Observational studies of inhaled corticosteroids in chronic obstructive pulmonary disease (COPD) have shown improved survival whereas randomized trials have not. It has been suggested that this difference may be due to immortal time bias. To investigate this further, we performed a prospective cohort study of patients with COPD, using time-dependent methods to determine whether use of inhaled corticosteroids more than 80% of the time reduced the risk of all-cause mortality and COPD exacerbations. Of 8,033 patients, 2,686 (33%) received inhaled corticosteroids. We did not find a significant reduction in mortality for average inhaled steroid use at either low (hazard ratio [HR], 0.96; 95% confidence interval [CI], 0.69–1.33) or medium/high dose (HR, 0.86; 95% CI, 0.67–1.10). Similarly, recent inhaled corticosteroid use was not associated with a reduction in mortality at low (HR, 0.80; 95% CI, 0.60–1.07) or medium/high doses (HR, 0.88; 95% CI, 0.71–1.09). There was no association between inhaled corticosteroid use and hospitalizations or exacerbations due to COPD. Patients using medium/high-dose inhaled corticosteroids did not have a significantly lower risk of COPD hospitalizations (HR, 0.85; 95% CI, 0.64–1.13) or COPD exacerbations (HR, 1.13; 95% CI, 0.94–1.36). In a time-dependent study of outpatients with COPD, adherence to inhaled corticosteroid use was not associated with a decreased risk of mortality or exacerbations.
Chronic obstructive pulmonary disease (COPD) is the fourth leading cause of death in the United States, and unlike other major causes of death, continues to increase (1, 2). The benefits of inhaled corticosteroids (ICS) in the treatment of COPD remain controversial. In four randomized controlled trials, inhaled corticosteroids did not slow the decline of FEV1, but their use was associated with slightly fewer exacerbations and less health care use (3–6). In these studies, patients treated with ICS did not have significant improvement in mortality compared with those treated with placebo (3–6).
Subsequently, several large observational studies have suggested that inhaled corticosteroids, as currently used in clinical practice, are associated with a significant reduction in mortality and exacerbations (7–9). One explanation for the lack of mortality benefit seen in the randomized controlled trials may be the lower mortality rate and smaller sample sizes compared with the observational studies. Others have suggested that the beneficial effect found in observational trials resulted from misclassification of ICS users and nonusers due to immortal person-time bias, a form of survival bias (10). This bias results from assessing ICS exposure after some subjects had died, so that patients who survive have a greater amount of time to receive inhaled corticosteroids. The previous observational studies of ICS in COPD had a limited assessment of medication use. These studies measured ICS use at baseline and monitored patients to ascertain death or hospitalizations. This approach implicitly assumes that medication use is constant, which may not be the case for many patients. In addition, prior studies did not ensure that patients acquired enough medications to take ICS regularly. To minimize this type of misclassification, pharmacoepidemiologic studies frequently require that users take the medication for 80% of the days during the assessment period (11, 12).
To address the concerns raised by prior observational studies, we used pharmacy refill records (13) in a time-dependent analysis to avoid immortal person-time bias. We sought to determine whether patients who regularly filled medications (11), as well as the length of exposure and dose of their inhaled corticosteroids, were associated with death and health care use among patients with COPD in a primary care setting. Some of the results of this study have been previously reported in the form of an abstract (14).
The patients for this analysis were identified on the basis of data collected as part of the Ambulatory Care Quality Improvement Project (ACQUIP) trial, a randomized controlled trial of quality improvement interventions in the primary care setting. To participate in the ACQUIP trial, patients were required to have an outpatient clinic visit in the general internal medicine clinic at one of seven participating Department of Veterans Affairs (VA) medical centers in the prior year, and an assigned primary care provider (15). Patients were enrolled from February 1997 through December 1999. As part of the study, inpatient and outpatient administrative and pharmacy data were regularly extracted from the Veterans' Health Information System Technology Architecture (VISTA) computerized medical record system. The study protocol was approved by the Institutional Review Board at the University of Washington (Seattle, WA).
