Rationale: Exacerbations are important events in chronic obstructive pulmonary disease. Preventing exacerbations is a key treatment goal. Observational data suggest that after a first exacerbation, patients may be at increased risk of a second exacerbation, but this has not been specifically studied. We hypothesized that exacerbations may cluster together in time, a finding that would have important implications for targeting preventative interventions and the analysis of clinical trial data.
Objectives: To assess whether exacerbations are random events, or cluster in time.
Methods: A total of 297 patients in the London chronic obstructive pulmonary disease cohort recorded daily symptoms and were assessed for a total of 904 patient-years. The observed timing of second exacerbations after an initial exacerbation was compared with that expected should exacerbations occur randomly.
Measurements and Main Results: The observed timing distribution of second exacerbations differed significantly (P < 0.001) from the expected exponential function (shape parameter of the fitted Weibull function, 0.966 [95% confidence interval, 0.948–0.985]), suggesting that more second exacerbations occurred sooner than later and that exacerbations cluster together in time. Twenty-seven percent of first exacerbations were followed by a second recurrent event within 8 weeks. Approximately one third of exacerbations were recurrent exacerbations. Although initial exacerbations were milder than isolated events, they were not less likely to receive treatment, and under-treatment of initial events is not a plausible explanation for exacerbation recurrence. Recurrent exacerbations contribute significantly to overall exacerbation frequency (rho = 0.81; P < 0.0001).
Conclusions: Exacerbations are not random events but cluster together in time such that there is a high-risk period for recurrent exacerbation in the 8-week period after an initial excerbation.
It is assumed that exacerbations in chronic obstructive pulmonary disease are random events. There is observational evidence to suggest that this may not be true, but the question has never been specifically studied.
This study demonstrates that exacerbations of chronic obstructive pulmonary disease cluster together in time. This finding has implications for the targeting of preventative interventions and the analysis of data from clinical trials.
It is possible to target preventative therapies at patients who have a high risk of frequent exacerbations; this is an important principle in the management of COPD (6). However, aside from the recognition that exacerbations are seasonal (7), existing work has not considered the concept that there may be periods during which patients are at heightened exacerbation risk.
Observational data have suggested that the risk of a second exacerbation is higher in the period immediately after an initial exacerbation. For example, in a study of outpatient prednisone, 27% of patients on active treatment required another physician assessment within the following 30 days (8). This is consistent with the findings of a UK national audit in which 34% of 1,221 hospitalized exacerbations were readmitted in the subsequent 3 months (9).
We hypothesized that after a first exacerbation there may be a high-risk period for a recurrent, second exacerbation and that exacerbations are not random events but tend to cluster together in time. If confirmed, this would have important implications for the targeting of preventative interventions and the analysis of exacerbation data in clinical trials.
This analysis comprises data collected from the London COPD cohort during the period October 1, 1995 through September 30, 2007. Our patients recorded daily, on diary cards, any worsening in their respiratory symptoms. The 297 patients who had completed data for at least 365 days were included.
COPD was defined as FEV1 < 70% predicted for age, height, and sex; FEV1/FVC ratio < 70%; and minimal or no β2-agonist reversibility (< 15% and/or < 200 ml). Patients with significant respiratory diseases other than COPD were not recruited.
At recruitment, a full medical history and examination was obtained. Measurements were made of FEV1, FVC, and PEF on a rolling seal (Sensor Medic Corp., Yorba Linda, CA) or a Vitalograph Gold Standard (Vitalograph Ltd, Maids Moreton, UK) spirometer. Arterialized ear lobe blood gases were also measured (model 278 Blood Gas Analyzer; Ciba-Corning, Medfield, MA or Rapid lab 348, Bayer Health Care, Newbury, UK).
Ethical approval was granted from East London and the City and Royal Free Hospital ethics committees. All patients provided written informed consent.
All patients recorded daily, on diary cards, any increase in respiratory symptoms, which were classified as major (dyspnea, sputum purulence, and sputum volume) or minor (colds [nasal discharge/congestion], wheeze, sore throat, and cough). This daily recording of symptoms was used to precisely define the onset and recovery of exacerbations as described below.
