| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Division of Clinical Epidemiology, Royal Victoria Hospital, McGill University Health Centre, and the Depts of Epidemiology and Biostatistics and of Medicine, McGill University, Montreal, Canada
CORRESPONDENCE: S. Suissa, Division of Clinical Epidemiology, Royal Victoria Hospital, 687 Pine Avenue West, Ross 4.29, Montreal, Canada, H3A 1A1. Fax: 1 5148431493. E-mail: samy.suissa@clinepi.mcgill.ca
Keywords: biases, cohort studies, drug effectiveness, epidemiology, immortal time, pharmacoepidemiology
Received: June 3, 2003
Accepted October 5, 2003
This research was funded by grants from the Canadian Institutes of Health Research (CIHR) and Fonds de la recherche en Santé du Québec (FRSQ). The acquisition of the database was funded by grants from AstraZeneca, Boehringer Ingelheim and GlaxoSmithKline. The author is the recipient of a Distinguished Investigator award from the CIHR. This study is based on de-identified data provided by the Saskatchewan Dept of Health. The interpretation and conclusions contained herein do not necessarily represent those of the government of Saskatchewan or the Saskatchewan Dept of Health.
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionAcknowledgementsReferences |
|---|
This bias was assessed using a population-based cohort of 3,524 newly treated COPD patients from Saskatchewan, Canada, observed from 1990–1999. Regular users of bronchodilators or ICS were followed for 3 yrs, during which time 860 deaths occurred. Cox's proportional hazards model was used to compare the hierarchical intention-to-treat approach employed in the recent study, a technique subject to bias from two sources of immortal time, with the conventional intention-to-treat approach and the according-to-treatment approach.
The adjusted rate ratio of death using the hierarchical intention-to-treat approach was 0.66 (95% confidence interval (CI) 0.57–0.76) for ICS use relative to bronchodilator use, compared with 0.75 (95% CI 0.62–0.90) with the conventional intention-to-treat approach. Conversely, the rate ratio was 0.94 (95% CI 0.81–1.09) with the according-to-treatment approach, which accounts for both sources of immortal time.
In this study, regular inhaled corticosteroid use in chronic obstructive pulmonary disease was not found to reduce all-cause mortality. Suggestion of this benefit from a previous observational study is the result of bias from unaccounted immortal time in its cohort design and analysis.
The use of inhaled corticosteroids (ICS) in the treatment of chronic obstructive pulmonary disease (COPD) is highly debated, with important divergence in the data from randomised trials and observational studies 1. Recent randomised trials have found that ICS treatment confers little or no benefit on lung function 2–7. Although not part of the primary objective, two of these studies found a reduction in exacerbations or healthcare resource use among patients with moderate 7 or severe COPD 6. Conversely, four recent observational studies have suggested that these drugs are extremely effective at preventing COPD morbidity and mortality.
Three of the observational studies, using an identical design, evaluated ICS dispensed after hospital discharge for COPD and found them to be very effective at preventing COPD morbidity and mortality 8–10. The cohort design employed by these studies defined time zero (date of hospital discharge) independently of ICS exposure and used an intention-to-treat analysis. However, the current author has shown that this approach can lead to an important bias resulting from the introduction of immortal time and its misclassification with respect to exposure 11. The study was replicated, using a different cohort of COPD patients, and a rate ratio (RR) of 0.69 was found for ICS, subject to bias. The RR became 1.00 when these sources of bias were properly attended to 11.
The fourth observational study used a different cohort design that appears to have accounted for immortal time 12. This cohort study found that regular ICS use, given to newly diagnosed COPD patients, not necessarily after COPD hospitalisation, reduced the rate of all-cause mortality by 38–52% 12. Unlike the other observational studies, however, time zero for cohort entry was defined by ICS exposure and the exposure was defined hierarchically. As a result of these definitions, it can be hypothesised that this study, using the intention-to-treat approach, also generated immortal time that was unaccounted for and that may have biased the findings.
In the present paper, it is shown that the intention-to-treat method used by this recently published study, by its approach to cohort formation and hierarchical exposure definition, introduces a bias involving immortal time that leads to significant distortion of the results. The nature of the bias is described and illustrated, and its impact is assessed using data from another population-based cohort of patients with COPD.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionAcknowledgementsReferences |
|---|
55 yrs of age. Subjects with any prescriptions for bronchodilators or other asthma drugs, including ICS, nasal corticosteroids, cromones (cromoglycate and nedocromil) and ketotifen, during the 5-yr period prior to cohort entry, were excluded. The cohort members were followed until December 31, 1999.