Patients were eligible for the present analysis if they had either an outpatient clinic visit or an inpatient hospitalization with a primary or secondary ICD-9 discharge diagnosis of COPD: 491.x (“chronic bronchitis”), 492.x (“emphysema”), and 496 (“chronic airways obstruction, not elsewhere classified”). To ensure that patients were being treated for COPD on entry into the cohort, we included only patients who were using at least one pulmonary medication during the 90-day period before the index visit. Subjects were 45 years of age or older and had participated in the ACQUIP trial for at least 1 year, ensuring that baseline data on comorbidities and medications were available.
All prescriptions that were filled during the study period for the following pulmonary medications were obtained from the VISTA database: inhaled corticosteroids (beclomethasone, triamcinolone, flunisolide, fluticasone, and budesonide), β-adrenergic agonists, ipratropium bromide, oral theophyllines, oral corticosteroids, and commonly used antimicrobials. Information collected included medication name and category, date on which the prescription was filled, medication dose, and supply duration (days).
The study period was divided into 90-day intervals and the percentage of days for which a medication was prescribed was calculated for each medication in each interval by dividing the total days supplied by 90 and multiplying by 100 (13). Prescriptions filled before the beginning of the interval could contribute if the days supplied extended into the interval period. Prescriptions filled at any point after the start of the interval contributed days from the date dispensed forward. For all medications except prednisone, patients were considered to be actively taking the medication (“user”) if prescriptions were filled for more than 80% of days during the interval (16) (Figure 1)
. Patients were considered chronic users of prednisone for the given time period if they received prescriptions for more than 50% of days during each interval (17).For users of ICS, an average daily dose was calculated for each time interval by dividing the total number of micrograms (based on canisters dispensed) by the number of days prescribed during the interval. To model dose, all ICS preparations were converted to triamcinolone dose equivalents (2× beclomethasone, 1× flunisolide, and 4× fluticasone) (18). There were no prescriptions for budesonide. The doses of β-adrenergic agonists and ipratropium bromide inhalers were calculated as the number of canisters prescribed during each 90-day interval. All other medications were considered as dichotomous variables and doses were not calculated.
The primary predictor for this analysis was use of ICS, defined as use for 80% of days during each time interval. We further categorized ICS users by dose (in triamcinolone equivalents) into low (less than 400 μg/day) and medium/high (more than 400 μg/day). The average daily dose exceeded 1,000 μg/day in only 1.2% of the 90-day intervals in the study (18).
All other pulmonary medications were considered potential markers of disease severity and were therefore included as potential confounders. Additional measures of disease severity included hospitalizations or outpatient visits for COPD in the year before the onset of the study, defined by a primary discharge diagnosis of COPD (ICD-9: 491, 492, or 496). To measure the extent of comorbid disease, the Deyo adaptation of the Charlson comorbidity score (excluding the lung disease category) was calculated for each patient on the basis of inpatient and outpatient ICD-9 CM codes in the year preceding the onset of the study (19, 20). The Charlson score is a weighted index of 19 chronic medical conditions that is predictive of mortality, postoperative complications, and length of hospital stay (19, 20). For example, myocardial infarct contributes one point, lymphoma two points, and metastatic cancer six points.
The primary outcome was death from any cause. Information about deaths was obtained from the VA Beneficiary Identification and Record Locator Subsystem (BIRLS), a database that records the date of death for veterans whose families file for the veteran's death benefit. The BIRLS database has been found to identify 98.8% of deaths among Medicare-eligible patients (21).
The secondary outcomes were either a hospitalization or outpatient exacerbation for COPD. Hospitalization for COPD was defined as a primary discharge diagnosis code (ICD-9) for COPD occurring during the follow-up period. An outpatient exacerbation was defined as an outpatient visit with a primary ICD-9 code for COPD combined with a new prescription for a 14 (or fewer)-day course of either prednisone or a commonly used antibiotic (amoxicillin, sulfa drugs, cephalosporins, quinolones, tetracyclines, and macrolides) that was dispensed and filled within 24–48 hours of the clinic visit.
