HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Permissions Request Permissions Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Search for Related Content PubMed PubMed Citation

The pharmacoepidemiology of COPD: Recent advances and methodological discussion

	Introduction

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

Chronic obstructive pulmonary disease (COPD) threatens an emerging public health crisis. The two major drivers for this are the ageing of the world's population and the impressive, if deplorable, success of the multinational tobacco companies at forcing open world markets. Although the World Health Organization estimates that COPD is the sixth most common cause of death worldwide and that by 2020 it will be the third most common, COPD is also an orphan condition that overwhelmingly affects the poor and has been broadly ignored by scientists and by governments. Although the cost-effectiveness of smoking cessation, pulmonary rehabilitation, and long-term oxygen therapy in the later stages of the disease are all high, there is little active treatment that can currently be offered.

Recently, randomised controlled trials, specifically of inhaled corticosteroids (ICS) in patients with COPD have failed to show any modification of the decline in lung function associated with the disease. Analyses have, however, suggested that these drugs may reduce the number of exacerbations, which are related to quality of life. These findings remain controversial, since in some cases they are based on secondary analyses, have inevitably been carried out in selected populations, the results appear to be sensitive to the design of the trials and confidence intervals are often wide. More recently still, evidence has begun to emerge that the use of ICS and possibly long-acting β₂-agonists may also reduce mortality as well as exacerbations in these patients. Exacerbations of COPD, particularly those leading to hospitalisation, are an important cost driver in the healthcare system and may account for three-quarters of the additional costs of a patient with COPD. With the lack of good alternative remedies, such an effect would clearly be important not only to patients but also to the health services.

Much of the new evidence comes, however, not from experimental studies, but from observational studies based on a variety of administrative databases. Although these databases have had an important part in the development of information about drug effects since the 1960s, interpretation of the data contained in them is less straightforward. Most recently, S. Suissa proposed a particular problem with some of the estimates of the effects of ICS on COPD from these sources, the so-called problem of "immortal time". This bias arises when time is allocated to the control (or treatment) group as "incident-free" even though no incident could have occurred during the period because of the definitions used in the study. It was suggested that some of the reports that ICS are effective in reducing mortality in patients with COPD can be accounted for by this bias. Combined with the surprisingly large effects that have been estimated using these methods, this bias has cast doubt on the robustness of these reports' findings.

The issues raised are not only relevant to the current debate on the effectiveness of ICS in COPD but affect a wide range of common problems in health policy. Use of databases is important, not simply as a substitute for randomised clinical trials that would be too large or expensive to undertake. They are also important for a broader understanding of the effectiveness of drug treatments. Very often, trials that are undertaken raise unrealistic hopes for effectiveness in the broader population, either because they are carried out in a highly restricted patient group or because use of the medication in practice is restricted by availability or by patient compliance. These issues need to be explored and one method of doing this is through the use of administrative databases. Robust interpretation of these data is therefore key to adequate policy-making.

The symposium reported here brought together a group of scientists who had experience in this field and many who had used different databases to explore the issue of the effectiveness of ICS and, in most instances, long-acting β₂-agonists to improve the outcomes of patients with COPD. The result was not only an excellent review of what is known of the effectiveness of these drugs, but also an important review of the methods, problems, and potential pitfalls of the uses of administrative databases for pharmacoepidemiological research.

The idea for the symposium arose from a discussion between S. Suissa of McGill University, Montreal, Canada, and J.B. Soriano of GlaxoSmithKline, Greenford, UK, and was financially supported by GlaxoSmithKline. The proceedings will be of interest to those interested in the treatment of COPD and also to any interested in the appropriate use of administrative databases in pharmacoepidemiology.

	Epidemiology of COPD: overview and the US perspective

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
W.M. Vollmer

Summary
The burden of chronic obstructive pulmonary disease (COPD) is increasing in the USA. The prevalence of COPD is now almost equal in males and females. Although objectively measured, airflow limitation is now generally believed to provide the most accurate estimates of disease; international guidelines on the diagnosis and treatment of COPD do not agree on standards for objectively defined measures. Cigarette smoking continues to be the primary risk factor for COPD, however, work-related exposures may be an important contributor to the overall burden of COPD. Approximately 7% of the adult population in the USA has low lung function; 70% of these adults with low lung function have never had a diagnosis of any obstructive lung disease.

Introduction
Chronic obstructive pulmonary disease (COPD) is a leading cause of morbidity and mortality worldwide and results in a substantial economic and social burden to society. In the USA, it is currently the fourth leading cause of death 1, and the World Health Organization (WHO) estimated that there were 2.74 million deaths worldwide from COPD in the year 2000 2. In 1993, the cost of COPD to the US economy was an estimated $23.9 billion, including $14.7 billion in direct medical costs and an additional $9.2 billion in indirect costs 1.

The global burden of COPD has also been increasing and is expected to continue to increase in the coming decades. According to the Global Burden of Disease Study, COPD, ranked twelfth worldwide in 1990 in terms of its impact on disability-adjusted life-yrs, is projected to rank fifth by the year 2020 3, 4. In the USA, death rates for COPD have climbed steadily over the past 40 yrs. While rates have begun to stabilise for males in recent years, they are, if anything, increasing for females 1. A similar pattern of increase is seen if trends in COPD-related healthcare utilisation, rather than mortality, are examined 5. What is even more striking is that these trends, at least in the USA, fly in the face of declining mortality from cardiovascular disease 2.

The two main reasons for these patterns are the increased consumption of cigarettes, especially in developing countries, and among females and the elderly. It has long been known that cigarette smoking is the primary risk factor for COPD 6. According to the third National Health and Nutrition Examination Survey (NHANES III) carried out in the USA between 1988–1994, current cigarette smokers are 3–5-times more likely than never-smokers to have airflow limitation and to report chronic respiratory symptoms 7, 8 (table 1). The WHO estimates that there are now 1.1 billion smokers worldwide 9 and this figure is expected to increase to 1.6 billion by 2025 10.

Prevalence of chronic obstructive pulmonary disease
The WHO Global Burden of Disease Study used data from a variety of published and unpublished sources to estimate the prevalence of COPD in various countries and regions throughout the world 3, 4. While the methodologies varied and some of the estimates are admittedly imprecise, a few general patterns emerged. First, the prevalence tended to be highest in countries where cigarette smoking is common, and the prevalence generally tends to be higher in males than in females. Recently, a different pattern has emerged in the USA with prevalence of COPD being almost equal in males and females 1, 5, 7. This likely reflects the increase in smoking among females that has occurred in the USA since World War II.

Estimates of the prevalence of COPD will depend on the definition and criteria used to define it 12. Estimates based on self-report of respiratory symptoms are very nonspecific and likely result in overestimates of disease, while estimates based on physician diagnosis will tend to lack sensitivity since mild disease is often undiagnosed. Objectively measured airflow limitation is now generally believed to provide the most accurate estimates of disease, but, even here there is a lack of consensus, since the American Thoracic Society, European Respiratory Society, and the Global Initiative for Chronic Obstructive Lung Disease (GOLD) definitions all differ. Celli et al. 13 recently compared a number of objectively defined measures and reported that prevalence estimates sometimes varied by as much as 100%. They recommended the GOLD clinical definition of a forced expiratory volume in one second (FEV₁)/forced vital capacity (FVC) ratio <0.70 on the basis of its simplicity and accuracy 12. While it may be simple, its accuracy is still in question. Hardie et al. 14 recently reported on 71 asymptomatic, nonsmoking adults, selected as a random sample of adults of >70 yrs of age. Thirty-five per cent had a prebronchodilator ratio <0.70 and this increased to 50% among those of >80 yrs. One-third of those >80 yrs actually met GOLD stage II criteria despite having no history of smoking or any apparent symptoms.

Some of the best prevalence data for COPD come from the USA NHANES III study. This large probability sample of the US population included a subsample of >16,000 adults for whom pulmonary function tests, a complete medical history, and self-reported diagnostic data were available. COPD (defined as the presence of airflow limitation) was estimated to be present in 24% of current smokers, 13% of exsmokers, and 7% of never-smokers (table 1). These estimates were very similar for males and females 7, 8. The prevalence of physician-diagnosed COPD, defined as chronic bronchitis or emphysema, increases steadily in both males and females through the mid-to-late adult years and, in this latter age range, tends to be greater for males than for females (table 2).

The NHANES III data can also be used to estimate the contribution of occupation to COPD 16. Using the GOLD stage II criterion of an FEV₁/FVC ratio <0.70 and an FEV₁ <80% of predicted, the fraction of COPD that may be attributable to work among individuals aged 30–75 yrs has been estimated as 19.2% overall and 31.2% among never-smokers.

Conclusion
Overall, the NHANES data suggest that 7% of the adult population in the USA have low lung function and that this is closely related to cigarette consumption and increasing age. These data also suggest that 70% of adults with low lung function have never had a diagnosis of any obstructive lung disease and that work-related exposures may be an important contributor to the overall burden of COPD.

	Epidemiology of COPD: a European perspective

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
G. Viegi

Summary
Mortality rates from chronic obstructive pulmonary disease (COPD) vary across Europe. However, the continuing elevated prevalence rates of current smoking among males and the increasing trend observed in females in the last decades predicts an increase of COPD mortality in many countries in the coming years. The real prevalence of COPD within a given population may vary widely depending upon the tools used for its identification: reported respiratory symptoms, medical diagnosis and abnormal lung function. Even when the diagnosis is based on an objective tool like spirometry, largely variable prevalence rates are found within the same population. In view of the different criteria endorsed by different scientific societies, it is clear that further research is needed to reach a standardised criterion for airways obstruction. Patients underestimate their own morbidity and may therefore be undertreated. The cost of COPD, largely driven by exacerbations, is expected to increase considerably in the future, reflecting the previous smoking habits of an ageing population. The impressive prevalence in current smokers <45 yrs of age in most countries highlights the need to improve the quality of prevention; early detection and screening programmes may be useful in this population of smokers.

Introduction
Chronic obstructive pulmonary disease (COPD) is a very important cause of mortality and morbidity in Europe 17. Although COPD and asthma are considered different entities with respect to pathophysiological and cellular conditions 18, it is well recognised that chronic persistent asthma may have the feature of irreversible airflow obstruction, thus being encompassed within the term COPD, as clearly shown in the nonproportional Venn diagram published in the 1995 American Thoracic Society (ATS) guidelines 19. Further, many available statistics on COPD mortality and morbidity data are based upon the combination of chronic bronchitis, emphysema and asthma; codes 490–493 of the International Classification of Diseases, ninth revision (ICD-9) 20, or codes J40–47 of ICD-10 21. Sometimes, code 496 ICD-9 is also used to include within COPD the airflow obstruction not classified elsewhere.

Mortality
On a worldwide scale, according to the estimates by Murray and Lopez 22, if tobacco epidemics will not change their trends, in the interval 1990–2020, COPD will jump from the sixth to the third rank among the leading causes of death. At the European level, huge differences in the mortality rates have already been depicted in the Appendix "Epidemiology" of the European Respiratory Society (ERS) Consensus Statement on COPD 17. It has been reported that mortality rates for COPD vary more than five-fold among the European countries and that they increase greatly with age and are considerably lower in females 23. More recent data encompassing the period 1993–1997, provided by the World Health Organization 24, show a wide range in the mortality rates from >70 in Hungary to <10 in Greece (per 100,000 population of 35–74 yrs of age). Analogous figures for females range between 40 in Scotland to <5 in Greece.

The comparison of mortality rates among different countries depends upon the relative weight of relevant risk factors in the different populations, but it is also linked to technical factors such as the use of different reference populations for standardising the rates, and the use of different codes for reporting the same disease (e.g. code 491 ICD-9 is used more in south Europe, code 496 ICD-9 in the north).

Current trends in COPD in the UK 25 differ from those in many other countries, because in the past COPD was much more common than in other countries undergoing a smoking epidemic at the same time, and peak cigarette consumption in males and females occurred >25 yrs ago. Male mortality from COPD has been falling for 30 yrs, while female mortality has risen steadily during the same period. A strong socioeconomic gradient in morbidity and mortality persists.

In Italy, of 36,834 deaths that occurred in 1998 for respiratory diseases 26, approximately one-half have been caused by COPD (codes 490-493 ICD-9). The number of deaths stratified by sex, standardised mortality rates per 100,000 people (with the world population as reference), and male/female ratios for all respiratory diseases and COPD are reported in table 4.

Among the factors that have been related to an increased risk of mortality (or of lower survival) for COPD in the general population, epidemiological data from Denmark have stressed the role of forced expiratory volume in one second (FEV₁) and chronic mucus hypersecretion. For subjects with an FEV₁ 40% at baseline, 5-yr survival after subsequent hospitalisation was only 28% 29. Within the framework of a 10–12-yr follow-up, chronic mucus hypersecretion was found to be a significant predictor (relative risk (RR) 3.5) of COPD-related death with pulmonary infection implicated 30.

Morbidity
Although COPD represents one of the main health issues of the present and the near future worldwide, some of its features are still undefined from a social and economic point of view. The real prevalence of COPD within a given population may vary widely depending upon the tools used for its identification: reported respiratory symptoms, medical diagnosis and abnormal lung function. Community surveys in countries of both northern and southern Europe 23 indicate that 4–6% of the adult population suffer from clinically relevant COPD. The prevalence increases greatly with age, however, two-thirds have only a mild reduction in lung function.

According to the Italian National Statistics Agency multipurpose survey on households, performed in 1999–2000, 4.4% of the Italian population (4.8% males, 3.9% females) suffered from chronic bronchitis and/or emphysema and/or respiratory failure. The highest rates have been found in the elderly >64 yrs of age (total 14.1%, males 18.3%, females 11.2%) 31.

Another source of routinely collected statistics is the hospital discharge standard form. Data pertaining to the year 2000 in Italy show that 20.6% of discharges for respiratory diseases are caused by Diagnosed Related Group 88 - COPD (126,927 cases). Total hospital days were 1,159,995 with an average length of stay of 9.4 days 32.

A dynamic multistate life table model was used to compute projections for the Netherlands 33. Changes in the size and composition of the population caused COPD prevalence to increase from 21 per 1,000 in 1994 to 33 per 1,000 in 2015 for males, and from 10 per 1,000 to 23 per 1,000 for females. Changes in smoking behaviour reduce the projected prevalence to 29 per 1,000 for males, but increase it to 25 per 1,000 for females. Total years of life lost increase by >60%, and disability-adjusted life-yrs lost increase by 75%. Costs rise 90%; smokers cause 90% of these costs. The model demonstrates the unavoidable increase in the burden of COPD, an increase that is larger for females than for males. The major causes of this increase are past smoking behaviour and the ageing of the population. According to the authors, changes in smoking behaviour will have only a small effect in the near future.

It is interesting to point out that among industrialised countries, Japan shows extremely low prevalence rates of COPD; in 1999, it was estimated that 212,000 people (139,000 males) were affected by COPD with a prevalence of 0.17% in the general population 28. One of the reasons that account for these values is the long delay in the uptake of tobacco smoking in Japan for cultural and socioeconomic reasons after the second World War.

The importance of sex, ageing and tobacco smoking in the development of COPD has been examined, in Italy, by Viegi et al. 34, using data collected, through questionnaire, in two longitudinal surveys carried out in the rural area of Po Delta (northern Italy) and in the urban area of Pisa-Cascina between 1980–1993. Data on prevalence rates of chronic bronchitis and emphysema (medical diagnosis) and of some respiratory symptoms, stratified by sex and smoking habit were obtained. The prevalence rate of chronic bronchitis was lower than that of chronic cough and phlegm, symptoms on which the diagnosis of chronic bronchitis is based 35. It confirms an underestimate of the frequency of such disease, when only medical diagnoses are considered 36.

The underestimate of COPD prevalence, possibly 25–50% and higher, has been found by several investigators 17, 37, 38. Two cross-sectional studies of respiratory symptoms and diseases in two population samples (5,700 subjects aged 35–36, 50–51 and 65–66 yrs) living in the same areas in northern Sweden were performed 6 yrs apart through a postal questionnaire 39. Lung function measurements were performed in stratified samples. Of the subjects diagnosed with chronic bronchitis, only 25% in 1986 and 23% in 1992 had been diagnosed prior to the study as having chronic bronchitis, emphysema or COPD. Chronic airflow limitation, defined as FEV₁/forced vital capacity (FVC) <70% and FEV₁ <80% of predicted value, was found in 171 subjects in 1986–1987 (12% of the examined subjects), and 166 subjects in 1993–1994 (11%). In 1986–1987, 26% of the subjects with chronic airflow limitation had been diagnosed as having chronic bronchitis or emphysema prior to the survey, while a diagnosis of asthma, chronic bronchitis or emphysema, or use of asthma medicines, was found in 58%. The corresponding figures in 1993–1994 were 31% and 63%, respectively.

Large differences in the prevalence of physician-diagnosed chronic bronchitis have been found in a postal survey conducted in 1996 in three countries 40: 10.6% in Tallinn, Estonia, 3.4% in Helsinki, Finland and 3.0% in Stockholm, Sweden. A representative sample of 14,076 French individuals of 25 yrs completed a self-administered questionnaire 41. Prevalence rates of chronic bronchitis and of chronic cough and/or expectoration were 4.1% and 11.7%, respectively; in individuals with comorbidity, these figures were 10.4% and 24.4%, respectively. Smoking was associated with an increased frequency of chronic bronchitis. In subjects with chronic bronchitis, 44.6% had spirometry or peak expiratory flow measurements, 24% were diagnosed as having chronic bronchitis and 7.2% received care.

Even when the diagnosis is based on an objective tool like spirometry, largely variable prevalence rates are found within the same population in view of the different criteria endorsed by different scientific societies. For instance, Viegi et al. 42 have shown in adults of 25 yrs of age (n=1,727 in 1988–1991) that the prevalence rates of airflow obstruction range from 11% with the ERS criteria 17 to 18% with the "clinical" criteria (later labelled as the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria Stage I) 43 to 40.4% with the 1986 ATS criterion 44. Corresponding figures for subjects aged 25–45 yrs and subjects of 46 yrs of age were as follows: ERS 10.8 and 12.2%, clinical 9.9 and 28.8%, and ATS 27 and 57%, respectively. When considering only moderate/severe obstruction, the rates were as follows: ERS 0.4 and 3.6%, clinical 0.3 and 4.4%, and ATS 0.5 and 5.2%, respectively. The trend was confirmed after stratifying for smoking habit and the presence/absence of respiratory symptoms/diseases. The highest specificity and predictive value for any respiratory symptom/disease was shown by the ERS, and the lowest was shown by the ATS criterion, while the reverse was true for sensitivity; overall accuracy was slightly lower for the ATS criterion. Multiple logistic regression models indicated a higher number of significant associations with known risk factors for airways obstruction according to clinical and ATS criteria than ERS criterion. The authors concluded that further research was needed to reach a standardised and epidemiologically consistent criterion for airways obstruction.

Probably, such a goal has not yet been achieved, even after the introduction of the GOLD criteria. Its ability to provide information of prognostic value in COPD patients has been questioned by Vestbo and Lange 45. Its applicability to the whole population regardless of age, has been criticised by Hardie et al. 14.

An interesting experience on early detection of COPD or asthma in a random sample from the general population aged 25–70 yrs has been carried out in 10 general practices located in the eastern part of the Netherlands within the framework of the Detection, Intervention, and Monitoring Programme of COPD and Asthma (DIMCA) 46. There was a two-stage protocol involving screening and a subsequent 2-yr monitoring of all subjects with positive results of screening. All known COPD and asthma patients were excluded. Of those eligible, 1,155 subjects (66%) participated in the screening stage, and 384 subjects (64% of those with positive screening results) participated in the monitoring stage. During the second stage, 252 subjects were detected with objective signs of COPD or asthma at an early stage. Smoking status as a screening criterion was neither sensitive nor specific. By extrapolation, 7.7% of the general population showed persistently reduced lung function or increased bronchial hyperresponsiveness (BHR). Another 12.5% of the general population showed a rapid decline in lung function (>80 mL·yr^–1) in combination with signs of BHR, and a further 19.4% of the general population showed mild objective signs of COPD or asthma.

A promising approach in early detection of COPD in high-risk populations using spirometric screening also comes from a Polish experience 47 on 11,027 smokers of >39 yrs with a smoking history of >10 pack-yrs. Spirometric signs of airway obstruction were found in 24.3% of the screened subjects: mild 9.5%, moderate 9.6%, and severe obstruction 5.2%. In addition, the same research group 48 was able to demonstrate in a subgroup of screened smokers that, after a minimal antismoking intervention, those with abnormal lung function had a nearly doubled quitting rate at 1 yr compared with those with normal spirometry.

An assessment of the international variation in the prevalence of chronic bronchitis and its main risk factor, smoking, has been performed in 35 centres from 16 countries on 17,966 subjects (20–44 yrs of age), randomly selected from the general population, in the frame of the European Community Respiratory Health Survey 49. The median prevalence of chronic bronchitis was 2.6%, with wide variations across countries (0.7–9.7%). The prevalence of current smokers ranged 20.1–56.9%, with a median value of 40%. Current smoking was the major risk factor for chronic bronchitis, especially in males. Only 30% of the geographical variability in prevalence could be explained by differences in smoking habits, suggesting that other environmental and/or genetic factors may play an important role.

Recently, the first international survey estimating the burden of COPD in the general population was published 50. The Confronting COPD International Survey aimed to quantify morbidity and burden in COPD subjects in 2000. From a total of 201,921 households screened by random-digit dialling in the USA, Canada, France, Italy, Germany, the Netherlands, Spain and the UK, 3,265 subjects with a diagnosis of COPD, chronic bronchitis or emphysema, or with symptoms of chronic bronchitis, were identified. The mean age of the subjects was 63.3 yrs and 44.2% were female. Subjects with COPD in North America and Europe appeared to underestimate their morbidity, as shown by the high proportion of subjects with limitations to their basic daily life activities, frequent work loss (45.3% of COPD subjects <65 yrs of age reported work loss in the past year) and frequent use of health services (13.8% of subjects required emergency care in the last year), and thus may be undertreated. There was a significant disparity between subjects' perception of disease severity and the degree of severity indicated by an objective breathlessness scale. Of those with the most severe breathlessness (too breathless to leave the house), 35.8% described their condition as mild or moderate, as did 60.3% of those with the next most severe degree of breathlessness (breathless after walking a few minutes on level ground).

Some relevant issues in the natural history
The Danish research group of the Copenhagen City Heart Study has elucidated the relationship of mucus hypersecretion and COPD morbidity. Among males, chronic mucus hypersecretion was associated with an excess FEV₁ decline of 22.8 mL·yr^–1 (95% confidence interval (CI) 8.2–37.4) and with subsequent hospitalisation due to COPD after adjusting for age and smoking (RR 2.4, 95% CI 1.3–4.5) 51. Socioeconomic factors operating from early in life affect the adult risk of developing COPD independently of smoking in both females and males 52.

In a follow-up of 8,955 adults, elevated plasma fibrinogen was associated with reduced FEV₁ and an increased risk of COPD hospitalisation rates 53. In a 21-yr follow-up on 9,187 adults, ₁-antitrypsin MZ heterozygotes had a slightly greater rate of decrease in FEV₁ and were modestly over-represented among persons with airway obstruction and COPD. In the population at large, MZ heterozygosity may account for a fraction of COPD cases (2%), similar to the percentage of persons with COPD who have the severe but rare ZZ genotype 54. Among subjects with severe disease (FEV₁/FVC <0.7) in a 14-yr follow-up of COPD patients, there was a significant risk ratio modification between effect of baseline body mass index (BMI) and weight change 55. In the normal-to-underweight (BMI <25), best survival was seen in those who gained weight, whereas for the overweight and obese (BMI 25), best survival was seen in stable weight. A high proportion of subjects with COPD experienced a significant weight loss, which was associated with increased mortality.

The Norwegian research group from Bergen tested the comparability of telephone and postal survey questionnaires for respiratory symptoms and risk factors 56. Furthermore, it demonstrated the use of biomarkers, like ₁-antitrypsin and calprotectin, and lung function tests different from FEV₁, like diffusing capacity 57. This group has provided a major contribution in understanding the role of occupational exposure on the development of COPD 58. In a 25-yr follow-up on 951 subjects from a random sample of 1,933 males aged 22–54 yrs, the decline in FEV₁ was associated with age, body height and smoking. Accelerated decline in FEV₁ was observed in subjects exposed to sulphur dioxide gas and metal fumes at work. The adjusted decline in FEV₁ increased progressively in subjects exposed to increasing numbers of occupational agents.

Among the environmental risk factors, an increasing body of evidence is accumulating on air pollution, especially urban air pollution 59 whose acute increases (mainly the particulate matter) have been related to short-term health effects (i.e. mortality and hospital admissions) in patients suffering from COPD. Beyond the acute effect, chronic exposure to air pollution seems related to lung function impairment and development of COPD. The few cross-sectional studies performed have shown an increase of self-reported diagnosis of chronic bronchitis and emphysema, breathlessness, and mucus hypersecretion and lower levels of lung function in the more polluted areas. The only cohort study in adults showed a faster decline of lung function. The great importance for public health knowledge of air pollution is due to its ubiquitous nature that renders the whole general population at risk.

