Risk for Death Associated with Medications for Recently Diagnosed Chronic Obstructive
Pulmonary Disease
Todd A. Lee, PharmD, PhD; A. Simon Pickard, PhD; David H. Au, MD, MS; Brian Bartle, MPH; and Kevin
B. Weiss, MD, MPH, MS
16 September 2008 | Volume 149 Issue 6 | Pages 380-390
Background: Concerns exist regarding increased risk for mortality associated with some chronic obstructive pulmonary
disease (COPD) medications.
Objective: To
examine the association between various respiratory medications and risk for death in veterans with newly diagnosed
COPD.
Design: Nested
case–control study in a cohort identified between 1 October 1999 and 30 September 2003 and followed through
30 September 2004 by using National Veterans Affairs inpatient, outpatient, pharmacy, and mortality databases;
Centers for Medicare & Medicaid Services databases; and National Death Index Plus data. Cause of
death was ascertained for a random sample of 40% of those who died during follow-up. Case patients were categorized
on the basis of all-cause, respiratory, or cardiovascular death. Mortality risk associated with medications was
assessed by using conditional logistic regression adjusted for comorbid conditions, health care use,
and markers of COPD severity.
Setting: U.S. Veterans Health Administration health care system.
Participants: 32 130 case patients and 320 501 control participants in the all-cause mortality analysis. Of
11 897 patients with cause-of-death data, 2405 case patients had respiratory deaths and 3159 case
patients had cardiovascular deaths.
Measurements: All-cause mortality; respiratory and cardiovascular deaths; and exposure to COPD medications, inhaled corticosteroids,
ipratropium, long-acting β-agonists, and theophylline in the 6 months preceding death.
Results: Adjusted
odds ratios (ORs) for all-cause mortality were 0.80 (95% CI, 0.78 to 0.83) for inhaled corticosteroids, 1.11
(CI, 1.08 to 1.15) for ipratropium, 0.92 (CI, 0.88 to 0.96) for long-acting β-agonists, and 1.05 (CI, 0.99
to 1.10) for theophylline. Ipratropium was associated with increased cardiovascular deaths (OR, 1.34
[CI, 1.22 to 1.47]), whereas inhaled corticosteroids were associated with reduced risk for cardiovascular
death (OR, 0.80 [CI, 0.72 to 0.88]). Results were consistent across sensitivity analyses.
Limitations: Current smoking status and lung function were not measured. Misclassification of cause-specific mortality
is unknown.
Cohort
Patients were eligible for inclusion if they received a diagnosis
of COPD (International Classification of Diseases, 9th Revision [ICD-9], codes 491.x, 492.x, or 496) between
1 October 1999 and 30 September 2003 at 2 or more outpatient visits within 12 months or were admitted to
the hospital with a primary diagnosis of COPD. Patients had to be 45 years of age or older when they received
their first eligible diagnosis, have used Veterans Health Administration health care services for at least 1 year
before their first COPD diagnosis, and have received respiratory medications. We excluded patients with a
diagnosis of asthma. We followed patients from the date of their second eligible outpatient visit or
their inpatient visit until death or 30 September 2004.
Case Patients
We identified all deaths that occurred during follow-up by using
the Veterans Affairs Vital Status database, a combination of Veterans Affairs, Medicare, and Social Security
Administration mortality data that captures approximately 98% of veteran deaths. Of these, 40% was randomly sampled
and we attempted to determine cause of death. This sample was estimated to provide more than 80% power
to detect odds ratios of 0.85 or lower or 1.15 or higher for each medication class. We ascertained cause of
death by using National Death Index Plus data from the National Center for Health Statistics. We defined 4 groups of case patients on the basis of cause of death: respiratory,
cardiovascular, respiratory or cardiovascular, and all-cause mortality. We defined respiratory as death
due to a respiratory system disease (ICD-10 codes J00 to J99) and cardiovascular as death due to ischemic
heart disease (ICD-10 codes I20 to I25), cardiomyopathy, cardiac arrest, or arrhythmias (ICD-10 codes I42
to I51). The index date for case patients was their death date.