Patients who obtain their care at the VA hospital may be hospitalized at non-VA facilities, and the likelihood of admission to a VA hospital is influenced by how far away the patient lives from the facility (22). We therefore calculated the straight-line distance from the patient's home to the VA hospital in miles to adjust for confounding by distance to the VA hospital (23). Using the postal codes for the patient's residence and the VA medical center, the corresponding longitude and latitude were determined and the straight-line distance was calculated (Geographic Data Technology, Lebanon, NH).
We performed a prospective cohort study, with the index date defined as the date of the first inpatient or outpatient visit that was assigned an ICD-9 code of COPD. Baseline medications were assessed for the 90-day period before the index date. Baseline comorbidity and health care use variables were calculated on the basis of data from the 1-year period before the index visit. Baseline variables for patients who were prescribed and received ICS at any time during the study period were compared with patients who had never received ICS, using the Student t-test for continuous variables and the χ2 test for categorical variables. All tests of significance were two sided.
We created univariate and multivariate models using Cox proportional hazards methods to estimate the hazard ratios (HRs) for death and COPD exacerbations. Three separate approaches were taken for modeling the inhaled medication exposure variables as illustrated at the top of Figures 2, 3, and 4
. For each model, the same nonmedication covariates were entered en bloc. The index date that defined the initial time point for the survival modeling was defined as the date of their first visit for COPD. Patients were censored if they had not died by the end of the follow-up period on January 1, 2000. Stata statistical software (Stata, College Station, TX) was used for all statistical analysis and for regression model development.In the time-dependent analyses, hazard ratios were estimated by comparing covariates of those who had the event (e.g., death) with those without the event during the same 90-day time interval. To model the effect of ICS in a way that allows for the fact that exposure is varying over time, it is necessary to assess ICS use for each surviving individual at every point in time at which an event (e.g., death) occurs. The three analysis strategies were as follows.
Using methods similar to those applied in previous observational studies (7, 8), medication exposure was determined during the 90-day baseline period before the index date. Because time-dependent covariates were not used in this analysis, the proportional hazards assumption was checked graphically by assessing the log(-log) plots of the survival function, and by calculating Schoenfeld residuals and the associated test statistic (24). Using the test statistic, a p value less than 0.05 suggests that the proportional hazards assumption has not been met, and that proportional hazards analyses are not appropriate.
We performed a time-dependent analysis in which the exposure was the average use of each medication in the entire study period up to the interval during which an event occurred. The average use was calculated for each 90-day interval.
A time-dependent analysis was performed in which medication use was measured in the 90-day interval immediately preceding the current analysis interval, ensuring that all subjects had the same opportunity to receive the medication.
Although all patients had an ICD-9 code for COPD, some also carried a concomitant diagnosis of asthma, leading to potential misclassification. We therefore performed a sensitivity analysis, restricting the cohort to those without an ICD-9 code for asthma (493.x) in the year before entry into the study.
A total of 8,033 patients met the inclusion criteria for this analysis. Of these, 52% filled at least one prescription for an ICS, and 2,654 (33%) were prescribed an ICS for at least 80% of a 90-day interval during the study period. The most commonly prescribed inhaled corticosteroids in this sample were triamcinolone (45%) and beclomethasone (44%); flunisolide (6%) and fluticasone (5%) were prescribed less frequently, and budesonide was not used. The average daily dose of ICS used by patients, in triamcinolone equivalents, was 663 μg (± 639). Of patients who used ICS, 63% were using more than 400 μg/day.