Study cohort
The study cohort was defined by the start of regular treatment with ICS or bronchodilators. Unlike the previously published study, which specifically evaluated the effect of fluticasone proportionate with or without salmeterol relative to bronchodilators, this study will focus simply on the effect of all ICS in contrast to bronchodilators. This is necessary since the use of long-acting ß-agonists was very infrequent among cohort members. Nevertheless, the bias principle that is being illustrated will in fact gain in clarity with a simple comparison between two groups, namely ICS and bronchodilators, and not four groups.
As defined in the previous published study, regular treatment was taken to be the first dispensing of three or more prescriptions of ICS (in this study beclomethasone, fluticasone or budesonide) or three or more prescriptions of a bronchodilator from any one of the following classes: inhaled SABA, inhaled anticholinergics, xanthines or combined bronchodilators within a 6-month period. Long-acting ß-agonist use was ignored as it was too infrequent among cohort members. All subjects were then assumed to remain regular users of either bronchodilators or ICS until the end of follow-up, namely until the date of death, the date of emigration from the province, the date of end of coverage of the health insurance plan or December 31, 1999, whichever occurred first.
Outcome
The outcome was the occurrence of death from any cause during the follow-up. In conformity with the previously published study design, all deaths that occurred within 6 months of cohort entry were excluded.
Exposure
The previously published study used a hierarchical intention-to-treat definition for regular ICS exposure. Using this approach, the ICS group was first identified by the regular use of ICS, irrespective of whether a subject was already a regular user of bronchodilators, the reference exposure. The reference group for this hierarchical approach was thus composed of regular users of bronchodilators who were not regular users of ICS at any time during follow-up. This hierarchical approach may be biased by its exclusion of the immortal time that occurs during exposure to regular use of bronchodilators preceding the start of regular ICS exposure (fig. 1
). Indeed, subjects could not have died during this period since they had to survive to become regular users of ICS. In addition, subjects were likely to be regular users of bronchodilators during this immortal time period.
|
Another source of immortal time arises from the definition of time zero. The recently published study used the date of the first of the three exposure-defining prescriptions within a 6-month period. The time span between the first and the third prescription is also immortal, as the subject must survive from the first to the third prescription to satisfy the definition of regular use. To avoid this bias, time zero should be defined as the date of the third of the three prescriptions. This alternative definition was used in the following analyses.
Data analysis
First, person-time of follow-up was classified as exposed to either ICS or bronchodilators (reference group) for each of the three different exposure definitions. Rates of death were computed for each of the three approaches and crude RRs were estimated for regular ICS use relative to regular bronchodilator use. Cox's proportional hazards model with time-fixed and time-dependent exposure was used to crudely estimate these RRs, and they were adjusted for age at cohort entry and sex, as well as the occurrence of COPD hospitalisations and the use of other COPD medications dispensed during the 6-month period prior to cohort entry.
With respect to the definition of time zero, the time span between the first and third prescriptions defining time zero was first compared separately for bronchodilators and ICS, to verify if it differed for the two exposures. This alternative definition of time zero, namely the date of the third prescription instead of the first, was then used to estimate the crude and adjusted RRs for the three different approaches.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionAcknowledgementsReferences |
|---|
describes these cohorts, after excluding all subjects who were followed-up for <6 months after cohort entry or who died within this period. The subjects who used ICS at any time were younger, more likely to have used SABA and to have been hospitalised for COPD prior to cohort entry.
|
displays the crude rates of death estimated by the three different approaches. It should be noted that the crude RR of death for ICS exposure relative to bronchodilators goes from 0.53 for the hierarchical intention-to-treat approach, to 0.63 for the conventional intention-to-treat approach and to 0.69 for the according-to-treatment approach. The conventional intention-to-treat approach is based on a slightly larger number of subjects, since fewer subjects with <6 months of follow-up after cohort entry were excluded among those started on bronchodilators and followed by ICS. There were also fewer deaths because of the 3-yr follow-up limitation. Table 2 shows the impact of the immortal period of regular bronchodilator exposure occurring prior to the start of regular ICS use. This immortal time (followed by 0 deaths) in 646 subjects accounts for 1,168.5 person-years. The classification of this immortal time into the reference group decreases the rate of death in the reference group from 13.95 to 10.62 deaths·100–1·yr–1.
|
shows that the mean time span between the first and third prescriptions is 85.9 days for bronchodilators and 99.5 days for ICS. When restricted to the subjects who belonged to both groups, the calculation gives similar results. These different time spans correspond to a difference in immortal time of 
83.5 person-yrs between the ICS and bronchodilator groups.
|
shows the results of Cox's regression analysis for the three different approaches and both definitions of time zero. The crude RRs, obtained using the first prescription as defining time zero, are similar to the crude values estimated from the person-time approach in table 2
. Generally, adjusted analyses increase the RR towards unity, as do the analyses based on the third prescription taken to define time zero. The adjusted RR for the according-to-treatment approach based on this latter definition is 0.94 (95% confidence interval (CI) 0.81–1.09).
