The medical community has made up its mind that, although inhaled corticosteroids reduce exacerbations in patients with chronic obstructive pulmonary disease (COPD),¹ they do not affect disease progression.² Despite measurement of forced expiratory volume in one second (FEV₁) being widely available and a strong predictor of mortality,³ the emphasis has moved to softer outcome measures which do show changes with available treatments. Ten years ago many of us had different beliefs—several small studies using simple statistics suggested that the rate of decline in FEV₁ could be reduced by about 20 ml a year by inhaled corticosteroids.^4–6 This led to several large studies, the primary outcome of which was decline in FEV₁ and which were powered to detect a 20 ml/year difference between active and placebo treatments.^7–10 All failed to show significant differences in their primary outcome of FEV₁ decline between various inhaled corticosteroids and placebo.

In this issue Sutherland and colleagues have done a meta-analysis of those trials, and have shown that inhaled corticosteroids do slow the decline in FEV₁ significantly.¹¹ How should we respond to this meta-analysis? Are the differences found clinically meaningful? Was there anything wrong with our original hypotheses or analyses? And what should we make of a similar meta-analysis that appeared to come to opposite conclusions?¹²

Despite the lack of clinical licences, inhaled corticosteroids have been widely prescribed for patients with COPD. In some parts of the world more that 50% of patients were receiving these drugs “off licence” by the mid 1990s.^8,13 This has allowed various database studies to estimate the effect on mortality of prescribing inhaled corticosteroids.

There are problems in adequately controlling for confounders in non-randomised studies; despite this, the extent of the reduction in mortality seen in some of these studies was much larger than expected,^13–15 and the study showing the largest reduction in mortality comparing the regularity of inhaled corticosteroid prescription would appear free of immortal time bias.¹⁴ Could mortality be reduced by so much if inhaled corticosteroids were not altering disease progression? The problems of unknown confounding can be overcome with randomised trials, giving added importance to the current meta-analysis, which shows a mean reduction of 7.7 ml/year FEV₁ decline with inhaled corticosteroids. Is this enough to explain the mortality reduction suggested by the database studies?

The meta-analysis used the estimates of FEV₁ decline derived from mixed effects models used in the original trial analyses.^16,17 There are unexplained differences in the estimates of FEV₁ decline using these models and those observed using linear regression before study entry. For instance the EUROSCOP study had a six month run-in with no active treatment. In individuals who had not previously taken inhaled steroids the mean FEV₁ declined by 113 ml/year in those subsequently randomised to placebo. The mixed effects model for the three years on placebo gave an estimate of 69 ml/year.⁷ In contrast, the Copenhagen City lung study, where the patients had much less advanced COPD, had similar rates of decline during a 13 year pretrial period and in the three year placebo treatment period, at 52 and 49.6 ml/year, respectively.⁹

The best explanation at present divides those seen cross sectionally into two groups. In one the FEV₁ is declining rapidly—these patients preferentially drop out of long term studies; they show a meaningful reduction in the rate of FEV₁ decline with inhaled corticosteroids, but their failure to complete studies means that they are under-represented in the mixed effects analysis. The second group have arrived at similarly low values of FEV₁ but now have stable disease and no room for an FEV₁ response from treatment. The Isolde study supports this model,¹⁸ where those withdrawn from the study randomised to placebo started with a higher FEV₁ than those randomised to fluticasone; their FEV₁ declined by 95.3 ml/year compared with 74.4 ml/year in the fluticasone group. Those who completed the three years of the trial declined at 50.7 and 46.4 ml/year, respectively. The message appears to be in those dropping out of the study after randomisation to fluticasone propionate, rather than those completing it. Mixed effects models are conservative when, as here, those who drop out have a larger treatment benefit than those completing the trial, as they only contribute data up to the point of withdrawal, minimising the overall estimates of change.

Attempts have been made to identify those who will and will not benefit from inhaled corticosteroid treatment. Neither short term response to bronchodilator nor oral corticosteroid use predicts long term response.^19–21 The presence or absence of emphysema is also not related to short term corticosteroid response.²² Whether pathology is related to long term inhaled corticosteroid response is unknown, and needs investigation.