Control Participants
Selecting more than 5 control participants per case patient can
yield limited gains in efficiency; however, because we were assessing several medications simultaneously, we selected
up to 10 control participants per case patient. We randomly selected control participants for each case
patient from eligible patients who were alive at the time of the case event. We matched control participants
to case patients individually on the basis of sex, age category (45 to 54 years, 55 to 64 years, 65
to 74 years, 75 to 84 years, and 85
years of age), region of the country, and year of diagnosis. We assigned control participants the same
index date as their matched case patients.
Exposure
We defined exposure to respiratory medications as having received
medications in the 180 days preceding each patient's index date. We identified medication exposure to inhaled
corticosteroids, ipratropium, long-acting β-agonists, theophylline, and short-acting β-agonists.
We defined primary exposure as any exposure in the 180-day period before the index date. We created
mutually exclusive medication regimens on the basis of medication exposure. Exposure to short-acting β-agonists
was not considered as part of the regimen but was included as a covariate in the analysis.
Covariates
We identified covariates by using data from the year before diagnosis
date until the index date. We used pharmacy data to identify medication use, including exposure to systemic steroids,
antihypertensives, lipid-lowering medications, antiarrhythmics, and diabetes medications. We used inpatient
and outpatient diagnoses to identify comorbid conditions. We measured health care utilization as the
annual number of hospitalizations and outpatient physician visits. We identified COPD exacerbations during follow-up
and whether they were inpatient or outpatient by using a previously described algorithm
Statistical Analysis
We performed separate analyses for respiratory-specific, cardiovascular-specific,
and all-cause mortality. We used conditional logistic regression to estimate adjusted odds ratios (ORs) and
95% CIs. We included the variables that we considered clinically important in each of the regression
models. Specifically, we included measures of COPD-related severity in all of the models and included markers
of cardiovascular disease in the models for cardiovascular and all-cause mortality. We included any remaining
variables that changed OR estimates for respiratory medications by more than 10% in the final models.
We assessed model fit by using the Bayesian information criterion and the Wald test of likelihood ratios
and through examination of outlier effects with leverage and fit diagnostics. Adjusted odds ratios represented
risk for events in patients receiving medication compared with those who had not received inhaled corticosteroids,
ipratropium, long-acting β-agonists, or theophylline in the previous 6 months. We performed all
analyses with Stata/MP 10.0 for Windows (StataCorp, College Station, Texas).
We conducted several sensitivity analyses to evaluate the robustness
of our results. First, we restricted the comparison group to patients who were actively treated with a short-acting
β-agonist in the 180 days preceding the index date. Second, because veterans may use health care
services outside the Veterans Health Administration system, we restricted the analysis to patients 65 years of
age or older. We used Medicare health care utilization data on these patients to capture health care
utilization outside of the Veterans Health Administration system. Third, we examined dose response by
classifying those in the highest quartile of average daily dose into a high-dose group and the rest of those exposed
into a low-dose group. Fourth, to observe the effects of ipratropium independent of short-acting β-agonist
exposure, we excluded patients who received a combination of ipratropium and short-acting β-agonists
in a single inhaler. Fifth, to address the imbalance in prevalence of chronic heart failure between case patients
and control participants, we created analytic cohorts by matching on presence of chronic heart failure and
repeated our analyses. Finally, we used the array approach to estimate the effect that unmeasured confounding
could have had on point estimates of the association between medications and mortality. We varied the
level of risk associated with the unmeasured confounder and the prevalence in the medication groups relative to
the no-treatment groups to determine what level of differential exposure would change the conclusions
from the primary analysis. We focused on current smoking rates and COPD severity because we considered
these to be 2 of the most important and influential unmeasured confounders. We compared rates of smoking status
and COPD severity across treatment groups by using data from a recently published study
Results: We identified 145 020 patients who met inclusion criteria; of
whom, 32 130 died . We located cause-of-death data for 11 897 patients but could not match data for 955
potential case patients. Of the patients for whom cause-of-death data were available, 2405 deaths were respiratory
and 3159 were cardiovascular. Within the respiratory mortality analysis, case patients had lower rates
of hypertension and osteoarthritis, higher rates of chronic heart failure, and more COPD exacerbations. Among those
included in the cardiovascular mortality analysis, case patients had higher rates of cardiovascular comorbid
conditions and COPD exacerbations, which suggests more severe respiratory disease. Case patients had a similar
or higher prevalence of respiratory medication use than control participants. The 3 top regimens for
each group were no medication or short-acting β-agonists only, ipratropium only, and inhaled corticosteroids
and ipratropium.