As in our previous work (1, 3), exacerbation onset was defined as the first of two or more days on which the patient recorded two or more new or worsening symptoms, with one of which being a major symptom. Symptoms were disregarded in identifying exacerbation onset if recorded continuously in the preceding 5 days. A small proportion (3.2%) of exacerbations for which no diary-card symptoms had been recorded was identified by hospital admission data or by questioning the patient at clinic visits.
In this investigation of exacerbation clustering, it was important to ensure that a first exacerbation had completely recovered before a second had commenced. This permitted differentiation between true repeat exacerbations and the symptom variation that can occur in the recovery period from a first exacerbation (and that may require a second course of treatment). We therefore defined a second exacerbation as occurring only when there had been five consecutive days free of all recorded symptoms. Our rationale for selecting this 5-day period is described in the Discussion.
Exacerbation treatment was at the discretion of the attending physician, who may or may not have been part of the research team.
An annual exacerbation rate was calculated for each patient by dividing the number of exacerbations by the number of years of diary card data.
To investigate factors associated with exacerbation recurrence, exacerbations were categorized into three types: “isolated,” “initial,” and “recurrent” (Figure 1). All exacerbations were separated by at least 5 days on which no symptoms were recorded. An isolated exacerbation was neither preceded within 8 weeks nor followed within 8 weeks by any other exacerbation. An initial exacerbation was an exacerbation that was followed within 8 weeks of onset by the onset of a second exacerbation but which did not itself follow a first exacerbation within 8 weeks. A recurrent exacerbation was an exacerbation that had an onset within 8 weeks of a preceding exacerbation. It may or may not have been followed within 8 weeks by the onset of further recurrent exacerbations.
Exacerbation recovery time, a measure of exacerbation severity, was defined as the number of days after exacerbation onset that major lower airway symptoms (dyspnea, sputum volume, and sputum purulence) were still being recorded. If no major symptoms were recorded on a single day but this day was bracketed by days when major symptoms were present, the exacerbation was considered to be continuing throughout that period.
Seasonality was examined by comparing the incidence of the three exacerbation types in winter (November–January) with that in summer (June–August).
Data were analyzed using STATA 8.2 (Stata Corporation, College Station, TX). Normally distributed data are presented as mean and SD, and skewed data are presented as median and interquartile range (IQR). A P value ≤ 0.05 was considered statistically significant. Comparisons between groups were made using Wilcoxon rank-sum, t test, and chi-squared tests as appropriate.The key analysis was an assessment of whether exacerbations cluster in time. If exacerbations occurred randomly, then the probability of remaining exacerbation free after a first event would decay exponentially with time. To assess this, we compared the actual distribution of the timing of second exacerbations with that predicted by an exponential function using a methodology previously applied to the assessment of paroxysmal atrial fibrillation (10). The unit of analysis was the exacerbation rather than the individual patient. First, we calculated the interexacerbation intervals and, from these, the cumulative distribution of exacerbations in 7-day periods. To this we fitted a monoexponential function of the form F(t) = 1 – e−(kt), where t is time and k is the rate parameter. We also fitted a Weibull function, which has a similar form but with the additional term λ, called the shape parameter: F(t) = 1 – . When λ = 1, the Weibull function is equivalent to the exponential function, and the exacerbation rate would be constant over time, indicating an exponential decay. However, if λ < 1, then the exacerbation rate would fall over time, and there would be more early exacerbations than late, suggesting a temporal clustering of events. Conversely, if λ > 1, then the exacerbation rate would increase with time, and there would be a period after a first exacerbation when a second exacerbation would be less likely. These concepts are illustrated in Figure 2. The exponential and Weibull functions were fitted using the nl command in Stata 8.2, which uses an iterative procedure (modified Gauss-Newton method). The rate parameter had a starting value of 0.1, and the shape parameter had a starting value of 1. Iterative tolerances were set to 10−9.