Patients using ICS were slightly older than those who did not use ICS, and were more likely to be married and white (Table 1)
No Inhaled Steroids (n = 5,398) | Inhaled Steroids (n = 2,654) | p Value | |
---|---|---|---|
Mean age (SD) | 66.5 (9.9) | 67.2 (9.0) | 0.01 |
Male, % | 97.9 | 98.3 | 0.3 |
Married, % | 51.9 | 57.3 | < 0.01 |
White,* % | 79.9 | 84.6 | < 0.01 |
Mean Charlson index score (SD)† | 1.2 (1.7) | 1.1 (1.5) | < 0.01 |
Health care use during year before entry | |||
Office visits, mean (SD) | |||
Total | 4.1 (3.8) | 4.4 (4.0) | 0.02 |
COPD related‡ | 0.4 (1.1) | 0.9 (1.5) | < 0.01 |
Hospitalizations | |||
All, % | 22.4 | 23.1 | 0.4 |
COPD related,‡ % | 2.3 | 5.0 | < 0.01 |
Mean distance to VA hospital, miles (SD) | 46 (124) | 44 (95) | 0.4 |
Prior concurrent diagnosis of asthma§ | 8.1 | 20.0 | < 0.01 |
Pulmonary medication use in 90 d before entry | |||
Short-acting β2-adrenergics, % | |||
Inhaled | 86.4 | 91.2 | < 0.01 |
Nebulized | 1.4 | 6.6 | < 0.01 |
Anticholinergics, % | |||
Inhaled | 55.8 | 72.8 | < 0.01 |
Nebulized | 0.5 | 2.1 | < 0.01 |
Long-acting β2-adrenergics (MDI), % | 0.4 | 2.4 | < 0.01 |
Theophylline, % | 9.0 | 22.0 | < 0.01 |
Oral corticosteroids, % | 4.9 | 8.3 | < 0.01 |
During a mean follow-up of 544 (± 240) days, 1,052 (13%) patients died, 559 (7%) were hospitalized for COPD, and 1,287 (16%) made an outpatient visit for an exacerbation. In the last group, 70% received an antibiotic only, 16% received prednisone only, and 14% received both medications.
To compare these results with prior studies that have used baseline pharmacy data alone (7, 8), we first performed an analysis using only baseline prescription data to predict death, hospitalizations, and outpatient exacerbations. During the baseline period, 999 (12.3%) of patients were exposed to ICS. After adjusting simultaneously for all other pulmonary medications, as well as age, VA hospital site, prior outpatient and inpatient COPD visits, comorbidity, and distance to the VA hospital, we did not find a significant reduction in the risk of death for patients prescribed ICS compared with those who did not use ICS (HR, 0.87; 95% CI, 0.72–1.05). After stratifying by ICS dose, there was no association between low-dose ICS use (HR, 0.75; 95% CI, 0.53–1.05), or medium/high-dose ICS use (HR, 0.91; 95% CI, 0.73–1.13) and death (Figure 2). In this baseline analysis, nebulized β-agonists and oral corticosteroids were associated with a significantly increased risk of death (Figure 2).
We evaluated the Cox proportional hazards assumption graphically by creating log(-log) plots of the survival function, and we found that ICS use did not meet the proportional hazards assumption for mortality, but did meet the assumption for hospitalizations and exacerbations. This was confirmed by calculating Schoenfeld residuals and the associated test statistic, which also showed that the proportional hazards assumption for ICS was not met for mortality (χ2 = 6.09, p = 0.0136).
We then performed time-dependent analyses, again adjusting simultaneously for all other pulmonary medications, as well as age, VA hospital site, prior outpatient and inpatient COPD visits, comorbidity, and distance to the VA hospital. For average ICS use, we found no significant protective effect of ICS at either a low dose (HR, 0.96; 95% CI, 0.69–1.33) or medium/high doses (HR, 0.86; 95% CI, 0.67–1.10). Among other COPD medications, nebulized β-agonist use (HR, 1.60; 95% CI, 1.19–2.16) and chronic prednisone use (HR, 1.70; 95% CI, 1.46–2.12) were associated with an increased risk of death (Figure 3).
