From the authors:
We thank E. Marchand for his interest and constructive comments relating to our paper 1 and we will try to answer his queries. Our findings for inhaled corticosteroids are likely to be a class effect. Our initial approach to the General Practice Research Database (GPRD) was done in order to mimic a planned controlled trial and therefore fluticasone propionate was chosen. Our sensitivity analysis did not reveal different findings when all inhaled corticosteroids were included in the analyses as illustrated in figure 4 of the paper. We understand E. Marchand's concern about comparability between groups but disagree with his conclusions. Table 1 showed baseline characteristics, i.e. before the groups were defined according to their prescriptions, within a 6‐month period. This explains why less than one-half of the reference group seemed to be without inhaled bronchodilators; in the following 6‐month period, all of them received at least three prescriptions for short-acting β‐agonists, xanthines, anticholinergics or combined bronchodilators. Similarly, in the group receiving salmeterol in the 6‐month period defining drug exposure, only 23.9% received long-acting β‐agonists at baseline; for inhaled corticosteroids the figure was 67.3% at baseline. For this reason, all four groups received “active” treatment during the first 6 months and a well-sustained pattern of prescriptions throughout 3 yrs. However, this terminology should be used in clinical trials only, not in observational studies in pharmacoepidemiology 2. E. Marchand is right in pointing out the difference in age but we have tried to adjust for these potential confounders using both multivariate survival analyses and stratification on age. It is true that use of nebulised therapy was less frequent in the reference group but oxygen therapy did not, in fact, differ.
Thus, in response to E. Marchand's specific questions, we would make the following comments. 1) Based on the above, we doubt that recording and misclassification in the GPRD explain our findings. The exact circumstances at which a label of chronic obstructive pulmonary disease (COPD) is entered into the register cannot be seen from the register data but it is fair to hypothesise that the first labelling of COPD in a primary care database could serve as a proxy for incidence of clinical COPD. Since all patients were entered because of prescriptions for COPD drugs, as recommended by the current British Thoracic Society (BTS) and Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines 3, 4, we believe we have overcome the potential bias suggested by E. Marchand. 2) The specificity of our definitions of moderate and severe COPD were 75.0% and 81.4%, respectively 5. 3) We consider overall mortality a much more exact end-point than cause-specific mortality, especially since studies on causes and circumstances of death in COPD are generally difficult to perform 6, 7. We visually reviewed all information on the 1,334 deaths observed in this cohort. Without any validation study to lean on we can say that causes of death were respiratory in 33.8%, cardiovascular in 24.4%, cancer in 16.1%, other in 8.5%, and unknown in 17.2%. There was no difference in the pattern of causes of death in the four groups, but it was beyond our protocol to expect that any respiratory medication would reduce specific causes of death.
- © ERS Journals Ltd