Costs
In the above-mentioned DIMCA study 46, the costs involved in detection (lung function assessments, organisation, transportation, and patient time) were calculated for three different scenarios, as follows: 1) the detection of subjects with persistently decreased lung function or an increased level of BHR during 6 months of monitoring; 2) scenario 1 plus the detection of subjects with a rapid decline in lung function with signs of BHR during 12 months of monitoring; and 3) scenario 2 plus the detection of subjects with a moderate increase in the decline in lung function or signs of BHR during 24 months of monitoring. The average costs per detected case varied from US $953 (scenario 1) to US $469 (scenario 3). Thus, detection of COPD or asthma at an early stage by means of a two-stage protocol seems feasible at relatively little expense in comparison with other mass screening programmes.

Further, in a prospective, randomised consent trial 60, the utilisation of healthcare resources and cost were ascertained in two groups: a screened group (n=416) and a control group (n=462). During an average follow-up of 3.6 yrs, there were no significant differences in healthcare resource utilisation and cost between the screened subjects and the controls. Resource utilisation before screening was not significantly different from resource utilisation after screening. Within the screened group, positive subjects with signs or symptoms of obstructive airway disease consulted their general practitioners 3.7-times more frequently for respiratory reasons than negative subjects. As expected, the total healthcare cost due to respiratory disease in screen-positive subjects was 6.4-times higher. Overall, there were no indications that screening for obstructive airway disease led to increased cost, above that of average care.

The burden of asthma and COPD on the general population is considerable in the Netherlands 61. The main cost element of asthma is medication, whereas hospitalisation accounts for the largest proportion of costs for COPD. Consequently, the annual cost per patient of managing COPD is almost three-times as high as that for asthma. Together, the two respiratory conditions cost the Dutch healthcare system US $346 million for direct medical costs in 1993, amounting to 1.3% of the total healthcare budget. The burden of COPD is expected to increase considerably in the future, reflecting the previous smoking habits of an ageing population.

Within the framework of the Italian National Healthcare System, a cost-of-illness analysis of three pathologies affecting the lower respiratory tract (community-acquired pneumonia (CAP), COPD and asthma) was conducted in a large region of north-east Italy, Triveneto, between 1999–2000 62. Patients of both sexes 14 yrs of age were randomly selected from 28 centres of pneumology. Consumption of medical resources used during the follow-up period was valued according to market prices and published official tariffs. A total number of 1,068 patients (596 males and 458 females) were selected: 42.5% were affected by asthma, 46.3% by COPD and 11.2% by CAP. Mean cost per patient per year for patients affected by asthma and COPD ranged 608–2,457 and from 1,500–3,912, respectively, depending on illness severity. The mean cost per episode of CAP was 1,586.

Exacerbations are the key drivers in the costs of COPD in Sweden 63. Among 202 subjects with COPD (defined according to the British Thoracic Society and ERS criteria), at least one exacerbation was reported by 61 subjects, who were then interviewed regarding resource use for these events. The average healthcare costs per exacerbation were Swedish krona (SEK) 120 (95% CI 39–246), SEK 354 (252–475), SEK 2,111 (1,673–2,612) and SEK 21,852 (14,436–29,825) for mild, mild/moderate, moderate and severe exacerbations, respectively. Exacerbations account for 35–45% of the total per capita healthcare costs for COPD.

Conclusion
Two comprehensive reviews summarising the European perspective on COPD epidemiology were published in 2001 64, 65. Chronic bronchitis is too infrequently diagnosed, investigated, and cared for. It is a substantial health problem even in young adults. The impressive prevalence in current smokers of <45 yrs of age in most countries highlights the need to improve the quality of prevention. Even if the current decline in the prevalence of smoking continues in Europe, in the near future there will be an increase in the prevalence of COPD (with the increase probably higher among females than males), largely due to the ageing of the population. Keeping these statistics in mind, decision-makers allocating funds to healthcare services need to consider that the prevention of moderate-to-severe exacerbations could be very cost-effective and improve quality of life. There is also a need for intervention studies that aim to avoid weight loss in normal-to-underweight COPD patients.

	The burden of illness and economic evaluation for COPD

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
S.D. Sullivan

Summary
This article reviews the important factors to consider in the design of economic evaluations or cost-effectiveness analyses (CEAs) of chronic obstructive pulmonary disease (COPD) treatments. The relevant costs associated with COPD can be divided into direct (direct medical and direct nonmedical) and indirect (programme and productivity) costs. The differences between the human-capital and friction-cost approaches to evaluate the impact of productivity loss are discussed. Since the primary cost-driver for COPD is hospital care for exacerbations, this may be the major outcome measure of interest in COPD economic evaluation studies. Robust CEA evaluations that take into account all of these factors will aid decision-makers in evaluating COPD therapies.

Introduction
Given the rising prevalence of chronic obstructive pulmonary disease (COPD) worldwide, it is urgent that its economic burden is understood and that more robust evaluations of healthcare interventions are designed to reduce its incidence and impact. Studies designed for making decisions and policy must apply robust methods and report results in a standardised fashion.

Economic evaluations are often known less precisely as cost-effectiveness analyses (CEAs). Although these terms have subtle differences, for purposes of this article, and to avoid confusion, "CEA" will refer to studies designed to evaluate the incremental impact of a particular COPD therapy or programme (usually new) versus the conventional approach. In recent years, standardised methods for conducting and reporting these studies have been embraced 66–70.

Capturing relevant costs related to chronic obstructive pulmonary disease and its treatment
Economic studies of COPD must include all relevant costs associated with the illness. Such costs should include not only the intervention of interest (e.g. inhaled bronchodilator therapy), but also all components associated with therapy, such as nebuliser equipment. Such costs can be divided into direct (direct medical and direct nonmedical) and indirect (programme and productivity) costs.

Programme costs refer to costs associated with building the infrastructure needed to deliver the technology. Many studies fail to take into account programme costs when evaluating interventions. For example, an evaluation of a new intensive smoking cessation clinic should include clinic costs (such as rent for office space and staff costs) amortised across the patient group as well as the cost of associated therapies such as nicotine patches or buspirone. Direct medical costs include all medical goods and services used to treat the illness. Usually, these costs are the easiest ones to identify and are thus part of most economic studies.

Direct nonmedical costs include items related to care not directly linked to the healthcare system. Comprehensive evaluations of nonmedical costs are needed for COPD. Such costs can include hired caregiver expenses, costs to the family, lost wages of family caregivers, expenses associated with modifications to living facilities, and transportation and parking costs for patients visiting their physicians. As these costs usually are not reimbursed by health insurance and are difficult to track, they are often excluded from economic studies. As a result, almost no information exists on the value of direct nonmedical costs in COPD. This may be an important oversight, particularly for developing countries. For example, transportation costs may be one of the largest expenses for those who have to travel from remote areas to receive care.

Productivity costs refer to the value of lost wages resulting from illness and from seeking treatment. They are particularly difficult to estimate and are usually excluded from economic evaluations. Nevertheless, productivity is reduced by sporadic absences, visits to healthcare providers and premature mortality. Even more so than direct nonmedical costs, this may be a particularly important omission where COPD is concerned, especially for burden-of-illness studies in developing countries. The value of permanent work loss is particularly important for diseases with high rates of premature mortality such as COPD. As productivity in COPD is potentially important, the two major approaches to valuing productivity, human-capital and friction-cost, are reviewed here in some detail.

Capital and friction-cost approaches to valuing productivity costs
The traditional approach to evaluating the impact of productivity loss caused by illness is the human-capital approach. This term derives from the observation that a person's earnings over a lifetime reflect an investment in that individual through education, on-the-job training, and work experience 71, 72. As these investments influence that individual's value to the economy, productivity loss usually is valued using market wage rates. For those not working for a wage (e.g. homemakers), wages are valued at those that replacement workers would earn for their specific services.

The friction-cost method differs from the human-capital approach in that it allows for the replacement of an absent worker by other workers or by those in the unemployed pool. The friction-cost method values productivity as the loss incurred during the time between a person's absence from work or termination of employment and the time at which another worker fills that position 73–75. The time required for worker replacement is called the "friction period."

Unfortunately for researchers, there is no general agreement on whether the human-capital or friction-cost method is more valid for measuring the productivity costs of illness 76–79. Further complicating the matter is that the estimate will vary greatly depending on which method is applied. For example, in a study of schizophrenia's impact on productivity, the human-capital and friction-cost methods resulted in an 85-fold difference in the estimate of productivity cost 80.

Time horizon
All downstream effects related to treatment for COPD should be included in cost studies. These costs should be tracked or modelled during the time that the intervention is expected to affect the individual or group. Often in the case of COPD, this time horizon equals a person's lifetime.

Key factors influencing cost
The call to "capture all costs" must be tempered by the reality of the study design's limitations and the budget available for conducting economic analyses. As there is no comparison group in burden-of-illness evaluations, comprehensiveness is more important for these studies than for CEAs of healthcare programmes. For CEAs, the intervention of interest is likely to have a large effect on some aspects of healthcare utilisation (e.g. hospital days), but little on other aspects. Since CEAs are an incremental form of analysis, it is only necessary to measure healthcare items that are expected to vary between the intervention and control groups. Of course, it is difficult to predict beforehand what will vary as a result of the intervention. Studies have shown that the primary "cost-driver" for COPD is hospital care for exacerbations, accounting for 70% of all direct medical costs in developed countries for this disease 81–86. Thus, if the intervention is expected to influence hospital care significantly, this may be the only item necessary to measure accurately. Other items with more subtle effects (e.g. office visits, medications) will probably require a more comprehensive analysis.

Methodological issues for cost-effectiveness studies
Cost-effectiveness studies are now common in medicine and have been applied to several therapies for COPD. Nevertheless, outside of smoking cessation programmes 87–93, few high quality CEAs exist for widely used COPD therapies 94. In fairness to researchers, few therapeutic breakthroughs for this disease have occurred, and many traditional therapies are perceived as only modestly effective (it could be argued that minimally effective therapies are not cost-effective and should not be applied). More recently, however, several therapies have begun development for persons with COPD 95. Ever tightening health budgets will force payers to scrutinise the value for expenditures of these new therapies more closely. In this context, it is an opportune time to review the important issues involved for conducting robust CEA evaluations of treatments.

	Pharmacotherapy of COPD

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
L.M. Fabbri

Summary
There are multiple goals to manage in chronic obstructive pulmonary disease (COPD) patients. The benefits of current pharmacological treatments for COPD are discussed here. The use of theophyllines remains somewhat controversial in the management of stable COPD, while the use of long-acting bronchodilators (such as salmeterol, formoterol, or tiotropium) alone or in combination with short-acting bronchodilators or theophylline are effective maintenance treatments for COPD. Treatment with inhaled corticosteroids (ICS) reduces symptoms and the frequency of exacerbations and improves the quality of life, but does not influence the long-term decline of forced expiratory volume in one second. However, combination therapy with ICS and long-acting β₂-agonists improves lung function and symptoms and reduces rescue medication use and the frequency of moderate and/or severe COPD exacerbations. The Global Initiative for Chronic Obstrutive Lung Disease guidelines have recently been revised to recommend maintenance therapy with inhaled long-acting bronchodilators starting from moderate (stage II) COPD, and combination therapy with ICS starting from severe (stage III) COPD to prevent exacerbations.

Introduction
The main objectives of chronic obstructive pulmonary disease (COPD) management are the prevention of disease progression, the relief of symptoms, the improvement of exercise tolerance and health status, the prevention and treatment of exacerbations and/or complications, and the reduction of mortality and of side-effects from treatment 12.

The long-term therapy of moderate and severe COPD consists of pharmacological treatment, such as the regular use of bronchodilators, and of nonpharmacological treatment, such as rehabilitation and/or long-term oxygen in the presence of respiratory failure. The most recent COPD guidelines from the Global Initiative for Chronic Obstructive Lung Disease (GOLD) recognise, as part of the definition of the condition, that there is "an abnormal inflammatory response" in the lung to noxious gases or particles 12. This suggests the need for effective anti-inflammatory treatment in COPD. However, inhaled corticosteroids (ICS) have not been shown to have a consistent anti-inflammatory effect in patients with COPD, and thus, based on the results of clinical trials, treatment with ICS is recommended in some, but not all COPD patients, and, in particular, in patients with severe and very severe (stages III and IV, respectively) COPD and repeated exacerbations.

Bronchodilator medications, such as short-acting and long-acting β₂-agonists, anticholinergics, and theophylline, are central to the symptomatic management of COPD. Long-acting inhaled β₂-agonists, such as salmeterol and formoterol, have a duration of action of up to 12 h and significantly improve symptoms, exercise capacity, and health status in patients with COPD. The use of salmeterol (a long-acting β₂-agonist) in COPD patients has been shown to significantly reduce dyspnoea and to improve forced expiratory volume in one second (FEV₁) values after long-term treatment 96, 97, and to reduce dynamic hyperinflation 97. Formoterol, both a short- and long-acting β₂-agonist, demonstrates better spirometric efficacy than either ipratropium 98 or theophylline alone 99, and its efficacy improves when administered in combination with ipratropium 100. A new long-acting once-daily anticholinergic agent, tiotropium, produces benefits of equivalent or greater size than salmeterol or formoterol 101, and is likely to be a useful addition to treatment for COPD. Thus, tiotropium has been shown to provide significant bronchodilation in terms of FEV₁ response, and reduces dyspnoea and frequency of COPD exacerbations 102. Theophyllines remain somewhat controversial in the management of stable COPD. They have a slow onset of action and are used as a maintenance treatment rather than for rapid relief of symptoms.

Combination treatment with formoterol plus ipratropium provides better improvement of pulmonary function and a greater reduction in symptoms 103; similarly, combination treatment with salmeterol plus theophylline provides significantly greater improvements in pulmonary function, significantly greater reductions in symptoms, dyspnoea, and albuterol use, and significantly fewer COPD exacerbations 104. Taken together, these two studies suggest that combination therapy with long-acting bronchodilators with different mechanisms of action may, in fact, produce additive effects.

Whether ICS have an anti-inflammatory effect in patients with COPD remains controversial 105, 106. It is clear that these drugs do not modify the natural history of COPD, as measured by the rate of decline in FEV₁ 107–110. Data from studies on long-term effects of ICS provide evidence that regular treatment with ICS is only appropriate for symptomatic patients with severe to very severe COPD with an FEV₁ <50% predicted (stage III: severe COPD, and stage IV: very severe COPD), and for repeated exacerbations requiring treatment with antibiotics or oral corticosteroids 111, 112. These studies have shown that long-term treatment with ICS reduces symptoms and the frequency of exacerbations and improves the quality of life 107–110, 112.

The recent randomised controlled trials examining the benefits of combining ICS and inhaled long-acting β₂-agonists in the treatment of COPD have shown interesting results. The combination of fluticasone propionate and salmeterol improves lung function and symptoms, reduces the severity of dyspnoea and rescue bronchodilator use 111, 113, and reduces the frequency of moderate and/or severe COPD exacerbations 111. The combination of budesonide and formoterol reduces the mean number of severe exacerbations, improves FEV₁ and peak expiratory flow values, and reduces all symptom scores and the use of rescue β₂-agonists 114.

While ICS should be used only in patients with severe to very severe COPD, they are considered to be first choice maintenance treatment in mild, moderate and severe persistent asthma 115. Asthma may cause fixed airflow limitation and, thus, elderly asthmatics, in particular, may be misdiagnosed with COPD. The characteristics of asthmatics who develop fixed airflow limitation still fit the definition of asthma in terms of pathology 18, natural history 116, and response to treatment 117. These patients should be diagnosed and treated as asthmatics and not COPD patients. In this respect, it is recommended that asthma be excluded from the Venn diagram that is frequently used to illustrate the different components of COPD.

Conclusion
To conclude, on the basis of this evidence, the update of the GOLD guidelines on the management of COPD suggests regular treatment with inhaled long-acting bronchodilators (including tiotropium) and rehabilitation, starting from moderate (stage II) COPD. The evidence also suggests treatment with combined therapy with ICS starting from severe (stage III) COPD to prevent exacerbations.

	Possible sources of bias in observational studies of the effectiveness of inhaled corticosteroids in COPD

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
S. Suissa

Summary
The possible sources of bias that can arise from observational studies using computerised claims databases are discussed. The four classes of bias are selection (asthma versus chronic obstructive pulmonary disease) and confounding (indication, age, duration and severity of disease), choice of outcome (morbidity or mortality), timing of the drug exposure (i.e. cohort or case-control design), or time-related issues (incident or prevalent cohort and immortal time concerns).

Introduction
The observational studies conducted on the effectiveness of inhaled corticosteroids (ICS) in chronic obstructive pulmonary disease (COPD) have all used computerised claims databases that have several limitations. The possible sources of bias that can arise from the use of these databases can be classified into four types.

Sources of bias
The first class of bias is selection and confounding. One challenge is to deal with the differential diagnosis of COPD and asthma, which is particularly crucial because the effectiveness of ICS has been very well established in asthma. Thus, as the inclusion of asthma patients may exaggerate the effectiveness of ICS, the study population must have clear criteria to identify COPD patients and exclude asthma patients. In particular, definitions and criteria based on a physician's reported diagnosis, drug treatment and age must be carefully combined to optimise the diagnosis of COPD. The study design used, namely, cohort or case-control approach may also engender some form of selection bias. Thus, the design must be carefully selected as must the source population and the base cohort. An important source of bias is confounding by indication. The disease severity and its markers must be identified and quantified to control for any imbalance between the users of ICS and the nonusers. In particular, the use of health services and the profile of use of other drugs for COPD must be examined and analysed to provide the proper statistical adjustment and elimination of confounding. Two additional confounding factors that must be taken into account are age and duration of COPD. Finally, the approach to account for these factors can either be by matching cases and controls on these factors, by restricting the study to subjects who have or do not have some of these factors, and of course, by statistical analysis using a multivariate regression model that will control for these differences simultaneously. An important point of discussion should be the timing of these markers of disease severity with respect to the outcome and the ICS exposure. For instance, should disease severity be evaluated at the start of disease or at the time of the outcome under study?

The second source of bias is related to the choice of the outcome, morbidity or mortality. Morbidity can be evaluated from exacerbations, outpatient or emergency room visits, as well as hospitalisation. An exacerbation can be identified in some databases by the use of drugs, such as the simultaneous treatment with oral corticosteroids and antibiotics, or indicated by a diagnostic code posed by a physician or during a hospitalisation. With respect to mortality, an issue raised by the studies conducted, to date, is the use of all-cause mortality as an outcome, as opposed to death due to COPD. Since medications would be expected to be more specific to the outcome of COPD death, studies that would focus on all-cause mortality may provide an underestimate of the effect since other causes may not be affected by the medication under study. Nevertheless, studies focusing on COPD mortality should address the validity of the cause of death in death certificates, as well as the issue of other causes and underlying cause, since these patients may have several conditions at the time of death.

The third source of bias is related to the exposure. An important point is the timing of the drug exposure, in particular, whether exposure is selected at cohort entry or at the time of the outcome under study. This question relates to the choice of design, cohort or case-control. In addition, whether the effects are acute or whether regular treatment is required to attain the effectiveness under study needs to be considered with respect to drug exposure. Drug exposure also affects the choice of the reference group and whether this group can include patients who do not currently use ICS but who used them previously, or patients who are restricted to other drugs or classes of drugs, such as bronchodilators. With these classifications and the question of timing of use, concern must then be placed upon issues of exposure misclassification. For instance, patients who are not using ICS should not be classified as users and vice versa. Finally, the exposure and its timing will also relate to the analysis of the data, and, particularly, whether exposure is fixed, such as for the intention-to-treat approach, or time-dependent, such as that used in nested case-control analysis.

The last source of bias is that arising from time-related sources. The cohorts under study may be incident (based on newly diagnosed patients) or prevalent (patients well into their disease) cohorts. It is determined by whether patients at time zero already have had COPD for some time or have already been exposed to the drug under study for some time. It may be preferable to use incident cohorts where new treatment or new disease defines time zero for the cohorts. If this is not possible, the duration of prior COPD or prior drug use should be examined and accounted for in the analysis. The choice of time zero is important and may be taken as the date of first COPD diagnosis, the date of the first hospitalisation for COPD, the date of any hospitalisation for COPD, or the first time an ICS or a referent drug was used. Finally, in all cohort studies that involve time-dependent exposure, immortal time should be identified and accounted for 118–120. Immortal time periods, defined by follow-up times during which patients cannot, by definition, incur the outcome, have to be identified and accounted for with a proper analysis. In addition, studies that improperly exclude immortal time or do not account for it in the proper exposure group should be identified and assessed with respect to bias.

Discussion
SORIANO: There is evidence that COPD patients who have an asthma component die more frequently than COPD patients without an atopy or hyperresponsiveness component. COPD patients with an asthmatic component are a subgroup of patients with more severe COPD. This subpopulation is easily 30–50% of individuals with COPD.

FABBRI: According to Burrows et al. 116, this is not true, i.e. smokers with COPD and asthma have a 10-yr survival (65%) that is in between smokers with COPD without asthma and asthmatics. In fact, nonsmoking asthmatics with fixed airflow limitation have a life expectancy similar to normal subjects.

VIEGI: In contrast to L. Fabbri's suggestion that the Venn diagram be removed, I believe that, rather than removing, we should understand these complex relationships better. Also, should we consider COPD just as a smoking-related disease? Fifteen per cent of COPD is work-related. What about the contribution of air pollution? We cannot anticipate the prevalence of COPD as only related to smoking.

ERNST: There are certainly patients with "pure" asthma or COPD. There are a number of patients, however, that fall somewhere in the middle. It is not appropriate to pretend that everyone falls in that middle group, since this would not allow us to aim the correct treatment at the correct patients. While there will always be people with components of both asthma and COPD, I do not think that these are the majority of COPD patients.

BOURBEAU: We have to ask ourselves, what exactly is our question or intent in these studies. If we are trying to understand COPD as a complex disease in an epidemiological study, we may be interested in looking at different populations including the nonsmoker. But if you want to test an intervention, it is best to define a population of patients who most likely have COPD, so then it should be related to smoking.

HAGAN: Another factor to consider is the perspective of pharmaceutical industry. When we are designing randomised clinical trials (RCTs), we have to abide by the criteria requested by Regulatory Agencies. For example, forced expiratory volume in one second reversibility criteria are getting more and more stringent because we have to prove that these are pure COPD patients. But, in the real world, most COPD patients do not have pure COPD.

MAPEL: When wrestling with this issue of how to define COPD, there are three areas in our end-points that are problematic. First, in spirometry itself, we tend to fixate solely on airway obstruction and ignore dynamic hyperinflation. In a population of female smokers that we brought in for testing with no diagnosis, we performed complete lung volumes and we found that a large population of female smokers had normal spirometry but remarkably elevated residual volumes. The first change in objective measures with smokers is increased residual volumes. However, we missed that completely when we used spirometry as an end-point. Spirometry is really an illogical end-point. Most of the people in a study are still smoking. If you do not get away from exposure, then disease will progress. So we need to expand the end-points. Exacerbations and hospitalisation rates, for example, are important and exciting end-points that are starting to be used. Pathological end-points are also important. The Hattatua study focused on chronic bronchitis patients without an asthma component (pure COPD) and found significant reductions in subepithelial mast cells. There are probably some inflammatory mechanisms in COPD that will be affected by corticosteroids. However, even in the pathological studies, we are seeing null results because they are looking at the wrong end-points. The truth is that we know remarkably little about the pathology of the disease.

HAGAN: It is interesting to note that the Hattatua study in patients with pure COPD used exactly the same entry criteria as the Inhaled Steroids in Obstructive Lung Disease in Europe trial. That does provide some pathological basis for selection criteria that we use in industry when selecting patients for COPD studies.

ERNST: But, I hope that we are not sending a message that we should use poorly reproducible surrogate end-points, such as mast cells or residual volumes. I would hope that we would be going towards outcomes such as hospitalisations and exacerbations that actually have an impact.

VIEGI: I think we need to go back to the basic issues. We cannot still use reference equations that were collected 40 yrs ago. Although there are attempts to standardise, there is still high technical and population variability. We cannot just use one reference equation. We need to recognise that when we start to measure lung function, we should check which is the best reference equation for our clinic setting and population.

FABBRI: The Global Initiative for Obstructive Lung Disease (GOLD) guidelines are becoming increasingly evidence-based. If you want to issue recommendations for treatment based on evidence, you should specify the entry criteria of the study you cite. And the entry criteria of most of the studies I presented excluded atopy, history of asthma, etc. However, you still may have the problem of the mixed population. But, we need studies of that mixed population, too.

BOURBEAU: In trying to make a diagnosis of COPD as epidemiologists, we will not be able to do better than what our current understanding is of the disease. We have a poor understanding of the phenotyping of COPD. There are many phenotypes of the disease, but it is too early to distinguish what they are. We should try to distinguish what is an asthma population, a COPD population and the in-between questionable population. When we are doing a pharmacoepidemiological study, we should probably look at these populations of patients differently and try to validate (what we have done very little of in pharmacoepidemology studies so far) from the different administrative databases what is asthma and what is COPD and then, in our conclusions, we should be able to speak the same language. This will evolve over the next 10 yrs as our understanding of the disease increases and we hope that pharmacoepidemiology studies will also evolve with that understanding.

PRICE: L. Fabbri said that the guidelines are becoming increasingly evidence-based and that recommendations will have to be made based on RCTs. Yet, the challenge from the industry perspective is that to meet regulatory requirement they will have to study patients from narrower and narrower groups. The industry is trying to produce more generalisable data and more pragmatic trials with different populations in spite of having to produce more regimented trials for registration purposes. We just had a new evidence-based asthma guideline produced in the UK. One of the challenges of the evidence-based hierarchy as it has been implemented in our asthma guidelines is that if you have a discrepancy between an observational database and an RCT, the RCT wins, rather than viewing the evidence as complementary and trying to look at understanding why the discrepancies may occur. In COPD, it will be particularly important that we look for generalisability, and we need to have a way of handling within the guidelines that breadth of data rather than a straight evidence-based hierarchy. Are there plans to encompass these kinds of data as complementary within the hierarchy or will they stay as inferior?

