Can Outpatient Pharmacy Data Identify Persons With Undiagnosed COPD?

, , , , ,
The American Journal of Managed Care, July 2010, Volume 16, Issue 7

Algorithms based on managed care pharmacy data can efficiently identify persons at risk for undiagnosed chronic obstructive pulmonary disease.

Objective: To develop and validate a method for identifying persons with undiagnosed chronic obstructive pulmonary disease (COPD) using outpatient pharmacy data.

Study Design: Case-control analysis of managed care administrative data with clinical validation by spirometry and standardized questionnaires.

Methods: Patients with a new diagnosis of COPD were matched to 3 control subjects by age and sex. Outpatient pharmacy utilization for the 2 years prior to the initial diagnosis was captured. Drugs associated with an eventual diagnosis of COPD were identified using conditional logistic regression, and then entered into a predictive algorithm using discriminant function analysis. The algorithm was tested in a second population from the same health plan and externally validated using 2 large multicenter databases. This system was clinically validated by testing 100 individuals identified by the algorithm with spirometry plus health status and respiratory symptoms questionnaires.

Results: COPD patients used significantly more antibiotics, cardiac medications, and respiratory drugs than their matched controls. The final algorithm identified COPD patients with a sensitivity of 60% and specificity of 70%, without the benefit of knowing any patient's smoking history. Of the first 100 persons identified by the algorithm as being at risk and recruited for testing, 25 were proven to have previously undiagnosed COPD.

Conclusions: Pharmacy utilization increases in the years prior to initial COPD diagnosis. Algorithms based on pharmacy utilization can efficiently identify persons at risk for undiagnosed COPD.

(Am J Manag Care. 2010;16(7):505-512)

Chronic obstructive pulmonary disease (COPD) is a leading cause of morbidity and mortality, and increases healthcare utilization years before it is finally diagnosed, but widespread application of spirometry to screen for COPD has not been practical. We describe the development and validation of a system that uses pharmacy data to identify persons at risk for COPD. The benefits of this system are:

  • It can improve the efficiency of screening for undiagnosed COPD.
  • It will facilitate referral of patients for appropriate confirmatory testing.
  • It may help satisfy recent mandates for improvements in the detection and treatment of chronic diseases.

It is estimated that more than 12 million people in the United States have chronic bronchitis or emphysema, but the true prevalence of chronic obstructive pulmonary disease (COPD) is undoubtedly higher because many patients are not aware they have the disease until substantial lung function has been lost.1-3 Analysis of data from the Third National Health and Nutrition Examination Survey, which examined and tested more than 14,000 adults from across the United States, found that 63% of individuals with low lung function were undiagnosed.4 A high proportion of those with undiagnosed airway obstruction were negatively affected by chronic respiratory symptoms and COPD-related physical impairment.5

Even though clinical trials such as the Lung Health Study have demonstrated that early diagnosis and aggressive interventions in COPD can improve long-term outcomes,6-9 screening programs for COPD have not gained widespread acceptance.10,11 Population-level screening with spirometry is costly and of uncertain effectiveness, and was not supported by a recent US Preventive Services Task Force investigation.11,12 The task force estimated that 455 adults between 60 and 69 years of age would have to be screened to defer 1 exacerbation, assuming optimum response to therapy.12 Likewise, spirometry is not recommended as a tool for airflow obstruction screening in asymptomatic patients.13 Thus, efficient mechanisms for identifying persons at risk for undiagnosed COPD are needed.14

Patients with COPD are known to have increased utilization of inpatient and outpatient healthcare services and prescription medications, and this utilization typically increases years before the diagnosis of COPD is made.15,16 We have previously demonstrated that an algorithm based on comprehensive healthcare utilization data can be used to identify persons at risk for COPD.17 However, comprehensive healthcare data are difficult to obtain and integrate, even in managed care systems. Pharmacy databases are more readily available and, if valid for screening purposes, could substantially increase the number of patients available for screening.

The purpose of this study was to determine whether outpatient pharmacy utilization can be used to identify patients at risk for undiagnosed COPD. The primary objective was to develop an algorithm for doing so using administrative pharmacy data. A secondary objective was to clinically validate this system by testing 100 individuals identified by the algorithm as being at risk with spirometry and respiratory symptom questionnaires.

