© 2003 American Thoracic Society
Inhaled glucocorticoids in COPDImmortal time biasTo the Editor:Dr. Suissa's current work (1) argues that immortal time bias was responsible for the results our group had reported earlier (2, 3). To make his case, he makes several assumptions about our analyses, which, we believe, are flawed. First, in our study (2), the 90-day window (for exposure) was selected purposefully and not whimsically, as Dr. Suissa implies. While it is true that most patients receive their prescription for medications at or near the time of hospital discharge, the actual filling of these drugs may be delayed for several days or weeks. This lag time can arise because patients may already possess these drugs (from filled prescriptions that occurred before hospitalization). Because each canister generally lasts 60 to 100 days, we chose a 90-day window period to avoid misclassification of these patients as nonusers. Implicit in this exposure definition is the assumption that patients who received inhaled corticosteroids within 90 days of discharge were also more likely to be users between time zero and 90 days than those who did not fill a prescription in the first 90 days. We tested this assumption by reanalyzing the original Ontario data. We found that patients classified as "users" of inhaled steroids were far more likely to have received inhaled corticosteroids before hospitalization than "nonusers." Indeed, over 70% of users had received at least one dispensation of inhaled corticosteroids within 1 year before their index hospitalization, whereas only a third of nonusers had such exposure (odds ratio, 4.7; 95% confidence interval [CI] 4.4 to 4.9). In contrast to the time-fixed model, a time-dependent analysis would classify all person-time occurring before the dispensation date as "nonexposed" time. Because it is likely that many patients were still using inhaled steroids (from their filled prescription before the index hospitalization) during this "nonexposed" time, a time-dependent analysis would bias the results towards the null value. Second, immortal time bias, if it were present, should apply to all anti–chronic obstructive pulmonary disease (COPD) medications (and not just to inhaled corticosteroids). A time-fixed analysis showed that neither short acting ß2-agonists nor ipratropium was significantly associated with mortality/hospitalization. A time-dependent analysis, on the other hand, demonstrated increased mortality/hospitalization with these medications. There is, however, no credible evidence that bronchodilators increase all-cause mortality in COPD (4, 5). An alternate explanation for Dr. Suissa's finding is protopathic bias, which can arise in situations where medications are used to treat acute symptoms of COPD (6). In clinical practice, bronchodilators and corticosteroids are often used during exacerbations. Accordingly, dispensation of these drugs may be a marker of clinical instability, which can falsely elevate mortality rates associated with these drugs. Protopathic bias can make an effective treatment appear nonefficacious or even harmful (6). This may explain why in Dr. Suissa's time-dependent analysis, receipt of inhaled corticosteroids had no effect on hospitalization/mortality rate, whereas receipt of inhaled ß-agonists increased the rate by 67%, a finding that is inconsistent with other previously published reports (4) and contravenes clinical logic and experience. A time-dependent analysis should, therefore, not be used to evaluate the effects of chronic pharmacologic therapies when they may also be used to treat acute exacerbations. Third, according to Dr. Suissa's argument, immortal time bias would be abolished if patients who had an event within the first 90 days of follow-up were excluded from the analysis. Using the Ontario data, we performed an analysis excluding all subjects who had a hospitalization or death within the 90-day window period. In this analysis, we found that inhaled steroids were still associated with a significant improvement in mortality (relative risk, 0.79; 95% CI, 0.71 to 0.88) (Figure 1) . Similar results were observed when hospitalization was combined to all-cause mortality.