FABBRI: When we discussed the criteria for grading evidence within the GOLD Scientific Committee, there was a strong suggestion to downregulate Cochrane reviews, post hoc analyses, and meta-analyses, with the understanding that these studies may help to generate hypotheses but do not provide evidence and that the evidence is provided only by RCTs. The hypotheses generated by post hoc analyses or meta-analyses and Cochrane reviews should then be properly tested in RCTs.

SORIANO: The reality is that we do not know the general epidemiology of COPD within the community. We still do not have a Framingham study in COPD. The majority of COPD patients are managed by general practitioners (GPs) and we know that GPs have been treating COPD patients with asthma drugs for a number of years. Probably, pharmacoepidemiological studies will help define what the outcomes are in COPD patients from the general population.

VOLLMER: We are all aware that individuals who enroll in RCTs are not representative of the general population. Entry criteria are often highly restrictive and participants highly motivated. While I have enormous respect for the value of RCTs, I also have respect for what can be learned from the large databases that we are beginning to collect from real-life experience. The trick with looking at these databases is to figure out the proper analytical methods to use. I am convinced that we need to find a way to marry these two sources of evidence.

WEISS: As chairman of the guidelines development committee for the American College of Physicians, I find myself asking the question, "How do we step away from the RCT, so that we can incorporate these other pieces of data?" We do not want to end up saying that, because of lack of data, it is best left up to the physician's best judgment. What kinds of questions should be addressed in non-RCT databases that won't be addressed in RCTs? Question-asking may be one of the most important ways to begin these discussions.

Methodological issues
PRICE: There may be issues of bias in terms of what analyses are being done and what actually gets published. We can get around that by registering clinical trials and data analysis plans. This is important to capture what was done versus what was reported. Another issue is that of clustering and the effect of centres in RCTs. In the General Practice Research Database, some GPs are much better at using their systems than others.

DAVIS: We can use this as a matching factor.

SUISSA: Matching does not resolve this issue. In certain studies, the drug may not be used appropriately or measurements themselves may be appropriate in some centres but not in others. If you then match on centres, your result will be attenuated towards a null effect, because you will have created all kinds of measurement errors, whereas in the centres where data are well collected, you may be closer to the truth. Therefore, rather than matching on the centre, a stratified analysis by these centres will provide more accurate estimates.

VOLLMER: It is not just measurement error, but variation in practice patterns.

WEISS: This relates to the clustering of effects above the individual level that have to be accounted for and most of those that we perceive right now are in the health system design whether it be in the actual practice of providers, the way they practice as a group, or how this system is financed. So we have to ask, is it important to account for these?

SULLIVAN: What about major health systems changes, like payments to hospitals that can affect the rate of hospitalisations or exacerbations, independent of anything going on with the disease or treatment?

SUISSA: That also speaks to time-related bias. It would be important that a patient in January 2001 gets compared with all patients in the database in January 2001, so that they are all subject to the same rate at that point in time. If the drug will increase or decrease this rate, it will be assessed at the same time point.

BURNEY: But, it is not just time. What about from institution to institution and the way they manipulate the data?

SULLIVAN: You may have local variations in these factors.

SUISSA: Definitely, stratification on the region would then be important.

VOLLMER: You also need to account for in-migration. These individuals have no prior history in the database and could be incorrectly classified as incident cases when they first present for care. This can be partially addressed by treating those who seek care during an initial enrolment period, say 6 months, as having prevalent disease.

STURKENBOOM: Confounding by indication or by contraindication is also a consideration. It is not so much about the severity of COPD but about the severity of all types of comorbidities.

MAPEL: A difficult confounder that we have spent time wrestling with is comorbid illnesses, particularly heart disease. The single most common death in COPD patients has been cardiac arrest. How you deal with that will greatly affect the results. If you use it as an exclusion criterion, you end up wiping out half of your population. But if you use Charlson index or a similar technique to adjust for that, it changes your results.

ERNST: The comorbidity problem is a big problem especially in COPD. We have looked at cause of death in patients in our COPD cohort and the most common cause of death is cardiovascular. Even among those who are hospitalised with a primary diagnosis of COPD, the primary cause of death is cardiovascular. A lot of this represents misclassification.

VOLLMER: Adjusting for severity is also difficult, since we will typically have very imprecise tools for assessing it. Furthermore, the measures that are generally available to us for defining severity are inevitably closely related to the same measures we would use to assess current level of control, which is an outcome. This creates the potential for overadjusting in our analyses.

VIEGI: Besides the time dimension we should also use the space dimension. There is overwhelming evidence coming from recent air pollution epidemiology data. Differential exposures to air pollution can represent an important source of variability when we compare studies on drug efficacy coming from different populations. Another issue comes from the problem of compliance. We know from P. Burney's study that there can be an inverse relationship between the rate of compliance and hospitalisation. Those who do not take medications are more at risk of being hospitalised.

DAVIS: The issue of patient care in observational studies is a potential source of bias. Some may argue that patients on ICS are getting better because they are getting better care overall.

TAYLOR: The checklist of potential biases that S. Suissa has presented really provides us with ways to try to minimise bias, but, they focus mostly on internal validity. However, it is also important to look at external validity, which is particularly important for observational databases.

FABBRI: We often have problems with the definitions of exacerbation, particularly for hospitalisation, when you may see a diagnosis for exacerbation or respiratory failure linked to COPD, but, in fact, we may face a complex that goes from heart failure, thromboembolism, or pneumonia. In other words, worsening of symptoms of COPD (particularly dyspnoea, but also cough and occasionally sputum) may be due to complications rather then exacerbations of the underlying disease.

ERNST: I am more concerned about all the hospitalisations for heart failure, which are actually COPD, right heart failure, or cor pulmonale. In my clinical experience, the diagnosis of heart failure is often made when it is actually COPD.

	The Ontario and Alberta experience with administrative databases in COPD research

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
D.D. Sin

Summary
Administrative databases were used from two different provinces in Canada to evaluate the relationship between inhaled corticosteroid (ICS) therapy and clinical health outcomes in patients with chronic obstructive pulmonary disease (COPD), who were previously hospitalised for their disease. In the first study, health databases from Ontario, Canada, (n=22,620) were used and it was found that elderly patients (65 yrs), who received at least one dispensation of ICS within 90 days of hospital discharge, had a combined 26% lower adjusted relative risk (RR) for respiratory hospitalisation and all-cause mortality than those who did not receive these medications (RR 0.74, 95% confidence interval (CI) 0.71–0.78).

In the second study, administrative databases in Alberta, Canada, (n=6,740) were used to evaluate the long-term "effects" of ICS among elderly COPD patients and to determine whether the survival benefits were dose-dependent. It was found overall that patients who received at least one dispensation of ICS during follow-up (average follow-up 32 months) had a 25% relative reduction in the risk for all-cause mortality (RR 0.75, 95% CI 0.68–0.82) compared with those who did not receive any ICS during follow-up. Patients on medium (501–1,000 µg·day^–1 of beclomethasone equivalent) or high-dose therapy (>1 mg·day^–1 of beclomethasone) had lower risks for mortality than those on low doses (RR 0.77, 95% CI 0.69–0.86 for low dose; RR 0.48, 95% CI 0.37–0.63 for medium dose; and RR 0.55, 95% CI 0.44–0.69 for high dose).

Methods
Study design
In the first study, health databases from Ontario, Canada (total population 11 million) were used to identify all elderly patients (65 yrs), who had a primary hospitalisation for chronic obstructive pulmonary disease (COPD) exacerbation (International Classification of Diseases, ninth revision (ICD-9) codes 490.x, 491.x, 492.x and 496.x) between 1992–1997 121. Patients who died within 30 days of discharge or during the initial hospitalisation, or were transferred to another acute care facility for active treatment were excluded. The remaining cohort was then divided into two mutually exclusive groups based on whether or not they received at least one dispensation of an inhaled corticosteroid (ICS; beclomethasone, budesonide, triamcinolone, and flunisolide) within 90 days of hospital discharge. These patients were followed from the date of hospital discharge until the date of their death, a repeat respiratory hospitalisation, or 1 yr of completed follow-up, whichever was earliest. A Cox proportional hazards model was used to compare the risk of all-cause mortality or repeat respiratory hospitalisation during the follow-up period between those who did and did not receive ICS within 90 days of discharge. In this model, the factors adjusted for were age (as a continuous variable), sex, modified Charlson comorbidity score 122, use of other anti-COPD medications (short-acting β₂-adrenergics, ipratropium bromide, antibiotics and oral corticosteroids), and history of emergency or office visit for asthma within the year previous to the initial hospitalisation.

In the second study, which used administrative health databases from Alberta, Canada, the COPD cohort was identified by searching through the hospital discharge abstracts from Alberta's version of the Canadian Institute for Health Information (CIHI) 123. Similarly to the Ontario study, residents of 65 yrs of age, who had at least one hospitalisation for COPD as the most responsible diagnosis between 1994–1998 using ICD-9 codes 490.x, 491.x, 492.x, and 496.x were included. Patients who died during the index hospitalisation were excluded. All hospital visits occurring after the first hospital visit were censored for each study patient in order to avoid double counting. The cohort was then divided into five mutually exclusive categories based on the use of ICS: 1) those who did not receive any ICS; and 2) those who received low dose (500 µg·day^–1 of beclomethasone or equivalent); 3) medium dose (501–1,000 µg·day^–1); 4) high dose (>1,000 µg·day¹); or 5) an indeterminate dose of ICS. The last category contained patients who received only one dispensation of ICS during the follow-up period. Although the average daily dose for these individuals could not be calculated, they most likely received an average daily dose that was lower than that in the low-dose category. Since Alberta Blue Cross provides data on the quantity of medications dispensed rather than the daily dose, the average daily dose of ICS was imputed by determining the total dose of these medications dispensed for each patient for the first two prescriptions and dividing the total dose of the first dispensation by follow-up time (in days) between these two doses. The calculated average dose was rounded to a clinically plausible dose 124. To allow cross-comparisons between different ICS preparations, all formulations were converted into beclomethasone dipropionate equivalents based on equivalency calculations suggested by the Canadian Asthma Consensus Report 125. The study patients were then followed for 3 yrs following discharge from the initial hospitalisation for COPD, or until the date of their death, whichever was the earliest. A Cox proportional hazards model was used to compare the survival rate between those who did and did not receive ICS during follow-up and across the various dose categories, adjusted for age, sex, Charlson comorbidity scores 122, admittance to an intensive care unit (ICU) during the initial hospital stay (yes or no variable), and the filling prescriptions of various pulmonary medications including short-acting β₂-agonists, ipratropium bromide, oral corticosteroids and oral theophyllines. A series of sensitivity analyses were conducted to determine the robustness of the main findings to a different set of conditions and assumptions.

Databases used
Cohort identification
For the first project 121, study patients were identified using discharge abstracts from Ontario's version of the CIHI database 126; for the second project, Alberta's version of the CIHI database was used to identify the study patients. The CIHI Discharge Abstract Database contains data on hospital discharges from all Canadian provinces except for Quebec and parts of Manitoba. Each abstract has information on patient demographics, contents of the hospital stay (including date of admission, date of discharge, the most responsible diagnosis and up to 15 secondary diagnoses) and disposition 126. In the CIHI Discharge Abstract Database, the most responsible diagnosis is defined as "the one diagnosis, which describes the most significant condition of the patient which causes his stay in hospital" 127.

The CIHI employs a continuous quality assurance programme to ensure that the information contained within the Discharge Abstract Database is of excellent quality 128. In >85% of the cases, the most responsible diagnosis in the CIHI Discharge Abstract Database matches the (most responsible) diagnosis obtained through independent chart audits performed by trained medical analysts 128. For most common conditions in the CIHI databases, the false-positive rate ranges 0–11% and the false-negative rate ranges 0–13% 128. False-positives are most commonly observed among ambulatory sensitive conditions such as depression and diabetes. COPD is not an ambulatory sensitive condition, and, as such, would be expected to have a very low false-positive rate (between 5–7%) but a slightly higher false-negative rate between 10–12%. These internal CIHI estimates are similar to those reported independently by Rawson and Malcolm 129. This group showed that a CIHI coding of COPD in the most responsible diagnosis field had a sensitivity of 94% in identifying COPD cases compared with primary data abstracted from patient charts 129. The sensitivity and specificity may be improved by increasing the minimum inclusion age of the study population, as COPD-related hospitalisation does not occur in appreciable numbers until >55 yrs of age with a majority of patients being >65 yrs 81, 130. In general, studies that probe the principal diagnostic field for COPD patients in the CIHI database may miss 10–12% of the total pool of available COPD patients. However, among those identified with COPD, the diagnosis is likely to be correct.

Identification of medications
Alberta Blue Cross and Ontario Drug Benefit plan for seniors
In Alberta, Canada, seniors (those 65 yrs of age) receive direct government subsidies for purchase of prescription medications, which are listed on the government's formulary of approved drugs. In most circumstances, the individual is asked to pay for 30% of the total cost of the drug, up to a maximum of $25 per drug. This fee is waived if a senior has a very low income 131. This database should be accurate and reliable since this information serves as a basis for financial reimbursement to pharmacies to cover 70% of total cost of the medications. The Ontario Drug Benefit (ODB) Programme extends similar coverage for senior residents living in Ontario 132. Although the provincial government of Ontario recently instituted a small co-payment system for prescription medications, during the period of the first study (1992–1997), prescriptions were dispensed free of charge to patients, aged 65 yrs. As with the Alberta Blue Cross database, the ODB should be accurate and reliable as the provincial government has a rigorous set of quality controls in place to ensure that the information within this database could be used for surveillance and reimbursement purposes. In both provinces, all commonly used formulations of ICS are covered under provincial drug plans.

Vital statistics
In Ontario, two sources of data were used to confirm inhospital deaths: CIHI Discharge Abstract Database and Registered Persons Database of Ontario. In Alberta, the Alberta Health Insurance Plan Registry as well as the CIHI Discharge Abstract Database were used. Using these sources, >95% of deaths occurring in hospitalised patients were able to be cross-validated. For deaths occurring outside the hospital setting, the Registered Persons Database was used for Ontario patients and the Alberta Health Insurance Plan Registry for Alberta patients. Both of these databases are likely to be comprehensive and accurate because the annual emigration rate from Ontario and Alberta is generally <0.5% and <2.0%, respectively 133.

Results
In Ontario's study, there were 22,620 study patients who met the inclusion and exclusion criteria. The mean age of the study patients was 75 yrs; approximately one-half were female. When adjusted for various factors, it was found that those patients who had received ICS within 90 days of discharge had a relative risk (RR) of 0.74 (95% confidence interval (CI) 0.71–0.78) for combined hospitalisation and all-cause mortality. The RR was 0.71 (0.65–0.78) for all-cause mortality and 0.76 (0.71–0.80) for repeat respiratory hospitalisation. There was a significant relationship between the use of ICS and survival (fig. 1).

View larger version (11K): [in this window] [in a new window]   Fig. 1.— The relationship between inhaled corticosteroids (ICS) and survival. Cox proportional hazard model. COPD: chronic obstructive pulmonary disease. - - -: ICS; –––: no ICS.

In Alberta's study, data from 6,740 patients were used. The mean age of the study population was 76 yrs. Of these patients, 3,661 (54.3%) were males; 3,744 (55.6%) had no comorbidities. Overall, 3,343 (49.6% of total) patients received an ICS during the study period. Of these patients, 2,011 (61.2%) used low-dose therapy, 318 (9.7%) used medium-dose therapy, 413 (12.6%) used high-dose therapy, and 601 (18.0%) received an indeterminate dose. After adjustments for age, sex, comorbidities, ICU stay, and use of other pulmonary medications, a 25% reduction in the overall mortality rate was observed in those who received ICS compared with those who did not (RR 0.75, 95% CI 0.68–0.82). Patients dispensed low-dose therapy had a 23% (RR 0.77, 95% CI 0.69–0.86) reduction in the mortality rate compared with those who did not receive any ICS. Those on medium-dose therapy experienced a 52% reduction (RR 0.48, 95% CI 0.37–0.63), while those on high-dose therapy experienced a 45% relative reduction in the mortality rate compared with those who did not receive any ICS (RR 0.55, 95% CI 0.44–0.69). Patients on indeterminate doses did not experience any significant decline in their all-cause mortality rate (RR 0.88, 95% CI 0.76–1.03; p=0.108). A significant relationship was also found between survival and the number of canisters of ICS received by patients during a given year (fig. 2). Interestingly, no such relationship was found between ipratropium bromide and survival (fig. 3). With short-acting β₂-adrenergics, there was a slight increase in mortality risk along the dispensation gradient, which may reflect confounding by indication or severity (fig. 4).

View larger version (10K): [in this window] [in a new window]   Fig. 2.— The relationship between inhaled corticosteroids and survival according to the average number of canisters received during a given year. RR: relative risk.

View larger version (9K): [in this window] [in a new window]   Fig. 3.— The relationship between inhaled (short-acting) anticholinergic and survival according to the average number of canisters received during a given year. RR: relative risk.

View larger version (10K): [in this window] [in a new window]   Fig. 4.— The relationship between inhaled short acting β2-agonists and survival according to the average number of canisters received during a given year. RR: relative risk.

Since survivor treatment selection bias was a potential concern for the analysis, sensitivity analyses were performed including only certain subgroups 134. For instance, a subgroup analysis was performed excluding all patients who had a follow-up time of 90 days or less and, thus, had a lower probability of receiving ICS than the rest of the cohort. In this analysis, ICS were associated with a 43% lower risk of mortality (RR 0.57, 95% CI 0.51–0.63). Using cut-offs of 6, 9 or 12 months made little difference to the overall findings (RR 0.57 for 6 months, RR 0.58 for 9 months, RR 0.58 for 12 months) suggesting that survivor treatment selection bias was not a significant concern for the analysis.

Interpretation
There are several important advantages of using administrative databases for research. Research studies based on administrative databases: 1) contain large sample sizes; 2) are population-based (unlikely to suffer from selection bias); 3) are relatively inexpensive to conduct; 4) can generate data quickly; 5) have very good follow-up of cohort (outmigration is usually <2%·yr^-1 from most provinces in Canada); 6) reflect "real-life" experiences (and do not suffer from "volunteer" bias or "Hawethorne" effects); 7) are able to determine whether effectiveness applies to subpopulations, which are generally excluded in randomised controlled trials (RCTs; e.g. elderly, females, certain racial groups, patients with multiple comorbidities); and 8) have the power to evaluate the effects of an intervention on hard clinical outcomes (such as mortality and hospitalisation), rather than relying on intermediate measurements, which may not adequately predict hard clinical outcomes and may suffer from ascertainment bias.

There are several important limitations: 1) diagnostic misclassification is possible as investigators have to rely on self or physician diagnosis of COPD; 2) administrative databases may not contain physiological and/or biochemical measurements, thus, adjustments for severity are problematic; 3) confounding by indication or severity can be problematic; 4) while administrative databases allow investigators to determine which study patients did and did not receive certain medications, patients' compliance with these medications are rarely found in these databases; 5) exposure misclassification may also be problematic. For example, if some patients who initially used ICS decide not to use them later on, while some initial nonusers become users, then exposure misclassification may occur. However, this type of "bias" is usually nondifferential in nature and pulls the RR towards the null value; 6) studies that use a longitudinal analytical design (e.g. Cox proportional hazards model) are susceptible to survivor treatment selection bias. For example, patients living longer are more likely to be exposed to ICS than those who die earlier. This bias will make ICS more efficacious than they really are. The longer the follow-up period, the more likely this bias is to be present. It is important, therefore, to perform sensitivity analyses. One method is to perform a similar type of analysis using other anti-COPD medications. If this bias is operative, then other drugs should also demonstrate a similar survival benefit. Another method of checking for this bias is to shorten the follow-up period. A third method is to exclude patients who died early on in the follow-up period to ensure that in the remaining cohort, all study patients had a reasonable chance of receiving ICS. In the Alberta study, all of these sensitivity analyses were conducted and it was found that the results were materially unaffected by changes in the follow-up period or in the exposure period; and 7) data from research studies based on administrative records are best used to complement findings from RCTs or to generate novel hypotheses that should be tested in large RCTs.

Discussion
ERNST: I am curious about the magnitude of effect on the reduction of all-cause mortality in this study (29%) versus what you may see in cardiovascular trials (33–36%) with simvastatin versus placebo, for example. How could ICS have an effect similar to antiplatelet therapy for myocardial infarction?

SIN: That is an excellent point. Do ICS reduce mortality? What is the magnitude of that reduction? I think observational studies are much stronger to answer the former rather than the latter. Our observational study suggests that they reduce mortality. Is there a plausible mechanism by which mortality can be reduced? Your point that the single most frequent cause of COPD mortality is cardiovascular is a germane one. So, if ICS have this powerful effect on mortality reduction, they have to have some effect on the cardiovascular system, but I do not know what that mechanism is.

ERNST: Did you try to look at the number of patients who may have been labelled as asthmatic?

SIN: No we did not do that. But, I think we are getting back to that whole business of what is COPD and what is asthma in the elderly? I do not think anybody has an answer to that.

BOURBEAU: First, the magnitude of effect of ICS, assuming that there is a true effect, could very well be magnified. It is known that observational studies may show a higher magnitude of effect than an RCT, which we usually refer to as the gold standard. Second, I also have some concerns in these studies with the diagnosis of COPD based only on the primary discharge diagnosis. With regard to COPD, we could exclude patients with associated asthma as a secondary diagnosis to obtain a strict definition of COPD. As it is, part of the effect observed could be influenced by patients with asthma. We do not know about this for sure. It would be of great value to validate the diagnosis of COPD versus asthma used from databases and see how it corresponds to the definition that we use in clinic.

ERNST: I think there is a problem with what we are doing with nonexposed time. In the Alberta cohort, a patient could get a prescription anytime in the last 3 yrs. Let us talk, for example, about a patient who is exposed in the last 6 months and dies on the last day of follow-up. The first 2.5 yrs, when the patient had no ICS, is counted as exposed time in the intention-to-treat analysis. I do not see how this can be correct.

SIN: I think that is the survivor treatment selection bias. Greater than 80% receive their first dispensation within the first 100 days. If they did not receive it within that time-frame, then it was highly unlikely that they would receive it. That is why you need to do secondary analyses to make sure your data are robust. That is why we looked at ipratropium and short-acting β-agonists. If survivor treatment selection bias was the main culprit here, it should also be evident in those drugs.

ERNST: I disagree. I think these drugs are prescribed very differently. They are the standard after being discharged from the hospital with COPD. There is less confounding by indication with those drugs. Perhaps we should call this immortal time when they are not exposed and they are not at risk for the outcome, because if they had the outcome before exposure they would be in the nonexposed group. So I think the follow-up time is going in the wrong group.

WEISS: Is this a direct treatment effect? What were the care pattern differences that may also be highly associated with the treatment pattern? You controlled for primary versus specialty. How did you do that? That assignment may have created some ambiguity and added noise to the model. That may have explained why you saw such a big effect size in your hazard ratios because you are picking up a care difference.

SIN: We looked at the principle care provider at the time of discharge of these patients.

WEISS: That may not be appropriate, because it really asks the question of who your system is assigning as the care provider. Whoever did the abstract discharge for that day and signed as the care provider was the doctor on record. So it is that care process that may be so closely associated with that dispensing moment that is really important.

FABBRI: What do you think is the strength of your own data? Based on your mortality data, would you be ready to issue a recommendation to tell your government to put this drug on the formulary to prevent COPD death?

SIN: I do not think you can make firm recommendations from observational studies. You have to look at the totality of evidence and formulate a rational recommendation. Our study has to be put into the pool of other studies published in this area. Right now, for moderate-to-severe patients, ICS appear to have a positive protective effect on reducing exacerbations and hospitalisation. More work needs to be done on mortality rates. If indeed, ICS produce an effect on mortality, there has to be some biological mechanism to explain that. So we cannot make any definitive statement about mortality. But, we should not dismiss the data because we do not understand how it works.

FABBRI: Do you think an observational study would be sufficient in this case?

SIN: No, I would also do an RCT.

WEISS: But, as soon you say let us do an RCT, then you are saying let us wait 7 yrs to get the results on this! Can we further investigate the biases in this study, explore how generalisable this is in other observational studies and see if one can converge the observational study environment on a similar set of findings. This will take just a few years, rather than 7 yrs for an RCT. The convergence of the observational realm would be a near approximation to the experimental realm.

STURKENBOOM: The results suggest that the way to look at responsiveness is to look at the number of canisters dispensed per year. I do not quite agree with that. Patients who take 10 canisters·yr^-1 are those who are surviving up to 10 canisters·yr^-1. By doing so, patients with 10 canisters·yr^-1 are forced to have a lower mortality rate. This is again the issue of immortal time. Using data during follow-up to define exposure is not the right way, and you should probably use a time-dependent approach.

HAGAN: There may be selection bias in the opposite direction. I would expect that salbutamol-ipratropium patients would have milder COPD while those getting ICS would be sicker. Therefore, I would have expected the death rate to go the other way around.

McLAUGHLIN: The same thing goes for treatment versus no treatment. In an observational database, if you are not receiving treatment, physicians believe you do not require treatment and you are probably less severe.

HAGAN: And if you do get a better effect from the ones that you thought were sicker patients, then something must be happening.