METHODS

Our system was developed using data from the Lovelace Health Plan (LHP), a health maintenance organization (HMO) serving New Mexico. The LHP is the insurance component of Lovelace Health System, an integrated staff and network model medical system that included 4 hospitals and 18 primary care clinics at the time of this study. The research protocol was reviewed and approved by Lovelace Health System’sOffice of Research Administration and the Institutional Review Board of the Ardent Health System. All persons who were recruited to participate in the clinical validation portion of the project underwent an approved patient consent protocol and were required to review and sign an informed-consent form.

Initial Algorithm Development

All LHP members were randomly assigned to 1 of 2 groups, the Algorithm Development Group or the Lovelace Test Group. All patients in the Algorithm Development Group who had 1 inpatient or at least 2 outpatient claims with a COPD diagnosis based on International Classification of Diseases, Ninth Revision(ICD-9) coding (chronic bronchitis [code 491.x], emphysema [code 492.x], or COPD [code 496]) were identified. In a previous chart review of LHP members with COPD, we found that more than 95% of persons identified by this method had at least 2 forms of evidence in their medical records supporting the diagnosis of COPD.18 The database was searched back through January 1, 1990, for the date of the very first COPD diagnosis, which was designated as the initial diagnosis or index date. The COPD patients had to be continuously enrolled for at least 2 years prior to the index date to be included in the Algorithm Development Group. Each COPD patient was matched by age on the index date and sex with up to 3 non-COPD controls, who also were required to have at least 2 years of enrollment in the LHP.

To identify drugs that were associated with COPD in the years prior to diagnosis, we developed conditional logistic regression models with COPD status as the dependent variable and age, sex, and medication use as classified by American Hospital Formularies Service (AHFS) pharmacologic therapeutic classifications as the independent variables. An AHFS category of interest was included in the model only if at least 30 COPD cases had received a drug within that category, so that rarely prescribed drugs would not contribute disproportionate weight to the final model. For a category to be included in the model, we also required utilization of the category to be 2 times higher among COPD cases than among controls, so that commonly prescribed drugs that were only slightly more commonly used among COPD patients would not contribute disproportionately to the model.

Discriminant function analysis (the STEPDISC procedure in SAS; SAS Institute, Inc, Cary, NC) was then used to determine which factors best discriminated between the COPD patients and their matched controls. Variables that were identified in the logistic regression models as being strongly associated with COPD were entered or removed in the discriminant function algorithm according to their respective F scores. Only factors that contributed significantly to the model with an F score of 50 or more were kept.

Initial Algorithm Testing and Validation

The initial algorithm was applied to the Lovelace Test Group. This group was restricted to persons aged 40 years and older who were continuously enrolled in 2003, resulting in a test population of 48,635 patients. For testing purposes, all persons who had 1 or more inpatient or outpatient claims with an ICD-9 code consistent with COPD were classified as COPD patients. The algorithm was applied using the DISCRIM procedure in SAS. The performance of the algorithm was described in terms of its ability to identify known COPD patients (sensitivity), its ability to exclude non-COPD patients (specificity), and how often those identified as being at risk for COPD by the algorithm were confirmed in the database as having the disease (positive predictive value).

Final Algorithm Development and External Validation

To validate the algorithm in other healthcare systems and to further improve its performance, we also applied it to 2 additional managed care databases. The first was the Constella Group MCO-I database, which contains healthcare utilization information from more than 3 million covered lives from HMOs in the southeastern and midwestern United States. The second was the Ingenix Lab-Rx, a United Health Care database, which contains longitudinal, patient-level information on more than 5 million covered lives. In both databases, COPD patients were identified using the same inclusion and exclusion criteria as those used in the Lovelace Algorithm Test Group.