Similarly, an adjusted time-dependent analysis of recent ICS use in the 90-day interval before death did not show a decreased risk of death for low-dose ICS use (HR, 0.88; 95% CI, 0.60–1.07) or medium/high use (HR, 0.88; 95% CI, 0.71–1.09). Again, nebulized β-agonist use (HR, 1.52; 95% CI, 1.18–1.95) and prednisone use (HR, 2.00; 95% CI, 1.68–2.38) were associated with an increased risk of mortality (Figure 4).
In baseline analyses, there was no difference in hospitalizations between those using inhaled corticosteroids and those not using ICS (HR, 0.85; 95% CI, 0.67–1.06). Similarly, using time-dependent methods, we found no protective effect of ICS whether average or recent ICS use was measured. Results for all models did not differ significantly, and only the models of average use of ICS before hospitalization are presented (Table 2)
Hospitalizations | Outpatient Exacerbations | |||||
---|---|---|---|---|---|---|
Variable | HR* | 95% CI | HR* | 95% CI | ||
Inhaled corticosteroids† | ||||||
None | 1.0 | Reference | 1.0 | Reference | ||
< 400 | 1.10 | 0.77, 1.59 | 1.05 | 0.81, 1.37 | ||
> 400 | 0.85 | 0.64, 1.13 | 1.13 | 0.94, 1.36 | ||
Inhaled β-agonist‡ | ||||||
< 1 | 1.0 | Reference | 1.0 | Reference | ||
1–2 | 1.18 | 0.95, 1.46 | 1.24 | 1.08, 1.43 | ||
> 2 | 1.77 | 1.37, 2.28 | 1.47 | 1.23, 1.76 | ||
Inhaled ipratropium‡ | ||||||
< 1 | 1.0 | Reference | 1.0 | Reference | ||
1–2 | 1.31 | 1.05, 1.62 | 1.43 | 1.25, 1.65 | ||
> 2 | 1.59 | 1.21, 2.11 | 1.36 | 1.12, 1.66 | ||
Nebulized β-agonist | 1.89 | 1.35, 2.64 | 1.59 | 1.24, 2.05 | ||
Nebulized ipratropium | 1.08 | 0.63, 1.92 | 1.24 | 0.82, 1.89 | ||
Prednisone | 1.45 | 1.13, 1.87 | 1.26 | 1.04, 1.52 | ||
Theophylline | 1.12 | 0.91, 1.39 | 1.15 | 1.00, 1.34 | ||
Salmeterol | 1.30 | 0.80, 2.13 | 1.46 | 1.02, 2.10 |
To determine whether the results obtained in this analysis were due in part to inclusion of patients with a concomitant diagnosis of asthma, we restricted the analyses to those without a prior concurrent asthma diagnosis. Restriction did not significantly affect the results in any of the models. For example, when modeling the average ICS use before the event (Table 3)
Death | Hospitalizations | Outpatient Exacerbations | |||||||
---|---|---|---|---|---|---|---|---|---|
HR* | 95% CI | HR* | 95% CI | HR* | 95% CI | ||||
Inhaled corticosteroids† | |||||||||
None | 1.0 | Reference | 1.0 | Reference | 1.0 | Reference | |||
< 400 | 1.01 | 0.72, 1.43 | 1.25 | 0.84, 1.86 | 1.07 | 0.80, 1.43 | |||
> 400 | 0.94 | 0.72, 1.21 | 0.81 | 0.58, 1.11 | 1.12 | 0.91, 1.37 |
We examined the relationship between inhaled corticosteroid use and mortality, using three different approaches to model ICS use and length of exposure. When replicating prior studies using baseline medication use, we did not find a significant reduction in mortality or hospitalizations. Similarly, for the analyses using time-dependent methods, we did not find a protective effect for either mortality or hospitalizations associated with either average or recent ICS use, even among those with medium/high daily ICS dose. This suggests that in a large cohort of outpatients with COPD, after adjusting for measures of comorbidity and COPD severity, there is no significant effect of ICS on either mortality or COPD exacerbations.
