Guidelines define an exacerbation of chronic obstructive pulmonary disease (COPD) as an acute deterioration in clinical symptoms (dyspnoea, cough, and/or sputum production) in the course of the disease that goes beyond the daily fluctuations in the symptoms and requires a change of medication (1). COPD exacerbations are etiologically and clinically heterogeneous events (2). Viral and/or bacterial respiratory tract infections constitute the most common causes of acute exacerbations, but in about 30% of cases no cause can be identified (3, 4). The following conditions, in particular, need to be considered in the differential diagnosis of an exacerbation: pneumonia, heart failure, pulmonary embolism, and pneumothorax.

Because of the dubious clinical relevance of mild exacerbations, the severity of an exacerbation is generally also determined. The need for systemic corticosteroid and/or antibiotic therapy has become accepted as a criterion for a moderate exacerbation and for in-patient treatment as a criterion for a severe exacerbation, at least in the context of randomized clinical trials (5-9). However, health care systems and consequently out-patient or in-patient treatment options vary in different countries. In addition, the previous strictly event-based definition underestimates the true extent of exacerbations, so that other definitions of exacerbation focus more on symptoms (10, 11). Generally, different symptom or event-oriented definitions of an exacerbation make a comparative evaluation of clinical studies challenging (4).


COPD EXACERBATIONS FROM THE CLINICAL PERSPECTIVE

Basic principles:
• Acute deteriorations increase with COPD severity (4, 5, 11)
• Previous exacerbations are predictive of further exacerbations (12-14)
• COPD exacerbations can follow a prolonged course (4)
• Repeated, severe exacerbations, i.e., requiring hospitalization, are associated with significantly higher mortality (15-18)
• COPD exacerbations are a common cause of hospitalizations and hence a major cost factor (8, 19, 20)
• COPD exacerbations are associated with increased morbidity, mortality, and reduced quality of life (12).

The main factors in repeated COPD exacerbations are summarised in Table 1. COPD exacerbation frequency and severity indicate clinical instability as well as progression of the disease process and are therefore relevant for prognosis (21). For this reason, prevention of acute exacerbations constitutes an important treatment goal. Drugs of choice for the long-term therapy of COPD are inhaled bronchodilators (long-acting anticholinergics and/or ß2 sympathomimetics), which have also been shown to prevent exacerbations (22-25). In patients with FEV1 <50% and with repeated exacerbations, additional inhaled corticosteroid treatment is recommended (26).

Table 1. Clinical and physiological effects of frequent COPD exacerbations.

BODE score - Composite score consisting of body-mass index, respiratory tract obstruction (FEV1), degree of dyspnoea (modified Medical Research Council dyspnoea scale) and exercise capacity, measured by means of the 6-min walking test (8, 19).

METHODOLOGICAL ASSESSMENT
OF COPD EXACERBATION
General considerations

In order to establish whether there is a significant difference between two or more interventions in terms of a specific endpoint, the recorded data must be described by a suitable statistical model and a test of the hypothesis performed. For a sensible interpretation of the test results, not only do the assessed intervention and the concerned endpoint need to be clearly defined, but also the analyzed groups should be comparable with one another and ideally differ only in the form of treatment. What is important here is to establish whether, and to what extent confounders, such as age or genetic predisposition, and bias, such as a systematically biased patient selection, exert an effect on the study results.

Problems of group comparability

In contrast to the analyses of administrative databases (27, 28) and observational studies (29), randomized clinical trials have the advantage that they can minimize or even eliminate the effect of selection and particularly of confounders by a random allocation of study participants to treatment groups. This advantage, however, applies only at the beginning of the study and not necessarily to its subsequent course. Even studies with a randomized design are subject to effects which, as systematic errors, can falsify the outcome. In this context, two possibilities had to be considered:

• Disproportionate drop-out of the study population (a drop-out rate unequally distributed between the treatment groups);
• Clinically determined switch of study participants from one treatment group to another, when, for example, it becomes unacceptable to leave certain patients in the treatment group to which they have been allocated. In most trials, the drop-out rate in the placebo arm of controlled studies is disproportionately increased compared with the active treatment arms (5, 30, 31).

