Health-related quality of life
Improvement on the St George's Respiratory Questionnaire (SGRQ) was greater with LABA therapy than with placebo (MD -2.32, 95% CI -3.09 to -1.54; I2 = 50%, P = 0.007; Analysis 1.1), based on data from 11,397 people in 17 studies. Results were analysed using a random-effects model because heterogeneity was high, and the outcome was downgraded from high to moderate quality for this reason. Heterogeneity could not be explained by differences between the effects of LABA drug and dose (test for subgroup difference, I2 = 0%, P = 0.45). The difference between LABA and placebo was significant for all three LABA agents separately, but important heterogeneity was noted within the formoterol 12 μg and salmeterol 50 μg subgroups (I2 = 57% and 48%, respectively).
More people taking a LABA showed clinically important improvement of at least four points on the SGRQ (OR 1.58, 95% CI 1.32 to 1.90; I2 = 86%, P = 0.0007; Analysis 1.2) based on 1871 people in three studies. Data were insufficient to allow conclusions regarding the difference between formoterol and salmeterol.
Four salmeterol studies reporting the Chronic Respiratory Disease Questionnaire (CRQ) were analysed separately and showed a significant benefit of LABA therapy (MD 3.10, 95% CI 1.22 to 4.98; I2 = 0%, P = 0.58; Analysis 1.3) with no heterogeneity. Because so few trials reported the CRQ and the number of people achieving a minimally clinically important difference (MCID) on the SGRQ of 4 units, we did subgroup and sensitivity analyses on the SGRQ continuous data only.
A test for subgroup differences suggested that some of the heterogeneity within the 12 trials which randomised people to either formoterol or placebo may be explained by differences in study duration (I2 = 75%, P = 0.02; Analysis 2.1). The effect of twice-daily formoterol 24 μg was significant when analysed separately but lacked precision because only three studies compared the dose with placebo (MD -2.32, 95% CI -4.52 to -0.13; I2 = 0%, P = 0.76; Analysis 3.1). No difference was detected between the two three-month trials (Dahl 2001; Wadbo 2002) and the year-long Rossi 2002 study. With so few studies, no heterogeneity was evident between or within subgroups. Unlike the formoterol 12 μg analysis, trial duration did not appear to be an important source of heterogeneity in the salmeterol included trials (I2 = 0%, P = 0.41; Analysis 4.1).
The overall pooled effect and drug subgroup effects were largely consistent with the original analysis after removal of five studies identified as having a high proportion of ICS use during the trial (see Included studies—Use of inhaled corticosteroids (ICS) during the study period). The magnitude of improvement due to LABA therapy was reduced after the studies were removed, but somewhat tighter confidence intervals and less between-study heterogeneity were noted (MD -1.53, 95% CI -2.08 to -0.98; I2 = 34%, P = 0.12; Analysis 5.1).
Four of the eight studies rated as having high risk of bias reported the SGRQ: three from the formoterol 12 subgroup and one from the salmeterol subgroup. After these studies were removed, the pooled difference between LABA and placebo remained significant but was again reduced (MD -1.60, 95% CI -2.15 to -1.05; I2 = 42%, P = 0.04; Analysis 6.1).
Severe COPD exacerbations (requiring hospitalisation)
When data from seven studies were combined (nine comparisons), the number of people hospitalised for COPD exacerbations was lower among LABA-treated participants than in those receiving placebo (OR 0.73, 95% CI 0.56 to 0.95; Analysis 1.4). Although little evidence was found for explicit selective outcome reporting within trials, the outcome was downgraded for publication bias and was rated as having moderate quality; more than half of the studies did not report the outcome of high interest to patients and clinicians, and the estimate might have been different if all studies could have been included in the analysis (see Included studies—Outcomes and analysis structure). No heterogeneity was noted between trials, and the results of a test for subgroup differences between drugs/doses were not significant (I2 = 57.5%, P = 0.10).