The baseline characteristics of the 297 patients are reported in Table 1, which demonstrates that the cohort are typical of symptomatic patients with COPD attending primary and secondary care (mean FEV1, 45.4%). The study is based on a total of 903.7 years of daily follow-up data, with each patient contributing a median of 2.16 (IQR, 1.36–3.87) years.
|FEV1, % predicted||45.4||18.1|
The number of patients and exacerbations included in the analysis is summarized in Figure 3. There were a total of 2,462 exacerbations during the study. Two hundred seventy-three events that would otherwise have met our symptomatic definition of exacerbation were not separated by 5 days free of all symptoms from the preceding exacerbation. These episodes may have represented symptom variation during recovery of this first exacerbation, and the following analysis is therefore based solely on the remaining 2,189 exacerbations. Of these, because not all patients had experienced a subsequent exacerbation during the period of analysis, we were able to calculate 1,923 interexacerbation intervals.
The distribution of the interexacerbation intervals is illustrated in Figure 4. In weeks 0 through 2, the observed number of recurrent exacerbations was low. This is because of our requirement for two consecutive days of symptoms to define an exacerbation and five consecutive days free of symptoms to define recurrence. It is therefore impossible to have a recurrent exacerbation in the week after the onset of a first. From week 3 to week 8, the data suggest that there were more recurrent exacerbations than would be expected were these events occurring randomly.
We have confirmed this observation using two separate analyses. First, the shape parameter (λ) of the Weibull function was 0.966 (95% confidence interval [CI], 0.948–0.985). This was significantly different from 1.000 (P < 0.001), and therefore the actual distribution of interexacerbation intervals was significantly different from the expected exponential function. Because λ < 1, there were a greater number of shorter interexacerbation intervals than expected, which is consistent with a higher risk of subsequent second exacerbation earlier than later after a first event. Therefore, exacerbations do cluster together in time. The effect size would be greater (λ lower) if the small number of exacerbations occurring in weeks 1 and 2 had not been included in the analysis. In support of our primary analysis, between week 3 and week 8 there were 633 exacerbations, which is 103 (19.4%) more than that predicted by the exponential function (paired t test; P = 0.040).
We also repeated the analysis using only those exacerbations that met a Health-Care Utilization definition of exacerbation (n = 1157 interexacerbation intervals). When defined by the need for antibiotics and or systemic corticosteroids, the shape parameter (λ) was 0.965 (95% CI, 0.949–0.981), indicating that our findings are robust using this alternative approach, which is commonly applied in clinical trials.
Of the 2,189 exacerbations, 1,084 (49.5%) were isolated, occurring at a median rate of 1.16 (IQR, 0.70–1.67) per patient per year; 410 exacerbations (18.7%) were initial, occurring at at a median rate of 0.18 (IQR, 0.00–0.66) per patient per year; and 695 exacerbations (31.8%) were recurrent, occurring at at a median rate of 0.18 (IQR, 0.00–0.89) per patient per year. Of all exacerbations that did not follow a preceding exacerbation within 8 weeks (isolated + initial, or “first” exacerbations), 27.4% (410/1494) were associated with subsequent recurrence within 8 weeks.
Of the 410 episodes in which an initial exacerbation was followed by one or more recurrent exacerbations, there were 254 episodes of exacerbation pairs (initial followed by one recurrent), 87 episodes of three exacerbations, 37 episodes of four exacerbations, and 32 episodes of five or more sequential exacerbations (initial followed by four or more recurrent exacerbations).
Isolated exacerbations were on average 25% more severe than initial exacerbations, as assessed by the duration of symptoms (5 [IQR, 2–10] vs. 4 [IQR, 2–8] days; P = 0.038). The median recovery time for recurrent exacerbations was 5 (IQR, 3–8) days, which was not significantly different from that for isolated (P = 0.630) or initial exacerbations (P = 0.072).
There were no significant differences in hospitalization rates between isolated (6.1%), initial (5.4%), and recurrent (6.3%) exacerbations (all P > 0.5).