SORIANO: This is called negative confounding by indication. They will be more severe. In our studies, the individuals who are treated versus those who are not treated eventually get a better outcome.

STURKENBOOM: That suggests you are more severe, but if you are more severe, then you are not being treated. If you find an effect, the real effect should be even bigger, but here it can be the other way around.

SORIANO: The solution is that it is better to compare not taking the drug with taking another drug. You partly overcome this problem when you use a reference group that includes treatment with another drug.

PRICE: Some patients will have had asthma. Are there any data on accuracy of hospital diagnosis of COPD upon admission?

SIN: Regardless of diagnostic issues, if the data are indeed true and ICS reduce all-cause mortality, that in itself is a powerful observation, even if it is driven by asthmatics.

VOLLMER: The issue of is it asthma or is it COPD has been a recurrent theme in this discussion. I am not sure I would worry too much about what diagnosis these patients really have. The fact is that they have been diagnosed as having COPD and are being treated accordingly. Thus, we can fairly clearly ask what happens when doctors treat their "COPD" patients with ICS. It is an interesting question, though not quite the same perhaps as studying the effect of ICS therapy in those who truly have COPD.

DAVIS: Did your crude values for adjustment of comorbidities change much? Are these really confounders? Are they associated both with the outcome we are interested in and the probability of getting an ICS?

SIN: With the separate analyses that we did, we were fairly confident that heart disease was not a major confounder.

	The UK General Practice Research Database Experience in COPD pharmacoepidemiology

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
J.B. Soriano

Summary
The UK General Practice Research Database (GPRD), an automated database of primary care covering a total population in excess of 3.4 million inhabitants (5.7% of the population) in the UK, provides a unique source of data to investigate the epidemiology of chronic obstructive pulmonary disease (COPD) from a population perspective and the effects of drugs on COPD-related outcomes. After formal validation of COPD and asthma definitions in this database and the description of the clinical epidemiology and trends in incidence and prevalence of disease, the results of two recent pharmacoepidemiology studies of COPD in the GPRD are comprehensively discussed.

Introduction
The UK General Practice Research Database (GPRD) database has been described in detail elsewhere 135–137, and has been utilised previously to obtain epidemiological trends of chronic obstructive pulmonary disease (COPD) incidence and prevalence in the UK during the 1990s 138, 139 (fig. 5). The GPRD is the world's largest computerised database of anonymised longitudinal patient records from general practice, containing >35 million patient-yrs of data. It is a significant resource for the following areas of study: clinical epidemiology, drug safety, pharmacoepidemiology, health outcomes, health service planning and pharmacoeconomics. It is used regularly by academics, the UK Department of Health, the Office for National Statistics (ONS), medicines regulatory authorities and the pharmaceutical industry. All research conducted within the database requires a protocol submitted and approved by the Scientific and Ethical Advisory Group of the GPRD. The GPRD bibliography now contains over 200 scientific papers and editorials published in top biomedical journals, demonstrating the quality and wide applicability of the GPRD data.

View larger version (15K): [in this window] [in a new window]   Fig. 5.— Prevalence trends of physician-diagnosed chronic obstructive pulmonary disease during the 1990s, by sex and age. : females; : males. –––: >65 yrs; - - -: 45–65 yrs; : 20–44 yrs. Adapted from 139.

In November 1993, Reuters Health Information acquired VAMP Ltd. In 1994, Reuters decided to donate the database to the Department of Health, while it continued its interest in the provision of practice management software. The database was renamed GPRD at this time. In 1995, Reuters launched Vision, a major new Windows-based practice management software application, which has subsequently become the dominant practice software used by GPs in the GPRD scheme.

Since 1994, the Secretary of State for Health has owned the database. Between 1994–1999, the database was managed by the Department's Statistics Division and operated by the ONS. In 1999, the Medicines Control Agency (MCA) took over management of the GPRD, relocated GPRD's operations from ONS to the MCA, and initiated a major redevelopment programme to enable broader research usage of the data both within the UK and overseas.

Methods
The chronic obstructive pulmonary disease General Practice Research Database definition
The identification of incident COPD patients in primary care and chest clinics is a medical challenge in the UK. Furthermore, there is no single COPD diagnosis code in the GPRD, therefore, specific searches had to be conducted to identify both the Oxford Medical Information System codes based upon the International Classification of Disease (ICD)-8 coding system and Read codes that determine COPD diagnoses within the database.

To demonstrate the ability of the GPRD database codes to provide a good differential diagnosis of COPD versus asthma in the GPRD, a validation study was conducted. Briefly, 300 questionnaires were mailed to a random sample of GPRD surgeries in charge of 225 patients with a diagnosis of COPD and an age- and sex-matched group of 75 patients with asthma. The response rate was 85.7%. The validation study found that the definition of COPD was able to satisfactorily distinguish COPD from asthma and describe different levels of severity among COPD patients, with a sensitivity of a correct COPD diagnosis of 71.2 and 80.3% in moderate and severe COPD patients, respectively 140. The specificity of the definitions of moderate and severe COPD was 75.0 and 81.4%, respectively.

However, a key question in any COPD clinical or epidemiological study is the reliability of the differential diagnosis of COPD versus asthma. Therefore, as a further safeguard, only patients >50 yrs of age have been included in the GPRD COPD analyses. Other demographic features, such as smoking history are supportive; indeed, the high mortality rate observed would itself make frequent misdiagnosis of asthma unlikely.

Finally, examination of cause of death revealed that only nine patients (three in the treated and six in the comparison group) had asthma recorded as their cause of death. Moreover, there were only 298 (6.4%) patients with an historical asthma diagnosis before the age of 50 yrs and the mortality rate differences from the study were similar between the total population and the population restricted to those without any evidence of asthma. For these reasons, it is believed that misclassification of asthma as COPD did not occur frequently in these GPRD studies. The efforts to take asthma into account at least match the efforts of other recent pharmacoepidemiological studies in COPD patients, where no validation study was conducted and asthma exclusion criteria were only based on restricting the population to older ages. Hence, from the double diagnosis by GP and hospital doctors at discharge, after several days of hospital admission to consider that patients were not asthmatic, there is considerable confidence in the validity of the GPRD diagnosis of COPD.

When studying COPD, the choice for study design in the GPRD has been the cohort rather than cross-sectional or case-control designs. The GPRD allows researchers to identify individuals from birth or registration within the database up to death or censoring. The advantages of identifying incident COPD, severity and duration of the condition have been presented elsewhere. In the two GPRD pharmacoepidemiology studies, individuals followed up for 3 yrs after COPD diagnosis 141, and those individuals followed up for 1 yr after a COPD hospital admission discharge 142 were examined.

Drug exposure
In studies regarding the beneficial effects of drugs in pharmacoepidemiology, it is recommended to compare a given drug versus a reference group exposed to another drug, rather than to a group of nonusers, to avoid or reduce confounding by indication i.e. treatment may be chosen for an individual patient because of the presence or absence of specific features of disease, and patients who have a diagnosis in their record but remain untreated are likely to be less severe 143.

The GPRD approach has been to compare potential drug exposure groups with current recommendations in clinical practice. Current British Thoracic Society guidelines for COPD, published in 1997, recommend short-acting bronchodilators for all symptomatic patients, but state that there is insufficient evidence for use of inhaled corticosteroids (ICS) or long-acting β-agonists 144. Therefore, all drug comparisons are categorised to simulate an "intention-to-treat" analysis, and compared with a reference group of physician-diagnosed COPD patients who received three or more prescriptions over an initial 6-month period of one of the following groups of drugs: short-acting β-agonists, xanthines, anticholinergics or combined bronchodilators, but no use of ICS or long-acting β-agonists since diagnosis with COPD.

In contrast with research that evaluated the effect of ICS alone, this study focused on the effect of the combination of long-acting β-agonists and ICS due to physiological, clinical and statistical evidence. There seems to be at least an additive effect of combined long-acting β-agonists and ICS use in respiratory disease 111, 145, and their molecular mechanisms of action are different 145. Combined use of ICS and long-acting β-agonists for COPD is currently being assessed in ongoing randomised controlled trials (RCTs) (Trial of Inhaled Steroids and Long-Acting β₂-Agonists 111 and Towards a Revolution in COPD Health trials). Finally, from the statistics perspective, comparison with other drug classes (i.e. short-acting bronchodilators, xanthines, anticholinergics, oxygen and other) would have required adjustment for multiple comparisons beyond the initial primary analysis.

Outcomes
The outcome of interest for the two studies was the occurrence of death 141, or of a first severe COPD exacerbation, defined either as rehospitalisation for a COPD-related medical condition or death 142. Due to the competing risks of these two outcomes, the main outcomes are presented separately for COPD rehospitalisations and all-cause mortality within a year.

Covariates
Asthma
Since a substantial number of COPD patients in the GPRD have at some time had asthma mentioned in their record, this is included as a potential confounder variable in the analysis.

Smoking
Information on tobacco use was categorised as nonsmoker (including never- and exsmoker), unknown or current smoker based on the database medical and prevention files of the patient. Pack-yrs of smoking were estimated in the identified smokers.

Oral steroids
Courses of oral steroids during the follow-up period, a potential confounder of the association between ICS and mortality 146, were included as a yes/no categorical variable.

Comorbidities
Baseline comorbidities were identified from the database medical file 12 months prior to the initiation of therapy and categorised into a modified Charlson comorbidity index 122. COPD was excluded from the list.

Year of entry
The year of entry into the drug exposure category was also included in the multivariate analysis in an effort to control for prescribing trends.

Statistical analysis
Standard survival analysis was used for comparison of groups according to drug exposure 147. Duration of follow-up was defined as the time period between COPD diagnosis and death or censoring; or from discharge from a first COPD hospitalisation and the next COPD hospitalisation, death or censoring. An immortal person-time period of 180 or 30 days was used to ensure that classification into treatment groups was based on a reasonable period of observation 119, 144. Therefore, in all treatment groups, patients who failed to complete a postdiagnosis of 180 days or a postdischarge 30-day period of observation for drug use were excluded from analyses.

Crude Kaplan-Meier and adjusted Cox survival estimates were then obtained for each drug exposure group. Relative risks were obtained using Cox's proportional hazards model, with adjustments for sex, year of entry, age, smoking status (non-, unknown and current smoker), comorbidities (absence, 1, 2 or 3+ comorbidities), oral corticosteroids and concomitant asthma mention in the patient's record.

Results
A total of 1,045 COPD patients who were regular users of salmeterol or fluticasone propionate (FP; alone or combined) were compared with 3,620 COPD patients. During year 1, the mean number of salmeterol prescriptions was 8.59, 0.13 and 7.03 in the combined salmeterol and FP, FP-only and salmeterol-only groups, respectively. Similarly, the mean number of FP prescriptions was 9.19, 8.64 and 0.05 in the combined salmeterol and FP, FP-only and salmeterol-only groups, respectively. These usage patterns were also maintained in the second and third year of follow-up.

Baseline characteristics of all drug exposure groups are shown in table 5. Users of FP and/or salmeterol were more often female, diagnosed at an earlier age, and were more often categorised as having severe COPD (p<0.05). In the baseline period of 6 months prior to initiation of pharmacotherapy, users of FP and/or salmeterol received more prescriptions of ICS, xanthines, anticholinergics, oral corticosteroids and combined bronchodilator products (p<0.05). Baseline use of general and COPD-related health services were similar between the groups, whereas a history of comorbidities was more common in the comparison group. There were no significant differences in baseline and demographic characteristics within the subgroups of patients using combined FP and salmeterol, FP only and salmeterol only (table 5).

View larger version (11K): [in this window] [in a new window]   Fig. 6.— Three-year Cox-adjusted survival function of chronic obstructive pulmonary disease patients by therapy with combined fluticasone propionate (FP) and salmeterol (–– - ––), FP only (), salmeterol only (– – –) versus comparison (–––––). #: p<0.0008 versus comparison; ¶: p<0.028 versus comparison. Adapted from 141.

View larger version (14K): [in this window] [in a new window]   Fig. 7.— One-year Cox-adjusted survival function of time to rehospitalisation or death in chronic obstructive pulmonary disease (COPD) users of inhaled corticosteroids (ICS) and long-acting β-agonists (–– - ––), ICS only (), long-acting β-agonists only (– – –) versus comparison (–––––). Adapted from 142.

Interpretation
The GPRD has stringent mechanisms of quality control but relies on the good practice of participating GPs who are invited to enter all significant morbidity events on each individual patient in the computer record, irrespective of whether the event occurred in the surgery, at a visit, or over the phone. All diagnoses and procedures communicated to the GP as a result of a hospital or other specialist visit (inpatient, outpatient, or in an accident and emergency unit) must be recorded when the GP is informed. Compared with data from an RCT and similar to other administrative and research databases, the GPRD lacks valid, complete information on respiratory function, weight, alcohol and tobacco consumption. Differential diagnosis or drug dosages are often difficult to assess, but major diagnoses, use of health services, and medication are registered with an acceptable degree of validity 148. Offsetting these drawbacks are the very large sample size and the extensive information of medical events and treatments that are available from practices in all parts of the UK. More extensive information on the advantages and limitations of the GPRD can be found elsewhere 149.

The GPRD framework is currently in transition to a full-featured online version that should reduce the current time gap between individual data entry and availability of cleared data for research from 12–15 months to weeks only. Additionally, there are renewed efforts to standardise drug dosing and to reinforce regular entry of tobacco and alcohol consumption information. Due to competition with other new systems, trends of attrition in participating practices and the final number of surgeries enrolled should determine the representativity of GPRD of the current UK population.

By reducing the gap time of GPRD data availability from 12–15 months to weeks, the GPRD has the potential to become a powerful tool for postmarketing drug safety studies. Its use for the early detection of safety signals in the general population of recently released respiratory drugs seems feasible. The opportunity for selection of population controls is most attractive.

Other current ongoing or planned COPD efforts in the GPRD include the automatic determination of cause of death in COPD individuals, the assessment of patterns of comorbidities and its changes, the determination of risk of fractures in COPD patients according to use of respiratory drugs, and the development of the clinical epidemiology and natural history of lung/bronchial cancer in COPD patients to enable future pharmacoepidemiology studies.

Discussion
COULTAS: I would like to mention the possibility of diagnostic misclassification, confounding by indication, and the effect of socioeconomic status in this study. You tried to validate by GP self-report of diagnosis, but with no objective information. As part of a randomised trial I am conducting, I took administrative data from our primary care network to identify patients aged 45 yrs with COPD-related ICD-9 code (GP diagnosed). We randomly selected persons and sent them postcard questionnaires about symptoms and smoking, and conducted spirometry in the home. We excluded those diagnosed with asthma. We excluded 43% based on lack of symptoms or smoking and 15% based on spirometry. Of the 426 persons with a clinical diagnosis of COPD, moderate-to-severe COPD was confirmed in only 42%. These results suggest that COPD may be overdiagnosed when using administrative data that rely on GP diagnosis.

I also attempted to look at confounding by indication by comparing the GPRD reference group with our validated COPD data. The use of short-acting β-agonists and anticholinergics in the reference group in the GPRD population is quite low. Is there something about that population that the GPs were treating differently? Is it possible that these patients had severe underlying concomitant disease? I believe this is indirect evidence of confounding by indication.

SORIANO: With regard to socioeconomic data, to ensure confidentiality, the pharmaceutical industry does not have access to geographic area or other proxies of socioeconomic information.

ERNST: How do you account for immortal time?

SORIANO: We start assessing the window of exposure within 90 days. For example, an individual is discharged from the hospital, goes to the GP next week and is prescribed an ICS, then dies within 1 week. That individual is excluded because the death occurred within the 90-day immortal time period from hospital discharge. In a sensitivity analysis, by changing the window of exposure (30 or 90 days), excluding or including deaths or COPD hospitalisation 30 or 90 days after discharge, the results were maintained or even more beneficial for combination treatment.

ERNST: Which of these analyses showed more benefits?

SORIANO: The combination became more beneficial when we included deaths that occurred within the first 90 days, but I have no explanation for this.

PRICE: Was the index date the same in all four groups? It is likely that you have reduced your power rather than increased it.

SORIANO: This could be a secondary effect of the use of ICS, long-acting β-agonists, or both. Although we included the year of entry, it was a factor that was irrelevant. It did not explain any change within the 3 yrs of follow-up. We did many analyses by practice, matching one individual in the FP, salmeterol or combination group with an individual in the same practice in the reference group and the results remained the same. That is why we think the analyses and results are robust.

BOURBEAU: In this database, the drug is collected from what the physician reported that (s)he is prescribing and not the actual filled prescription, unlike D.D. Sin's database. So you cannot extrapolate and say that this is representative of what the patient is taking over the year. In D.D. Sin's database, you still do not know if the patient is taking the drug, but it is one step closer to patient drug compliance.

PRICE: In our practice, we found that >95% of patients were filling their prescriptions for ICS across the board. This system works in the UK because they can get the prescriptions when they ask for them. They do not come to see us each time to get a prescription, so they have already chosen to get that prescription issued to them, which is different than what is seen in the USA. There are some data on refill rates for computer prescriptions, which suggest that >90% of prescriptions are filled 150.

SIN: There is likely to be a gap between prescription dispensation and medication consumption. However, in practical terms, that gap is likely to bias the result toward the null value. I think I am less bothered by the gap issue than by some other issues.

	The Saskatchewan Experience

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
J. Bourbeau

Summary
The present study used the Saskatchewan administrative database to assess the effect of inhaled corticosteroid (ICS) therapy on hospitalisation related to chronic obstructive pulmonary disease (COPD) exacerbation. A nested case-control strategy within this cohort of patients (n=843) was used to allow assessment of exposure to ICS in relation to the timing of a rehospitalisation, and tight matching of subjects to compare patients of similar severity. Unlike the results from previous studies, no benefit of ICS on reducing severe exacerbations requiring hospitalisation among patients with COPD could be shown.

Introduction
The Saskatchewan administrative database has been the source of numerous pharmacoepidemiological studies in the past. Studies that are very well known are those in patients with asthma 151, 152. These studies have been conclusive in showing a reduction of the risk of hospitalisation for patients prescribed inhaled corticosteroids (ICS) compared with those not on ICS. More recently, the Saskatchewan database has been used for the construction of a cohort of chronic obstructive pulmonary disease (COPD) patients after first hospitalisation, using a similar research design as in the asthma studies, to assess the effect of ICS in preventing hospital readmission 153. In this COPD study, the use of ICS, even in patients that were prescribed high doses (800 µg·day^-1 of beclomethasone equivalent), could not be demonstrated to reduce the risk of severe acute exacerbations requiring hospital admission.

The Saskatchewan database
The computerised administrative databases of the Saskatchewan universal healthcare insurance system were the source of information for this COPD study. The administrative database of the Saskatchewan is unique in Canada as it includes all patients in the general population receiving medications commonly prescribed for the treatment of COPD and hospitalised for this condition. Since 1975, with the exception of registered First Nations Canadians, members of the Royal Canadian Mounted Police and Armed Forces, Saskatchewan residents have been issued a unique health services number, which is used to obtain healthcare, including reimbursement of prescription drugs, in the universal health programmes offered by the provincial government. One of the advantages of this database is that it has been used extensively in the past and has been shown to be both comprehensive and valid 129. In this database, the assessment of the use of any drugs is considered complete because the provincial drug plan is the primary insurer of Saskatchewan formulary benefits (therefore, claims must be processed through the drug plan's online system) and the drugs of interest are listed on the formulary. Furthermore, the information related to the drugs represent dispensed prescription by the patient, which is likely to be more accurate than patient- or physician-reporting drug prescription. Finally, the prescription drug database is submitted to several validation checks.

The diagnostic and treatment classification system used in the Saskatchewan COPD study was the World Health Organization International Classification of Diseases, ninth revision codes 490, 491, 496. A previous validation study of the Saskatchewan databases has shown the hospitalisation diagnosis of COPD to be quite accurate 129. However, it has been recognised that this is true so long as no attempt is made to differentiate asthma from COPD and that both disorders are considered together. This confusion in the diagnosis with a risk of misclassification bias is common to all administrative databases although it is rarely recognised by the reader. Everyone will understand that including an asthmatic in a cohort of COPD patients may favourably influence the response, especially with pharmacological treatment such as ICS. The consequence will be that a clinician may generalise the results in their practice to a population of COPD patients that is completely different than the population originally under study. In the Saskatchewan COPD study, the importance of this problem has been recognised and the confusion limited. The cohort has been limited to the subjects with onset of treatment after the age of 55 yrs and excluded patients with prior asthma therapy, specifically those prescribed cromolyn, nedocromil, or ketotifen, nasal or oral ICS in the prior 5 yrs. Patients in the final cohort were those who were discharged from a first hospitalisation for COPD.

Methods
In the Saskatchewan COPD study, one main difference with other pharmacoepidemiological studies was the use of a nested case-control strategy within the original cohort of COPD patients (after their first hospitalisation for exacerbation). This is an important point of this study. It has allowed the assessment of exposure within specific windows of time in relation to the outcomes of interest. For example, the window of time of exposure could be examined just a few months before hospitalisation for acute exacerbation. These patients would be expected to benefit more from a drug treatment such as ICS than those who have stopped taking their medication long before the study outcome. Using a cohort analysis, as in most pharmacoepidemiology studies in COPD, would be restrictive in this matter. Patients defined in the exposed group will still be considered to be exposed without knowing if they have stopped their medication during follow-up. In addition, patients defined in the nonexposed group could very well start to take ICS during the follow-up. These potential biases related to inappropriate allocation of exposure, possibly present in cohort study design, were brought to attention in a recent publication by Suissa 120. In the same paper, another potential bias called "immortal time bias" was explicitly developed; although, this bias was not recognised until recently.

In the Saskatchewan study, a conditional logistic regression with matched case-control sets was used, and adjustments for age, sex, number of hospitalisations for health problems other than COPD (to account for comorbidities), calendar year oral and inhaled bronchodilators, oral corticosteroids, and antibiotics in the year prior to rehospitalisation were made. Confounding by indication remains a common bias in this study as it is present in all observational studies.

Results/interpretation
Using the Saskatchewan database with a nested case-control strategy within a cohort of COPD patients, hospitalised for exacerbation, did not show ICS to reduce the risk of rehospitalisation. These results are, therefore, not consistent with previous reports of a significant benefit of ICS in reducing rehospitalisation for COPD 121, 141. Recently, many potential biases that could be seen in certain observational studies have been brought to attention 120. There are still uncertainties regarding the effects of ICS and results from other pharmacoepidemiology studies should be interpreted very cautiously.

Large ongoing randomised controlled trials will help to clarify this and other issues. It is also vital that understanding of the mechanisms responsible for inflammatory response in COPD and what treatments are effective in suppressing inflammation are improved.

Discussion
VOLLMER: My first comment regards the definition of the study population and whether or not these patients are truly COPD patients. Because there was no screening for asthma in this study, we cannot be sure that these patients do not have an asthma component.

BOURBEAU: I am not so worried about there being asthmatics in the group. This would have been more of a concern if we had demonstrated a protective effect of ICS on hospitalisation. But, what kind of COPD patient have we selected for here? If these results are true, then we do not want to generalise this information to the entire population of COPD. The message here is that you cannot show a protective effect in preventing a second hospitalisation in this particular population of COPD, when patients are treated with ICS after a first hospitalisation.

VOLLMER: Is overmatching a concern? Matching was done to control for severity (based on medications). Why not match on severity at time of first hospitalisation rather than just prior to the index event?

SUISSA: The matching issue would be relevant if the exposure was defined at the time of first hospitalisation. If you adjust for the very recent covariates, then, yes, there would be overmatching because these would be in the causal pathway. But in this study, we have the reverse. The exposure is very close to the time of the event, in fact, "current use" and the covariates come in the year before. What is the strength of the association between the exposure and all of these covariates on the outcome? We have to look at the crude effect. If the crude effect was protective with respect to the exposure and this adjustment made it go away, then one explanation could be that it is overmatching but, in fact, the crude effect was above one and the adjustment effect made it equal to one.

BOURBEAU: With regard to the suggestion of adjusting for the severity of the disease at the time of hospitalisation, I would not expect that it would be particularly helpful here. If you look at the way a COPD exacerbation is treated in the hospital across provinces, treatment is very similar so I would not expect to see much variation.

SUISSA: In matched studies, the population is defined by the cases, and the cases are more severe than the ones who are not readmitted to the hospital. In this way, when you match the cases with the controls from the cohort, you are actually selecting only the more severe ones. Therefore, the final sample will represent a more severe group than the entire cohort.

SORIANO: The mean age of this population was 76 yrs. Is this an age that is too old to show effectiveness?

ERNST: The average age of patients in D.D. Sin's study was the same.

SORIANO: The average age in our General Practice Research Database (GPRD) study was 69 yrs. Is it possible that your null hypothesis was negative and that your results should be inconclusive rather than negative?

BOURBEAU: I think we are not comparing the same populations here. The GPRD looked at a population of general practitioners. We are looking at a population of COPD patients that have been hospitalised. They will be older, for sure. I would expect a younger average age in the GPRD study. The average age of a COPD patient in most of the clinical trials is <65 yrs.

ERNST: I do not think that the conclusion of this study is that ICS do not work in COPD. The conclusion was that we were unable to show an effect in that population. It is important to understand why other investigators are showing an effect and what we can attribute these differences in results to.

WEISS: One of the interesting aspects of this study is the attempt to control severity in a way that is much more specific to COPD. That is a defining characteristic of this study versus using a number of comorbidities or the use of oxygen. We should not overlook that when discussing overmatching issues. What would it look like if we had those same kind of covariates in your cohort studies?