No evidence of a statistically significant effect of trial duration was observed in the formoterol 12 μg trials (six studies across three duration subgroups). Similarly, no significant difference was observed between the three-month Dahl 2001 study and the 12-month Rossi 2002 study comparing formoterol 24 μg with placebo. Only one six-month study comparing salmeterol and placebo reported hospitalisation rates, so no assumptions could be made regarding relative effects of trial duration.
The pooled effect of LABA versus placebo decreased in magnitude and was no longer significant after four studies with a high percentage of participants taking inhaled corticosteroids (six comparisons) were removed. This effect was based on only three remaining trials (OR 0.85, 95% CI 0.62 to 1.16; Analysis 5.2), and the change in effect could be explained by another variable. None of the eight studies rated as having high risk of bias reported the number of participants with exacerbations requiring hospitalisation, so the outcome does not appear in the sensitivity analysis.
Moderate COPD exacerbations (requiring a course of antibiotics and/or oral steroids)
LABA treatment reduced the number of people who had one or more exacerbations requiring a course of antibiotics, oral steroids or both compared with placebo (OR 0.73, 95% CI 0.61 to 0.87; I2 = 8%, P = 0.37; Analysis 1.6), with minimal heterogeneity. The quantity of data for the outcome was similar to that provided for severe COPD exacerbations, so the outcome was downgraded for publication bias for the same reasons and was rated as having moderate quality. No evidence suggested important subgroup differences on the basis of which LABA/dose was used (I2 = 0%, P = 0.68). The three studies that compared formoterol 12 μg with placebo showed no significant benefit (OR 0.78, 95% CI 0.56 to 1.07; I2 = 7%, P = 0.34; Analysis 2.3) with little heterogeneity, and no evidence was found of differential effectiveness related to study duration. The higher-dose formoterol 24 μg did show a significant reduction relative to placebo (OR 0.57, 95% CI 0.38 to 0.88; Analysis 3.3), although this was based on a total of 124 events in one study (Rossi 2002). Four studies that compared salmeterol with placebo showed an overall reduction in moderate exacerbations with the study drug (OR 0.75, 95% CI 0.60 to 0.94; I2 = 40%, P = 0.17; Analysis 4.3), with some non-significant heterogeneity and no observed differences related to study duration.
Moderate/severe COPD exacerbations (requiring hospitalisation or a course of antibiotics/oral steroids or ER visit)
Around half of the studies did not report rates of moderate and severe COPD exacerbations as separate outcomes, as we had defined in the protocol, but rather reported a composite rate, which could not be combined with the other data; therefore we decided to present these data separately in the review. As with the other two exacerbation outcomes, the outcome was rated of moderate quality after it was downgraded for publication bias.
LABA treatment did not significantly reduce the number of people with moderate or severe exacerbations compared with placebo (OR 0.88, 95% CI 0.76 to 1.02; I2 = 0%, P = 0.80; Analysis 1.5), as determined on the basis of seven studies with 1142 participants. No between-study or between-drug heterogeneity was observed. Since the outcome was added, post hoc and subgroup analyses have been reported for the other two exacerbation outcomes, and we did not perform separate subgroup analyses of trial duration for this outcome.
Study deaths were relatively uncommon; therefore the analyses were conducted using Peto odds ratio, as this method does not require adjustment for zero cells. LABA treatment did not significantly reduce mortality compared with placebo (OR 0.90, 95% 0.75 to 1.08; I2 = 21%, P = 0.21; Analysis 1.7), and a degree of between-study heterogeneity was noted. The outcome was downgraded for imprecision and was rated as having moderate quality because the confidence intervals for the pooled effect included important benefit and potential harm. A test for subgroup differences between drugs was not statistically significant (I2 = 55%, P = 0.11). Results of a test for subgroup differences related to the duration of the formoterol trials were not significant (Analysis 2.4). In the two formoterol 24 μg trials that reported all-cause mortality, only one death occurred in the LABA group of Rossi 2002; the confidence intervals were too wide to allow interpretation of direction or magnitude of effects (Analysis 3.4) or assumptions based on trial length. Four studies with no events did not contribute to the salmeterol analysis (Analysis 4.4), and the effect was largely influenced by the large Calverley 2007 [TORCH] study (93% of total weight). This large study was unique because it was much longer than the other trials (at three years), and investigators logged mortality for all participants, regardless of how long they stayed in the study. Between-trial heterogeneity was not significant in the salmeterol analysis (I2 = 29%, P = 0.22); a test for subgroup differences regarding trial duration also was not significant (I2 = 0%).