In the ISOLDE study, 57% of patients treated with inhaled glucocorticoids (ICS) completed the study, but only 47% in the placebo group (5). For Calverley et al. (32), the drop-out rate was 41% in the placebo group, but only 29% in the budesonide/formoterol-treated group, while the approximately 20% rate of non-COPD-related side effects and associated discontinuations was twice as high in the treatment than in the placebo group. Kesten et al. (33) describe higher drop-out rates in the placebo groups than in the tiotropium treatment arm. In the TORCH study the proportion of study discontinuations was also highest in the placebo group with 44.2% (31).

This raises the question as to the extent of differential distortion occurring in the placebo group compared with the active group as a result of the higher drop-out rate. If, as may be suspected, those patients from the placebo group with COPD instability have a higher discontinuation rate, there will inevitably be an increased number of healthier and/or clinically more stable patients remaining. Those patients most likely have a comparably lower exacerbation risk, better FEV1, and a smaller annual FEV1 decline. This placebo-related selection can potentiate or attenuate the difference from the active treatment group depending on the study design. The observed group difference between the placebo and active treatment groups could be seen as a result of different health status or study design-related selection pressure. Such problems of the group comparability are a confounding factor in individual studies and cannot be eliminated in the quantitative analysis of individual studies, or meta-analyses.

Meta-analyses

Meta-analyses are quantitative summaries of outcomes of individual studies in which diverse hypotheses have been tested (34). They are viewed as a useful supplement to clinical decision-making and, like other types of study, are governed by clear methodological guidelines (35). Meta-analyses reduce the random errors introduced by the smaller case numbers of individual studies, as a result of which significant results can be obtained even for small differences. Small studies are subject to a greater random error, which is readily apparent in the higher variance and the size of the confidence interval, as an expression of the dispersion of the effect size than studies with larger case numbers. In addition, as described above, all studies can be affected by systematic errors and confounders with a distorting effect on the result, which affects the risk estimate. Meta-analyses do not compensate for systematic errors, so that false positive or negative results can occur. Therefore, meta-analyses, even of randomized clinical trials, should be interpreted with caution. The minimum requirement for a meta-analysis is that the studies included are defined in the same way in terms of treatment (exposure) and target event (primary and/or secondary parameters, in this case exacerbation). It must also be ensured that all relevant studies are included in the meta-analysis to prevent ‘publication bias’, which generally occurs if small studies with statistically non-significant results are ignored. Some of these points are exemplified below.

In the Cochrane analysis by Nannini et al. (36) a reduction in the exacerbation rate from 1.84 to 1.42 per year was calculated on the basis of four studies in which two combination products had been tested (fluticasone/salmeterol and budesonide/formoterol; study duration 6-12 months) with a total sample size of 2986 patients. The authors point out that the clinical outcome cannot be quantified, as the target event (exacerbation) was defined differently in the four studies. In addition, various drug doses were tested. The meta-analysis by Alsaeedi et al. (37), who also included studies which differed in terms of the ICS products and doses, found a significant heterogeneity in the study results. Here again the different ICS effects in respect to the outcome parameter exacerbation can be attributed to the heterogeneity of the studies included in the evaluation. In the authors’ view, this is due to differences in the definition of exacerbations. However, the studies cited differ quite substantially in other respects, like the proportion of active smokers which ranged between 38 and 90%, and the study durations varied between 6 and 40 months (37).

There is a broad spectrum of treatment modalities and definitions of exacerbations in both individual studies and their meta-analyses, so that the requirement of including only studies with the same exposure and target event is fulfilled to limited extent. The number of study drop-outs in the reference groups and their effect on outcome are largely ignored in the meta-analyses. For example, in the meta-analysis from Salpeter et al. (38) the precise definition of exacerbation in the individual studies was quite unusual, because it has been solely stated as an event that resulted in either exclusion from the study, hospitalization, or death. The absence of the precondition of group comparability necessary for the calculation makes the meta-analytical assessment of treatment efficacy problematic.