Regarding individual symptoms, there were no differences on the day of onset between initial and recurrent exacerbations in the presence of any of the major or minor symptoms. However, coryzal symptoms, sore throat, and cough were more likely to be present in isolated than in initial or recurrent exacerbations (Table 2). Dyspnea was more commonly reported in recurrent than in isolated exacerbations.
|Isolated||Initial||Recurrent||Isolated vs. Initial||Isolated vs. Recurrent||Initial vs. Recurrent|
|Number of exacerbations||960||383||654|
We examined the rates of prescription of oral corticosteroids and antibiotics for isolated, initial, and recurrent exacerbations to investigate whether exacerbation recurrence may be associated with under-treatment of the preceding event. Treatment was prescribed at the discretion of the attending physician. In our practice, amoxicillin is commonly used as a first-line antibiotic, with ciprofloxacin reserved as a second-line agent.
The results of this analysis are reported in Table 3, which demonstrates that treatment rates for initial and isolated exacerbations were similar. This suggests that recurrence is not due to under-treatment of an initial event. However, we found that recurrent exacerbations were less likely to be treated than isolated exacerbations but that when recurrent exacerbations were treated they were more likely to receive ciprofloxacin than initial and isolated events.
|Isolated||Initial||Recurrent||Isolated vs. initial||Isolated vs. recurrent||Initial vs. recurrent|
|Number of exacerbations||994||377||647|
|Oral corticosteroids and/ or any oral antibiotic||72.3%||67.9%||65.7%||0.106||0.004||0.469|
|Any oral antibiotic||64.9%||62.6%||59.4%||0.430||0.023||0.305|
|First-line oral antibiotic (amoxicillin)||65.7%||64.1%||49.5%||0.681||<0.001||0.001|
|Second-line oral antibiotic (ciprofloxacin)||8.8%||8.3%||15.3%||0.810||0.003||0.017|
There was a significant correlation in individual patients between the total annual number of exacerbations and the number of recurrent exacerbations per year (rho = 0.81; P < 0.001) (Figure 5).
Frequent exacerbators (i.e., patients whose annual exacerbation frequency was greater than the cohort median of 1.97/yr) largely comprised patients who had experienced recurrent exacerbations (84.6%).
Patients who experienced one or more recurrent exacerbations were younger than those who did not (66.5 [8.2] vs. 68.9 [8.1] years; P = 0.010) but did not otherwise differ in sex, smoking status, FEV1, FVC, or FEV1/FVC ratio.
Exacerbations were more common in winter (November–January) than the summer months (June–August). However, the relative proportions of isolated, initial, and recurrent exacerbations did not vary by season (47.3 vs. 52.9%, P = 0.067; 19.3 vs. 16.5%, P = 0.233; and 33.4 vs. 30.6%, P = 0.330).
The major finding of this study is that exacerbations cluster together in time, with a high-risk period for recurrent exacerbation in the 8 weeks after a first exacerbation. This has important implications for the targeting of preventative interventions and the analysis of data from clinical trials.
This is the first analysis specifically designed to assess the timing of exacerbation recurrence. We have reported that approximately 27% of first exacerbations are associated with a second discrete exacerbation over the subsequent 8 weeks. This is consistent with a UK national audit of COPD outcomes, in which 34% of 1,221 hospitalized patients with exacerbations of COPD were readmitted in the subsequent 3 months (9).
Almost one third of all the exacerbations in this analysis were recurrent, following a previous exacerbation within 8 weeks, despite full recovery of the preceding event. To date, exacerbation preventative strategies must be targeted widely. These data suggest that it may be particularly important, regardless of exacerbation frequency, to target patients after an initial exacerbation. Our work is therefore the first to identify a specific high-risk period for exacerbation occurrence, which may be used in addition to strategies targeting high-risk patients (“frequent exacerbators”). Our finding of a high-risk period for recurrent exacerbation may also be important in guiding patient follow-up. Although the WHO/NHLBI GOLD document suggests follow-up at 4 to 6 weeks after a hospitalized exacerbation (11), specific guidance on exacerbation recurrence and prevention is not discussed in this article. Given our previous study reporting that early therapy improves outcome at exacerbation (12), it seems appropriate to facilitate monitoring and access to care in the period immediately after a first exacerbation.