Figure

So that the algorithm developed at Lovelace could be more readily applied to other databases, we rebuilt the entire algorithm using National Drug Code (NDC) identifiers in the Ingenix database. This also had the benefit of finding other drugs that might have been missed by the relatively crude AHFS classes. We again required the COPD cases to have had at least twice the rate of utilization of a particular NDC code as controls (or a simple ratio of 0.66 COPD to control prescription users), and we required that at least 30 COPD cases had at least 1 fill for any individual NDC code for that code to be included in the model. Potential biases caused by these arbitrary criteria were examined in a post hoc sensitivity analysis. The original analyses were repeated with a criterion that the drug utilization rate be 2.27 times higher in cases than in controls (or a simple ratio of 0.75 COPD to control prescription users) for the drug to be included in the algorithm. We also examined the impact on algorithm performance of requiring 100, 200, and 500 patients, rather than 30, to have received a given drug in order for the drug to be included in the model ().

Clinical Validation

The final algorithm was clinically validated using the LHP database. All individuals aged 40 to 90 years who were enrolled in the LHP in 2008 and who did not have a diagnosis of COPD within the past 2 years were eligible for inclusion. The algorithm was applied to the LHP pharmacy database, producing a list of persons at risk for having COPD. Patients were then randomly selected from this roster and sent a letter explaining the purpose of the study and asking them to return a postcard enclosed with the letter giving us permission to contact them by phone. Those who returned the postcard participated in a brief phone interview to review their eligibility. Only those who did not report a previous COPD, chronic bronchitis, or emphysema history and had at least a 10 pack-year exposure to cigarettes were brought in for testing. The testing included spirometry performed using equipment and techniques recommended by the American Thoracic Society, a modified version of the American Thoracic Society Respiratory Disease Questionnaire, the St. George’s Respiratory Questionnaire (SGRQ), and the SF-36.

Statistics

Normally distributed values were compared using the t test, nonnormally distributed values were compared using the Wilcoxon rank-sum test, and proportions were compared using the c2 test. A P value of .05 or less was defined as a statistically significant difference.

RESULTS

Table 1

Table 2

Compared with their age- and sex-matched controls, COPD patients in the Algorithm Development Group had substantially increased utilization of respiratory medications, antibiotics, and cardiovascular agents (). Control group patients were actually more likely to have used other medications, but no medications were negative predictors for COPD. We explored whether the total number of prescription fills was a better discriminant factor between having and not having COPD, as opposed to a simple categoric designation for use of that type of agent during the observation period. We found that among control group patients who had a prescription fill for a respiratory, antibiotic, or cardiovascular agent, the mean number of prescription fills was very similar to that among COPD patients (). Therefore, we did not use the number of prescription fills as a predictive factor in the model.

Table 3

Application of the initial model based on AHFS classifications to the Lovelace Algorithm Development popu-la-tionyielded a sensitivity of 48.3% (). The major limitation to this system is that the sensitivity is limited by the proportion of patients who had a prescription fill for 1 of the drugs of interest. In this population, 51.7% of the COPD patients did not have a prescription fill for either a cardiovascular agent, an antibiotic, or a respiratory agent; thus, they could not be detected by the pharmacy utilization algorithm.

Table 4

The final algorithm based on NDC codes was applied to the Lovelace Test Group and the Constella and Ingenix databases, yielding similar performance in all 3 populations (). The algorithm’s sensitivity was actually higher in the Constella and Ingenix databases due to higher documented medication utilization in their populations. The positive predictive value was highest in the Constella group due to the higher prevalence of COPD in that population. When the population was limited to persons aged 65 years and older, the algorithm’s sensitivity remained the same, but the positive predictive value substantially improved due to the higher prevalence of COPD among older individuals (Table 4).

Examination of changes to the inclusion criteria revealed that increasing the ratio of prescription use for cases to greater than 2.27 times that of controls (simple ratio >0.75) significantly reduced the sensitivity of the algorithm, but improved the positive predictive value (Figure). Increasing the minimum prescription criterion from 30 to 200 or 500 did not have a significant effect on the performance of the algorithm (Figure).

Clinical Validation

Table 5

Of the first 100 persons who consented to clinical testing, 25 were found to have both airflow obstruction and chronic respiratory symptoms (). By Global Initiative for Chronic Obstructive Lung Disease criteria, 7 had mild disease, 15 had moderate disease, and 3 had severe disease.14 Another 20 patients did not have airflow obstruction by the Global Initiative for Chronic Obstructive Lung Disease criteria, but they did have a forced vital capacity reduced below 80% of predicted. There were no significant differences between those confirmed as having COPD and those without airflow obstruction in terms of demographics, smoking history, SGRQ scores, or SF-36 subscales (Table 5).