	Immortal time bias in cohort studies of inhaled corticosteroid effectiveness

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
S. Suissa

Summary
Immortal time bias, which can arise in observational studies using computerised databases, refers to a period of follow-up time in a cohort study during which no outcome events can occur. Data from the Saskatchewan chronic obstructive pulmonary disease (COPD) cohort using rehospitalisation for COPD as the outcome, and the design of the Sin and Tu study 121 were used to illustrate immortal time bias. The misclassified analysis based on a 90-day exposure period definition produced a rate ratio of 0.69. When the immortal person time (30.2 person-yrs) was correctly allocated to the nonuse group, the crude rate ratio increased to 0.82. Using the Cox proportional hazards regression model analysis with the misclassified exposure produced an adjusted rate ratio of 0.71 (95% confidence interval (CI) 0.55–0.91), while with the properly classified exposure, the adjusted rate ratio was 1.13 (95% CI 0.87–1.47). This bias increased with increasing length of the exposure period. This bias acts by artificially increasing the rate of the outcome among "unexposed" patients and, once corrected, resulted in no association between inhaled corticosteroid (ICS) use and COPD readmission. These data illustrate that immortal time bias is present in observational studies that evaluate the effectiveness of ICS in COPD and suggest that their reanalysis is warranted to account for this bias.

Introduction
There has been a recent surge in observational studies of drug effectiveness in asthma and chronic obstructive pulmonary disease (COPD). A large number of these studies employed a cohort approach and were conducted using computerised databases. The cohort design offers several options to define exposure and to analyse the resulting data. The preferred approach of these recent studies has been to use an intention-to-treat definition of exposure with an analysis where the exposure remains time-fixed. Most studies presented in this publication use this approach. A bias that is little known in pharmacoepidemiology and that has only been briefly documented is that of "immortal person time" 118, 119. It refers to a period of follow-up time in a cohort study during which no outcome events can occur. The improper account of such immortal person time will produce a biased estimate of the rate ratio.

To describe the role of immortal time bias in these recent studies, the approach taken by Sin and Tu 121 and replicated in several studies presented in this report was used. Briefly, the study by Sin and Tu 121 employed a retrospective cohort design to verify whether the use of inhaled corticosteroids (ICS) after discharge from hospital for COPD was effective at reducing the risk of COPD readmission or all-cause death. All 22,620 patients of >65 yrs of age admitted to hospital for COPD in Ontario, Canada, between April 1992 and March 1997 were identified from Ontario's health insurance database. The patients were followed from the date of discharge for up to 1 yr, or earlier if they were readmitted or died, in which case follow-up ceased at those points. The 11,481 patients who filled at least one prescription for an ICS during the first 90 days after discharge, or less if they had an event and were followed for <90 days, were classified as users. The remaining 11,139 who did not were considered as nonusers. An intent-to-treat analysis was performed on the basis of this classification using the proportional hazards regression model, accounting for several covariates, including comedication. The adjusted rate ratio of COPD readmission or all-cause death was 0.74 (95% confidence interval (CI) 0.71–0.78) for ICS use relative to nonuse. The adjusted rate ratio of COPD readmission was 0.76 (95% CI 0.71–0.80), while for all-cause death it was 0.71 (95% CI 0.65–0.78).

Immortal time bias is introduced in this design by the definition of exposure. A subject is considered exposed when an ICS is dispensed at any time during the 90-day period after discharge. Hence, to be exposed, a patient must survive until they receive that first prescription in that 90-day period. Thus, the span between the date of discharge and the date of the first prescription of ICS is immortal. Since no outcome events can occur during this immortal period, the survival function will necessarily be distorted. Moreover, this immortal period is considered "exposed" although the patient could not, in fact, become exposed until the first prescription in that 90-day period was dispensed. The question is then but to what extent this immortal time biases the results.

Data from the Saskatchewan COPD cohort was used to replicate the Sin and Tu 121 design and illustrate the immortal time bias. The impact of this bias on the rate ratio of COPD outcomes was quantified for ICS use. Since the exact replication of the Sin and Tu study 121 has already been published 120, for this illustration the study was replicated using another outcome, namely, hospitalisation for COPD.

Methods
The source cohort was described previously 120 and in J. Bourbeau's presentation (see above). Briefly, COPD patients were identified using the computerised databases from the universal health insurance programme of Saskatchewan, Canada 154. Focus was placed upon the cohort of patients who were hospitalised for COPD between January 1, 1990 and December 31, 1997. Cohort entry was taken as the date of discharge of the first hospitalisation with a primary diagnosis of COPD. All subjects were followed for up to 1 yr, their first readmission with a primary diagnosis of COPD, or death from any cause during the 1-yr follow-up, whichever occurred first. As done by Sin and Tu 121, all subjects who died within 30 days of cohort entry were excluded.

A subject was considered "exposed" if they received an ICS during the first 90 days of follow-up and "unexposed" otherwise. To illustrate the bias, the length of the 90-day time period selected to determine exposure was varied. Cox's proportional hazards models were used to estimate the rate ratio for the time-fixed exposure definition of Sin and Tu 121. The time-dependent definition of exposure that classified a subject as unexposed prior to filling the first ICS prescription was also considered.

Results
The cohort included 979 subjects, of whom 299 were rehospitalised for COPD during the 1-yr follow-up. During the first 90 days of follow-up, 39% were dispensed an ICS. As in the Sin and Tu 121 study, the users of ICS were much more likely to have received inhaled β-agonists, ipratropium bromide, oral corticosteroids, antibiotics, and xanthines during this same 90-day period.

Table 8 compares, using the person-time analysis, the rate of readmission between users and nonusers of ICS. The misclassified analysis based on the 90-day definition of exposure produces a rate ratio of 0.69. When the immortal (and unexposed) person time preceding exposure, which amounts to 30.2 person-yrs, is correctly allocated to the nonuse group, the crude rate ratio increases to 0.82.

The bias arises from misclassifying unexposed immortal time as exposed. To be considered exposed, a subject must be dispensed an ICS at anytime during a 90-day period after discharge. Thus, the exposed subject must necessarily survive until they receive their first prescription in the 90-day period. The span between the date of discharge and the date of the first prescription of ICS is thus immortal. Moreover, this immortal period is considered exposed by the intention-to-treat approach employed by Sin and Tu 121, although the patient could not in fact become exposed until the first prescription in that 90-day period was dispensed. As a result, the rate is underestimated in the exposed group and overestimated in the unexposed group. This will produce an underestimate of the rate ratio comparing the exposed to the unexposed.

The bias increases with increasing length of the exposure period. It was taken to be 90 days in the study of Sin and Tu 121, although no justification was offered for this choice. Clearly, the opportunity for longer immortal time increases as the exposure window increases. As a result, the biased rate ratios decreased from 1.05 to 0.57 as the length of the exposure window increased from 15 to 365 days. With the correct time-dependent analysis, the rate ratios were stable.

These analyses indicate that immortal time bias is present in observational studies of the effectiveness of ICS in COPD and may explain the apparent benefit found in those studies 121, 123, 142. These studies should present a reanalysis of their data using the proper time-dependent approach before they can be considered as part of the evidence concerning the effectiveness of ICS in COPD. In addition, other cohort studies using a similar design should also be assessed for the possibility of immortal time bias 141.

Discussion
SIN: If a drug (ICS) is effective in producing beneficial outcomes, you would think that they would be most beneficial during a time period when a patient is at the greatest risk of developing these outcomes. For most drugs, if you look at the survival curves, they diverge relatively quickly, then the effect is not as pronounced later on. So, one way to address this concern is to eliminate that period and increase immortal time, but the risk in doing that is that you may miss subtle but powerful effects during the period when the patient is most susceptible in developing those outcomes. We have to be careful about the assumptions that we make. There are pros and cons to every approach.

SUISSA: I had the same concern. Should it take a certain amount of time before the effect is seen? That is why I replicated these analyses using different time windows and examined the effect of the definition of exposure on the estimate of the rate ratio. If we used a 365-day time window, the equivalent rate ratio for exposed versus unexposed would be 0.57. If we used a 15-day time window, the rate ratio is 1.05. We looked at 30 days and 90 days and we noted that there was a gradient. When the time window was very short, the rate ratio was high and as the time window increases, rate ratio decreases.

SIN: Or perhaps using the 15-day window results in more exposure misclassification because you are not giving them sufficient time to fill their prescription.

SUISSA: It is hard to know. But, the results remain the same when you vary this time window, with time-dependent approach that is not subject to this immortal time bias.

SORIANO: Such a short window of exposure may create drug misclassification. This problem happens when you compare one drug exposure versus no drug exposure. I think that this would not happen if you used another drug as a reference group.

SUISSA: There is also disagreement about whether patients are exposed or unexposed during this immortal time.

VOLLMER: Yes, but, if you look at dispensing data for the 2–3 months prior to the first admission, you should be able to make some reasonable hypothesis about who is likely to be exposed during the immortal time.

SUISSA: So you would define exposure prior to the initial hospitalisation? That is interesting. If subjects were actually using ICS before hospitalisation, the inclusion of this information may help attenuate the bias from immortal time and exposure misclassification. This would need to be investigated in these studies.

Acknowledgement. Some of the data in this article of immortal time bias in COPD studies have already been published in the Americal Journal of Respiratory and Critical Care Medicine 120. This is a transcription of an oral presentation given at the workshop. Data presented in this study are based on de-identified data provided by the Saskatchewan Department of Health. The interpretation and conclusions contained herein do not necessarily represent those of the Government of Saskatchewan or the Saskatchewan Department of Health.

	The USA Lovelace Experience: examining systematic biases that affect the relationship between inhaled corticosteroids and survival in COPD

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
D.W. Mapel

Summary
To examine the relationship between the use of inhaled corticosteroids (ICS; with or without the use of salmeterol) and survival in patients with chronic obstructive pulmonary disease (COPD), information from the administrative databases of two regional health maintenance organisations, the Health Plan and Kaiser-Permanente, were used in a retrospective cohort analysis (n=1,685). After adjustment for age, medication use, sex, COPD severity, and comorbidities, COPD patients with mild asthma had significantly better survival than those without asthma (hazard ratio 0.672, p=0.0138). Hazard ratio estimates remain remarkably similar regardless of the time required to achieve the exposure criteria or changes in the duration of the exposure period, suggesting that the measurement errors are minor and that the positive association between ICS with or without salmeterol is robust. The survival benefit observed for ICS and salmeterol use did not appear to be significantly confounded by differences in smoking behaviour or comorbid conditions.

Introduction
The New Mexico Chronic Obstructive Pulmonary Disease (COPD) Outcomes Project was started to examine a variety of clinical issues, including how the diagnosis of COPD is usually made, the primary factors affecting utilisation among persons with this disease, and the relationships between airflow obstruction, quality of life and healthcare costs. The project has been conducted in three healthcare centres that represent the major types of delivery systems in the area: 1) Lovelace Health Systems, a regional for-profit system that provides care largely through the Lovelace Health Plan (LHP), its staff-and-network model health maintenance organisation (HMO); 2) the Albuquerque Veterans Administration Medical Center; and 3) the University of New Mexico Hospital, which serves as a tertiary referral centre and the primary care provider for the area.

Last year, the resources and experience that had been developed in examining COPD outcomes were used to conduct a project designed to examine the relationship between use of inhaled corticosteroids (ICS), with or without use of salmeterol, and survival. One of the goals of the project was to see if the results of the General Practice Research Database study could be reproduced 141. To improve the power of the study and the generalisability of the results, the LHP database was merged with that of the Kaiser-Permanente, Georgia (KP-GA), which is an HMO of a similar size based in Atlanta, GA.

One of the major challenges of any clinical or epidemiological study is to identify the biases affecting the relationship between the exposure and the outcome of interest and to control for these biases in the analysis when possible, or describe their potential influence when not possible. Systematic biases can usually be classified as either selection biases, measurement errors or confounding. One of the goals of the New Mexico COPD Outcomes Projects was to describe the biases that are likely to affect COPD clinical studies, especially those that are prone to affect cross-sectional or retrospective analyses. In this discussion, the major systematic biases considered in the COPD Survival Project are presented, and how they are likely to have affected the results and the results of similar projects from other databases.

Methods
Selection biases
The most problematic issue in working with COPD is that it is not a disease. It is a clinical syndrome encompassing several diseases that share a common physiological phenomenon, forced expiratory airflow obstruction. Emphysema and chronic bronchitis are traditionally accepted as the two major disease entities included in COPD because they are both caused by cigarette smoking and because most COPD patients have both to varying degrees. The heterogeneity caused by just these two diseases would be enough to make this syndrome a challenge, but practicing clinicians treating COPD also tend to include asthma, particularly asthma that has a large fixed airflow obstruction component, or asthma patients who smoke. Therefore, even if the situation is simplified by saying that each COPD patient either has or does not have each of these three lung diseases, eight possible categories are still available to stratify patients into (n^x=2³=8). Based on the current understanding of the pathophysiology and clinical presentation of these three diseases, it may be expected that persons with emphysema are likely to have a different response to ICS than those who have asthma, and that chronic bronchitis may or may not have a response depending on what inflammatory marker or clinical outcome is chosen. Thus, even in randomised, prospective, clinical trials for COPD, the categorisation of these patients is highly problematic, the results are still susceptible to selection biases and misclassification errors, and positive treatment effects can be overlooked if appropriate end-points are not selected.

To help understand the COPD population at Lovelace and how clinicians use these diagnostic terms associated with COPD, a detailed abstraction of the medical records of 200 randomly selected COPD patients who were treated by the LHP in 1998 was conducted, followed by an abstraction of every available medical record of the 2,600 COPD patients treated by the LHP in 1999. In these abstractions, the diagnostic term most commonly used by their primary caregiver or pulmonologist to describe each patient's lung disease (the primary diagnosis) has been captured, along with any other terms that have been used to characterise the disease (the secondary diagnosis). The majority of patients (62%) were simply labelled "COPD" without further elaboration. Emphysema is a term that was uncommonly recorded by clinicians (4%), which is interesting because it is the term most commonly listed by LHP COPD patients when asked about their lung disease. Chronic bronchitis, or chronic bronchitis with COPD, was the most commonly used term used for 20% of the LHP cohort, and asthma or asthma with COPD was used as the primary diagnosis for 9%. Almost all of these asthma patients were either never-smokers who had severe asthma with a fixed airflow obstruction component, or asthma patients who were current or former smokers. The remainder of this population (5%) was comprised of persons with lung diseases that may be associated with airflow obstruction but are not usually thought of as COPD, such as cystic fibrosis, bronchiectasis, pulmonary fibrosis, and obstructive sleep apnoea.

Clearly, there are misclassification errors that could affect how LHP COPD patients respond to ICS, but can the direction in which these errors will go be predicted? For example, ICS are well established as first-line therapy for asthma because they have been shown to improve a number of outcomes including survival. However, in previous longitudinal studies of COPD, persons with increased airway reactivity and responsiveness to bronchodilators (i.e. asthma features) have had significantly accelerated decline in lung function and poorer clinical outcomes. It is impossible therefore to predict a priori whether the benefits imparted by use of ICS in this "COPD with asthma" subpopulation will overcome their predisposition towards worse survival. Another factor that makes prediction of misclassification errors difficult is the problem of continued cigarette use. It is known from the chart abstraction that up to one-half of the LHP COPD patients were documented as using cigarettes at least occasionally at some point during the study year. Smoking status is highly correlated with age and stage of lung disease, so that younger COPD patients with mild disease are more likely to still be smoking than older COPD patients with severe disease. As previously noted, the COPD patients with asthma in this cohort tended to be either younger patients who were still smoking, or older persons with a fixed baseline airflow obstruction considered "mild" in COPD but severe in asthma, so an asthma patient's mortality risk could potentially be worse than that of an older generic "COPD" patient with moderate disease who managed to quit smoking more than a decade ago.

To adjust for having concomitant COPD and asthma in the analysis of ICS and survival, two different approaches were used. The first was to adjust for the presence and severity of asthma in the Cox proportional hazards model. Patients who had a primary discharge diagnosis of asthma were labelled as "severe" asthma, patients who had one or more emergency department encounters (not admitted) for asthma were labelled as "moderate" asthma, and those who had two or more clinic visits coded as asthma were labelled as "mild" disease.

Results
After adjustment for age, medication use (ICS and salmeterol), sex, COPD severity, and comorbidities, COPD patients with mild asthma had significantly better survival than those without asthma (hazard ratio (HR) 0.672, p=0.0138). Patients with moderate asthma (HR 0.997, p=0.99) and severe asthma (HR 0.97, p=0.91) had the same survival rate as those with no asthma. Thus, it appears that a concomitant diagnosis of asthma made by a clinic physician is likely to identify a patient who has a significant likelihood of better survival even after adjustment for confounding factors. Furthermore, because asthma is also a heterogeneous condition, some attention needs to be made to adjusting for the severity of the asthma component of the disease to avoid problems with residual confounding.

Another way of dealing with the asthma problem is to simply eliminate all patients who have any mention of asthma in their clinical record to see how this affects the HR estimates. In the database, this reduced the number of available patients by one-half (n=840), and most of the eliminated patients were from the ICS- and salmeterol-treated groups. However, the HR estimates for ICS, salmeterol, and combined ICS/salmeterol use changed very little, and the combination of ICS and salmeterol continued to be significantly associated with improved survival despite the substantial reduction in power (table 11).

Another concern is that patients with more severe COPD may be more or less likely prescribed ICS or salmeterol; therefore, disease severity could be a selection bias. Disease severity in COPD is usually described as per cent of predicted forced expiratory volume in one second (FEV₁), or stage of disease per the American Thoracic Society, European Respiratory Society, or Global Initiative for Chronic Obstructive Lung Disease staging systems. That poses a problem for most administrative databases, because, typically, very few clinical data outside of International Classification of Diseases, ninth revision (ICD-9) or ICD-10 codes are included. Furthermore, in cross-sectional or retrospective studies, it is unlikely that many COPD patients will have spirometry tests available that were obtained during the time interval of interest. An earlier project that was conducted using the LHP database suggests, however, that healthcare utilisation and the presence of comorbid conditions such as heart disease or cancer are better predictors of outcome than the degree of airflow obstruction. Using the clinical data abstracted from the chart review of 2,600 COPD patients (1,100 of whom had spirometry data), multivariate models were developed that identified the clinical characteristics present in calendar year 2000 that best predicted a poor outcome, such as high healthcare costs or death, in calendar year 2001. In bivariate analysis, per cent of predicted FEV₁ was only weakly associated with high future healthcare costs (p=0.05). Actual healthcare utilisation, such as the number of inpatient, outpatient, and urgent care visits for COPD during the year 2000, the use of supplemental oxygen, or the presence of a serious comorbid condition such as heart disease, were much stronger independent predictors of a poor outcome in the following year (p<0.001). In multivariate models that included age, healthcare utilisation, and the presence of serious comorbid conditions, per cent of predicted FEV₁ was no longer a significant predictive factor (p>0.10). Therefore, when trying to adjust for disease severity in longitudinal or prospective analyses of COPD, the best predictor of future behaviour is previous behaviour, and FEV₁ data are not essential.

To adjust for disease severity in this mortality study, the natural distribution of utilisation for inpatient, emergency department, and outpatient service encounters for COPD were examined in this cohort; then the group was stratified in each of these areas. It was found that after adjustment for age, health plan, medication use, comorbidities, and asthma, patients with 20 outpatient services for COPD in the year prior to exposure had significantly poorer survival (HR 3.5, p<0.0001), than those with 3–19 encounters (HR 2.33, p<0.0001).

Measurement errors
Although electronic pharmacy databases have the advantage of being able to describe when a specific drug was dispensed during a specific time interval, they still have limitations, and there are still potential systematic errors in their analysis that may affect the exposure/outcome relationship. Therefore, there are several potential measurement errors that must be considered when using pharmacy data.

Previous studies have set an arbitrary minimum exposure period of 90 days' worth of respiratory drug fills to help establish that their use is causally associated with survival 19, 141. This criterion implies several assumptions that are probably not accurate. First, it may be inferred that all patients have been fully compliant with their ICS or salmeterol treatment for 90 consecutive days and that they continue to be compliant with their treatment throughout the follow-up period. Neither is likely. Patients may take >1 yr to accumulate 90 days' worth of exposure, and if they stop their inhalers at day 91, they are still considered to be in the treated group. Secondly, it is known when a prescription fill is made, but in the Lovelace database, the number of inhalers dispensed or how the patient was instructed to take the medication is not known. Pharmacists generally dispense a 1-month supply at the initial fill and a 90-day supply on refills, but it is left to their discretion to interpret the prescription and decide how many inhalers will be needed for this time interval. Finally, the 90-day criterion is arbitrary, and it is not known how a shorter or longer exposure interval may affect the outcome.

Without further examination, it could be assumed that most of these factors are biases towards a null effect. Noncompliant patients who were dispensed just enough ICS or long-acting β-agonist to fulfill the minimum exposure criterion but who never really used the treatment regularly should have an outcome more similar to the never-treated group. Any observed difference between the observed and treated group is therefore likely to be a true effect because of the heterogeneity of the exposure among the treated.

The exposure measurement problem was examined in several ways. First, the mean and median number of days required to accumulate the 90 days of exposure by drug group were calculated, assuming that all prescription fills were for 30 days initially and 90 days at follow-up. The mean±sd and median cumulative time to reach 90 days of exposure in the exposed groups (mean 353±345, median 213; p<0.0001) are highly variable and overlap that of the comparison (i.e. short-acting bronchodilator) group (mean 276±274, median 169; p<0.0001). The combined ICS and long-acting β-agonist group is the only one that is statistically different from the comparison group (mean 576±450, median 421; p<0.0001 for both), and it is not surprising that it took substantially longer to fulfill the criteria in the combined group because there had to be at least 90 days of overlapping fills to be included. What effect the longer time to full exposure may have had in the ICS and long-acting β-agonist group is unclear, but it does not appear likely that the longer exposure time could solely explain the enhanced survival benefit for the combination of the two. Also, note that the mean and median time to achieve 90 days exposure in the ICS (mean 281±267, median 182) and comparison groups (mean 276±274, median 169) were very similar, and therefore unlikely to explain the survival benefit seen with ICS alone.

A similar comparison was conducted after eliminating all COPD patients with asthma (table 12). Again, the HRs have not substantially changed, and statistically significant categories continued in spite of a rather small study population.

View this table: [in this window] [in a new window]   Table 13 Comparison of hazard ratios (HRs) for patients with 60 days (n=1405), 90 days (n=1685), or 180 days (n=1162) of drug exposure

Confounding
Cigarette smoking can be both a confounding factor (there is a direct relationship between the number of cigarettes smoked and death) and an effect modifier (ICS may have a survival benefit only for those who have quit smoking). Measuring smoking exposure is problematic in any study due to the problems with self-reported behaviours, inconsistent medical record documentation, variability in inhalation patterns, and differences in the smoke contents among different brands of cigarettes. Current smoking is associated with a higher risk of most adverse events than former smoking, but as previously noted, there is a strong relationship between smoking status and age. In many longitudinal studies, current smoking status is associated with a survival advantage, which is usually due to the effects of residual confounding and measurement errors, and not some inherent health benefit from cigarettes.

For the survival analysis, all available medical records were abstracted and pack-yr smoking estimates were obtained whenever they were recorded or were estimated from the provided information. The pack-yr smoking histories of 73% of the cohort were able to be estimated. When examined in a survival model that included age, sex, drug treatment, and comorbid illnesses, smoking pack-yrs were not independently associated with survival, and they did not have a significant effect on the ICS/salmeterol survival relationship (data not shown). The lack of a relationship between smoking and survival in this model most likely indicates that smoking effects on survival are accounted for by other factors in the model. In any case, the survival benefit observed for ICS and salmeterol use does not appear to be significantly confounded by differences in smoking behaviour.

The other major confounding issue of interest is that of the presence of comorbid conditions. Comorbid illnesses could affect a physician's decision about whether or not to prescribe an ICS or salmeterol. For example, a physician may be reluctant to prescribe an ICS to a COPD patient who also has diabetes, or prescribe salmeterol to a COPD patient who also has arrhythmias. The Sin and Tu 121 study described a significantly higher prevalence of comorbid conditions in the non-ICS-treated group, although the difference was small and probably not clinically relevant.

Comorbid illnesses were adjusted for using the Deyo modification of the Charlson Index. One limitation of this method is that the Charlson Index was validated on a hospitalised population, thus its validity when used on an outpatient population is uncertain. This problem was addressed by calculating Charlson Index scores based on inpatient and outpatient codes separately, then including both in the model.

The Charlson Index was not a significant independent factor in the models unless the exposure criterion was cut to 60 days, and that is only seen in the outpatient score. This suggests that comorbidities are more likely to be responsible for the early deaths and that it is very reasonable to have a 90-day exposure requirement to eliminate any bias that may be introduced by other serious illnesses.

Interpretation
In many retrospective or cross-sectional surveys, the epidemiologist's challenge is to eliminate or adjust for systematic biases so that an exposure/disease relationship can be seen. Even with comprehensive population-based longitudinal data, such as that found in the LHP and KP-GA databases, the problems are often overwhelming, and large numbers of cases are needed to overcome signal variability. In examining the relationship between ICS and salmeterol use and survival in COPD, the starting point was a very strong unadjusted association, and the major potential sources of bias that could explain this relationship were examined. After adjusting for all of these factors, the relationship remains, and in some ways is even stronger. Owing to the limitations of this database and the analysis methods, the study cannot "prove" a causal relationship between ICS use, with or without salmeterol, and survival in COPD. Nevertheless, the strength of the relationship, the robustness of the findings, the similarities between the results and those of studies from other countries, and the similarities in these findings and those of randomised clinical trials suggest that this is a true relationship. Hopefully, the data from current randomised clinical trials, coupled with a better understanding of the inflammatory mechanisms in COPD, will resolve this question in the near future.