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionAcknowledgementsReferences |
|---|
The bias results from overlooking two sources of immortal time. The bias is first introduced in the hierarchical definition of exposure. From a cohort of all newly diagnosed COPD patients, the hierarchy of exposure first assessed subjects from the date of diagnosis to the end of follow-up to identify the exposed, namely those who received three or more prescriptions of fluticasone proportionate or salmeterol during a 6-month period. The unexposed reference group is then defined by the subjects who did not receive any fluticasone proportionate or salmeterol during follow-up, but received three or more prescriptions of a short-acting bronchodilator during a 6-month period. Consequently, a large number of exposed subjects could have used short-acting bronchodilators prior to their start of fluticasone proportionate or salmeterol, as confirmed in table 1 12. This is plausible because of the stepped-care approach recommended in treatment guidelines for this disease, i.e. treatment should start with bronchodilators and corticosteroids should be added when the disease worsens 16. Thus, several subjects from the exposed group were in fact unexposed before switching to this exposure status. More importantly, however, this pre-exposure time during which subjects were unexposed is an immortal period. These subjects, in switching from unexposed status to exposed status, will necessarily do so alive. Had they died before switching, they would, by definition, have belonged to the unexposed group. Thus, the bias occurs because valid unexposed person-time of follow-up with no deaths is not accounted for in the reference rate of death. This results in an artificial increase in the rate of death of the reference group, leading to a spurious appearance of effectiveness. The magnitude of the bias will then depend on the actual pattern and duration of stepping-up therapy from short-acting bronchodilators to the addition of ICS and/or long-acting ß-agonists. In the present study, analysis of the COPD cohort indicates that accounting for this first source of immortal time changes the rate ratio from 0.66 to 0.85.
The previously published study focussed on fluticasone and salmeterol, suggesting that the combination is more effective than either drug, which are, in turn, more effective than short-acting bronchodilators alone. In this context, the hierarchical approach will accentuate the bias, as subjects receiving the combination could have first been exposed to short-acting bronchodilators, the reference, then to the long-acting bronchodilator or the ICS, before eventually being dispensed the latter two together during the same 6-month period. All immortal time spent in the reference exposure, as well as in the individual long-acting bronchodilator and ICS exposures, are not accounted for in the analysis. The present analysis suggests that the bias from unaccounted immortal time bias resulting from the hierarchical approach will lead to a hierarchy of effects.
The second source of immortal time arises from selecting time zero as the time of the first of the three prescriptions that define exposure. The time from the first to the third prescription is also immortal, as the subjects must have three prescriptions to satisfy the definition of regular use and are not included in the study if they die after the first or second. To avoid this bias, time zero should be defined as the date of the third of the three prescriptions, as has been done in previous similar studies of asthma and COPD 14, 17. The analysis of the present COPD cohort indicates that accounting for this second source of immortal time changes the biased RR from 0.66 to 0.71 and the RR corrected for the first source of immortal time from 0.85 to 0.94. In essence, the two sources of immortal time turn out to have an important impact on the RR of death for ICS use, changing it from a highly significant 0.66 (95% CI 0.57–0.76) to a nonsignificant 0.94 (95% CI 0.81–1.09).
The observational study presented here is also subject to the usual biases. ICS are expected to be prescribed to patients with more severe disease, who are at higher risk of death. Thus, the finding of a RR of 0.94 for ICS, even after adjustment for several confounding factors, may in fact suggest a modest benefit that must be investigated in large randomised trials. The diagnosis of COPD, from administrative databases, was only based on age and bronchodilator use. In the absence of information on smoking, symptoms and spirometry measures, subjects in the cohorts may have been misclassified as COPD. The unintended, yet possible, inclusion of asthma patients as COPD by these administrative criteria could lead to an apparent but fallacious beneficial effect of ICS in COPD. In addition, exposure was crudely defined to be consistent with the previously published cohort study 12. Once a subject was allocated to an exposure group, they remained there and were considered exposed to that drug category until the end of follow-up, irrespective of whether they stopped or continued for the entire 3-yr follow-up. This assumption will produce exposure misclassification that tends to attenuate the RR towards unity.