Most COPD exacerbations are caused by tracheobronchial infection (13). The mechanisms of exacerbation recurrence remain unexplored, and it is not known whether recurrence is due to persistence of an existing organism or to acquisition of a new pathogen. The failure to eradicate bacteria with exacerbation therapy has been associated with an incomplete recovery in inflammatory markers (14), and we have recently reported a relationship between elevated serum C-reactive protein during the recovery period of a first exacerbation and shorter time to the next (15). It is therefore possible that exacerbation recurrence relates to persistence of inflammation. This has implications for the duration of exacerbation therapy. The major benefit of systemic corticosteroids at exacerbation is in accelerating the recovery of FEV1 (16), but if failure to suppress inflammation is a risk factor for subsequent recurrent exacerbation, then it may be that therapies should be assessed against, and guided by, their effects on airway infection and the acute-phase response.
The differences in symptoms between isolated and initial exacerbations are intriguing; symptoms more typical of viral infection are significantly more common during isolated events. Although our symptom data suggest that initial exacerbations were less severe than isolated events, this was not reflected in reduced prescription of antibiotics or oral corticosteroids. Therefore, exacerbation recurrence is not related to a failure to treat the first exacerbation, nor does it reflect the possibility that initial exacerbations were more likely to be unreported to health care professionals (17, 18). It may be that the nature and duration of treatment for the first exacerbation predisposes to the timing of a second: we have reported that exacerbation therapy is associated with suppression of serum IL-6 concentration to below baseline levels, with a subsequent rebound after cessation of treatment (15). Such a steroid withdrawal effect has been associated with an increased risk of exacerbation in clinical trials (19).
A second important implication of our findings is that most current analyses of clinical trial data, including Poisson and Binomial regression analyses, make the underlying assumption that exacerbations are random events (20, 21). We have demonstrated that this is not true. This may be particularly important in trials using time to the next exacerbation as an outcome measure because our data suggest that this variable is affected by the timing of a preceding event over at least an 8-week period. The important phenotype of the frequent exacerbator appears similar, but not identical, to patients who experience recurrent exacerbations. Given the important relationship between exacerbation frequency and health status (3), it is likely that recurrent exacerbations are responsible for a considerable burden of ill health.
There is an important distinction to be made between exacerbations that are recurrent after successful treatment of a first episode and treatment failure of a first exacerbation. The latter might be termed “relapse.” We have been careful to include only recurrent exacerbations in this analysis. Only by using daily diary card data, which confirms that a first exacerbation has completely recovered, is it possible to differentiate relapse from recurrence. Much of the existing work on exacerbations has therefore likely considered relapse. Factors affecting exacerbation recurrence have not been previously studied.
We believe that our methodology is robust and that the findings are generalizable for the following reasons. (1) We have used our usual validated, symptomatic definition of exacerbation and repeated the analysis using a Health-Care Utilization approach. (2) The patients included in this analysis were typical of symptomatic patients with COPD attending primary and secondary care.
It is also important to clarify some features of our methodology. By ensuring a 5-day symptom-free period between exacerbations, we have tried to exclude exacerbation relapse. It was necessary to arbitrarily define the length of this period to perform the analysis. Selecting a shorter period would increase the risk of classifying relapse as recurrence; selecting a longer period would not permit the detection of new events occurring shortly after full recovery of a first exacerbation, which was the aim of the analysis.
Addressing concerns that using exacerbations (rather than patients) as the unit of analysis may have biased the results in favor of frequent exacerbators, we have separately examined clustering in frequent and infrequent exacerbators (divided about the cohort median of 1.97/yr). The shape parameter (λ) was significantly different from 1.000 in infrequent and frequent exacerbators (λ = 0.155; 95% CI, 0.103–0.207 and λ = 0.825; 95% CI, 0.817–0.834, respectively), suggesting that temporal clustering of exacerbations is not solely a phenomenon seen in patients prone to frequent exacerbations.