After we adjusted for differences in covariates, both inhaled corticosteroids
and long-acting β-agonists were associated with reduced odds of death (OR, 0.80 [95% CI, 0.78 to 0.83]
for inhaled corticosteroids and 0.92 [CI, 0.88 to 0.96] for long-acting β-agonists), whereas ipratropium
was associated with an increased risk (OR, 1.11 [CI, 1.08 to 1.15]). For cause-specific mortality, theophylline
exposure was associated with a statistically significant increase in respiratory deaths compared with the
unexposed group (OR, 1.71 [CI, 1.46 to 2.00]). Although point estimates indicated a more than 10% increase
in the odds of respiratory death associated with long-acting β-agonists (OR, 1.12 [CI, 0.97 to
1.30]) and a more than 10% decrease with inhaled corticosteroids (OR, 0.88 [CI, 0.79 to 1.00]), neither was
statistically significant. With respect to cardiovascular death, ipratropium exposure was associated with a 34%
increase in the odds of cardiovascular death (OR, 1.34 [CI, 1.22 to 1.47]), whereas inhaled corticosteroid
exposure was associated with a 20% decrease in the odds of a cardiovascular death (OR, 0.80 [CI, 0.72
to 0.88]). Long-acting β-agonists (OR, 0.97 [CI, 0.84 to 1.11]) and theophylline (OR, 1.16 [CI, 0.99 to 1.37])
were not associated with statistically significant risks in cardiovascular deaths.
A recently published study of veterans reported the prevalence of
current smokers to be 11% among those receiving short-acting β-agonists or not treated, 20% among those receiving
ipratropium, 18% among those receiving inhaled corticosteroids, and 16% among those receiving long-acting
β-agonists. With current smoking associated with a relative risk for death of 1.5, these estimates
would result in adjusted risk ratios of 0.77 for inhaled corticosteroids, 1.08 for ipratropium, and 0.90 for long-acting
β-agonists. To reduce the association between ipratropium and all-cause mortality to an adjusted OR
of 1.0, the proportion of current smokers in the ipratropium-treated group would have to reach 35% relative
to the 11% observed in the no-treatment group. To reduce the association between cardiovascular death and
ipratropium to an odds ratio of 1.0, the prevalence of current smoking in the ipratropium group would have had
to be at least 80%.
For COPD disease severity, we assumed an increase of 1.4 in the
risk for death for those with moderate to severe COPD relative to those with mild COPD, based on the NHANES (National
Health and Nutrition Examination Surveys) data in Maninno and colleagues' study. From the published
pilot study, we found the prevalence of moderate to severe COPD to be approximately 20% in the untreated group;
45% to 55% in those receiving inhaled corticosteroids, ipratropium, or long-acting β-agonists; and 70% in
those treated for COPD. However, COPD was not newly diagnosed in the patients in this study, and the
differential estimates are therefore probably higher than those for patients with newly diagnosed COPD.
Accounting for severity as an unmeasured confounder would result in an all-cause mortality risk ratio of 0.90 for
theophylline, 1.02 for ipratropium, 0.72 for inhaled corticosteroids, and 0.83 for long-acting β-agonists.
For respiratory death, if the estimated risk for COPD-related death is increased nearly 4-fold in those
with moderate to severe disease compared with those with mild disease, the risk associated with theophylline
would be reduced to 0.88 if the prevalence of moderate to severe disease was 70% in the theophylline group
and 20% in the no-treatment group.
Among medication regimens, those that included theophylline were
associated with increased risk for respiratory death. For cardiovascular death, ipratropium alone (OR, 1.42 [CI,
1.27 to 1.59]) and ipratropium plus theophylline (OR, 1.47 [CI, 1.09 to 1.98]) were associated with
increased risk, whereas the presence of inhaled corticosteroids with ipratropium reduced the risk for
cardiovascular death (OR, 1.04 [CI, 0.90 to 1.22]; P < 0.001 for Wald test compared with ipratropium
alone). In the all-cause mortality group, inhaled corticosteroids were consistently associated with
reduced odds of death when used alone or in combination with other medications, whereas ipratropium and ipratropium
plus theophylline were associated with elevated risk for death.