DISCUSSION

The purpose of this project was to see whether managed care pharmacy data could be used to efficiently identify patients who are at risk for having undiagnosed COPD. As expected, we found that COPD patients had significantly increased pharmacy utilization in the years before they were initially diagnosed with COPD. Using a discriminant function algorithm, we were able to identify a set of factors that are significantly predictive of a COPD diagnosis, and then were able to show the validity of this system in 2 other databases populated with more than 3 million adults. We clinically validated the algorithm by using it to identify and recruit a cohort of LHP patients at risk for undiagnosed COPD. Of the first 100 tested, 25% were proven to have COPD based on their spirometry and history of chronic respiratory symptoms, which is a proportion similar to the algorithm’s positive predictive value in all persons aged 65 years and older (29%). Therefore, we conclude that pharmacy utilization data can be used to efficiently identify a large number of patients who are at risk of having COPD and facilitate their referral for appropriate diagnostic testing.

The excess utilization of antibiotics, respiratory, and cardiovascular drugs in patients with undiagnosed COPD was not unanticipated. In an early case-control examination of healthcare utilization in COPD, we found medication utilization was doubled or more in all of these categories.19 The clinical course of COPD is characterized by frequent respiratory infections20-22 and significant cardiovascular complications.23-26 In fact, accelerated decline in forced expiratory volume in the first second of expiration predicts coronary artery disease in smokers and nonsmokers9 as well as all-cause mortality.27-29 The fact that we were able to identify these undiagnosed COPD cases based on their drug utilization suggests that we are capturing patients who have entered into more complicated stages of the disease and that these are the patients who might benefit from more aggressive intervention.

The major limitation to this approach is that the sensitivity of the algorithm is limited to those who have used the medications of interest. Only half of the COPD patients in our study populations had used the drugs included in the final algorithm during the study year, so the sensitivity is limited to approximately 50%, which is far below the usual sensitivity required of a population screening test. However, we do not propose this method as a substitute for spirometry, the accepted standard diagnostic test for COPD. Rather, we propose use of this algorithm as a tool for efficiently identifying persons who either have COPD or are at high risk of developing COPD, and referring these persons for diagnostic spirometry testing. Our clinical validation study demonstrates that if our database screening approach is then followed by standardized screening questionnaires, the positive predictive value of the combination is sufficient to make this a very cost-effective case identification system.

There are other limitations to consider when applying this approach and algorithm to other populations. The sensitivity of the test was relatively consistent from group to group, as expected, but the positive predictive value of any screening test depends on the prevalence of the disease of interest in the target population. We observed this factor when comparing the LHP group with the Constella and Ingenix populations, and when the results were stratified by age (Table 4). For the algorithm to be efficient, it must be used in appropriate target populations. Also, we did not have smoking history in our databases, so we were not able to examine how information about smoking could have improved the performance of the model. Patients who did not have insurance that includes a pharmacy benefit, or who otherwise could afford prescription medications, would not be detected. The prevalence of comorbidities and use of medications are known to increase with severity of disease, so it is likely that this system is biased toward finding individuals with more advanced COPD. It is not known whether repeated longitudinal application of this system would eventually capture those who did not use a medication of interest during the study year. Finally, COPD patients in managed care systems are likely to have different clinical characteristics and utilization behaviors than patients in other healthcare systems, so this system requires additional validation in other populations.

Our clinical validation study did confirm that the algorithm could select a population with a high prevalence of COPD. Because we clearly notified potential participants that this was a clinical research study, it is possible that selection biases were introduced, but it is difficult to predict how a study population would differ from ones recruited for disease management, public health screening, or quality improvement programs. Due to patient confidentiality and research protection policies, we were not able to systematically reevaluate the study participants who had normal spirometry to see why the algorithm selected them. However, they did have SGRQ scores that were as poor as those of the COPD patients and there were a high proportion who had limited forced vital capacity, so it is likely that their clinical presentations were similar to those of patients with COPD. We did learn from pulmonologists affiliated with the health plan that other serious respiratory complaints including pulmonary fibrosis, pulmonary hypertension, and obstructive sleep apnea were diagnosed in these non-COPD patients as a direct consequence of participating in the study. Therefore, although the prevalence of COPD among participants was 25%, the overall yield of newly diagnosed serious but treatable lung diseases detected by this system was much higher.