The optimal dose of inhaled corticosteroids in COPD is still unknown. None of the large randomised trials used more than one dose. The reanalysis of smaller studies showed some evidence that beclomethasone dipropionate 800 μg/day was less effective than ⩾1500 μg/day.⁴ The observational studies also show less efficacy with doses of <500 μg/day compared with larger doses,¹⁵ and less effect when fluticasone prescriptions were repeated less regularly compared with 12 times a year.¹⁴ In the current meta-analysis, the lung health study—using triamcinolone 1.2 mg daily (equivalent to about 600 μg beclomethasone dipropionate)—showed less effect on FEV₁ decline, at 2.8 ml/year v 9.9 ml/year for the studies using budesonide 800 μg/day or fluticasone propionate 1 mg/day. Ten millilitres a year is still a small effect. However, it could be interpreted as reducing the excess FEV₁ decline caused by COPD (over and above the 30 ml/year caused by healthy aging) by about 30%, from 30 to 20 ml a year for the higher dose studies. There is a proportionally bigger effect for those who have stopped smoking, as the absolute change in FEV₁ decline seems to be reduced similarly in those who continue to smoke and in those who have stopped completely.⁹ On current evidence doses of beclomethasone dipropionate or equivalents of ⩾800 μg/day should be used; any increased benefit for higher doses remains to be proven.

The reduction in exacerbations of COPD with inhaled corticosteroids is more impressive in those with FEV₁ <50% of predicted.¹ The current meta-analysis also shows a greater effect on FEV₁ decline in this group, the mean reduction being 18.3 ml/year, close to the 20 ml/year suggested by the preliminary studies.¹¹

Meta-analyses are only as good as the studies included, and any selection bias from publication or inclusion bias. The studies in this analysis were regarded as high quality. There was no evidence of significant statistical heterogeneity in the higher dose studies, nor in the studies that enrolled subjects with an FEV₁ <50% of predicted. There was evidence of publication bias, with a lack of small negative studies identified from the funnel plot. The authors comment that this is unlikely to have influenced the results because of the number of large negative studies included.¹¹ Meta-analyses are also reliant on the quality of the numerical information that can be extracted from the studies on which they are based; where the original studies do not present all the required data—that being the mean annual decline in lung function and its standard error in this case—meta-analyses include a degree of subjective guesswork.

The problems that this can introduce are well demonstrated if one compares the results of the present (Sutherland) study¹¹ with those of a previous meta-analysis undertaken by Highland et al,¹² which sought to answer the same question and arrived at a different conclusion. Highland found a reduction of 5 ml per year in the decline in FEV₁ in the inhaled steroid group compared with placebo, which was not statistically significant (p=0.11). The two meta-analyses used data from an almost identical set of studies, five of six being common to both, and the sixth comprising overlapping data. The differences partly depend on the results of the study of early COPD, which most agree genuinely shows no benefit from inhaled budesonide in a mainly asymptomatic group.⁹ Highland presumed a 3.1 ml/year greater rate of FEV₁ decline in the budesonide group in their analyses, while Sutherland correctly interpreted the data as showing a 3.1 ml/year benefit in the budesonide group.

Sutherland and Highland also arrived at different approximations to the standard error (using information such as the p value) for some of the remaining studies. If Highland et al had used the results as extracted by Sutherland (the latter seeming to correspond more closely in most instances to those shown in the original papers), their findings would have been more equivocal (with a difference of 5.5 ml/year and a p value of 0.07). In the EUROSCOP study,⁷ which was included in both meta-analyses, two versions of the effect of inhaled steroid on lung function decline were presented, one based on the three year decline in FEV₁ in those who completed the study, used by Sutherland et al in their meta-analysis, and the other from a mixed effects model including all study subjects from nine months of treatment onwards, used by Highland. This, together with the one differing study, explain any residual discrepancy between the two meta-analyses. Their findings are therefore more consistent with respect to the size of effect of inhaled steroid in COPD than appears at first sight—their differing messages to some extent serve to demonstrate what can happen when one draws different conclusions depending on whether a p value is to the right or left of 0.05.

The differences in the meta-analyses raise the question of a separation of the short term improvement in FEV₁ from any effect on subsequent FEV₁ decline. The short term effect had been missed in the initial smaller and shorter studies.^4–6 It was the size and duration of the four main studies^7–10 which allowed the short term effect to be separately identified. It was also the cause of these studies’ reduced power to identify any long term effect, as the first 6–9 months’ data were excluded from the calculation of long term decline. It is possible that the effect on exacerbation reduction is related to a one-off improvement in FEV₁ from inhaled corticosteroids, which is maintained for as long as they are taken and accounts for relapses once they are stopped.^23,24 However, the increasing benefit of treatment on health related quality of life over time⁸ would favour a long term disease modifying effect over and above any one-off effect.

This meta-analysis is a welcome addition to the work on inhaled corticosteroids in patients with COPD. It is no longer ethical to do more long term placebo controlled studies in this condition. New studies should concentrate of the optimal dose, the optimal stage of the disease for starting regular treatment with inhaled corticosteroids, and the optimal combinations of long acting bronchodilators, inhaled corticosteroids, and other treatments. FEV₁ decline remains a valid but difficult endpoint; validation of the existing mixed effects models is required for studies with differential dropouts of patients with the most rapidly progressing disease.