Non-fatal serious adverse events; all-cause
All studies included in the analysis reported this outcome with participants as the level of analysis (i.e. number of people who had serious adverse events as opposed to the number of adverse events in total). When findings of all studies were pooled, no difference was observed between LABA and placebo (OR 0.97, 95% CI 0.83 to 1.14; I2 = 34%, P = 0.06; Analysis 1.8). Heterogeneity was significant at P = 0.1, and unexplained differences between the two formoterol doses showed opposite directions of effect. A test for subgroup differences indicated that some of the heterogeneity may be explained by these differences in individual drugs/doses, which were significant (I2 = 83%, P = 0.002). It is unclear whether formoterol 12 μg significantly increases rates of serious adverse events, as the confidence interval touched the line of no effect (OR 1.20, 95% CI 1.00 to 1.43; I2 = 14%, P = 0.32; Analysis 2.5), and differences in trial duration were not statistically significant. Three formoterol 24 μg studies showed that serious adverse events were lowered by LABA use (OR 0.53, 95% CI 0.36 to 0.79; I2 = 0%, P = 0.67; Analysis 3.5) with no significant heterogeneity, although confidence intervals were quite wide. Studies that compared salmeterol 50 μg with placebo showed no significant differences between groups (OR 0.94; 95% CI 0.83 to 1.06; I2 = 13%, P = 0.33; Analysis 4.5) based on nine studies with 1608 events. Some statistically insignificant heterogeneity between trial results was noted and could not be explained by reliable differences in trial duration.
Predose forced expiratory volume in one second (trough FEV1)
The predose FEV1 of participants taking LABA was 73 mL higher at the end of the trials than that of participants taking placebo inhalers (95% CI 48 to 98; I2 = 71%, P < 0.0001; Analysis 1.9); this finding was based on data from 6125 participants in thirteen studies that reported the outcome. A large degree of heterogeneity was noted, so the analysis was downgraded for inconsistency. Half of the studies did not report the outcome or reported the outcome in a way that could not be entered in meta-analysis; therefore it was downgraded for publication bias and was rated as having low quality. A test for subgroup differences suggested that the heterogeneity may be accounted for by differences between formoterol 12 μg and salmeterol 50 μg (no formoterol 24 studies reported trough FEV1) (I2 = 84%, P = 0.01). Individually, formoterol 12 μg (MD 45 mL, 95% CI 29 to 60) and salmeterol 50 μg (MD 101 mL, 95% CI 60 to 142) were associated with improved predose FEV1 relative to placebo.
No heterogeneity was observed between the formoterol 12 μg studies (I2 = 0%, P = 0.57), and no observable differences related to trial duration were reported. In the salmeterol 50 μg studies, heterogeneity was substantial (I2 = 69%, P = 0.003), and a test for subgroup differences of study duration suggested that the benefits of LABA treatment over placebo become less distinct over time (I2 = 87%, P = 0.0005; Analysis 4.6). Some of the heterogeneity may have been introduced by within-subgroup variation in the recruited populations. For example, Rennard 2001 split the population by high and low reversibility to compare the effects of salmeterol on different participant groups, and the data entered into the analysis represent the population as a whole.
Withdrawal from study treatment
Withdrawal rates were higher for placebo than for LABA treatment (OR 0.74, 95% CI 0.69 to 0.80; I2 = 0%, P = 0.75; Analysis 1.10), as determined on the basis of data from all studies except Watkins 2002. The formoterol 24 μg studies contributed the least data to the analysis and showed no difference between LABA and placebo, although subgroup differences between drugs and doses were not significant (I2 = 0%, P = 0.66). No heterogeneity between trials was noted, and no evidence showed significant subgroup differences between dose categories. No clear effects of trial duration were evident.