Discussion
WEISS: To bring up the patient care issue, I would like to present data from the US Veterans Administration dataset, which is a very stable cohort of patients with a high prevalence of COPD. We looked at patients with COPD as a primary outpatient visit and ended up with a dataset of 100,000 patients. We found that the mean number of days that patients went from hospitalisation to hospitalisation without any intervention by a physician visit was 71 days. That is terrible care. Is the health system watching over them during that time period? Are we looking too narrowly at outcomes? It is important to think about not just one drug stopping those events, but instead, a whole cascade of care activities happening or not happening to that patient that may, in fact, be more likely driving what that outcome is.

DAVIS: This analysis looked at mortality. Confounding by indication is a problem, but in this case we are seeing the reverse pattern in that you would assume that those taking ICS would be the sicker patients.

WEISS: We do not know that. We have not delved into the care patterns enough to know who these people are and why they are not getting the drugs. Maybe they are healthy or maybe they are terminal, that is the piece we do not understand. The problem remains about confounding by indication or by severity. What are the best measures of controlling for severity? I think that spirometry data will make the difference in the analysis. Although this study used some spirometry data, we do not really have a good way to work with spirometry in these databases, yet. These data can be very messy and it is not what you would see in a randomised controlled trial. For those of us who do not have spirometry data to work with, does the Charlson index help, especially if applied to outpatient data?

SIN: I think your spirometry data, particularly FEV₁, would help a lot, despite all the noise that could easily be put aside by the sheer power of the sample size. Secondly, Charlson is never used to control for severity, more for morbidities. Finally, in J. Bourbeau's study 155, all indicators seem to show that patients diagnosed with COPD and who are receiving ICS have greater severity of disease, not less. I think that empiric evidence suggests that confounding by indication, while we cannot fully explain that away, is unlikely to whittle away the differences.

WEISS: It is just one study and in just one population. We need more studies.

ERNST: With regard to the confounding by severity issue, I think there are differences between patients in the reference groups in this study. There are those who are getting short-acting bronchodilators and who are being treated according to the usual care of 5 or 10 yrs ago versus those patients who are now getting combination therapy, which may be usual care for 2005. I think there are differences in the practice patterns of physicians prescribing these different medications and I think there are differences in their patient populations.

VIEGI: It is surprising to see that smoking does not affect survival.

MAPEL: Smoking activity between the treatment groups was the same. So, in this study it does not explain the difference in survival. Also, how do you include a term for current or exsmoking? That is problematic because even in a chart abstraction, it is poorly reported, so how will you classify it? Perhaps, if we had a larger population it would have made a difference. But for this analysis, the differences in smoking behaviour do not explain survival benefit.

SORIANO: G. Viegi's point is very important. Others have explained the paradoxical effect of smoking in COPD patients in clinical trials. When COPD patients quit smoking, they die more frequently. The reason is that if you are a long-term smoker, you will only quit your addiction when you have a life-threatening event.

SUISSA: In this study, the group with the combined medications has a shorter follow-up, 1,000 days. The reference group has 1,800 days. I suspect this is because to enter the group with two drugs, you have to have first survived that period. Is it possible that this period of immortal time that was excluded outright should be considered and perhaps classified in either the ICS only, long-acting β-agonist only, or the unexposed group? The short follow-up is a clear indicator that there could have been substantial immortal time before then.

To understand this, let us say that someone starts 90 days of short-acting bronchodilator and then 2 yrs later they actually start long-acting β-agonist and ICS at the same time. There is a 2-yr period where nothing happened until they received their dual medications. At that point, they get classified in the double-drug group. Let us assume that they had died before they received this dual therapy. Where would they have been classified? In the reference short-acting bronchodilator group. But, because they did not die and they made it to the point where they could receive two medications, immortal time has been created, and this patient will now be classified into the combined treatment groups. Therefore, by excluding such immortal time from the reference exposure group, we overestimate the rate of death in the reference short-acting bronchodilator group and that creates a drop in its survival curve.

SIN: I would be a bit concerned about that patient who suddenly got 90 days of the combination towards the end. There is probably a reason why it was prescribed. It may suggest a worsening of the condition.

ERNST: I think the effect is independent of a confounding by indication. I think all that time before the combination therapy is dispensed is counted in the wrong group.

SORIANO: Yes, but the patient is older.

SUISSA: As a matter of fact, because of these confounding issues, i.e. the patients are older and they are getting worse, we would have expected the graph to be reversed. The ones getting both drugs are presented as having a worse prognosis than the ones getting one. We are not seeing this either crudely or after adjustments for age and severity. Therefore, I believe that the unaccounted immortal time issue is causing the bias we discussed.

MAPEL: What gets confusing is when you look at the combination group. They are different patients. They are 4–5 yrs younger. There is additional time that is required to get that dual exposure. If we started from the very first day they got a prescription and we did not throw anybody out who died early, it would end up looking the same.

ERNST: But you would still have the time before they got there to worry about. You are making the problem smaller but it is still there in this type of analysis.

WEISS: It is the cohort analysis that is limiting. The case control may allow you more flexibility.

	The USA PharMetrics Experience in COPD: hospitalisation risk and COPD medication

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
T. McLaughlin

Summary
The hospitalisation risk of various initial treatment regimens for chronic obstructive pulmonary disease was compared using the PharMetrics Integrated Outcomes Database® in a retrospective observational cohort study (n=4,038). Patients treated with inhaled corticosteroids (ICS) and salmeterol had the greatest hospitalisation risk reduction (74%, hazard ratio (HR) 0.26, 95% confidence interval (CI) 0.14–0.46) followed by patients treated with ICS and ipratropium (IPR) (45%, HR 0.55, 95% CI 0.41–0.74) and those treated with ICS (29%, HR 0.71, 95% CI 0.61–0.82). This benefit remained after excluding patients with concomitant asthma in all the cohorts except the ICS plus IPR cohort (HR 0.72, 95% CI 0.49–1.05).

Introduction
The objective of this study was to compare the hospitalisation risk of various initial treatment regimens for chronic obstructive pulmonary disease (COPD) in a retrospective observational cohort study. The PharMetrics Integrated Outcomes Database® was used for this analysis. This database consists of administrative claims data from enrollees of >20 large, managed, care plans across the USA and is representative of a commercially insured population. All covered medical claims (inpatient, outpatient, pharmacy) are captured for enrollees during the study period. Data were available for January 1997 to December 2000. The results of this analysis were presented at the American College of Chest Physicians' 2002 Annual Meeting in San Diego, CA, USA.

Methods
Inclusion/exclusion criteria
Patients of 45 yrs of age enrolled in one of 24 different managed care plans across the USA during 1998–1999 with a primary diagnosis of COPD (International Classification of Diseases, ninth revision clinical modification 491, 492, 496 were identified). Study patients were required to have 24 months of continuous enrolment centred on the index prescription claim for one of the following: ipratropium (IPR) or IPR plus albuterol fixed combination, salmeterol, or inhaled corticosteroid (ICS).

Study design
Five therapy cohorts were identified: IPR, salmeterol, ICS, ICS plus IPR, and ICS plus salmeterol based on medication utilisation within the first 60 days following the index date. Patients were excluded from the analysis if they received other treatments for COPD including cromolyn, theophylline and leukotriene-modifying agents, 12 months prior to their initial prescription.

Statistical analysis
Cox proportional hazard analysis comparing all the cohorts to the IPR cohort was performed, investigating the time to first COPD hospitalisation adjusting for age, sex, concomitant asthma and other respiratory disorders, previous hospitalisation, managed care plan, baseline oral steroid use and physician specialty.

Results
Of 4,038 patients identified, 1,975 (48%) were on IPR, 1,122 (27%) were on ICS, 416 (10%) were on ICS plus IPR, 291 (7%) were on salmeterol and 234 (6%) were on ICS plus salmeterol. Compared with IPR, all other treatment cohorts had a lower risk of COPD hospitalisation during the study period, except for salmeterol (hazard ratio (HR) 0.82, 95% confidence interval (CI) 0.59–1.13). Patients treated with ICS had a 29% lower risk of hospitalisation (HR 0.71, 95% CI 0.61–0.82), ICS plus IPR patients had a 45% lower risk (HR 0.55, 95% CI 0.41–0.74), while ICS plus salmeterol was better with a 74% risk reduction (HR 0.26, 95% CI 0.14–0.46). This benefit remained after excluding patients with concomitant asthma in all the cohorts except the ICS plus IPR cohort (HR 0.72, 95% CI 0.49–1.05).

Interpretation
The results of this analysis suggest that initial treatment with ICS, alone or in combination with salmeterol, compared with IPR, was associated with a significant decrease in the risk of COPD hospitalisation during the first 12 months of initial therapy independent of concomitant asthma diagnosis. The combination of ICS plus salmeterol, as an initial treatment for COPD, was associated with the greatest decrease in risk, suggesting that further work should focus on this combination to confirm these findings.

Discussion
PRICE: I think the outcomes we have been using, hospitalisations and deaths, may not be the best way forward if we are going to try to answer some of these questions of potential confounders. I would like to put to rest the issue of misdiagnosis and underdiagnosis of COPD. We are conducting a study in the UK in one general practice centre (12,000 patients), in patients of >50 yrs who are on respiratory medications. We compared diagnoses before and after formal spirometric assessment. The pre-diagnosis of asthma dropped substantially and the pre-diagnosis of COPD increased dramatically.

In another large database (43,000 patients), patients diagnosed with respiratory disease underwent spirometry and reversibility and were subsequently relabelled based on this information. We found that a substantial number of patients got relabelled as COPD patients, but very few patients got relabelled with asthma.

Also, we are getting a consistent theme that patients are clear about when they need to see a physician about their COPD. The decision for hospitalisation is really about seeing a doctor they do not know. That worries me about using hospitalisation as an outcome because we have lots of variables inputting into that, including the patient, the doctor, and the system. The decision to see the physician is more about the patient and less about the physician and the system. I would argue that healthcare consultations for COPD may be actually a stronger outcome to use. There is also a lot more of them, which will increase power and you can compare them before exposure.

Another potential confounding issue is age. In the Glenfield database, there were few individuals <60 yrs who fit into the Global Initiative for Chronic Obstructive Lung Disease 2 or 3 severity category. I think it is important that patients of <60 yrs be examined separately in these administrative databases. No one has given me a plausible reason why we are preventing hospitalisation and deaths using ICS and long-acting β-agonists in those younger patients if their COPD is unlikely to be severe.

HAGAN: I would like to challenge D. Price's paradigm on age. Of COPD patients who have a hospitalisation, 25% are dead in the next year and 2–3 die in the next 3 yrs. The benefit of using ICS is to reduce exacerbations, which drive deterioration of lung function and health status leading to hospitalisation. So if there is a benefit with ICS, they must be introduced early. So the opposite of what you are saying may be true. Younger patients would benefit most from ICS.

PRICE: In terms of exacerbation, if you are arguing for using these drugs in an earlier stage of the disease, then we need to be using outcomes that are actually going to show a difference.

SORIANO: I would also like to challenge the statement that "We cannot diagnose or use any individual with COPD before the age of 60." From a public health point of view, we are seeing younger patients with COPD. If you screen for COPD in younger patients and get a smoker to quit smoking early in life, you will change the natural history of COPD in that patient.

PRICE: But, when we are trying to understand the databases, I think we actually have a lot more asthma in the databases. We just have to do stratified analyses and we should be suspicious about the data we see in the younger age groups.

One of the major struggles that we are having as a group is that we are not sure whether our datasets are truly comparable. I would like to find an outcome that we could compare with pre-exposure. By looking at hospitalisation and deaths, we are making it almost impossible to compare.

WEISS: I agree that it would be nice to walk forward from mortality and hospitalisation and into a more detailed look at care. That could be exacerbations, at least by some definition that we would create. We were saying earlier that all-cause mortality made sense because half of COPD-related deaths are cardiovascular events. D.D. Sin was also hinting that he is looking into how well ICS may prevent cardiovascular deaths. What are we proposing that the ICS are doing for the other half of deaths? What is going on with these ICS that may be life-saving? Are they preventing respiratory failure or pneumonias? I think we need to look at those questions because if we are thinking that these are the outcomes, we better have some mechanistic way of linking exposure to outcome.

SULLIVAN: What is it about combination therapy from a mechanistic standpoint that makes it better than salmeterol or ICS alone?

SORIANO: Looking at the survival curves for these data, I am not convinced that the combination treatment with IPR and ICS is significant. However, I am pretty convinced that the addition of salmeterol is very significant. As we have heard, it is possible that this could be due to a design issue, that the use of two drugs could be expanding and biasing the effect. However, it should be noted that there is molecular research that shows complementary mechanisms of action between long-acting β-agonists and ICS.

ERNST: Part of why the combination of IPR is very different from that of salmeterol is that the IPR group is likely a very different population that marks a group of older people with COPD. It is a much more COPD-specific drug.

FABBRI: If you are thinking about mechanism and start speculating about the interactions between COPD and cardiovascular pathophysiology, you have to be careful before translating a molecular interaction into a clinically relevant effect. I have shown you the results of several clinical trials demonstrating an almost identical effect on moderate and/or severe exacerbations of drugs with completely different mechanisms of action, for example, long-acting β₂-agonists, anticholinergics, and ICS.

VIEGI: In the last 10 yrs, there has been a lot of information about air pollution studies and fine particles. They have prompted a series of studies on the mechanisms involved. Those who die from air pollution are the sick and the elderly. These particles are proinflammatory agents that give rise to a cascade of inflammatory cytokines that have an important effect on cardiac arrhythmias. This type of research may be useful in understanding the results of the study on the beneficial effect of ICS and long-acting β-agonists from a mechanistic point of view.

SULLIVAN: I would like to discuss this issue of treatment concomitance and exposure. There is a window defined within, which needs to have the presence of two of the combination products to define combination exposure. But, there is never any evaluation downstream about whether the patient continues on both medications. Is it truly concomitant consumption or just an overlap of two prescriptions?

McLAUGHLIN: That is a limitation of this study design. The solution is that we would have to follow all patients for a longer period of time and ensure that they are adherent to medications.

ERNST: Another solution is to change the study design. If you can use a nested case-control approach, you can see the pattern of exposure in relationship to the events of interest. We did this for asthma years ago showing that it was not the number of prescriptions of β-agonists that was associated with adverse outcomes, but rather the pattern of use. There are all sorts of ways of doing this if you are willing to give up the classic design.

McLAUGHLIN: Would the nested case-control answer S. Sullivan's question about true concomitant use?

ERNST: You can look over the last year and define regularity of use in both products and then you can compare with another group of compliant patients, if you want. Starting from the event and going back to the exposure allows you to do all those things without creating immortal time. When you try to do those things in the proportional hazard model, you are adding immortal time as you are waiting for these things to happen.

SUISSA: It is important to note that to be able to receive these two medications, the patient needs to survive at most 60 days. That 60-day period is crucial and appears to impact on the analysis, if you look at the early part of the survival curves. The subjects with the full 60-day period had zero events during that time. That is what defined them into coming into the study, whereas the others are allowed to have events during that time period. That may explain the large discrepancy at the beginning of follow-up and would be important to report.

McLAUGHLIN: We actually did exclude all patients who had an event within the first 60 days and the results were similar.

SIN: L. Goldman recommends the use of propensity scores for large datasets with lots of variables to adjust for residual compounding. None of these studies used propensity adjustment. What does the group think about using propensity scores to further adjust for various confounding factors and should epidemiologists mandate propensity scores for future studies that use observational data?

SUISSA: It would not help at all in our context. There is little difference between adjusting for the many variables that were adjusted for in all of these studies and adjustment by propensity score measures. Propensity scores were designed specifically for studies that have small numbers of subjects (e.g. 150 subjects) and where there are many potential confounding variables. With studies using these large databases, propensity scores would not be helpful.

	The Integrated Primary Care Information Experience in COPD in the Netherlands

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
M.C.J.M. Sturkenboom

Summary
Using an intent-to-treat analysis approach, a retrospective study (n=1,096) compared the survival rates of treatment with salmeterol, fluticasone, fluticasone plus salmeterol, and a reference group utilising data from the Integrated Primary Care Information database in the Netherlands, a longitudinal observational database consisting of computer-based patient records of >150 general practitioners in the Netherlands. Treatment with salmeterol plus fluticasone (adjusted risk ratio (RRadj) 0.37, 95% confidence interval (CI) 0.21–0.67) or fluticasone alone (RRadj 0.34, 95% CI 0.18–0.66) resulted in significant survival advantage compared with the reference group alone. This result did not change in a series of sensitivity analyses that aimed to study the effect of quality care, disease misclassification, or exposure misclassification.

Introduction
The Dutch system of healthcare is based on general practitioners (GPs) who practice in the community but not in the hospital, referring ambulatory patients to specialists for outpatient or inpatient care. Specialists report their findings to the GP, who acts as a gatekeeper. Approximately 90% of the patients' presenting problems are addressed by the GP. Full-time staff physicians who are specialists of various kinds provide hospital care. Medical care, including prescription drugs, is essentially paid for by a combination of public and private insurers. The public insurers are regional agencies collectively called the Sickfunds. They provide insurance coverage for 60% of the population, i.e. generally those who fall below an annual income level (25,000). A flat fee per year reimburses the GP for Sickfund patients; for privately insured patients the GP is reimbursed on a fee for service basis. Patients should be registered with one GP but are free to change, which happens infrequently and nearly always because the patient moves out of the area. When a patient transfers, so does the record. More than 75% of the patients will visit their GP at least once a year 156.

The Dutch hypothesis has claimed for several years that asthma and chronic obstructive pulmonary disease (COPD) were obstructive lung diseases with a similar origin and recommended that these diseases be called chronic nonspecific lung disease (CARA) 157. Therefore, many GPs have diagnosed patients with CARA without specifying asthma or COPD in the past. Starting from the second half of the 1990s, GPs began distinguishing COPD and asthma but the change in habit has been slow. The Dutch hypothesis is problematic for the retrospective identification of COPD patients and requires specific identification and validation algorithms.

The Dutch GPs' treatment guidelines for COPD recommend smoking cessation as a first intervention 158. Pharmacological treatment should start with short-acting bronchodilators (such as an anticholinergic or short-acting β₂-agonists such as salbutamol, terbutaline, or fenoterol). If effectiveness is not satisfactory after 2 weeks, a change of bronchodilator should be considered. If the second is not effective, both bronchodilators should be given together. If use of short-acting bronchodilators is not effective in ameliorating nightly dyspnoea, a long-acting β-agonist should be prescribed. In case of insufficient effect, the addition of xanthine derivatives should be considered, which is usually limited to severe cases. Initiation of xanthines should be performed by a pulmonary physician. The use of inhaled corticosteroids (ICS) or acetylcysteine is not generally recommended and inhalational steroids should be reserved for patients with an atopic constitution or patients with >3 exacerbations per year.

The type of care and the referral of patients to specialists are based on the severity (classified by European Respiratory Society criteria) of COPD, level of dyspnoea, diagnostic problems, and the effectiveness in controlling the disease. Patients should be referred to a specialist in the following circumstances: 1) COPD is suspected in subjects <50 yrs of age; 2) doubt about COPD or heart failure as the origin of worsening dyspnoea; 3) forced expiratory volume in one second <50%; 4) progressive worsening under maximum treatment; 5) unintended loss of weight; 6) >2 exacerbations per year; and 7) an indication for oxygen therapy (long-term oxygen therapy).

The GP receives letters from the specialist that report on the findings and undertaken actions. Specialists often issue prescriptions that are not registered by the GP but which will be repeated by the GP. These (first) specialist prescriptions may go unnoticed in a database that is based on GP records.

In the Netherlands, the "healthy" elderly may transfer to homes for the elderly, in which they remain rather independent and keep their own GP. If they become ill, care-dependent patients are transferred to nursing homes where they receive care from an internal physician. This feature of Dutch healthcare may lead to the loss of patients in the end stage of life, particularly those who are chronically ill. If a patients' registration with the GP is (erroneously) not terminated upon transfer to a nursing home, the occurrence of death may be missed or the date of death may be reported with a delay.

Methods
Database
A retrospective study was conducted that utilised data from the Integrated Primary Care Information (IPCI) database in the Netherlands, a longitudinal observational database, with data from computer-based patient records of >150 GPs in the Netherlands.

In 1992, the IPCI was started by the Department of Medical Informatics of the Erasmus University Medical School (MIEUR), Rotterdam, initially in collaboration with IMS but independently from 1999 onwards. IPCI is a longitudinal observational database that contains data from computer-based patient records of a selected group of GPs throughout the Netherlands that voluntarily chose to supply data to the database 159. Practitioners control usage of their data and only receive a minimal reimbursement. The collaborating GPs are comparable with other Dutch GPs regarding age and sex.

As of December 2002, there are 93 active practices belonging to 118 GPs that are providing ongoing data to the database. The first practice was recruited into the IPCI project in 1994. Practices have therefore been supplying data for varying periods of time. The database now contains information on 485,000 patients. This is the cumulative number of patients who have been part of the dynamic cohort of registered patients. Turnover occurs as patients move and transfer to new practices. The records of "transferred out" patients remain on the database and are available for retrospective study with the appropriate time periods. As of December 2002 there were >370,000 active patients registered with the collaborating GPs, 49.1% were male, 57% were insured through the Sickfund, and the mean±sd age was 37.7±21.9 yrs. On average, patients were only 1 yr younger than the average of the Dutch population in 2001. In addition, the percentage of persons insured through private insurance was higher than the Dutch average.

The database contains identification information (date of birth, sex, patient identification, insurance, date of registration and transferring out, date of death), notes (subjective and assessment text), prescriptions, and indications for therapy, physical findings, referrals, hospitalisations, and laboratory values, which have been stored directly onto computer. MIEUR has implemented a research-specific module in the software that requires linkage of an indication to each prescription. The International Classification of Primary Care (ICPC) is the coding system for patient complaints and diagnoses, but diagnoses and complaints can also be entered as free text that is available as raw data 160. Prescription data such as product name, quantity dispensed, dosage regimens, strength, and indication are entered into the computer to produce printed prescriptions 159. The National Database of drugs, maintained by the Z-index, enables the coding of prescriptions, according to the Anatomical Therapeutic Chemical (ATC) classification scheme recommended by the World Health Organization 161.

Data are downloaded on a monthly basis and the information is sent to the gatekeeper who ensures all information is anonymous before further access is provided. Access to original medical records (discharge letters of hospitals) and administration of questionnaires to GPs is possible through the gatekeeper after approval of the Steering Committee.

Data accumulated in the IPCI database have proven to be of high quality and suitable for epidemiological and pharmacoepidemiological research 159.

Study population and follow-up
For comparison of mortality between patients treated with fluticasone, salmeterol, or other bronchodilators, the population comprised all persons >50 yrs of age newly diagnosed with COPD who had received at least two prescriptions of salmeterol, or at least two prescriptions of a short-acting β-agonist, xanthines, anticholinergics, or combined bronchodilators, but had not used ICS or long-acting β-agonists since diagnosis with COPD. Patients were excluded if they had <6 months of follow-up. In a sensitivity analysis, patients with <6 months of follow-up were included.

Identification and definition of chronic obstructive pulmonary disease
All subjects who had a potential diagnosis of COPD were identified by an automated database search on ICPC-coded diagnoses (ICPC codes R91 and R95), markers (CARA, COPD) and string searches that included the words: "CARA", "bronchitis", "COPD", "obstructive" "lung disease" and "emphysema", "R91", and "R95".

In a second step, all records with denials of COPD (such as no COPD, COPD-) and records that indicated acute bronchitis or bronchiectasis were excluded. In a third step, a manual review of the records that included words such as family, father, mother, brother, etc. was conducted. All records that suggested COPD in a family member rather than the patient, and all records that did not clearly indicate the existence of COPD, chronic bronchitis, emphysema, or obstructive lung disease were excluded. Subsequently, patients were classified as having a diagnosis of COPD or not. Patients without any record of COPD but instead one indicating CARA, were classified as COPD if they were >45 yrs of age, if they were <45 yrs of age they were considered asthmatics. The first record consistent with COPD was used as the index date (onset of COPD).

To conduct sensitivity analyses, COPD patients were further divided into probable or possible categories. Since all COPD patients should be treated with bronchodilators (according to the recommendations), all possible COPD patients who had not been diagnosed by a lung physician and had no prescription for any type of bronchodilator were excluded.

Cases were classified as "probable" COPD if the medical record comprised a coded diagnosis of COPD or chronic bronchitis (R91, R95), or a specialist-based diagnosis of COPD (COPD diagnosis mentioned in specialist letter) and if they received at least one prescription for a bronchodilating drug (β-agonists, anticholinergics) or xanthine derivative (ATC: R03AC, R03AH, R03AK, R03BB, R03C, R03DA).

Subjects were classified as "possible" COPD if a specialist letter mentioned COPD, emphysema, or chronic obstructive lung disease but they had no GP prescription of a bronchodilator or xanthine derivative (these could be patients treated only by the lung physician). If patients had a mention of COPD but no specialist diagnosis and no use of bronchodilators, the patient was not considered a COPD patient.

Deaths
Death and the date of death were identified by looking at the population file that included the date of death, and by a search in the patient medical record on the terms "dead", "death", "died". A manual review was conducted to validate the occurrence of death in the person at stake and the date of death.

