• chronic obstructive pulmonary disease
• corticosteroids
• bronchodilators

The main results of the ISOLDE trial have been published¹ and are now part of the canon of knowledge about COPD—inhaled corticosteroids (ICS) do not change the rate of decline of FEV₁ in COPD. This finding is consistent with that of other similar large long term trials,^2–4 so the question has been settled. As in some of the other studies, there may be decreases in the frequency and severity of acute exacerbations, and quality of life may be modestly improved. These questions are being addressed in separate ongoing long term trials.

Despite the rule that the primary and secondary outcomes of a trial must be prestated and set in stone in the protocol, trials such as ISOLDE are so large and so carefully planned, executed, and monitored that the huge amounts of data they generate offer many opportunities to examine questions other than the prestated outcomes—questions that are scientifically important but which are unlikely ever to find funding as primary outcomes. In the absence of any methodological aspect of the trial that would invalidate it from being used to answer a different question, it is appropriate to mine the data for any and all other useful information they may provide. Sometimes disparaged as “data dredging”, the meticulous review of the data for other insights seems rather to be an economical use of a precious resource—namely, good data which should be exploited for all they are worth. Two important papers in this issue of Thorax illustrate this.

The paper by Burge et al⁵ bears on one of the most important remaining issues concerning the long term use of ICS in COPD—is it possible to identify “steroid responders” by a short preliminary course of oral prednisolone? The issue is important because a very large proportion of patients with COPD are already receiving long term ICS, often without a clear rationale or a solid experimental basis; this is a matter of some concern.

Inhaled corticosteroids are probably not innocuous, particularly in the relatively large dosages that are sometimes being used in COPD,⁶ and particularly when one considers that ICS may be used continuously for 10 or even 20 years in the COPD population who are older and in whom the common side effects of steroid therapy such as osteoporosis and muscular atrophy are already common. Potential adverse effects of ICS such as these might take many years of continuous use to become evident, and none of the published studies of ICS has exceeded 40 months, which is not nearly long enough to be confident of the safety of their long term use. By the widespread and indiscriminate use of ICS in COPD, might we be making a Faustian compact in which modest short term benefits in some patients are being traded for what may be an epidemic of serious adverse effects over the long term? This possibility, admittedly portrayed in its most disastrous form, makes it most important to determine whether there is an identifiable subgroup of patients with COPD who are not responsive to corticosteroid therapy. If so, these patients should perhaps be saved from even the potential risk of long term ICS use. Previous attempts to identify steroid responder and non-responder groups have been largely unsuccessful, as cited by Burge et al.⁵

The unique design of the ISOLDE study, which included a 2 week trial of oral prednisolone before randomisation into the main 3 year trial, made it possible to address this question with good statistical power. In the present report the authors found no statistical association between the change in FEV₁ resulting from the course of oral prednisolone and any outcome over the subsequent 3 years of treatment with ICS or placebo. This finding invalidates the traditional recommendation that one should perform a steroid trial before instituting ICS on a long term basis. It could be argued that the increase in FEV₁ during the short prednisolone trial might not be an appropriate way to identify responder and non-responder groups. Indeed, the FEV₁ does not seem to be a measurement that correlates well with clinically relevant outcomes.⁷ However, it is hard to know what other outcome(s) of a short trial of prednisolone would be better than the FEV₁, or even practical in the clinic. If the purpose of a preliminary prednisolone trial is to guide the clinician in his/her decision whether or not to administer ICS to the patient with COPD, the indicator of its success or failure must be one that clinicians can readily apply in their daily practice. This rules out the measurement of markers of COPD-type inflammation in airway secretions, cells, or breath condensates that might conceivably predict steroid non-responsiveness.⁸ One day, perhaps, but such assays are neither validated nor realistic in clinical practice at present.