We are not aware of any similar studies in the literature that have used pharmacy utilization data to identify persons at risk for COPD. In a prior study we developed an algorithm that used all healthcare utilization (inpatient, outpatient, and pharmacy data) to develop a similar predictive COPD algorithm.17 That algorithm had slightly better sensitivity and specificity; however, comprehensive healthcare utilization can be difficult to obtain or integrate into usable files. Therefore, even though the pharmacy-only algorithm is not quite as effective, it may be a more practical approach even when comprehensive medical claims data are available.

For patients aged 65 years and older, the Medicare Prescription Drug, Improvement, and Modernization Act of 2003 included a directive that programs “identify and monitor” enrollees with chronic conditions.30 Chronic obstructive lung disease is one such chronic condition and is among the 10 leading causes of mortality.31 Although the US Preventive Services Task Force did not support the use of spirometry to screen for COPD,11,12 a targeted system that can efficiently identify patients at risk for undiagnosed COPD is attractive from the perspective of secondary prevention, which serves to identify and treat persons who have preclinical disease or who have already developed risk factors. There are effective interventions for COPD, and the potential health gains from interventions are the greatest early in the course of disease and may improve the long-term prognosis of COPD. Smoking cessation is the cornerstone of early interventions and is highly cost-effective.32,33 The Lung Health Study has shown that an intensive smoking cessation program followed by 5 years of reinforcement leads to a substantial and significant reduction in all-cause mortality in people with mild to moderate airway obstruction.9 Recent studies have demonstrated that aggressive treatment can slow the decline in lung function characteristic of COPD,34,35 and episodes of acute decline in lung function that result in need for steroids or hospitalization can be reduced even among persons with moderate airflow obstruction.36 Large-scale, organized case-finding efforts aimed at early identification and early interventions in COPD are especially important because COPD is increasingly recognized as a disease state that is, in principle, preventable and treatable.37

CONCLUSION

We found that pharmacy utilization data can be used to help identify persons at risk for having undiagnosed COPD, even without any information about smoking status. This approach requires further clinical validation, and we are willing to provide the final algorithm with the specific NDC codes to interested health services researchers. We believe that this system has potential to be an efficient method for identifying patients at risk for COPD and facilitating their referral for appropriate diagnostic testing and therapy.

Author Affiliations: From the Lovelace Clinic Foundation (DWM), Albuquerque, NM; Lovelace Respiratory Research Institute (HP, MHR, JSH, FJF), Albuquerque, NM; and the Global Outcomes Research Group (JPM), Pfizer Pharmaceuticals, Inc, New York, NY. Ms Hurley is now with Hurley Consulting, Placitos, NM.

Funding Source: This research was supported by a research grant from Pfizer Global Pharmaceuticals, Inc.

Author Disclosures: Dr Mapel reports having previously served as a consultant to Pfizer. Mr Petersen, Ms Roberts, and Ms Hurley are em-ployees of a not-for-profit research organization that does externally funded studies in the form of grants and contracts focused on respiratory research. Dr Marton is an employee of Pfizer, the funder of the study and a manufacturer of medication approved for the treatment of COPD, and reports owning stock in the company. Dr Frost reports no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (DWM, MHR, JSH, FJF, JPM); acquisition of data (DWM, HP, JSH, FJF, JPM); analysis and interpretation of data (DWM, HP, MHR, JSH, FJF, JPM); drafting of the manuscript (DWM, MHR, FJF, JPM); critical revision of the manuscript for important intellectual content (DWM, HP, MHR, JSH, JPM); statistical analysis (DWM, MHR, FJF); provision of study materials or patients (DWM, FJF); obtaining funding (DWM, JSH, FJF); administrative, technical, or logistic support (DWM, JSH, FJF, JPM); and supervision (DWM).