Conclusion: The possible association between ipratropium and elevated risk for all-cause and cardiovascular death needs
further study.
Caution: Potential confounders, such as smoking status and disease severity, were not known.
Associations may not reflect causal relationships. In addition to the risk for cardiovascular death with ipratropium, we found an increased
risk for cardiovascular or respiratory death in patients exposed to theophylline that was primarily due
to increased respiratory deaths. Despite limited evidence on the value of theophylline for treating patients with
COPD, nearly 10% of patients received the drug. Although we attempted to control for differences in
disease severity by limiting the cohort to patients with recently diagnosed COPD and by adjusting for
COPD exacerbations, some of the increased risk associated with theophylline may be due to more severe disease in
patients exposed to theophylline. We therefore conducted a sensitivity analysis to evaluate the influence
of disease severity as an unmeasured confounder on disease outcomes. We found that the observed association
between theophylline and respiratory mortality could be completely explained by differences in disease severity
between the groups if the prevalence of moderate to severe disease was 60% in the theophylline-treated patients
relative to 20% in the comparison group. This was less than the 70%
prevalence of moderate to severe disease that we observed among our small cohort of patients with COPD
who received theophylline; however, this group was not restricted to patients with newly diagnosed COPD.
Therefore, differences in disease severity may account for some of the observed difference in risk for respiratory
death associated with theophylline; however, it is unclear whether a large enough difference in severity
would exist among those with newly diagnosed COPD to fully explain our findings.
In contrast, we found inhaled corticosteroid use was associated
with decreased risk for cardiovascular death. Our estimate of a 22% reduction in the risk for cardiovascular
death, combined cardiovascular and respiratory death, and all-cause mortality in patients exposed to
inhaled corticosteroids and long-acting β-agonists is consistent with the hazard ratio of 0.78 (CI,
0.57 to 1.06) reported by the TORCH trial for COPD-related deaths. Our results suggest that decreased risk for
cardiovascular events may be responsible for benefits associated with inhaled corticosteroids in patients
with recently diagnosed COPD. Previous reports have shown that inhaled corticosteroids reduce inflammatory
markers in patients with COPD; a hypothesized mechanism includes potential reduction in cardiovascular risk. However,
these results need to be considered in the context of other cause-specific mortality estimates. The patient-level
meta-analysis that found benefit versus all-cause mortality found no association between inhaled corticosteroids
and cardiovascular death (hazard ratio, 0.98 [CI, 0.59 to 1.62]), although the study was inadequately
powered to address this outcome. Of note, the TORCH trial the INSPIRE study and a recently published
observational study have all found an increased risk for pneumonia associated with use of inhaled corticosteroids.
We did not assess the risk for pneumonia in this cohort, although it needs to be considered in relation to
the mortality findings by patients and providers when making treatment decisions. Our study has limitations. Observational studies are susceptible to bias due to confounding
by indication, in which patients with more severe disease may be more likely to have events and therefore
are more likely to be exposed to the treatment of interest. We attempted to minimize concerns about this effect
by focusing on recently diagnosed disease and by controlling for markers of disease severity. In addition,
we conducted a sensitivity analysis to examine the potential effect of unmeasured confounding due to
disease severity. In a small sample of veterans with COPD, we found a substantial difference in the proportion
of patients with moderate to severe COPD between those who received only short-acting β-agonists or
no treatment and those who received other treatments. If this difference in severity was present in
our study, it would have been sufficient to explain the observed increase in risk for respiratory death associated
with theophylline. Severity of COPD as an unmeasured confounder could have also reduced the association between
ipratropium and all-cause mortality to 1.02. However, this cohort included patients with prevalent COPD
and probably overestimates the difference in the prevalence of severity by treatment group. We found
that for unmeasured severity to fully account for the observed differences, the prevalence of severe disease in
patients treated with theophylline would have to be 3 times that observed in those who received no treatment
or who received only short-acting β-agonists. For the association between all-cause mortality and
ipratropium, the prevalence of severe disease in the ipratropium group would need to be nearly 2.5 times that observed
in the patients who received no treatment or who only received short-acting β-agonists. Thus, differences
in severity between treatment groups would need to be very large to account for the associations that
we observed.