Four large randomized controlled trials of ICS in COPD had not shown a significant beneficial effect of ICS on decline in FEV1, but had not been designed to detect a reduction in mortality. Subsequently, in a large observational study, Sin and Tu found a 29% reduction in mortality for patients who received one prescription for ICS in the 90 days after discharge from the hospital for a COPD exacerbation (7). Soriano and co-workers also found a 38% reduction in mortality among outpatients who received three prescriptions for ICS in a 6-month period (8).
It has been suggested that one explanation for the surprising large reduction in mortality seen in these observational trials is immortal person-time bias (10). This bias occurs when the exposure is measured in terms of cumulative lifetime exposure and those who die or have an event will have less cumulative exposure. In such a biased analysis, subjects who die do not have the same opportunity to receive a prescription for ICS as a subject who lives. As a result, patients who receive ICS may appear to survive simply because they are alive to receive the medication. This underscores the importance of an accurate assessment of ICS use or exposure.
In this analysis, we have employed two methods to more accurately measure ICS use. First, as routinely performed in observational cohort studies in cardiovascular medicine (11, 13), we have defined users as those who received medications more than 80% of the time. This definition of ICS use is more likely to accurately represent true medication use and exposure than those previously used, and increases the likelihood that any effect seen is due to the medication itself. Second, as recommended by Suissa (10) and Samet (25), we have performed a time-dependent analysis in which ICS use can change over time. We have ensured that both the patients who live and those who die during a 90-day interval had the same opportunity to receive the medication in the preceding time periods to avoid immortal person-time bias.
Use of time-dependent methods is also important because in attempting to replicate previous studies (7, 8) that measured medication use at baseline, we found that this approach violated two different tests of the proportional hazards assumption for Cox regression analysis in our study population. These previous studies using Cox regression methods do not comment on the tests of proportionality, and our results suggest that the risk of adverse outcomes in patients who use ICS changes over time and that conclusions drawn from a baseline analysis should be viewed with caution. Employing time-dependent methods addresses violations in the proportional hazards assumption.
We observed a lower average daily dose of ICS use in this population than in clinical trials. This is consistent with observational studies that have found that 54% of patients with COPD underutilize inhaled medications (26), and that only 54% of patients were compliant with ICS in a large health maintenance organization setting (27). In another observational trial that found a significant protective effect of ICS on mortality, 60% of patients were using beclomethasone at less than 500 μg/day (9), suggesting that in actual practice most patients are using low-dose ICS. Two other prior observational studies of ICS and mortality did not report average daily dose (7, 8); however, it is possible that if patients in those populations were adherent with higher doses of ICS, it might explain the protective effect of ICS that was seen.
To determine whether we had enough power to detect the marked reduction in mortality seen in other studies, we performed power calculations based on our data for a two-sample comparison of survival curves based on Freedman/Peto formulas for a 1.5-year follow-up period and a two-sided α value of 0.05 (28). On the basis of the 12.5% baseline exposure to ICS in our study, there was a power of 0.89 to detect the 29% reduction in mortality described in the Sin and Tu article (7), and a power of 0.99 to detect the 38% reduction in mortality described in the article by Soriano and coworkers (8). On the basis of these calculations, there was adequate power in our study to detect the mortality benefit seen in the prior studies. These calculations would tend to underestimate the power in our study, as we used time-dependent methods incorporating data from multiple time intervals.
There are several potential limitations to this study. Similar to prior studies of COPD using administrative data, we were unable to confirm the diagnosis or severity of COPD spirometrically. We were able to include several measures of COPD severity including oral corticosteroid use, previous COPD exacerbations, and number of inhaled bronchodilators. Even though we adjusted for these characteristics, residual confounding from severity of COPD may exist. We were also not able to ensure that patients did not have asthma. The prevalence of concomitant asthma among patients using ICS in our study was significantly lower than in the study by Soriano and co-workers (20 versus 74%), suggesting that there was less misclassification of patients with asthma in our sample. Furthermore, restriction to those without a prior asthma diagnosis did not significantly change our results and the direction of the bias introduced by including patients with asthma would likely be toward improved outcomes with ICS.