Address correspondence to: Douglas W. Mapel, MD, MPH, Lovelace Clinic Foundation, 2309 Renard Pl SE, Ste 103, Albuquerque, NM 87106-4264. E-mail: doug.mapel@lcfresearch.org.

1. CDC. Deaths from chronic obstructive pulmonary disease--United States, 2000-2005. MMWR. 2008;57(45):1229-1232.

2. Mannino DM. COPD: epidemiology, prevalence, morbidity and mortality, and disease heterogeneity. Chest. 2002;121(5 suppl):121S-126S.

3. Mannino DM, Homa DM, Akinbami LJ, Ford ES, Redd SC. Chronic obstructive pulmonary disease surveillance—United States, 1971-2000. MMWR Surveill Summ. 2002;51(6):1-16.

4. Mannino DM, Gagnon RC, Petty TL, Lydick E. Obstructive lung disease and low lung function in adults in the United States: data from the National Health and Nutrition Examination Survey, 1988-1994. Arch Intern Med. 2000;160(11):1683-1689.

5. Coultas DB, Mapel D, Gagnon R, Lydick E. The health impact of undiagnosed airflow obstruction in a national sample of United States adults. Am J Respir Crit Care Med. 2001;164(3):372-377.

6. Anthonisen NR, Connett JE, Kiley JP, et al. Effects of smoking intervention and the use of an inhaled anticholinergic bronchodilator on the rate of decline of FEV1. The Lung Health Study. JAMA. 1994;272(19):1497-1505.

7. Anthonisen NR, Connett JE, Murray RP. Smoking and lung function of Lung Health Study participants after 11 years. Am J Respir Crit Care Med. 2002;166(5):675-679.

8. Murray RP, Connett JE, Rand CS, Pan W, Anthonisen NR. Persistence of the effect of the Lung Health Study (LHS) smoking intervention over eleven years. Prev Med. 2002;35(4):314-319.

9. Anthonisen NR, Skeans MA, Wise RA, et al. The effects of a smoking cessation intervention on 14.5-year mortality: a randomized clinical trial. Ann Intern Med. 2005;142(4):233-239.

10. Anthonisen NR, Woodlrage K, Manfreda J. Use of spirometry and respiratory drugs in Manitobans over 35 years of age with obstructive lung dis-eases. Can Respir J. 2005;2(2):69-74.

11. US Preventive Services Task Force. Screening for chronic obstructive pulmonary disease using spirometry: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med. 2008;148(7):529-534.

12. Lin K, Watkins B, Johnson T, Rodriguez JA, Barton MB; US Preventive Services Task Force. Screening for chronic obstructive pulmonary disease using spirometry: summary of the evidence for the U.S. Preventive Services Task Force. Ann Intern Med. 2008;148(7):535-543.

13. Qaseem A, Snow V, Shekelle P, et al; Clinical Efficacy Assessment Subcommittee of the American College of Physicians. Diagnosis and man-agement of stable chronic obstructive pulmonary disease: a clinical practice guideline from the American College of Physicians. Ann Intern Med. 2007;147(9):633-638.

14. Global Initiative for Chronic Obstructive Lung Disease (GOLD). Global Strategy for Diagnosis, Management and Prevention of COPD. December 2009. http://www.goldcopd.org/Guidelineitem.asp?l1=2&l2=1&intId=2180. Accessed December 30, 2009.

15. Jansson SA, Lindberg A, Ericsson A, et al. Cost differences for COPD with and without physician-diagnosis. COPD. 2005;2(4):427-434.

16. Mapel DW, Robinson SB, Dastani HB, Shah H, Phillips AL, Lydick E. The direct medical costs of undiagnosed chronic obstructive pulmonary dis-ease. Value Health. 2008;11(4):628-636.

17. Mapel DW, Frost FJ, Hurley JS, et al. An algorithm for the identification of undiagnosed COPD cases based on health care utilization data. J Manag Care Pharm. 2006;12(6):457-465.

18. Mapel DW, McMillan GP, Frost FJ, et al. Predicting the costs of managing patients with chronic obstructive pulmonary disease. Respir Med. 2005;99(10):1325-1333.