Exposure to drugs
In line with an earlier study conducted in the General Practice Research Database, similar criteria to classify drug exposure were applied 141. Basically an intention-to-treat approach was used, which had an additional advantage in this setting since it reduces the problem of exposure misclassification due to prescriptions written by a specialist. In the exposure groups all persons needed to be newly diagnosed with COPD and 50 yrs of age and treated with at least two bronchodilators (short-acting β-agonists, anticholinergics, xanthines or combined bronchodilators). Within this group of COPD persons regularly treated with bronchodilators, the following groups of patients were distinguished. 1) Reference: patients who never had a long-acting β-agonist (salmeterol/formoterol:R03AC12/R03AC13/R03AK06/R03AK07) and never had an ICS (R03BA). Start of follow-up was identified as the first prescription of bronchodilators after COPD diagnosis and the end of follow-up was death, transferring out or last data draw down, whichever date was earliest. 2) Salmeterol-only group: patients with at least two prescriptions for salmeterol and one or less than one prescription for fluticasone. 3) Salmeterol plus fluticasone: all patients with at least two prescriptions for salmeterol and at least one overlapping prescription for fluticasone. 4) Fluticasone-only group: all patients without use of long-acting β-agonists and at least two prescriptions for fluticasone.

Start of follow-up was defined as the first prescription of salmeterol after COPD diagnosis among persons who used salmeterol alone, the latest of salmeterol or fluticasone among persons with the combination treatment (although 75% of salmeterol and fluticasone combined prescriptions were single dose units), and as the first fluticasone prescriptions in those persons who were part of the fluticasone group.

Covariates
For the overall mortality study in COPD patients versus non-COPD patients, the (modified) chronic comorbidity score was assessed over 1 yr prior to the index date. For the treatment mortality study, the chronic disease score in the 1 yr prior to the index date, respiratory treatment in the 6 months prior to the index date, year of diagnosis, duration of COPD prior to entry, presence of asthma in the patients' history, the severity of COPD, and the number of GP visits in the 6 months prior to the index date, referrals to a lung physician, and the use of oral corticosteroids during follow-up were all assessed.

Severity of chronic obstructive pulmonary disease
All patients were considered to have at least moderate-to-severe COPD in the treatment mortality study, since criteria for entry required at least regular treatment with bronchodilators. Patients were considered to suffer from severe COPD at baseline if they used oxygen therapy or nebulised therapy prior to baseline, if they were judged by the physician to have severe COPD (obtained from clinical notes), or if they had more than two prescriptions for antibiotics indicated for respiratory tract infections in the 6 months prior to baseline.

Statistical analysis
Standard survival analyses (Kaplan Meier) and Cox proportional hazards were conducted to compare survival between treatment groups assuming an intention-to-treat analysis and considering the persons without long-acting β-agonists and ICS as the reference group. Patients were only included if they had a follow-up of at least 6 months after entry, since treatment effects may not be expected directly after start of treatment. The survival was defined as the time period from 6 months after entry until the time of death. Follow-up time was censored at 3 yrs. Cox proportional hazard estimates were calculated by adjusting for all covariates that were univariately associated with mortality (p<0.10)

Results
A total of 1,096 newly diagnosed COPD patients were eligible for entry in the study as they met the inclusion criteria and had >6 months of follow-up after entry. The reference cohort comprised 371 persons, the salmeterol-only group (n=109), the salmeterol/fluticasone group (n=433) and the fluticasone-only group (n=179). Most of the persons in the combination group used the single unit combination (seretide). The reference group comprised fewer females and was older, but less frequently suffered from severe COPD compared with the index cohorts. In addition, the reference cohort suffered slightly less from other chronic comorbidities (ns) and had a diagnosis of asthma less often in their history. Regarding respiratory treatment in the 6 months prior to baseline, the reference group had significantly lower use of ICS, short-acting β₂-agonist, oral steroids, mucolytics and combined products. The GP contact rate was equal between the groups but the reference group had been referred to a specialist less frequently (statistically significant), which is an indicator of lower severity.

Treatment with salmeterol and fluticasone (adjusted risk ratio (RRadj) 0.37, 95% confidence interval (CI) 0.21–0.67) or fluticasone alone (RRadj 0.34, 95% CI 0.18–0.66) resulted in significant survival advantage compared with the reference group alone. This result did not change in a series of sensitivity analyses that aimed to study the effect of quality care, disease misclassification or exposure misclassification.

Interpretation
Advantages
The advantages are the availability of indication (antibiotics and oral corticosteroids were considered only if they had a respiratory indication), the GP is gatekeeper and the IPCI database contains information on specialist diagnoses.

Disadvantages
The disadvantages are that there are no complete data on spirometry, smoking, and alcohol use; CARA diagnosis in the past complicates distinction between asthma and COPD; confounding by severity and inability to measure severity completely (however, the inability to adjust for severity would lead to results in the other direction and cannot explain the observed result); missing data on prescriptions from specialists; few patients on salmeterol alone; short follow-up; loss to follow-up for patients transferring to nursing homes.

Discussion
MAPEL: One of the challenges of trying to present this information is that you are presenting it to different audiences. Clinicians do not like hazard ratios. They like to see Kaplan Meier curves. "Gamblers understand risks, physicians understand ratios, but only epidemiologists and statisticians think they understand risk ratios." So in a pulmonary journal, how do I communicate this back to the practicing pulmonologist? That is one of the advantages of working with intention-to-treat analyses, because intuitively you can deal with that. It is a little more confusing to present time-dependent analyses. I agree that it should be done to double-check our validity, but the difficulty is in communicating this information to clinicians. You need to be able to show that graphically in a figure because looking at risk ratios are meaningless.

ERNST: Being a clinician, I think it is very easy to communicate the concept that patients who took drug A in the past year did well while patients who took drug B did not. To me that is more relevant. I want to know what caused my patient to die or end up in the hospital. I do that by taking a history and that is what the nested case-control analysis does. That is more intuitive to the clinician instead of these forced curves that look pretty but are all the same and no one quite understands them.

MAPEL: What if you adjusted for what happened before exposure?

ERNST: I want to know what has happened before the event. What is relevant is the time before that event. The randomised controlled trial paradigm that we are married to (the intention-to-treat analyses) does not allow you to do that.

WEISS: Perhaps the two greatest effect modifications in your model were the comorbidity scores and whether or not they had a pulmonary physician. There are primary care physicians who do not treat COPD well because they do not think there is a good way to treat it. When patients exacerbate, they send them to the hospital. Lung specialists are used to treating these patients and that probably has some effect on mortality. One simple way to look at this in your database would be to look at shared care versus exclusive primary care versus exclusive pulmonary care.

STURKENBOOM: The problem of not being able to identify prescriptions by lung physicians is one of the major reasons why I did not start out with a case-control study or a time-dependent analysis. If you do want to do an as-treated analysis, you need accurate data on treatment episodes. Regarding different care of persons who are treated by a lung physician, we are exploring that further.

	Use of routine databases in health technology assessment: a policy maker's perspective

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
R. Taylor

Summary
The role of observational studies versus that of randomised clinical trials (RCTs) in the assessment of a drug's potential effectiveness is discussed. There are special circumstances in which the design of an observational database, rather than an RCT, would be desired. Decision analytical models and checklists for RCTs/observational studies can be used to assist policy makers in determining a drug's efficacy and to assess the quality of database studies.

Introduction
The traditional skepticism held by policy makers towards observational evidence is neatly summed up by the following quotation from the late A. Cochrane, "Observational evidence is clearly better than opinion, but it is thoroughly unsatisfactory. All research on the effectiveness of therapy was in this unfortunate state until the early 1950s. The only exceptions were drugs whose effects on immediate mortality were so obvious that no trials were necessary, such as insulin, sulphonamide, and penicillin" 162.

It has long been recognised that observational designs may lead to the overestimation of a drug's treatment effect. For example, Sacks et al. 163 examined the effect of anticoagulants on the mortality in myocardial infarction patients by comparing meta-analyses of observational (historical control) studies compared with randomised controlled trials (RCTs). Despite correcting for potential biases, the observational studies overestimated drug benefit more than three-fold, compared with the magnitude of drug benefit estimated by the RCTs. Therefore, when it comes to addressing the policy question of a drug's potential effectiveness, the gold standard form of evidence is, and will continue to be, the RCT.

Nevertheless, there is growing recognition of the role of observational studies, and routine databases specifically, in assisting policy decisions about drug therapies. There will always be situations where undertaking an RCT is not possible, either for ethical or practical reasons. A nonexperimental or observational design will therefore, instead, need to be employed as an alternative. However, to conclude that the role of databases stops here would be to substantially underplay their potential value.

Policy makers, particularly at a local level (e.g. a hospital or primary care trust), often want to know the potential service implications of the introduction of a new drug. For example, "What will be the uptake of the drug in practice? How adherent will patients be in taking the drug? What will be the impact on a healthcare budget?" Although addressed to some extent by RCTs, such questions can be better answered by routine databases.

There are a number of situations where efficacy and effectiveness cannot be fully addressed by the RCT. These include the assessment of treatments for rare outcomes (requiring prohibitively long follow-up), subgroup effects in specific patient groups (which an RCT is usually underpowered to assess), inability to consider all possible comparators, and the practical difficulty in assessing long-term outcomes 164. Only by using information from routine databases can the findings of RCTs be effectively extended to address these additional important issues. Increasingly, decision analytical models are being used to provide a framework whereby the results of an RCT(s) and an observational study (or studies) can be combined to address the policy maker's question regarding a drug 165.

The focus of this workshop has been to discuss appropriate methodological approaches to ensure that routine databases provide reliable and robust conclusions. An important extension to this is the development of checklists that policy makers can easily apply to routine database studies to assess their quality. Although there are published checklists for RCTs and for a number of observational designs 166, 167, there is a lack of such instruments for routine databases. It is important that such checklists address not only the internal validity (i.e. the identification of biases) in routine databases, but also their external validity (e.g. how representative are the patients and clinicians covered by the database), and the quality of reporting (e.g. prestatement of study hypotheses).

To conclude, with the drive towards policy makers assessing the "real world" impact of drugs, there is likely to be an increased use of routine databases alongside RCTs. The future utility and success of such database evidence depends on a number of factors: 1) harmonisation of methods and reporting of database analyses; 2) the improved compatibility and linking of current information systems; and 3) the adequate resourcing and funding of database initiatives.

	The TORCH study: towards a revolution in COPD health

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 
G.W. Hagan

Summary
Some of the methodological concerns identified in the previously mentioned research on the efficiency of respiratory drugs in chronic obstructive pulmonary disease (COPD) are intrinsic by the observational nature of epidemiological research. A large randomised controlled trial is currently being conducted. This trial has been powered to demonstrate an effect of respiratory drugs on COPD survival if any. The design and methods of this trial are summarised below.

Methods
Post hoc analysis of data from the Inhaled Steroids in Obstructive Lung Disease in Europe (ISOLDE) study suggests a reduced mortality rate in chronic obstructive pulmonary disease (COPD) subjects randomised to fluticasone propionate (FP) compared with placebo during the 3 yrs postrandomisation 110, 168. Although ISOLDE was not designed to investigate mortality, and therefore the study should be regarded as a pilot, the data provide a rationale on which to base a definitive mortality study. No prospective mortality data are available for subjects treated with salmeterol either. However, it can be hypothesised that the properties of salmeterol, particularly its cytoprotective effects, and hence its potential to reduce infective exacerbations, could have a significant impact on subject survival. If this is the case, there could be, at least, an additive effect of salmeterol and FP in reducing mortality in subjects with COPD. Therefore, the objectives of the Towards a Revolution in COPD Health (TORCH) study are to investigate the long-term effects of the salmeterol/FP combination product at a strength of 50/500 µg twice daily compared with salmeterol 50 µg twice daily alone and FP 500 µg twice daily alone, on survival of COPD subjects over 3 yrs of treatment. The primary efficacy of end-point is all-cause mortality. Secondary efficacy end-points include rate of severe or moderate COPD exacerbations and quality of life assessed by the St George's Respiratory Questionnaire.

TORCH is a multicentre, randomised, double-blind, parallel-group, placebo-controlled study conducted wordwide. Inclusion criteria are as follows: male or female outpatients aged 40–80 yrs with a baseline forced expiratory volume in one second (FEV₁) of <60% of predicted normal, an established clinical history of COPD (per European Respiratory Society consensus statement), current or exsmokers with a smoking history of 10 pack-yrs, poor reversibility of airflow obstruction defined as <10% increase in FEV₁ as a percentage of normal predicted, 30 min after inhalation of 400 µg salbutamol via metered dose inhaler and spacer, and baseline FEV₁/ forced vital capacity ratio 70%. With 6,000 patients randomised, TORCH should have sufficient sample size to determine whether FP, alone or in combination with salmeterol, has an effect in survival in COPD patients treated during 3 yrs. The first results are expected around 2006.

Future perspectives

Discussion
SUISSA: The issue of immortal time bias is fundamental to all the studies presented. We have seen in these studies three different types of designs. In the Sin and Tu approach 121, there is clearly a problem of misclassified exposure before the subject actually starts to be exposed, which is compounded by it being a period of immortal time. To better quantify the proper rate ratio estimate, we should reclassify the patients into the correct exposure categories so that we can minimise the amount of bias from misclassification of exposure. With the approach that J.B. Soriano took with the General Practice Research Database, immortal time was dealt with by removing it rather than misclassifying it. This is an improvement, but I believe that immortal time must still be accounted for in the denominator of the reference group. In essence, we should be evaluating people at the beginning of the first of their treatments and not at the beginning of the treatment of interest because during that early excluded time, they had to have survived to get there. Somehow, that time has to be accounted for using time-dependent exposures either with Cox models or with the nested case-control approach.

The second issue is the relationship between exposure and outcome. In the cohort approach that emulates a randomised controlled trial, we have an exposure at baseline and we associate it with an outcome that occurs later. Somehow, we have to be able to make a solid link between the outcome that occurs 3 yrs later and a baseline exposure to the treatment that we are studying. For example, J.B. Soriano has presented data showing that patients who were exposed to long-acting β-agonists and inhaled corticosteroid (ICS) combination therapy continue to use this therapy throughout the follow-up. Another approach, the nested case-control approach that J. Bourbeau presented, associates the outcome with current exposure within a certain period of time close to the outcome under study.

I believe these two issues must be addressed before we can say that these studies have no bias in their estimates. I challenge those who have the data available, to return to the databases, reclassify the patients accordingly, and properly account for immortal time. When these new results can be presented to us, we may have more confidence in concluding that ICS are effective in preventing mortality and hospital readmission.

PRICE: Some combination of treatment and care is associated with different outcomes. We need to try and disentangle these two. One way is to think about our outcome variables so we can compare our outcomes in our different populations before and after. Where we may go with the observational work is to end up with two different grades of exacerbations, moderate (general practitioner consults) and severe (hospitalisation/deaths). We could use the moderate exacerbation as our way to compare our two groups before and after. Correcting for centres may be difficult because of the size of the numbers involved. We may look at the possibility of prospective rather than retrospective studies possibly using higher quality routinely recorded data.

WEISS: The issue may not be between centres, but, it may be more about the type of provider and the shared relationship between provider and patient. How much of that is correlated with the concept of severity? I have seen four different versions of severity presented here. How much should we push to get spirometry data entered into this discussion?

VOLLMER: When thinking about the intention-to-treat versus nested design, we need to remember that we have an exposure variable (medication status) that changes over time. We also need to think about how we control for severity, which gets back to the issue of overmatching.

WEISS: M.C.J.M. Sturkenboom raised the issue of coincident disease. In her analysis, heart disease and diabetes were discussed. We are beginning to ask questions about mental health and depression, since depression is a predictor of mortality and highly coincident with COPD. Perhaps we should be allowing ourselves to look at not just one disease, but, a cluster of diseases and how those clusters actually work over time.

BOURBEAU: We have been talking about pharmacological treatment, but we have not considered nonpharmacological treatment. There are issues such as education, self-management, and pulmonary rehabilitation that are recognised to be effective treatments. Our database does not collect everything we need. We need to join our database with a more clinical database and find a way to integrate the pharmacological and nonpharmacological treatments.

VIEGI: Since we already have data on younger patients (i.e. age >45 yrs, people who are still working), one of the potential outcomes could be reduction of absenteeism after proper management of disease. The other suggestion is to have an integration of industry and public health resources to manage this very important research.

BURNEY: It has always been true that experimental studies are good at saying what is true and observational studies are better at saying what is more important. Both of these have their relevance and interpreting one without the other is difficult.

	REFERENCES

TOP Introduction Epidemiology of COPD: overview... Epidemiology of COPD: a... The burden of illness... Pharmacotherapy of COPD Possible sources of bias... The Ontario and Alberta... The UK General Practice... The Saskatchewan Experience Immortal time bias in... The USA Lovelace Experience:... The USA PharMetrics Experience... The Integrated Primary Care... Use of routine databases... The TORCH study: towards... REFERENCES

 