To perform this study, several assumptions were made regarding medication use and pharmacy data within the VA system. First, we assumed veterans obtained all their medications through the VA. This is supported by the fact that there is a financial incentive of either no copay or minimal copay when prescriptions are filled in the VA system, and 98.5% of VA patients prescribed antihypertensives identified the VA system as the only source of medications (12). We also assumed that patients who filled prescriptions used the medication as directed during the prescribed interval, and did not save or lose medications, which may have overestimated medication use in this population. This bias could minimize a true benefit of ICS, but is also a limitation of all the previous observational studies.
In addition, veterans may obtain health care outside the VA, and we may therefore have missed COPD hospitalizations occurring at non-VA facilities. Although we adjusted for the distance to the VA hospital as a proxy for likelihood of being admitted to a VA facility, it is nevertheless possible that there was residual confounding by access to care delivered by VA facilities. This could result in misclassification of the outcomes, but is unlikely to introduce bias unless ICS is also associated with admission to non-VA facilities. Finally, veterans are generally sicker and have worse general health status than non-VA patients (29). This bias would increase the likelihood of finding a true effect of ICS because prior studies suggest the effect is more pronounced among more severely ill patients. Strengths of this design include a prospective cohort design in a large outpatient setting, measuring ongoing medication refills as well as dose of ICS, and using time-dependent covariates to estimate risk of adverse outcomes.
In conclusion, by defining ICS users as those who regularly fill prescriptions, and by using time-dependent methods, we did not find an association between length of ICS use or dose with either mortality or COPD exacerbations. Our results support the suggestion that the previously reported protective effect of inhaled corticosteroids on mortality seen in cohort studies may be due to residual confounding factors, such as immortal time bias. Because prior randomized controlled studies have used decline in pulmonary function as the primary outcome, our results further support performing a large randomized control trial of inhaled corticosteroids in COPD with outcomes of mortality and exacerbations.
V.S.F. has no declared conflict of interest; C.L.B. has no declared conflict of interest; J.R.C. has no declared conflict of interest; S.D.F. has no declared conflict of interest; P-O.B. has no declared conflict of interest; M.B.McD. has no declared conflict of interest; D.H.A. received funding support for validation of a health status instrument from Inspire Pharmaceuticals ($20,567 for 10/2001–1/2002).
1. | Mannino DM, Brown C, Giovino GA. Obstructive lung disease deaths in the United States from 1979 through 1993: an analysis using multiple-cause mortality data. Am J Respir Crit Care Med 1997;156:814–818. |
2. | Benson V, Marano MA. Current estimates from the National Health Interview Survey, 1995. Vital Health Stat 10 1998;199:1–428. |
3. | Pauwels RA, Lofdahl CG, Laitinen LA, Schouten JP, Postma DS, Pride NB, Ohlsson SV. Long-term treatment with inhaled budesonide in persons with mild chronic obstructive pulmonary disease who continue smoking. European Respiratory Society Study on Chronic Obstructive Pulmonary Disease. N Engl J Med 1999;340:1948–1953. |
4. | Lung Health Study Research Group. Effect of inhaled triamcinolone on the decline in pulmonary function in chronic obstructive pulmonary disease. N Engl J Med 2000;343:1902–1909. |
5. | Burge PS, Calverley PM, Jones PW, Spencer S, Anderson JA, Maslen TK. Randomised, double blind, placebo controlled study of fluticasone propionate in patients with moderate to severe chronic obstructive pulmonary disease: the ISOLDE trial. BMJ 2000;320:1297–1303. |