Statistical model

In addition to the inclusion of the target event and the drop-out rates, the selection of analytical method is crucial for the target outcome. At the same time, the properties of the target event must be taken into account when analyzing the frequency or distribution of that event. In many studies the number of exacerbations for all days of treatment is extrapolated from the exacerbation rate per patient to obtain the number of exacerbations per treatment year and per patient. Thereafter, a non-parametric procedure, such as Wilcoxon’s rank sum test, can be used to assess whether there is a statistically significant difference between the forms of treatment, as has been done in the ISOLDE study and the Lung Health study (39, 40). Alternatively, a t-test of the mean event rates can be performed in a parametric model, as done in more recent publications assuming a normal distribution.

However, exacerbation rates are not normally distributed, but generally occur 0 to 2 times a year. In addition, the proportion of study participants with more than 2 exacerbations per year decreases slowly and asymmetrically over the course of a study (41). It has, therefore, been recommended that a Poisson distribution should be used to model exacerbation rates, as this takes account of their specific properties (28, 41-43). A Poisson distribution refers to the expected number of events within a defined unit of time. It has to satisfy three assumptions:

• probability of the occurrence of an event is proportional to the duration of observation;
• number of events is in principle equally distributed over the whole observation period, i.e., the expected number of events per unit of time is constant;
• events occur independently of one another. This last assumption would be infringed for example in an epidemic as a result of an infectious disease, since here one event affects the next.

Since exacerbations are a) binomial i.e., either the patient has an exacerbation or not, b) they can develop independently of any previous exacerbations, and c) as a first event are distributed constantly over a period of time, exacerbation rates in principle follow the laws of a Poisson distribution. Any difference in the length of time in which the patients remain in the studies is ignored in this form of analysis, since a defined unit of time is used in the calculation and the individual probability of a subsequent event alters with the number of previous exacerbations.

Suissa (42) has introduced a refinement of the now current Poisson approach in most clinical exacerbation studies to analyze prophylactic ICS efficacy by minimizing COPD exacerbation rates. To illustrate and test his statistical approach, the author constructed one large and several small “randomized” studies from a COPD cohort in an administrative database (Saskatchewan Health). In this model calculation he assumed an exacerbation in the event of a prescription of an antibiotic with or without an additional prescription of an oral corticosteroid (43, 44). Patients who had already suffered an exacerbation before inclusion in the cohort served as a reference group. The virtual “treatment group” consisted of a random selection of COPD patients without exacerbation (i.e., without a previous prescription of an antibiotic and/or corticosteroid). In this figurative example, the mean observation period was predefined as 3-4 years. The author analyzed the results using three different analytical methods: a) Wilcoxon’s rank sum test, b) normal Poisson regression, and c) modified Poisson regression variant, designed to allow for individual variability in the observation periods and hence heterogeneity in the individual exacerbation rates by means of weighting. This last approach is primarily based on the observation that patients had different study durations due to high drop-out rates and thus had different probabilities of an exacerbation event. Further, this modified Poisson regression was designed to account for the inhomogeneous distribution of exacerbations within a cohort, in which many COPD patients suffer a number of exacerbations, while others suffer no acute events at all.

For example, it is not the same thing, either statistically or in the real world, when one patient in a study suffers 12 exacerbations and 12 participants suffer one exacerbation each. The overall model in the modified Poisson regression provides, therefore, for a weighting for both the different duration of observation and the different probabilities of exacerbation among study participants. This method demonstrated that the event rates (exacerbations) of treated and untreated participants were closer to one another under the weighted calculation than the unweighted calculation suggested (42). The author concluded that without weighting, as done in most statistical calculations of COPD drug studies, there is a statistical calculation error in the form of a falsely increased difference between active treatment and placebo. The risk situation of study participants with regard to an exacerbation can only be correctly reflected, if the patient’s time in the study, the time frame till the onset of the exacerbation as well as the tendency to multiple exacerbations are considered (42). Interestingly, there is a reduction in the exacerbation rate of the treatment group compared with the reference group. The weighted risk for the comparison of “treated” (patients without previous exacerbation) with “untreated” (patients with previous exacerbation) was 0.75 (25% reduction during treatment), the unweighted exacerbation risk based on the mean was 0.57, and the risk based on the median was 0.63. The weighted result was not significantly different. The two COPD studies which also used a Poisson model (32, 45) achieved similar, statistically not significant results. Conversely, a weighted analysis of a randomized study of exacerbation rates of salmeterol/fluticasone versus salmeterol, as done with Suissa’s recommendation (46) revealed only a “cosmetic“ effect on the overall result (47, 48). There was only a marginal change in the result in terms of the confidence interval from 0.65 (95% confidence interval: 0.57–0.76) to 0.65 (95% confidence interval: 0.54–0.79), with no change in the statistical result. However, this study was confined to patients with severe COPD and frequent exacerbations, so that the variability in the individual susceptibility to exacerbations was possibly smaller than in other studies (48). Alternatives to the Poisson regression, such as the negative binomial distribution, are currently being discussed, but these will not be addressed here (49).


