Inhaled steroids and mortality in COPD: bias from unaccounted immortal time

To the Editor:

In the March 2004 issue of the European Respiratory Journal, an article was published in which Suissa 1 claimed to replicate the design of our previously published study 2 in a different cohort in Saskatchewan. Our study had suggested that inhaled corticosteroids (ICS) with or without long-acting β₂‐agonists were associated with a reduction in all-cause mortality risk in chronic obstructive pulmonary disease (COPD) patients compared to short-acting bronchodilators alone. From his analysis in a different dataset, S. Suissa makes the categorical statement in the last line of the abstract that our published conclusion “is the result of bias from unaccounted immortal time in its cohort design and analysis”. This statement, astonishingly, totally omits any consideration of differences between the results of the Saskatchewan database and the General Practice Research Database we used.

We are well aware of an earlier paper by Suissa 3 on bias due to unaccounted immortal time, which clearly is irrelevant to our paper published in the European Respiratory Journal 2. However, S. Suissa now postulates that the association we found was due to a further “subtle” type of unaccounted immortal time bias. Our study design specifically addressed the issue of immortal time bias as defined in analytical epidemiology 4.

First, patient follow-up time in the cohort design only started a day after the immortal period of 180 days from the start of therapy (see p. 820 and figs 1 and 4 in our paper 2). S. Suissa suggests that because “regular treatment” was defined as at least three prescriptions of the relevant drug in the 180 days after the first prescription, cohort entry should be defined as the date of the third prescription and this has a significant impact in his analyses. The distinction may matter in the Saskatchewan database but in our study it was irrelevant as groups receiving ICS actually had shorter duration between first and third prescription than the control group (short-acting bronchodilators: 87.1 days; fluticasone: only 77.3 days; and fluticasone and salmeterol: 74.3 days). Thus, the theoretical distinction between the first and third prescription was without any relevance and seems difficult to justify. In our study, we also reported the number of prescriptions of the relevant drugs over the first 12 months after cohort entry, providing strong evidence that the initial pattern of prescribing in our groups was well maintained.

Secondly, we are unable to follow his reasoning on the “hierarchial” approach to treatment, which is implicit in the stepped care approach recommended in all major guidelines on COPD (and asthma) throughout the 1990s. Indeed, we are unaware of the circumstances that would lead to regular prescription of ICS in COPD without regular use of bronchodilators (see Suissa 1, p. 393, table 1, column 3). Comparison of table 1 in our study 2 and table 1 in Suissa 1 shows that drug use was more irregular in the Saskatchewan database with low use of the recommended bronchodilators. In spite of this, S. Suissa's own results using “very first regular exposure identified after diagnosis” in his “conventional intention-to-treat approach” still indicated a significant association between ICS use and mortality, rate ratio 0.75 (0.62–0.90). Furthermore, S. Suissa ignores that our study used two designs, a cohort approach for the main analysis and a nested case-control approach to explore a dose-response relationship, with both methods indicating an association with ICS. The latter design has been described previously by Suissa 5 as one that simplifies the cohort analysis when exposures vary over time and leads to valid estimates with negligible loss in precision.

Finally, Suissa 1 used a time-dependent exposure approach and obtained results, which suggested that inhaled corticosteroids were not better than bronchodilators at reducing the risk of death in chronic obstructive pulmonary disease patients. We are not surprised that the benefit of inhaled corticosteroids could not be established with the treatment switching approach. This methodology is known to be valid only if the reason for the switch to inhaled corticosteroids is unrelated to the patient's subsequent risk of death 6. In our setting, the switch to inhaled corticosteroids was unlikely to be independent of mortality risk. Clinical experience suggests inhaled corticosteroids would be prescribed to sicker patients who were no longer responsive to bronchodilator therapy alone.