a PRISMA flow diagram and in the Characteristics of excluded studies table.

We initially screened all studies using the following criteria.

1. Acute exacerbation of COPD?

2. Systemic corticosteroids as intervention?

3. Comparison of short‐duration (seven or fewer days) versus longer‐duration (longer than seven days) corticosteroid therapy?

4. Randomised controlled trial?

We noted other arms and reasons for study exclusion when appropriate. We determined exclusion using the criteria listed above and categorised the following as applicable.

1. Not different durations of steroids.

2. Not exacerbations of COPD.

3. Not treatment with systemic corticosteroids.

4. Review or other type of article.

We also assessed each study for accuracy of the diagnosis of COPD using the following criteria (GOLD 2014).

1. Did the patient have dyspnoea, chronic cough or sputum production?

2. Was there a history of exposure to risk factors including tobacco smoke; occupational dusts and chemicals; and smoke from home cooking and heating fuel?

3. Was a spirometry measurement of the FEV₁/FVC ratio less than 0.7 post bronchodilator?

Data extraction and management

We used a data collection form to document study characteristics and outcome data. Two review authors (DT, CW) extracted the following study characteristics from included studies.

1. Methods: study design, total duration of study, number of study centres and locations, study setting, withdrawals and date of study.

2. Participants: N, mean age, age range, gender, COPD diagnostic criteria, diagnostic criteria for exacerbation, baseline lung function, smoking history, inclusion criteria and exclusion criteria.

3. Interventions: systemic corticosteroid treatment dose/schedule, concomitant medications and excluded medications.

4. Outcomes: primary and secondary outcomes specified and collected and time points reported.

5. Notes: funding for trial and notable conflicts of interest of trial authors.

Two review authors (two of JW, DT, CW) independently extracted outcome data from included studies. Results presented in graphical form were extracted using the software Plot Digitizer. When results were available as means with interquartile ranges, standard deviation was based on the width of the interquartile range—approximately 1.35. standard deviations.

Data were entered into Review Manager (by JW, DT, CW) and were double‐checked by a second review author. We checked that data were entered correctly by comparing data presented in the systematic review against data provided in the study reports.

Assessment of risk of bias in included studies

Two review authors (of JW, CW, DT) independently assessed the risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We resolved disagreements by discussion or by involving another review author. We assessed risk of bias according to the following domains.

1. Random sequence generation.

2. Allocation concealment.

3. Blinding of participants and personnel.

4. Blinding of outcome assessment.

5. Incomplete outcome data.

6. Selective outcome reporting.

7. Presence of other bias(es).

We graded each potential source of bias as high, low or unclear, and provided a quote from the study report, together with a justification for our judgement, in the 'Risk of bias' table. We summarised risk of bias judgements across different studies for each of the domains listed. When information on risk of bias was related to unpublished data or correspondence with a trialist, we noted this in the 'Risk of bias' table.

When considering treatment effects, we took into account the risk of bias for studies that contributed to this outcome.

Measures of treatment effect

For continuous variables, data were analysed as mean differences (MDs) with 95% confidence intervals (CIs). Standardised mean differences (SMDs) with 95% CIs were used when different scales of measurement had been used for an outcome. The SMD is a statistic that expresses the difference in means between treatment groups in units of the pooled standard deviation. Dichotomous outcomes were analysed using Mantel‐Haenszel odds ratios with 95% CIs. When events were rare, we employed the Peto odds ratio. Scale data were entered with a consistent direction of effect.

We undertook meta‐analyses only when these were meaningful—when treatments, participants and the underlying clinical question were similar. When data were skewed, we described them narratively.

For 'time‐to‐event' outcomes such as log hazard ratios, the fixed‐effect generic inverse variance outcome was used to combine results. This method provides a weighted average of the effect estimates of separate studies (Higgins 2011). Number needed to treat for an additional beneficial outcome was calculated from the pooled odds ratio and its confidence interval, using baseline risk in the control group.

Unit of analysis issues

We analysed dichotomous data using participants as the unit of analysis. For continuous data, the MD based on change from baseline was preferred over the MD based on absolute values when both were available.

Dealing with missing data

We contacted investigators to obtain missing numerical outcome data when possible (e.g. when a study was reported as an abstract only). When this was not possible, and when missing data were thought to introduce serious bias, we performed a sensitivity analysis to explore the impact of including such studies in the overall assessment of results.

Assessment of heterogeneity

An assessment of possible heterogeneity, when the null hypothesis is that all studies are evaluating the same effect, was carried out for pooled effects using a Breslow‐Day test of heterogeneity; the Chi² test measures the deviation of observed effect sizes from an underlying overall effect. This test has low power for detecting true heterogeneity when studies have small sample size or are few in number, hence we used a P value of 0.10 (Deeks 2008). In addition, the I² statistic, which describes the percentage of total variation across studies that is due to heterogeneity rather than to chance (Higgins 2003), was used. Statistical heterogeneity was interpreted as follows: 0% to 40% might not be important, 30% to 60% may represent moderate heterogeneity and 50% to 90% may represent substantial heterogeneity (Higgins 2011).

We assessed clinical and methodological heterogeneity by recording differences in study design and participant characteristics between individual studies. When we found substantial heterogeneity, we reported this and explored possible causes by performing prespecified subgroup analysis.

Assessment of reporting biases

We tried to minimise reporting bias from non‐publication of studies or selective outcome reporting by using a broad search strategy. We checked references of included studies and relevant systematic reviews, and we contacted study authors to request additional outcome data. To assess the presence of publication bias, we visually inspected funnel plots when 10 or more studies contributed to the analysis for a specified outcome.

Data synthesis

Data were pooled using Review Manager 5.3 (RevMan 2014). We used a fixed‐effect model, and when heterogeneity could not be explained, we performed a sensitivity analysis using a random‐effects model. We presented the findings of our primary outcomes in a 'Summary of findings' table according to recommendations provided in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) (generated with the use of GradePro software).

Subgroup analysis and investigation of heterogeneity

When we detected statistically significant heterogeneity (P value < 0.10), several options were available. We explored the clinical diversity of studies through prespecified subgroup analyses by comparing the following factors.

1. Definition of COPD.

2. Aspects of study quality (e.g. concealment of allocation).

3. Characteristics of participants.

We considered random‐effects analysis methods (DerSimonian 1986) when heterogeneity was unexplained and study sizes were large, or when study findings were not pooled.

We performed subgroup analyses, stratified by the following variables, when possible.

1. Inpatient versus outpatient.

2. Studies that included participants previously treated with corticosteroids (inhaled and systemic).

Sensitivity analysis

In assessment of heterogeneity, possible causes arising from details of study design were considered. Sensitivity analyses were performed using a random‐effects model versus a fixed‐effect model for risk of bias and other potential confounders.