CONCLUSIONS
Correct investigation of acute COPD exacerbations and analysis of the extent to which pharmacological interventions can reduce these events depend considerably on the methodological and statistical approach adopted. They relate firstly to the precise definition and monitoring of exacerbations in order to obtain conclusive results about the effectiveness of treatment. Studies that failed to consider the time course of the distribution of exacerbations in the past have resulted in falsely increased differences between the exacerbation rates in the active treatment compared with placebo groups. The following points should be considered in assessing exacerbation studies:

• duration of observation duration, at least 6 months, preferably 1 year, to allow for seasonal variations, variable clinical severity and the wealth of potential triggers;
• definition of exacerbation, including severity, should be based on the current GOLD COPD guideline recommendations for the purpose of greater comparability;
• statistical analysis: the analysis of exacerbations by weighted Poisson regression is considered the standard to prevent falsely elevated exacerbation estimates;
• meta-analyses with exacerbation as an endpoint are not in principle to be rated more highly from the perspective of clinical evidence than prospective randomized individual studies and are dependent among other things on the selection, comparability, and analysis of the studies included.

Conflicts of interest: Adrian Gillissen has received consultant’s and speaker’s fees during the last three years and has obtained research grants from the following pharmaceutical companies: AstraZeneca, Asche Chesi, Bayer Vital, Boehringer Ingelheim, Jansen Cilag, GlaxoSmithkline, Novartis, Pohl-Boskamp. The St. George Medical Center has received grants for services performed during participation in single- and multi-center clinical phase I-IV trials organized by various pharmaceutical companies.
Thomas Glaab was an employee of Boehringer Ingelheim at the time of publication.
Roland Buhl has received reimbursement for attending scientific conferences, and/or fees for speaking and/or consulting from Chiesi-AstraZeneca, Talecris, Boehringer Ingelheim, GlaxoSmithKline, Janssen-Cilag, Novartis, and Pfizer. The Pulmonary Department at Mainz University Hospital received financial compensation for services performed during participation in single- and multi-center clinical phase I-IV trials organized by various pharmaceutical companies.
Michael Lewis was an employee of FISPE Epidemiology GmbH at the time of publication.
Heinrich Worth has received consultant’s and speaker’s fees during the last three years and has obtained research grants from the following pharmaceutical companies: AstraZeneca, Asche Chesi, Boehringer Ingelheim, and GlaxoSmithkline. The Hospital Furth has received grants for services performed during participation in single- and multi-center clinical phase III-IV trials organized by various pharmaceutical companies.