Fourth, Dr. Suissa's arguments, although appealing conceptually, are not supported by empiric evidence. Soumerai and coworkers (7) performed an observational study using a study design that was almost identical to the one we used and found that beta-blocker dispensation (within 90 days after hospital discharge for myocardial infarction) were associated with an approximately 40% reduction in mortality, a finding that was very similar to those observed in randomized controlled trials (8). Soumerai's study also demonstrated lack of effectiveness of calcium-channel blockers, which have also been "validated" by clinical trial results (9). Findings from other observational studies wherein our study design was used (10, 11) have largely matched those reported by clinical trials (12). Notably, in Dr. Suissa's analysis, he found, as we did, a significant reduction in hospitalization/mortality rate among users of inhaled corticosteroids than nonusers in a cohort that was much smaller and younger than the one we used. Even in the so-called "corrected" person–time analysis (adjusting for immortal time bias), users still had 21% lower event rate than nonusers. Dr. Suissa's findings from the Saskatchewan database, thus add to the growing body of evidence that inhaled corticosteroids are likely to have a beneficial effect on health outcomes of moderate-to-severe COPD patients (2, 3, 13, 14).
a University of Alberta Edmonton, Alberta, Canada REFERENCES
To the Editor: Dr Suissa's article (1) raises an important point about the handling of the period after discharge from hospital and the first prescription(s) dispensed as an outpatient in the paper of Drs. Sin and Tu (2), and in our study (3). Both deal with the effects of drug treatment on survival and rehospitalization of patients with chronic obstructive pulmonary disease (COPD) in the year after a hospital admission for COPD; our study was based on 4,263 such patients identified in the United Kingdom General Practice Research Database (GPRD). Although we deliberately simulated the 30-day immortal period and the 90-day period for establishing pharmacotherapy used in the Sin and Tu study (2), our comparison was not of treatment with inhaled corticosteroids versus no inhaled corticosteroids; instead, it was a comparison of four different drug-exposure groups. Compared with users of bronchodilators only, we found that the risk of COPD rehospitalization or death was reduced by 10% in users of long-acting ß-agonists only, by 16% in users of inhaled corticosteroids only, and by 41% in users of the combination of inhaled corticosteroids and long-acting ß-agonists. Reanalyzing our data with an immortal period of 90 days, the same outcomes were reduced by 5% in users of long-acting ß-agonists only, by 11% in users of inhaled corticosteroids only, and by 42% in users of the combination of inhaled corticosteroids and long-acting ß-agonists. Consequently, our conclusions are unchanged. The suggestion of Dr. Suissa that the entire period between discharge and the first outpatient prescription is a period of no treatment imposes a theoretical maximum. In the United Kingdom, an initial supply of treatment is commonly dispensed from the hospital, together with a recommendation that the "preventive" component should be continued; this practice seems to have been frequently adopted, because the mean number of prescriptions ordered by general practitioners for inhaled corticosteroids for each patient in the first 90 days after discharge was at least 2.18 (Table 2 of Reference 3), and the mean time from discharge to first prescription was less than 30 days in all groups. Unfortunately, hospital-issued prescriptions cannot be retrieved from the GPRD database so we cannot quantify how frequently drugs were or were not dispensed on discharge. We conclude that any bias in our analysis due to early misclassification of treatment would be considerably less than in the example analyzed by Dr. Suissa. Different approaches to analysis will continue, but we believe pharmacoepidemiologic studies are a good use of epidemiology to prepare for outcomes of ongoing COPD clinical trials by 2006.
a National Heart and Lung Institute Imperial College London, United Kingdom FOOTNOTES Conflict of Interest Statement: J.B.S. is currently an employee of GlaxoSmithKline R&D, manufacturer of respiratory drugs; J.V. has participated as a speaker in scientific meetings and courses organized and financed by various pharmaceutical companies (GlaxoSmithKline, AstraZeneca, Pfizer, and Boehringer-Ingelheim), in the steering committee of an ongoing trial sponsored by GlaxoSmithKline and has a spouse employed by GlaxoSmithKline; N.B.P. has been a consultant on COPD for GlaxoSmithKline for the last 3 years, and in that capacity has contributed to design, review of results, and writing up of a number of studies of treatment of COPD; and V.A.K. is currently an employee of GlaxoSmithKline R&D, manufacturer of respiratory drugs. REFERENCES
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