The present trial⁵ and all previous evidence seems to suggest, as Burge and colleagues state, that “patients with COPD cannot be separated into . . . corticosteroid responders and non-responders”. Whether this is because there really are no subgroups of steroid responders and non-responders, or whether there are but we have not yet discovered how to identify and differentiate them, is uncertain. I believe the right course is to keep an open mind about the possibility that there may be responder and non-responder subgroups and to continue to seek ways to identify and characterise them, if they exist. Meanwhile, I regretfully agree with the conclusion that corticosteroid trials are not diagnostically helpful in primary care.

The paper by Calverley and colleagues in this issue of Thorax,⁹ also based on data from the ISOLDE trial, addresses the important question of whether poor bronchodilator responsiveness is a valid criterion for the diagnosis of COPD or predicts disease progression. Although the ISOLDE population was selective due to the inclusionary criterion of a poor bronchodilator response, Calverley et al show that “classifying patients as [bronchodilator] ‘responders’ and ‘non-responders’ can be misleading and does not predict disease outcome”.

The problem surrounding the notion of non-responsiveness to bronchodilators in COPD is the legacy of a document generated by a group of academic lung specialists who met in 1958 to establish diagnostic criteria for chronic obstructive lung diseases. They defined what we now call COPD as “a group of diseases with persistent or irreversible obstructive lung disease”.¹⁰ However, it has been known for as long as spirometry has been routinely performed that this definition is flawed^11,12 and that patients with clinical features that every practising respiratory physician would call COPD are often capable of a “significant” bronchodilator response. The only thing one can confidently say about the bronchodilator responsiveness of patients with COPD is that it is not, on average, as great as that in patients with asthma. But the overlap between the two diagnoses in this respect is so great that they cannot be reliably distinguished on this basis.^13,14 Bronchodilator responses of patients with COPD can, in fact, be quite large with some current agents,¹⁵ nor might they be elicited by a bronchodilator challenge on a single occasion as shown by Calverley et al and previously by others.^16–18 If one still doubts that a bronchodilator response is typical in patients who meet the clinical criteria of COPD, one should look at fig 2 in the paper by Calverley and colleagues⁹ where what looks as though it might have been a perfect Gaussian distribution of bronchodilator responsiveness is sharply truncated on the right hand side by the prior exclusion of potential subjects who had a response to salbutamol of more than 10% of predicted FEV₁. Despite this, the mean FEV₁ response following salbutamol was 4–6% predicted, or about 170 ml, which is both statistically and clinically significant and was even greater when salbutamol and ipratropium were used together.

The implication which Calverley and colleagues make here, and which is made in another recent paper in almost identical words,¹⁹ is that the response of an individual COPD patient to a bronchodilator challenge on a single occasion does not predict whether the patient will or will not benefit from that agent subsequently. The size of the response in patients with COPD is not only inconsistent over time but, whatever limit one sets on it, is arbitrary—whether 12% of baseline²⁰ or 10% of predicted.¹ Perhaps equally important, they argue that a bronchodilator response should not be used as an exclusionary criterion in clinical trials involving patients who otherwise conform to the diagnostic criteria of COPD. I strongly agree. Doing so not only perpetuates a 45 year old misconception, but it irrationally excludes a subset of patients with COPD who happened to be tested on a day when they responded rather better than they may have responded on another day.

There is a divergence of views between regulatory agencies in Europe and North America on this matter which hurts both of us. In North America a limited response to bronchodilators is no longer required as an inclusionary criterion of COPD trials. Of course, as a North American, I say let us compromise—do it our way! But it would be best for practitioners on both sides of the Atlantic if we could be consistent in how we diagnose and admit patients to COPD trials. The paper by Calverley and colleagues suggests rather strongly that COPD trials should include all patients who meet the clinical criteria of COPD that we can all agree upon and should not exclude patients on the mistaken notion of bronchodilator non-responsiveness that goes back to an era when spirometric tests were very rarely performed.

Two papers, two important messages. Does the fat lady have another encore for us?