REFERENCES

1. Rabe KF, Hurd SS, Anzueto A, et al. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease. GOLD Executive Summary. Am J Respir Crit Care Med 2007; 176: 532-555.
2. Grabicki M, Parysek H, Batura-Gabryel H, Brodnicka I. Comorbidities as an element of multidimensional prognostic assessment of patients with chronic obstructive pulmonary disease. J Physiol Pharmacol 2008; 59(Suppl. 6): 297-302.
3. Aaron SD, Fergusson D, Marks GB, et al. Counting, analysing and reporting exacerbations of COPD in randomised controlled trials. Thorax 2008; 63: 122-128.
4. Burge S, Wedzicha JA. COPD exacerbations: definition and classifications. Eur Respir J 2003; 41: 46s-53s.
5. Burge PS, Calverley PMA, Jones PW, Anderson JA, Maslen TK. Randomised, double blind, placebo controlled study of fluticasone propionate in patients with moderate to severe chronic obstructive pulmonary disease: the ISOLDE trial. BMJ 2000; 320: 1297-1303.
6. 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.
7. Vincken W, Noord van JA, Kesten S, Bantje TA, Korducki L, Cornelissen PJG. Improved health outcomes in patients with COPD during 1 yr’s treatment with tiotropium. Eur Respir J 2002; 19: 209-216.
8. Wedzicha JA, Seemungal TAR. COPD exacerbations: defining their cause and prevention. Lancet 2007; 370: 786-796.
9. Cazzola M, MacNee W, Martinez FJ, et al. Outcomes for COPD pharmacological trials: from lung function to biomarkers. Eur Respir J 2008; 31: 416-469.
10. Rodriguez-Roisin R. Toward a consensus definition for COPD exacerbations. Chest 2000; 117: 398s-401s.
11. Langsetmo L, Platt RW, Ernst P, Bourbeau J. Underreporting exacerbation of chronic obstructive pulmonary disease in a longitudinal cohort. Am J Respir Crit Care Med 2008; 177: 396-401.
12. Miravitlles M, Ferrer M, Pont A, et al. Effect of exacerbations on quality of life in patients with chronic obstructive pulmonary disease: a 2 year follow up study. Thorax 2004; 59: 387-395.
13. Donaldson GC, Seemungal TAR, Patel IS, Lloye-Owen SJ, Wilkinson TM, Wedzicha JA. Longitudinal changes in the nature, severity and frequency of COPD exacerbations. Eur Respir J 2003; 22: 931-936.
14. Donaldson GC, Seemungal TAR, Bhowmik A, Wedzicha JA. Relationship between exacerbation frequency and lung function decline in chronic obstructive pulmonary disease. Thorax 2002; 57: 847-852.
15. Seneff MG, Wagner DP, Wagner RP, Zimmerman JE, Knaus WA. Hospital and 1-year survival of patients admitted to intensive care units with acute exacerbation of chronic obstructive pulmonary disease. JAMA 1995; 274: 1852-1857.
16. Fuso L, Incalzi RA, Pistelli R, et al. Predicting mortality of patients hospitalized for acutely exacerbated chronic obstructive pulmonary disease. Am J Med 1995; 98: 272-277.
17. Connors AF, Dawson NV, Tomas C, Harrell FE Jr, Desbiens N, Fulkerson WJ. Outcomes following acute exacerbation of severe chronic obstructive lung disease. The SUPPORT investigators (Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments). Am J Respir Crit Care Med 1996; 154: 959-967.
18. Soler-Cataluna JJ, Martinez-Garcia MA, Roman Sanchez P, Salcedo E, Navarro M, Ochando R. Severe acute exacerbations and mortality in patients with chronic obstructive pulmonary disease. Thorax 2005; 60: 925-931.
19. Cote CG, Dordelly LJ, Celli BR. Impact of COPD exacerbations on patient-centered outcomes. Chest 2007; 131: 696-704.
20. Bahadori K, Fitzgerald JM. Risk factors of hospitalization and readmission of patients with COPD exacerbation-systematic review. Int J Chron Obstruct Pulmon Dis 2007; 2: 241-251.
21. Anzueto A, Sethi S, Martinez FD. Exacerbations of chronic obstructive pulmonary disease. Proc Am Thorac Soc 2007; 4: 554-564.
22. Powrie DJ, Wilkinson TM, Donaldson GC, et al. Effect of tiotropium on sputum and serum inflammatory markers and exacerbations in COPD. Eur Respir J 2007; 30: 472-478.
23. Salpeter SR, Buckley NS, Salpeter EE. Meta-analysis: anticholinergics, but not beta-agonists, reduce severe exacerbations and respiratory mortality in COPD. J Gen Intern Med 2006; 21: 1011-1019.