1. National Heart, Lung and Blood Institute. Morbidity & mortality: chartbook on cardiovascular, lung, and blood diseases. www.nhlbi.nih.gov/nhlbi/seiin/other/cht-book/htm. Bethesda, US Dept of Health and Human Services, Public Health Service, National Institutes of Health, 1998.
2. Centers for Disease Control and Prevention. Trends in ischemic heart disease mortality - United States, 1980–1988. MMWR 1992;41:548–556.[Medline] [Order article via Infotrieve]
3. Murray CJL, Lopez AD. Evidence-based health policy – lessons from the Global Burden of Disease Study. Science 1996;274:740–743.[Free Full Text]
4. Murray CJLLopez ADeditors. The Global Burden of Disease: A Comprehensive Assessment of Mortality and Disability from Diseases, Injuries, and Risk Factors in 1990 and Projected to 2020Cambridge, Harvard University Press, 1996.
5. Mannino DM, Homa DM, Akinbami LJ, Ford ES, Redd SC. Chronic obstructive pulmonary disease surveillance – United States, 1971–2000. Morb Mortal Wkly Rep Surveill Summ 2002;51:1–16.
6. US Surgeon General. The Health Consequences of Smoking: Chronic Obstructive Pulmonary Disease. Publication No. 84-50205Washington DC, US Dept of Health and Human Services, 1984.
7. 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 Int Med 2000;160:1683–1689.[Abstract/Free Full Text]
8. National Center for Health Statistics. Plan and operation of the Third National Health and Nutrition Examination Survey. 1988–1994. Publication PHS 94-1308Washington DC, US Dept of Health and Human Services, 1996.
9. World Health Organization. World Health Report 1999. http://www.who.int/whr/1999/en/report.htm. Geneva, World Health Organization, 1999.
10. World Bank. Curbing the Epidemic: Governments and the Economics of Tobacco Control. http://www1.worldbank.org/tobacco/reports.htm. 1999.
11. Administration on Aging. http://www.aoa.gov/aoa/stats/AgePop2050.html. Accessed March 2003.
12. National Heart, Lung and Blood Institute/World Health Organization Workshop Report. Global Initiative for Chronic Obstructive Lung Disease: global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease. National Institutes of Health Publication No. 2701Washington DC, US Dept of Health and Human Services, Public Health Service, National Institutes of Health, National Heart, Lung and Blood Institute, 2001;. www.goldcopd.com. 2003.
13. Celli BR, Halbert RJ, Isonaka S, Schau B. Population impact of different definitions of airway obstruction. Eur Respir J 2003;22:268–273.[Abstract/Free Full Text]
14. Hardie JA, Buist AS, Vollmer WM, Ellingsen I, Bakke PS, Mørkve O. Risk of over-diagnosis of COPD in asymptomatic elderly never-smokers. Eur Respir J 2002;20:1117–1122.[Abstract/Free Full Text]
15. National Center for Health Statistics. Current estimates from the National Health Interview Survey, United States, 1995. Publication No. 96-1527Washington DC, Dept of Health and Human Services, Public Health Service, Vital and Health Statistics, 1995.
16. Hnizdo E, Sullivan PA, Bang KM, Wagner G. Association between chronic obstructive pulmonary disease and employment by industry and occupation in the US population: a study of data from the Third National Health and Nutrition Examination Survey. Am J Epidemiol 2002;156:738–746.[Abstract/Free Full Text]
17. Siafakas NM, Vermeire P, Pride NB, et al., on behalf of the Task Force. Optimal assessment and management of chronic obstructive pulmonary disease (COPD). Eur Respir J 1995;8:1398–1420.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
18. Fabbri LM, Romagnoli M, Corbetta L, et al. Differences in airway inflammation in patients with fixed airflow obstruction due to asthma or chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2003;167:418–424.[Abstract/Free Full Text]
19. American Thoracic Society. Standards for the diagnosis and care of patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med 1995;152:S77–S121.
20. World Health Organization. Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death. Volume I. Ninth revisionGeneva, World Health Organization, 1977.
21. World Health Organization. International Statistical classification of Diseases and Related Health Problems. Tenth revisionGeneva, World Health Organization, 1992.
22. Murray CJL, Lopez AD. Alternative projections of mortality and disability by cause 1990–2020. Lancet 1997;349:1498–1504.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
23. Gulsvik A. Mortality in and prevalence of chronic obstructive pulmonary disease in different parts of Europe. Monaldi Arch Chest Dis 1999;54:160–162.[Medline] [Order article via Infotrieve]
24. Hurd SS. International efforts directed at attacking the problem of COPD. Chest 2000;117:336s–338s.
25. Pride NB, Soriano JB. Chronic obstructive pulmonary disease in the United Kingdom: trends in mortality, morbidity, and smoking. Curr Opin Pulm Med 2002;8:95–101.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
26. Dati Istituto Superiore di Sanità. La mortalità per causa in Italia: 1980–1998. www.mortalita.iss.it.
27. Camilli AE, Robbins DR, Lebowitz MD. Death certificate reporting of confirmed airways obstructive disease. Am J Epidemiol 1991;133:795–800.[Abstract/Free Full Text]
28. Izumi T. Chronic obstructive pulmonary disease in Japan. Curr Opin Pulm Med 2002;8:102–105.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
29. Vestbo J, Prescott E, Lange P, Schnohr P, Jensen G. Vital prognosis after hospitalization for COPD: a study of a random population sample. Respir Med 1998;92:772–776.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
30. Prescott E, Lange P, Vestbo J. Chronic mucus hypersecretion in COPD and death from pulmonary infection. Eur Respir J 1995;8:1333–1338.[Abstract]
31. Italian National Statistics Agency. Indagine multiscopo sulle famiglie "Condizioni di salute della popolazione". 1999–2000.
32. Dati del Ministero della Salute. Ricoveri Ospedalieri–Anno 2000. www.ministerosalute.it.
33. Feenstra TL, van Genugten ML, Hoogenveen RT, Wouters EF, Rutten-van Molken MP. The impact of aging and smoking on the future burden of chronic obstructive pulmonary disease: a model analysis in the Netherlands. Am J Respir Crit Care Med 2001;164:590–596.[Abstract/Free Full Text]
34. Viegi G, Pedreschi M, Baldacci S, et al. Prevalence rates of respiratory symptoms and diseases in general population samples of North and Central Italy. Int J Tuberc Lung Dis 1999;3:1034–1042.[Web of Science][Medline] [Order article via Infotrieve]
35. CIBA Foundation Guest Symposium. Terminology, definitions and classification of chronic pulmonary emphysema and related conditions. Thorax 1959;14:286–299.[Medline] [Order article via Infotrieve]
36. Viegi G, Paoletti P, Carrozzi L, et al. Prevalence rates of respiratory symptoms in Italian general population samples, exposed to different levels of air pollution. Environ Health Perspect 1991;94:95–99.[Web of Science][Medline] [Order article via Infotrieve]
37. Lundback B, Nystrom L, Rosenhall L, Stjernberg N. Obstructive lung disease in northern Sweden: respiratory symptoms assessed in a postal survey. Eur Respir J 1991;4:257–266.[Abstract]
38. Stang P, Lydick E, Silberman C, Kempel A, Keating ET. The prevalence of COPD: using smoking rates to estimate disease frequency in the general population. Chest 2000;117:354s–359s.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
39. Lindstrom M, Jonsson E, Larsson K, Lundback B. Underdiagnosis of chronic obstructive pulmonary disease in Northern Sweden. Int J Tuberc Lung Dis 2002;6:76–84.[Web of Science][Medline] [Order article via Infotrieve]
40. Pallasaho P, Lundback B, Meren M, et al. Prevalence and risk factors for asthma and chronic bronchitis in the capitals Helsinki, Stockholm, and Tallinn. Respir Med 2002;96:759–769.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
41. Huchon GJ, Vergnenegre A, Neukirch F, Brami G, Roche N, Preux PM. Chronic bronchitis among French adults: high prevalence and underdiagnosis. Eur Respir J 2002;20:806–812.[Abstract/Free Full Text]
42. Viegi G, Pedreschi M, Pistelli F, et al. Prevalence of airway obstruction in a general population sample. European Respiratory Society vs American Thoracic Society definition. Chest 2000;117:339s–345s.
43. Pauwels RA, Buist AS, Calverley PM, Jenkins CR, Hurd SS. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease. National Heart, Lung and Blood Institute/World Health Organization Global Initiative for Chronic Obstructive Pulmonary Disease (GOLD) Workshop summary. Am J Respir Crit Care Med 2001;163:1256–1276.[Free Full Text]
44. American Thoracic Society. Evaluation of impairment/disability secondary to respiratory disorders. American Thoracic Society. Am Rev Respir Dis 1986;133:1205–1209.[Web of Science][Medline] [Order article via Infotrieve]
45. Vestbo J, Lange P. Can GOLD Stage 0 provide information of prognostic value in chronic obstructive pulmonary disease?. Am J Respir Crit Care Med 2002;166:329–332.[Abstract/Free Full Text]
46. van den Boom G, van Schayck CP, van Mollen MP, et al. Active detection of chronic obstructive pulmonary disease and asthma in the general population. Results and economic consequences of the DIMCA program. Am J Respir Crit Care Med 1998;158:1730–1738.[Abstract/Free Full Text]
47. Zielinski J, Bednarek M. Know the Age of Your Lung Study Group: Early detection of COPD in a high-risk population using spirometric screening. Chest 2001;119:731–736.[Abstract/Free Full Text]
48. Czajkowska-Malinowska M, Nowinski A, Gorecka D, Zielinski J. Effects of spirometric screening in the community on smoking cessation. Pneumonol Alergol Pol 2001;69:524–529.[Medline] [Order article via Infotrieve]
49. Cerveri I, Accordini S, Verlato G, et al. European Community Respiratory Health Survey (ECRHS) Study Group. Variations in the prevalence across countries of chronic bronchitis and smoking habits in young adults. Eur Respir J 2001;18:85–92.[Abstract/Free Full Text]
50. Rennard S, Decramer M, Calverley PM, et al. Impact of COPD in North America and Europe in 2000: subjects' perspective of Confronting COPD International Survey. Eur Respir J 2002;20:799–805.[Abstract/Free Full Text]
51. Vestbo J, Prescott E, Lange P. Association of chronic mucus hypersecretion with FEV₁ decline and chronic obstructive pulmonary disease morbidity. Copenhagen City Heart Study Group. Am J Respir Crit Care Med 1996;153:1530–1535.[Abstract]
52. Prescott E, Lange P, Vestbo J. Socioeconomic status, lung function and admission to hospital for COPD: results from the Copenhagen City Heart Study. Eur Respir J 1999;13:1109–1114.[Abstract]
53. Dahl M, Tybjaerg-Hansen A, Vestbo J, Lange P, Nordestgaard BG. Elevated plasma fibrinogen associated with reduced pulmonary function and increased risk of chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2001;164:1008–1111.[Abstract/Free Full Text]
54. Dahl M, Tybjaerg-Hansen A, Lange P, Vestbo J, Nordestgaard BG. Change in lung function and morbidity from chronic obstructive pulmonary disease in alpha1-antitrypsin MZ heterozygotes: A longitudinal study of the general population. Ann Intern Med 2002;136:270–279.[Abstract/Free Full Text]
55. Prescott E, Almdal T, Mikkelsen KL, Tofteng CL, Vestbo J, Lange P. Prognostic value of weight change in chronic obstructive pulmonary disease: results from the Copenhagen City Heart Study. Eur Respir J 2002;20:539–544.[Abstract/Free Full Text]
56. Brogger J, Bakke P, Eide GE, Gulsvik A. Comparison of telephone and postal survey modes on respiratory symptoms and risk factors. Am J Epidemiol 2002;155:572–576.[Abstract/Free Full Text]
57. Welle I, Bakke PS, Eide GE, Fagerhol MK, Omenaas E, Gulsvik A. Increased circulating levels of alpha1-antitrypsin and calprotectin are associated with reduced gas diffusion in the lungs. Eur Respir J 2001;17:1105–1111.[Abstract/Free Full Text]
58. Humerfelt S, Gulsvik A, Skjaerven R, et al. Decline in FEV₁ and airflow limitation related to occupational exposures in men of an urban community. Eur Respir J 1993;6:1095–1103.[Abstract]
59. Sunyer J. Urban air pollution and chronic obstructive pulmonary disease: a review. Eur Respir J 2001;17:1024–1033.[Abstract/Free Full Text]
60. van den Boom G, Rutten-van Molken MP, Folgering H, van Weel C, van Schayck CP. The economic effects of screening for obstructive airway disease: an economic analysis of the DIMCA program. Prev Med 2000;30:302–308.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
61. Rutten van-Molken MP, Feenstra TL. The burden of asthma and chronic obstructive pulmonary disease: data from the Netherlands. Pharmacoeconomics 2001;19: Suppl. 2, 1–6.
62. Dal Negro R, Berto P, Tognella S, Quareni L. Global Outcomes in Lung Disease Study Group. Cost-of-illness of lung disease in the TriVeneto Region, Italy: the GOLD Study. Monaldi Arch Chest Dis 2002;57:3–9.[Medline] [Order article via Infotrieve]
63. Andersson F, Borg S, Jansson SA, et al. The costs of exacerbations in chronic obstructive pulmonary disease (COPD). Respir Med 2002;96:700–708.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
64. Anto JM, Vermeire P, Vestbo J, Sunyer J. Epidemiology of chronic obstructive pulmonary disease. Eur Respir J 2001;17:982–994.[Abstract/Free Full Text]
65. Viegi G, Scognamiglio A, Baldacci S, Pistelli F, Carrozzi L. Epidemiology of chronic obstructive pulmonary disease (COPD). Respiration 2001;68:4–19.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
66. Drummond MF, Jefferson TO. Guidelines for authors and peer reviewers of economic submissions to the BMJ. The BMJ Economic Evaluation Working Party. BMJ 1996;313:275–283.[Free Full Text]
67. Siegel JE, Torrance GW, Russell LB, et al. Guidelines for pharmacoeconomic studies. Recommendations from the panel on cost-effectiveness in health and medicine. Panel on Cost-Effectiveness in Health and Medicine. Pharmacoeconomics 1997;11:159–168.[Web of Science][Medline] [Order article via Infotrieve]
68. Siegel JE, Weinstein MC, Russell LB, Gold MR. Recommendations for reporting cost-effectiveness analyses. Panel on Cost-Effectiveness in Health and Medicine. JAMA 1996;276:1339–1341.[Abstract/Free Full Text]
69. Birch S, Gafni A. On being NICE in the UK: guidelines for technology appraisal for the NHS in England and Wales. Health Econ 2002;11:185–191.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
70. Garber AM, Phelps CE. Economic foundations of cost-effectiveness analysis. J Health Econ 1997;16:1–31.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
71. Rice DP. Estimating the costs of illness. Am J Public Health 1967;68:424–440.
72. Rice DP.. Estimating the cost of illness. Health Economics Series 6. Publication no. 947.6Washington DC, US Government Printing Office, 1966.
73. Koopmanschap MA, van Ineveld BM. Towards a new approach for estimating indirect costs of disease. Soc Sci Med 1992;34:1005–1010.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
74. Koopmanschap MA, Rutten FF. The impact of indirect costs on outcomes of health care programs. Health Econ 1994;3:385–393.[Web of Science][Medline] [Order article via Infotrieve]
75. Koopmanschap MA, Rutten FF, van Ineveld BM, et al. The friction cost method for measuring indirect costs of disease. J Health Econ 1995;14:171–189.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
76. Rothermich EA, Pathak DS. Productivity-cost controversies in cost-effectiveness analysis: review and research agenda. Clin Ther 1999;21:255–267.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
77. Hutubessy RC, van Tulder MW, Vondeling H, Bouter LM. Indirect costs of back pain in the Netherlands: a comparison of the human capital method with the friction cost method. Pain 1999;80:201–207.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
78. Johannesson M, Karlsson G. The friction cost method: a comment. J Health Econ 1997;16:249–255.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
79. Koopmanschap MA, Rutten FF. A practical guide for calculating indirect costs of disease. Pharmacoeconomics 1996;10:460–466.[Web of Science][Medline] [Order article via Infotrieve]
80. Goeree R, O'Brien BJ, Blackhouse G, Agro K, Goering P. The valuation of productivity costs due to premature mortality: a comparison of the human-capital and friction-cost methods for schizophrenia. Can J Psychiatry 1999;44:455–463.[Web of Science][Medline] [Order article via Infotrieve]
81. Sullivan SD, Ramsey SD, Lee TA. The economic burden of COPD. Chest 2000;117: Suppl. 2, 5S–9S.[Abstract/Free Full Text]
82. Faulkner MA, Hilleman DE. The economic impact of chronic obstructive pulmonary disease. Expert Opin Pharmacother 2002;3:219–228.[CrossRef][Medline] [Order article via Infotrieve]
83. Ruchlin HS, Dasbach EJ. An economic overview of chronic obstructive pulmonary disease. Pharmacoeconomics 2001;19:623–642.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
84. Strassels SA, Smith DH, Sullivan SD, Mahajan PS. The costs of treating COPD in the United States. Chest 2001;119:344–352.[Abstract/Free Full Text]
85. Ward MM, Javitz HS, Smith WM, Bakst A. Direct medical cost of chronic obstructive pulmonary disease in the USA. Respir Med 2000;94:1123–1129.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
86. Jacobson L, Hertzman P, Lofdahl CG, Skoogh BE, Lindgren B. The economic impact of asthma and chronic obstructive pulmonary disease (COPD) in Sweden in 1980 and 1991. Respir Med 2000;94:247–255.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
87. Buck DJ, Richmond RL, Mendelsohn CP. Cost-effectiveness analysis of a family physician delivered smoking cessation program. Prev Med 2000;31:641–648.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
88. Curry SJ, Grothaus LC, McAfee T, et al. Use and cost effectiveness of smoking-cessation services under four insurance plans in a health maintenance organization. N Engl J Med 1998;339:673–679.[Abstract/Free Full Text]
89. Meenan RT, Stevens VJ, Hornbrook MC, et al. Cost-effectiveness of a hospital-based smoking cessation intervention. Med Care 1998;36:670–678.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
90. Fiscella K, Franks P. Cost-effectiveness of the transdermal nicotine patch as an adjunct to physicians' smoking cessation counseling. JAMA 1996;275:1247–1251.[Abstract/Free Full Text]
91. Lennox AS, Osman LM, Reiter E, et al. Cost effectiveness of computer tailored and non-tailored smoking cessation letters in general practice: randomised controlled trial. BMJ 2001;322:1396.[Abstract/Free Full Text]
92. Wasley MA, McNagny SE, Phillips VL, Ahluwalia JS. The cost-effectiveness of the nicotine transdermal patch for smoking cessation. Prev Med 1997;26:264–270.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
93. Weiss SJ, Jurs S, Lesage JP, Iverson DC. A cost-benefit analysis of a smoking cessation program. Eval Program Plann 1984;7:337–346.[CrossRef][Medline] [Order article via Infotrieve]
94. Molken MP, van Doorslaer EK, Rutten FF. Economic appraisal of asthma and COPD care: a literature review 1980–1991. Soc Sci Med 1992;35:161–175.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
95. Rennard SI. New therapeutic drugs in the management of chronic obstructive pulmonary disease. Curr Opin Pulm Med 2002;8:106–111.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
96. Mahler DA, Donohue JF, Barbee RA, et al. Efficacy of salmeterol xinafoate in the treatment of COPD. Chest 1999;115:957–965.[Abstract/Free Full Text]
97. Ramirez-Venegas A, Ward J, Lentine T, Mahler DA. Salmeterol reduces dyspnea and improves lung function in patients with COPD. Chest 1997;112:336–340.[Abstract/Free Full Text]
98. Dahl R, Greefhorst LA, Nowak D, et al. Inhaled formoterol dry powder versus ipratropium bromide in chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2001;164:778–784.[Abstract/Free Full Text]
99. Rossi A, Kristufek P, Levine BE, et al. Comparison of the efficacy, tolerability, and safety of formoterol dry powder and oral, slow-release theophylline in the treatment of COPD. Chest 2002;121:1058–1069.[Abstract/Free Full Text]
100. Wadbo M, Lofdahl CG, Larsson K, et al. Effects of formoterol and ipratropium bromide in COPD: a 3-month placebo-controlled study. Eur Respir J 2002;20:1138–1146.[Abstract/Free Full Text]
101. Hansel TT, Barnes PJ. Tiotropium bromide: a novel once-daily anticholinergic bronchodilator for the treatment of COPD. Drugs Today 2002;38:585–600.[CrossRef][Medline] [Order article via Infotrieve]
102. Casaburi R, Mahler DA, Jones PW, et al. A long-term evaluation of once-daily inhaled tiotropium in chronic obstructive pulmonary disease. Eur Respir J 2002;19:217–224.[Abstract/Free Full Text]
103. D'Urzo AD, De Salvo MC, Ramirez-Rivera A, et al. In patients with COPD, treatment with a combination of formoterol and ipratropium is more effective than a combination of salbutamol and ipratropium: a 3-week, randomized, double-blind, within-patient, multicenter study. Chest 2001;119:1347–1356.[Abstract/Free Full Text]
104. ZuWallack RL, Mahler DA, Reilly D, et al. Salmeterol plus theophylline combination therapy in the treatment of COPD. Chest 2001;119:1661–1670.[Abstract/Free Full Text]
105. Gizycki MJ, Hattotuwa KL, Barnes N, Jeffery PK. Effects of fluticasone propionate on inflammatory cells in COPD: an ultrastructural examination of endobronchial biopsy tissue. Thorax 2002;57:799–803.[Abstract/Free Full Text]
106. Culpitt SV, Maziak W, Loukidis S, Nightingale JA, Matthews JL, Barnes PJ. Effect of high dose inhaled steroid on cells, cytokines, and proteases in induced sputum in chronic obstructive pulmonary disease. Am J Respir Crit Care Med 1999;160:1635–1639.[Abstract/Free Full Text]
107. 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.[Abstract/Free Full Text]
108. Pauwels RA, Lofdahl CG, Laitinen LA, et al. 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.[Abstract/Free Full Text]
109. 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.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
110. Burge PS, 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.[Abstract/Free Full Text]
111. Calverley P, Pauwels R, Vestbo J, et al. Combined salmeterol and fluticasone in the treatment of chronic obstructive pulmonary disease: a randomised controlled trial. Lancet 2003;361:449–456.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
112. Jones PW, Willits LR, Burge PS, Calverley PM. Disease severity and the effect of fluticasone propionate on chronic obstructive pulmonary disease exacerbations. Eur Respir J 2003;21:68–73.[Abstract/Free Full Text]
113. Mahler DA, Wire P, Horstman D, et al. Effectiveness of fluticasone propionate and salmeterol combination delivered via the diskus device in the treatment of chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2002;166:1084–1091.[Abstract/Free Full Text]
114. Szafranski W, Cukier A, Ramirez A, et al. Efficacy and safety of budesonide/formoterol in the management of chronic obstructive pulmonary disease. Eur Respir J 2003;21:74–81.[Abstract/Free Full Text]
115. Update of the National Heart, Lung and Blood Institute/World Health Organization Workshop Report: Global Strategy for Asthma Management and Prevention, issued January 1995. Document no. 02-3659. Bethesda, National Institutes of Health, National Heart, Lung and Blood Institute, Global Initiative for Asthma, 2002.
116. Burrows B, Bloom JW, Traver GA, Cline MG. The course and prognosis of different forms of chronic airways obstruction in a sample from the general population. N Engl J Med 1987;317:1309–1314.[Abstract]
117. Kerstjens HA, Brand PL, Hughes MD, et al. A comparison of bronchodilator therapy with or without inhaled corticosteroid therapy for obstructive airways disease. Dutch Chronic Non-Specific Lung Disease Study Group. N Engl J Med 1992;327:1413–1419.[Abstract]
118. Walker AMeditors. Observation and Inference: an Introduction to the Methods of Epidemiology. Newton Lower Falls, Epidemiology Resources Inc., 1991.
119. Rothman KJGreenland Seditors. Modern Epidemiology. 2nd EdnHagerstown and Philadelphia, Lippincott-Raven, 1998.
120. 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.[Abstract/Free Full Text]
121. 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.[Abstract/Free Full Text]
122. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45:613–629.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
123. Sin DD, Man SFP. Inhaled corticosteroids and survival in COPD: does the dose of therapy matter?. Eur Respir J 2003;21:260–267.[Abstract/Free Full Text]
124. Rochon PA, Tu JV, Anderson GM, et al. Rate of heart failure and 1-year survival for older patients receiving low-dose beta-blocker therapy after myocardial infarction. Lancet 2000;356:639–644.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
125. Boulet LP, Becker A, Berube D, Beveridge R, Ernst P. Canada Asthma Consensus Report, 1999. Canadian Asthma Consensus Group. CMAJ 1999;161: Suppl. 11, S1–S61.[Free Full Text]
126. Canadian Institute for Health Information. http://secure.cihi.ca/cihiweb/dispPage.jsp?cw_page=home_e. Accessed February 2003. Last updated April 2003.
127. National Center for Health Statistics. International Classification of Diseases Ninth Revision, Clinical Modification. 5th EdnLos Angeles, Practice Management Information Corporation, 1999.
128. Richards J, Brown A, Homan C.. The quality study of the Canadian discharge abstract databaseIn:. Achieving Data Quality in a Statistical Agency: A Methodological Perspective. Proceedings of Statistics Canada Symposium, 2001.
129. Rawson NS, Malcolm E. Validity of the recording of ischaemic heart disease and chronic obstructive pulmonary disease in the Saskatchewan health care datafiles. Stat Med 1995;14:2627–2643.[Web of Science][Medline] [Order article via Infotrieve]
130. Lacasse Y, Brooks D, Goldstein R. Trends in the epidemiology of COPD in Canada, 1980 to 1995. Chest 1999;116:306–313.[Abstract/Free Full Text]
131. Government of Alberta. Alberta Health and Wellness. Health care coverage and services. Prescription drug coverage for seniors. http://www.health.gov.ab.ca/coverage/benefits/seniors.html. Accessed February 2003. Last updated June 23, 2003.
132. Ontario Ministry of Health and Long-Term Care. Ontario drug benefits. http://www.gov.on.ca/health/english/program/drugs/drugs_mn.html. Accessed February 2003. Updated daily.
133. Ministry of Industry. 1996 Census Technical Report: Mobility and Migration. Ottawa, Statistics Canada, 1999.
134. Glesby MJ, Hoover DR. Survivor treatment selection bias in observational studies: examples from the AIDS literature. Ann Intern Med 1996;124:999–1005.[Abstract/Free Full Text]
135. General Practice Research Database. GPRD: Excellence in public health research. www.gprd.com. Accessed February 5, 2003.
136. Lawson DH, Sherman V, Hollowell J. The General Practice Research Database. Scientific and Ethical Advisory Group. QJM 1998;91:445–452.[Abstract/Free Full Text]
137. Nazareth I, King M, Haines A, et al. Accuracy of diagnosis on general practice computer system. BMJ 1993;307:32–34.[Abstract/Free Full Text]
138. Hansell A, Hollowell J, Nichols T, et al. Use of the General Practice Research Database (GPRD) for respiratory epidemiology: a comparison with the 4th Morbidity Survey in General Practice (MSGP4). Thorax 1999;54:413–419.[Abstract/Free Full Text]
139. Soriano JB, Maier WC, Egger P, et al. Recent trends of physician-diagnosed COPD in women and men in the UK. Thorax 2000;55:789–794.[Abstract/Free Full Text]
140. Soriano JB, Maier WC, Visick G, Pride NB. Validation of general practitioner-diagnosed COPD in the UK General Practice Research Database. Eur J Epidemiol 2001;17:1075–1080.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
141. 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.[Abstract/Free Full Text]
142. Soriano JB, Kiri V, Pride NB, Vestbo J. Inhaled corticosteroids with/without long-acting beta agonists reduce the risk of re-hospitalisation and death in COPD patients. Am J Respir Med 2003;2:67–74.[Medline] [Order article via Infotrieve]
143. Strom BLed. Pharmacoepidemiology. 3rd EdnChichester, John Wiley & Sons, 2000.
144. The COPD Guidelines Group of the Standards of Care Committee of the British Thoracic Society. BTS guidelines for the management of chronic obstructive pulmonary disease. Thorax 1997;52: Suppl. 5, S1–S28.[Medline] [Order article via Infotrieve]
145. Kavuru M, Melamed J, Gross G, et al. Salmeterol and fluticasone propionate combined in a new powder inhalation device for the treatment of asthma: a randomized, double-blind, placebo-controlled trial. J Allergy Clin Immunol 2000;105:1108–1116.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
146. Schols AM, Wesseling G, Kester AD, et al. Dose-dependent increased mortality risk in COPD patients treated with oral glucocorticoids. Eur Respir J 2001;17:337–342.[Abstract/Free Full Text]
147. Hosmer DWLemeshow Seditors. Applied Survival Analysis: Regression Modelling of Time to Event DataChichester, John Wiley & Sons, 1999.
148. Jick H, Jick SS, Derby LE. Validation of information recorded on general practitioner based computerised data resources in the UK. BMJ 1991;302:766–768.[Abstract/Free Full Text]
149. Garcia Rodriguez LA, Perez Gutthann S. Use of the UK General Practice Research Database for pharmacoepidemiology. Br J Clin Pharmacol 1998;45:419–425.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
150. Beardon PH, McGilchrist MM, McKendrick AD, McDevitt DG, MacDonald TM. Primary non-compliance with prescribed medication in primary care. BMJ 1993;307:846–848.[Abstract/Free Full Text]
151. Blais L, Suissa S, Boivin JF, Ernst P. First treatment with inhaled corticosteroids and the prevention of admissions to hospital for asthma. Thorax 1998;53:1025–1029.[Abstract/Free Full Text]
152. Blais L, Ernst P, Boivin JF, Suissa S. Inhaled corticosteroids and the prevention of readmission to hospital for asthma. Am J Respir Crit Care Med 1998;158:126–133.[Abstract/Free Full Text]
153. Bourbeau J, Ernst P, Cockroft D, Suissa S. Inhaled corticosteroids and severe acute exacerbations of chronic obstructive pulmonary disease. Eur Respir J 2003;22:286–289.[Abstract/Free Full Text]
154. Downey W, Beck P, McNutt M, Stang M, Osei W, Nichol J.. Health databases in SaskatchewanIn: Strom BL, ed. Pharmacoepidemiology. 3rd EdnChichester, John Wiley & Sons, 2000; pp. 325–346.
155. Bourbeau J, McIvor A. Inhaled corticosteroid should not be prescribed to all chronic obstructive pulmonary disease patients. Can Respir J 2003;10:148–149.[Medline] [Order article via Infotrieve]
156. van der Lei J, Duisterhout J, Westerhof H, et al. The introduction of computer-based patient records in the Netherlands. Ann Intern Med 1993;119:1036–1041.[Abstract/Free Full Text]
157. Sluiter HJ, Koeter GH, de Monchy JG, Postma DS, de Vries K, Orie NG. The Dutch Hypothesis (chronic non-specific lung disease) revisited. Eur Respir J 1991;4:479–489.[Abstract]
158. Artsennet. Artsennet, medische informatie van en voor artsen in Nederland. www.artsennet.nl. Accessed March 2003. Updated daily.
159. Vlug AE, van der Lei J, Mosseveld BM, et al. Postmarketing surveillance based on electronic patient records: the IPCI project. Methods Inf Med 1999;38:339–344.[Web of Science][Medline] [Order article via Infotrieve]
160. Lamberts HWood Meditors. International Classification of Primary CareOxford, Oxford University Press, 1987.
161. de Smet P. The Dutch approach to computerized drug information: conceptual basis and realization. J Social and Admin Pharmacy 1988;5:49–58.
162. Cochrane AL.. Effectiveness and Efficiency: Random Reflections on Health ServicesLondon, Nuffield Provincial Hospitals Trust, 1972; pp. 20–25.
163. Sacks H, Chalmers TC, Smith H Jr. Randomized versus historical controls for clinical trials. Am J Med 1982;72:233–240.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
164. Black N. Why we need observational studies to evaluate the effectiveness of health care. BMJ 1996;312:1215–1218.[Free Full Text]
165. Weinstein MC, O'Brien B, Hornberger J, et al. Principles of good practice for decision analytic modeling in health-care evaluation: report of the ISPOR Task Force on Good Research Practices--Modeling Studies. Value Health 2003;6:9–17.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
166. Moher D, Schulz KF, Altman DA, for the Consolidated Standards of Reporting of Trials Group. The CONSORT statement: Revised recommendations for improving the quality of reports of parallel-group randomised trials. Lancet 2001;357:1191–1194.[CrossRef][Web of Science][Medline] [Order article via Infotrieve]
167. National Health Service Centre for Reviews and Dissemination. Undertaking Systematic Reviews of Research on Effectiveness: CRD's Guidance for Carrying Out or Commissioning Reviews. http://www.york.ac.uk/inst/crd/crdrep.htm. CRD Report 4. 2nd Edn. 2001. Accessed March 1, 2003.
168. Waterhouse JC, Fiskwick D, Burge PS, Calverley PMA, Anderson JA, on behalf of the Inhaled Steroids in Obstructive Lung Disease in Europe Trial Group. What caused death in the ISOLDE study?. Eur Respir J 1999;14: Suppl. 30, 387S.

This article has been cited by other articles:

				P. Tonnesen, L. Carrozzi, K. O. Fagerstrom, C. Gratziou, C. Jimenez-Ruiz, S. Nardini, G. Viegi, C. Lazzaro, I. A. Campell, E. Dagli, et al. Smoking cessation in patients with respiratory diseases: a high priority, integral component of therapy Eur. Respir. J., February 1, 2007; 29(2): 390 - 417. [Abstract] [Full Text] [PDF]

				K.F. Rabe and P.J. Sterk Rapid, reliable and responsive: for whom? Eur. Respir. J., January 1, 2004; 23(1): 1 - 2. [Full Text] [PDF]

This Article Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Permissions Request Permissions Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Search for Related Content PubMed PubMed Citation