Similar to disease severity, we could not ascertain smoking status
among our cohort, which may have introduced unmeasured confounding. However, it is probable that many of these
patients have a history of substantial smoking, given the clinical diagnosis of COPD. Our sensitivity
analysis of current smoking rates indicates that current smokers would have to account for a larger proportion
of those treated with ipratropium to reduce the odds ratio to 1.0 for either all-cause mortality or cardiovascular
death. We also conducted several other sensitivity analyses that strengthen our findings due to the
consistency of the results. We found consistent results across the subgroups of patients for each of
the analyses.
Our study is probably underpowered for many of the medications examined
in the regimen-based analysis of cause-specific mortality. Because of the low rates of exposure to some regimens,
we are unlikely to find statistically significant differences. However, for the most commonly used medications,
we found risks consistent with the main effects observed for each medication. We also found that combinations
of medications may reduce the risks associated with individual medication exposures. In addition, many
of the point estimates for the regimen-based analyses of cause-specific mortality were consistent with the all-cause
mortality analysis, in which the larger number of included case patients alleviated problems of statistical
power.
Our results are most applicable to men with recently diagnosed COPD
and may not apply to patients with more severe disease or to women. We used National Death Index Plus data to identify
cause of death, which has shown high concordance with coding by trained nosologists but may not be consistent
with adjudicated deaths in clinical trials. The TORCH trial investigators observed 52% agreement between adjudicated deaths and the site investigator's
coding of primary cause of death, and agreement increased to 67% compared with either primary or secondary cause
listed by the site investigator. How adjudicated deaths would relate to underlying cause defined in National
Death Index Plus data is unclear and thus it is difficult to know how often the cause of death we identified
by using death certificate data would differ from that ascertained by an adjudication committee. The
degree of misclassification may be related to severity—patients with more severe disease were more likely
to have COPD listed as a cause of death. This may raise concerns that individuals treated with ipratropium
had less severe disease than other patients and were more likely to be classified as having a non–COPD-related
cause of death. However, we also evaluated the association between medication exposure and all-cause mortality.
If the relationship between ipratropium and cardiovascular death occurred because deaths were more likely
to be classified as cardiovascular than respiratory in the ipratropium group owing to differences in disease
severity rather than to medication exposure, we would expect to find no association between ipratropium and all-cause
mortality. However, we found an association between ipratropium and all-cause mortality, which warrants concerns
about the safety of the medication in treating COPD.
Our study contributes important new evidence on the potential harms
associated with medications used in COPD. Ipratropium may increase the risk for cardiovascular death; however,
this risk may be attenuated by the concomitant use of inhaled corticosteroids, which were associated
with reduction in the risk for all-cause and cardiovascular death. The risk for death due to some medications
must be weighed against potential benefits of these medications that are not captured in observational database
studies, such as symptom relief, health status, or quality of life. It is not clear, however, that these
benefits would outweigh the increased risk for death. Given the risk observed in our study and in previous
studies of ipratropium, caution is warranted in the use of ipratropium alone in patients with recently diagnosed
COPD. It is also important to further examine this relationship through measurement of cardiovascular events
and to measure the effect of such factors as disease severity and smoking status.
Grant Support:
By the U.S. Department of Veterans Affairs Health Services Research and Development
(IIR 03-307).
Potential Financial Conflicts of Interest: Honoraria: T.A. Lee (AstraZeneca, Novartis), D.H. Au (GlaxoSmithKline). Consultancies:
K.B. Weiss (Merck & Co.). Stock ownership or options (other than mutual funds): D.H. Au (Pfizer).
Grants received: T.A. Lee (Altana, Aventis, AstraZeneca, Boehringer Ingelheim, Chiesi, GlaxoSmithKline,
Merck & Co., Novartis, Pfizer, Schering-Plough, Sepracor, University of Kentucky). Other: D.H. Au (Assessing the Impact of Recent Updates for Advair and Serevent Special Issues
Board).
