© 2004 American Thoracic Society
Inhaled Corticosteroids in COPD and MortalityInaccuracies?To the Editor:The observational study by Fan and colleagues (1) that assessed the effects of inhaled corticosteroids in chronic obstructive pulmonary disease (COPD) on all-cause mortality and other outcomes includes two inaccuracies that can affect the conclusions (1). First, the authors claim to have replicated the approach used in a previous observational study (2), an approach shown to be subject to immortal time bias (3). We contend that the method was not actually replicated. The authors defined cohort entry as the time of a first inpatient or outpatient visit for COPD and defined exposure as use of inhaled corticosteroids during the 90-day period before this cohort entry visit (see Figure 2 in Reference 1). This approach to exposure definition does not lead to immortal time. Instead, the approach that engenders immortal time and its bias defines exposure as use of inhaled corticosteroids during the 90-day period after cohort entry. Consequently, the time between cohort entry and the reception of the first subsequent prescription of inhaled corticosteroids is necessarily immortal because, by definition, no deaths can occur during this time span. Such immortal time induces a bias by both overestimating the death rate in the unexposed group and underestimating the death rate in the exposed group (3). Second, the authors use a time-dependent analysis to account for variations in inhaled corticosteroid use during the 1-year follow-up period. Although this is the proper approach to deal with such a design (3, 4), the authors use a crude version that may be insufficient. By using the trimester as the time unit to define exposure, there are only four points of time at which exposure values can change. The analysis is then based on exposure during the previous trimester, so that, for example, cases that occur on Days 95 and 175 of follow-up will both be classified according to their exposure on Days 1 through 90. This will result in exposure misclassification, which biases the rate ratio toward the null. A more precise time-dependent analysis should use the day as the unit of time instead of the trimester. This finer time unit was used in a study (5) that found the hazard ratio of readmission for COPD with current use of inhaled corticosteroids to be 1.02 (95% confidence interval: 0.85–1.22). Although the authors found a similar rate ratio, using finer time units will alleviate the uncertainty that the rate ratio of inhaled corticosteroid use on major COPD outcomes they report is biased and nonsignificant because of such exposure misclassification.
McGill University Montreal, Quebec, Canada FOOTNOTES Conflict of Interest Statement: S.S. has been reimbursed for attending several conferences and also has participated as a speaker in scientific meetings financed by various pharmaceutical companies (Schering-Plough, AstraZeneca, and GlaxoSmithKline) and has received funding for research grants from AstraZeneca ($89,000), Schering-Plough ($80,000), and GlaxoSmithKline ($159,000). REFERENCES
From the Authors: We thank Dr. Suissa for his comments emphasizing the importance of an accurate assessment of medication exposure in pharmacoepidemiologic studies. In observational studies, immortal time bias may result in misclassification of exposure, thereby biasing the results (1). The goal of our study was to determine whether length of exposure to, and dose of, inhaled corticosteroids was associated with mortality and exacerbations (2). Our aim was not to recreate immortal time bias, and, as Dr. Suissa notes, our conclusions did not suffer from this bias. A time-dependent analysis such as the one we performed is an appropriate method to account for variations in medication use over time (3, 4). Dr. Suissa raises the concern that medication use is reassessed only four times in our analysis. In fact, over a median follow-up of 622 days, exposure was recalculated a median of eight times per patient. Calculating exposure more frequently might further improve the precision of medication exposure; however, any effect of this nondifferential misclassification would likely be small. Our method is a significant improvement over baseline analyses, and our results are similar to others using time-dependent methods (3, 5). More importantly, we believe that our method of medication assessment actually improves classification of inhaled corticosteroid use by ensuring that patients have a sufficient exposure to result in a clinical effect. Inhaled steroids most likely must be taken regularly, in adequate doses, and for a period of weeks to months to have a biologic effect on underlying inflammation (4, 6). By requiring that patients have enough prescriptions to be adherent to inhaled corticosteroids over a 90-day period, our method lends biologic plausibility to any hypothesized effect on exacerbations or mortality. In contrast, Dr. Suissa and colleagues defined exposure as a single prescription filled at any time in the 30 to 60 days before readmission to the hospital (5). Using this method, a patient prescribed a medication the day before hospitalization is classified as a user, even though they may have used the medication for only 1 day. This approach, although providing more frequent assessments of medication use, could actually result in misclassification bias because the medication would not have sufficient time to have an effect. Determining superiority of one method over another seems difficult without direct comparison. Identifying methods to minimize biases is important to advancing the field of pharmacoepidemiology, and we appreciate Dr. Suissa's comments and the work he has done to advance this field.
a University of Washington Medical Center Seattle, Washington FOOTNOTES Conflict of Interest Statement: V.S.F., J.R.C., C.L.B., and S.D.F. have no declared conflict of interest; D.H.A. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript; D.H.A. has shares of Pfizer stock that was purchased in November 1999. REFERENCES
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