24. Dusser D, Bravo ML, Iacono P. The effect of tiotropium on exacerbations and airflow in patients with COPD. Eur Respir J 2006; 27: 547-555.
25. Cazzola M, Matera MG. Long-acting ß2 agonists as potential option in the treatment of acute exacerbations of COPD. Pulm Pharmacol Ther 2004; 16: 197-201.
26. GOLD Executive Committee. Global initiative for chronic obstructive lung disease. www goldcopd com; 2007.
27. 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.
28. 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.
29. Soriano JB, Kiri VA, Pride NB, Vestbo J. Inhaled corticosteroids with/without long-acting ß-agonists reduce the risk of rehospitalization and death in COPD-patients. Am J Respir Med 2003; 2: 67-74.
30. Rabe KF. Treating COPD - the TORCH trial, p values, and the Dodo. N Engl J Med 2007; 356: 851-854.
31. Calverley PMA, Anderson JA, Celli BR, et al. Salmeterol and fluticasone propionate and survival in chronic obstructive pulmonary disease. N Engl J Med 2007; 356: 775-789.
32. Calverley PMA, Boonsawat W, Cseke Z, Zhong N, Peterson S, Olsson H. Maintenance therapy with budesonide and formoterol in chronic obstructive pulmonary disease. Eur Respir J 2003; 22: 912-919.
33. Kesten S, Plautz M, Piquette CA, Habib MP, Niewoehner D E. Premature discontinuation of patients: a potential bias in COPD clinical trials. Eur Respir J 2007; 30: 898-906.
34. Petitti D. Meta-analysis: decision analysis and cost-effectiveness analysis. In Methods for Quantitative Synthesis in Medicine. Oxford University Press, New York, 1994.
35. Spitzer W O. The challenge of meta-analysis. J Clin Epidemiol 1995; 48: 1-4.
36. Nannini L, Lasserson T, Poole P. Combined corticosteroid and long acting beta-agonist in one inhaler for chronic obstructive pulmonary disease. Cochrane Database Syst Rev 2003; 4: CD003794.
37. Alsaeedi A, Sin D, McAlister F. The effects of inhaled corticosteroids in chronic obstructive pulmonary disease: a systematic review of randomized placebo-controlled trials. Am J Med 2002; 113: 59-65.
38. Salpeter SR, Buckley NS, Ormiston TM, Salpeter EE. Meta-analysis: effect of long-acting ß-agonists on severe-asthma exacerbations and asthma-related deaths. Ann Intern Med 2006; 144: 904-912.
39. Burge PS, Calverley PMA, Jones PW, Spencer S, Anderson J. Randomised, placebo-controlled study of fluticasone propionate in patients with moderate to severe chronic obstructive pulmonary disease; the ISOLDE trial. Br Med J 2000; 320: 1297-1303.
40. Scanlon PD, Connett JE, Waller LA, Altose MD, Bailey WC, Buist AS. Smoking cessation and lung function in mild-to-moderate chronic obstructive pulmonary disease. The Lung Health Study. Am J Respir Crit Care Med 2000; 161: 381-390.
41. Calverley PMA. Minimal clinically important difference - exacerbations of COPD. COPD 2005; 2: 143-148.
42. Suissa S, Ernst P, Vandemheen KL, Aaron S. Methodological issues in therapeutic trials of chronic obstructive pulmonary disease. Eur Respir J 2008; 31: 927-933.
43. Suissa S. Statistical treatment of exacerbations in therapeutic trials of COPD. Am J Respir Crit Care Med 2007; 173: 842-846.
44. de Melo MN, Ernst P, Suissa S. Rates and patterns of chronic obstructive pulmonary disease exacerbations. Can Respir J 2004; 11: 559-564.
45. Szafranski W, Ramirez A, Menga G, et al. Efficacy and safety of budesonide/formoterol in the management of chronic obstructive pulmonary disease. Eur Respir J 2003; 21: 74-81.
46. Suissa S. On the correct statistical analysis of exacerbation rates in clinical trials. Am J Respir Crit Care Med 2007; 176: 522-523.
47. Kardos P, Wencker M, Glaab T, Vogelmeier C. Impact of salmeterol/fluticasone propionate versus salmeterol on exacerbations in severe chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2007; 175: 144-149.
48. Kardos P, Wencker M, Glaab T, Vogelmeier C. On the correct statistical analysis of exacerbation rates in clinical trials (letter). Am J Respir Crit Care Med 2007; 176: 522-523.
49. Keene ON, Calverley PMA, Jones PW, Vestbo J, Anderson JA. Statistical analysis of exacerbation rates in COPD: TRISTAN and ISOLDE revisited. Eur Respir J 2008; 32: 17-24.