19. Mapel DW, Hurley JS, Frost FJ, Petersen HV, Picchi MA, Coultas DB. Health care utilization in chronic obstructive pulmonary disease. A case-control study in a health maintenance organization. Arch Intern Med. 2000;160(17):2653-2658.

20. Sethi S, Mallia P, Johnston SL. New paradigms in the pathogene-sis of chronic obstructive pulmonary disease II. Proc Am Thorac Soc. 2009;6(6):532-534.

21. Lange P. Chronic obstructive pulmonary disease and risk of infection. Pneumonol Alergol Pol. 2009;77(3):284-288.

22. Sapey E, Stockley RA. COPD exacerbations, 2: aetiology. Thorax. 2006;61(3):250-258.

23. Rodríguez-Roisin R, Soriano JB. Chronic obstructive pulmonary disease with lung cancer and/or cardiovascular disease. Proc Am Thorac Soc. 2008;5(8):842-847.

24. Halpin D. Mortality in COPD: inevitable or preventable? Insights from the cardiovascular arena. COPD. 2008;5(3):187-200.

25. Sin DD, Anthonisen NR, Soriano JB, Agusti AG. Mortality in COPD: role of comorbidities. Eur Respir J. 2006;28(6):1245-1257.

26. Buch P, Friberg J, Scharling H, Lange P, Prescott E. Reduced lung function and risk of atrial fibrillation in the Copenhagen City Heart Study. Eur Respir J. 2003;21(6):1012-1016.

27. Schünemann HJ, Dorn J, Grant BJ, Winkelstein W Jr, Trevisan M. Pulmonary function is a long-term predictor of mortality in the general population: 29-year follow-up of the Buffalo Health Study. Chest. 2000;118(3):656-664.

28. Bang KM, Gergen PJ, Kramer R, Cohen B. The effect of pulmonary impairment on all-cause mortality in a national cohort. Chest. 1993;103(2):536-540.

29. Hole DJ, Watt GC, Davey-Smith G, Hart CL, Gillis CR, Hawthorne VM. Impaired lung function and mortality risk in men and women: findings from the Renfrew and Paisley prospective population study. BMJ. 1996;313(7059):711-715.

30. Centers for Medicare and Medicaid Services. CMS Legislative Summary. April 2004. Summary of H.R. 1 Medicare Prescription Drug, Improvement, and Modernization Act of 2003, Public Law 108-173. Enacted December 8, 2003. Washington, DC. http://www.cms.hhs.gov/MMAUpdate/downloads/PL108-173summary.pdf. Accessed November 30, 2009.

31. Heron MP, Hoyert DL, Murphy SL, Xu JQ, Kochanek KD, Tejada-Vera B. Deaths: final data for 2006. Natl Vital Stat Rep. 2009;57(14):1-134.

32. Faulkner MA, Lenz TL, Stading JA. Cost-effectiveness of smoking cessation and the implications for COPD. Int J Chron Obstruct Pulmon Dis. 2006;1(3):279-287.

33. Warner KE. Cost effectiveness of smoking-cessation therapies. Interpretation of the evidence—and implications for coverage. Pharmacoeconomics. 1997;11(6):538-549.

34. Celli BR, Thomas NE, Anderson JA, et al. Effect of pharmacotherapy on rate of decline of lung function in COPD: results from the TORCH Study. Am J Respir Crit Care Med. 2008;178(4):332-338.

35. Soriano JB, Sin DD, Zhang X, et al. A pooled analysis of FEV1 decline in COPD patients randomized to inhaled corticosteroids or placebo. Chest. 2007;131(3):682-689.

36. Jenkins CR, Jones PW, Calverley PM, et al. Efficacy of salmeterol/fluticasone propionate by GOLD stage of chronic obstructive pulmonary disease: analysis from the randomized, placebo-controlled TORCH study. Respir Res. 2009;10:59.

37. Celli BR, MacNee W, ERS TF. Standards for the diagnosis and treatment of patients with COPD: a summary of the ATS/ERS position paper [published correction appears in Eur Respir J. 2006;27(1):242]. Eur Respir J. 2004;23(6):932-946.