6. | Vestbo J, Sorensen T, Lange P, Brix A, Torre P, Viskum K. Long-term effect of inhaled budesonide in mild and moderate chronic obstructive pulmonary disease: a randomised controlled trial. Lancet 1999;353:1819–1823. |
7. | Sin DD, Tu JV. Inhaled corticosteroids and the risk of mortality and readmission in elderly patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2001;164:580–584. |
8. | Soriano JB, Vestbo J, Pride NB, Kiri V, Maden C, Maier WC. Survival in COPD patients after regular use of fluticasone propionate and salmeterol in general practice. Eur Respir J 2002;20:819–825. |
9. | Sin DD, Man SF. Inhaled corticosteroids and survival in chronic obstructive pulmonary disease: does the dose matter? Eur Respir J 2003;21:260–266. |
10. | Suissa S. Effectiveness of inhaled corticosteroids in chronic obstructive pulmonary disease: immortal time bias in observational studies. Am J Respir Crit Care Med 2003;168:49–53. |
11. | Psaty BM, Heckbert SR, Koepsell TD, Siscovick DS, Raghunathan TE, Weiss NS, Rosendaal FR, Lemaitre RN, Smith NL, Wahl PW, et al. The risk of myocardial infarction associated with antihypertensive drug therapies. JAMA 1995;274:620–625. |
12. | Steiner JF, Koepsell TD, Fihn SD, Inui TS. A general method of compliance assessment using centralized pharmacy records: description and validation. Med Care 1988;26:814–823. |
13. | Steiner JF, Prochazka AV. The assessment of refill compliance using pharmacy records: methods, validity, and applications. J Clin Epidemiol 1997;50:105–116. |
14. | Fan VS, Bryson CL, Au DH, Bridevaux PO, McDonell MB, Fihn SD. Inhaled corticosteroid use over time and risk of death in COPD patients [abstract]. Am J Respir Crit Care Med 2003;167:A89. |
15. | Au DH, McDonell MB, Martin DC, Fihn SD. Regional variations in health status. Med Care 2001;39:879–888. |
16. | Wannemacher AJ, Schepers GP, Townsend KA. Antihypertensive medication compliance in a veterans affairs healthcare system. Ann Pharmacother 2002;36:986–991. |
17. | Fan VS, Curtis JR, Tu SP, McDonell MB, Fihn SD. Using quality of life to predict hospitalization and mortality in patients with obstructive lung diseases. Chest 2002;122:429–436. |
18. | NHLBI/WHO Workshop Report: Global Strategy for Asthma Management and Prevention. Issued January, 1995. NIH Publication No 02–3659. 2002 report is available from: www.ginasthma.com |
19. | Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373–383. |
20. | Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45:613–619. |
21. | Fleming C, Fisher ES, Chang CH, Bubolz TA, Malenka DJ. Studying outcomes and hospital utilization in the elderly: the advantages of a merged data base for Medicare and Veterans Affairs hospitals. Med Care 1992;30:377–391. |
22. | Burgess JF Jr, DeFiore DA. The effect of distance to VA facilities on the choice and level of utilization of VA outpatient services. Soc Sci Med 1994;39:95–104. |
23. | Phibbs CS, Luft HS. Correlation of travel time on roads versus straight line distance. Med Care Res Rev 1995;52:532–542. |
24. | Grambsch P, Therneau T. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994;81:515–526. |
25. | Samet JM. Measuring the effectiveness of inhaled corticosteroids for COPD is not easy! Am J Respir Crit Care Med 2003;168:1–2. |
26. | Dolce JJ, Crisp C, Manzella B, Richards JM, Hardin JM, Bailey WC. Medication adherence patterns in chronic obstructive pulmonary disease. Chest 1991;99:837–841. |
27. | Kelloway JS, Wyatt RA, Adlis SA. Comparison of patients' compliance with prescribed oral and inhaled asthma medications. Arch Intern Med 1994;154:1349–1352. |
28. | Freedman LS. Tables of the number of patients required in clinical trials using the logrank test. Stat Med 1982;1:121–129. |
29. | Kazis LE, Miller DR, Clark J, Skinner K, Lee A, Rogers W, Spiro A III, Payne S, Fincke G, Selim A, et al. Health-related quality of life in patients served by the Department of Veterans Affairs: results from the Veterans Health Study. Arch Intern Med 1998;158:626–632. |