As chronic obstructive pulmonary disease is one of the major causes of death, and as both mortality and morbidity due to chronic obstructive pulmonary disease have increased over the last two decades, the clinical implications of this observational study are important 18, 19. While observational studies are a valuable addition to randomised controlled trials to demonstrate treatment effectiveness, they must be properly conducted to minimise bias 20. Care must be taken in evaluating cohort studies with a hierarchical approach to exposure definition, as they can be subject to biases from unaccounted immortal time. This bias can give the false appearance of an important drug benefit, as can be seen with the apparent effectiveness of combination therapy with fluticasone and salmeterol to reduce mortality in chronic obstructive pulmonary disease. Randomised controlled trials are essential to verify whether these drugs, alone or in combination, are beneficial in chronic obstructive pulmonary disease. The important bias identified in this study, implies that the effectiveness of inhaled corticosteroids in chronic obstructive pulmonary disease remains debatable 21, 22.
|
Acknowledgements |
|---|
|
TOPAbstractMethodsResultsDiscussionAcknowledgementsReferences |
|---|
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionAcknowledgementsReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  J. A. Falk, O. A. Minai, and Z. Mosenifar Inhaled and Systemic Corticosteroids in Chronic Obstructive Pulmonary Disease Proceedings of the ATS, May 1, 2008; 5(4): 506 - 512. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M Sridhar, R Taylor, S Dawson, N J Roberts, and M R Partridge A nurse led intermediate care package in patients who have been hospitalised with an acute exacerbation of chronic obstructive pulmonary disease Thorax, March 1, 2008; 63(3): 194 - 200. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Suissa Immortal Time Bias in Pharmacoepidemiology Am. J. Epidemiol., February 15, 2008; 167(4): 492 - 499. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Suissa, R. McGhan, D. Niewoehner, and B. Make Inhaled Corticosteroids in Chronic Obstructive Pulmonary Disease Proceedings of the ATS, October 1, 2007; 4(7): 535 - 542. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Mikaeloff, G. Caridade, S. Assi, M. Tardieu, S. Suissa, and on behalf of the KIDSEP study group of the French Hepatitis B vaccine and risk of relapse after a first childhood episode of CNS inflammatory demyelination Brain, April 1, 2007; 130(4): 1105 - 1110. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Suissa Statin Use and Risks for Death and Hospitalization in Chronic Heart Failure JAMA, March 14, 2007; 297(10): 1057 - 1057. [Full Text] [PDF]
|
|
|
|
 |
|
 S. Setoguchi, R. J. Glynn, J. Avorn, H. Mogun, and S. Schneeweiss Statins and the Risk of Lung, Breast, and Colorectal Cancer in the Elderly Circulation, January 2, 2007; 115(1): 27 - 33. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. Macie, K. Wooldrage, J. Manfreda, and N. R. Anthonisen Inhaled Corticosteroids and Mortality in COPD. Chest, September 1, 2006; 130(3): 640 - 646. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Suissa Observational Studies of Inhaled Corticosteroids in Chronic Obstructive Pulmonary Disease: Misconstrued Immortal Time Bias Am. J. Respir. Crit. Care Med., February 15, 2006; 173(4): 464 - 464. [Full Text] [PDF]
|
|
|
|
|
|
J. F. Donohue Combination Therapy for Chronic Obstructive Pulmonary Disease: Clinical Aspects Proceedings of the ATS, November 1, 2005; 2(4): 272 - 281. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. M. Samet Inhaled Corticosteroids and Chronic Obstructive Pulmonary Disease: New and Improved Evidence? Am. J. Respir. Crit. Care Med., August 15, 2005; 172(4): 407 - 408. [Full Text] [PDF]
|
|
|
|
|
|
V. A. Kiri, N. B. Pride, J. B. Soriano, and J. Vestbo Inhaled Corticosteroids in Chronic Obstructive Pulmonary Disease: Results from Two Observational Designs Free of Immortal Time Bias Am. J. Respir. Crit. Care Med., August 15, 2005; 172(4): 460 - 464. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. Barber "Chronic obstructive pulmonary disease": the diagnostic last refuge of the intellectually challenged? Eur. Respir. J., October 1, 2004; 24(4): 710 - 711. [Full Text] [PDF]
|
|
|
|
|
|
S. Suissa From the Author Eur. Respir. J., July 1, 2004; 24(1): 191 - 192. [Full Text] [PDF]
|
|
|
|
|
|
V.A. Kiri, J. Vestbo, N.B. Pride, and J.B. Soriano Inhaled steroids and mortality in COPD: bias from unaccounted immortal time. Eur. Respir. J., July 1, 2004; 24(1): 190 - 191. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