A further important consideration in our methodology was the decision to define recurrent exacerbations as occurring within 8 weeks of an initial event. There were two reasons to select an 8-week period. (1) Visual examination of the data in Figure 4 suggested that by 8 weeks the number of exacerbations occurring had approximated the expected exponential function. (2) Rounded up to the nearest whole week, our previous description of a relationship between C-reactive protein and exacerbation recurrence had also been assessed over an 8-week period (15).
In summary, this is the first study to report that exacerbations cluster together in time, with a high-risk period for recurrent exacerbation in the 8-week period immediately after a first exacerbation. This concept that exacerbations are not random has important implications for the analysis of clinical trial data and identifies a specific high-risk period for recurrent exacerbation during which preventative interventions might best be targeted.
|1.||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.|
|2.||Kanner RE, Anthonisen NR, Connett JE. Lower respiratory illnesses promote FEV1 decline in current smokers but not ex-smokers with mild chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2001;164:358–364.|
|3.||Seemungal TA, Donaldson GC, Paul EA, Bestall JC, Jeffries DJ, Wedzicha JA. Effect of exacerbation of quality of life in patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med 1998;157:1418–1422.|
|4.||Soler-Cataluña JJ, Martínez-García MÁ, Román Sánchez P, Salcedo E, Navarro M, Ochando R. Severe acute exacerbations and mortality in patients with chronic obstructive pulmonary disease. Thorax 2005;60:925–931.|
|5.||Sullivan SD, Ramsey SD, Lee TA. The economic burden of COPD. Chest 2000;117:5S–9S.|
|6.||National Institute for Clinical Excellence (NICE). Chronic obstructive pulmonary disease: national clinical guideline for management of chronic obstructive pulmonary disease in adults in primary and secondary care. Thorax 2004;59(Suppl I).|
|7.||Donaldson GC, Wedzicha JA. COPD exacerbations 1: epidemiology. Thorax 2006;61:164–168.|
|8.||Aaron SD, Vandemheen KL, Hebert P, Dales R, Stiell IG, Ahuja J, Dickinson G, Brison R, Rowe BH, Dreyer J, et al. Outpatient oral prednisone after emergency treatment of chronic obstructive pulmonary disease. N Engl J Med 2003;348:2618–2625.|
|9.||Roberts CM, Lowe D, Bucknall CE, Ryland I, Kelly Y, Pearson MG. Clinical audit indicators of outcome following admission to hospital with acute exacerbation of chronic obstructive pulmonary disease. Thorax 2002;57:137–141.|
|10.||Gillis AM, Rose MS. Temporal patterns of paroxysmal atrial fibrillation following DDDR pacemaker implantation. Am J Cardiol 2000;85:1445–1450.|
|11.||Global Initiative for Chronic Obstructive Lung Disease. Global strategy for the diagnosis, management and prevention of chronic obstructive lung disease. Updated 2007. Available at: (accessed June 19, 2008).|
|12.||Wilkinson TM, Donaldson GC, Hurst JR, Seemungal TA, Wedzicha JA. Early therapy improves outcomes of exacerbations of chronic obstructive pulmonary disease. Am J Respir Crit Care Med 2004;169:1298–1303.|
|13.||Hurst JR, Wedzicha JA. The biology of a chronic obstructive pulmonary disease exacerbation. Clin Chest Med 2007;28:525–536.|
|14.||White AJ, Gompertz S, Baylet DL, Hill SL, O'Brien C, Unsal I, Stockley RA. Resolution of bronchial inflammation is related to bacterial eradication following treatment of exacerbations of chronic bronchitis. Thorax 2003;58:680–685.|
|15.||Perera WR, Hurst JR, Wilkinson TM, Sapsford RJ, Mullerova H, Donaldson GC, Wedzicha JA. Inflammatory changes, recovery and recurrence at COPD exacerbation. Eur Respir J 2007;29:527–534.|
|16.||Davies L, Angus RM, Calverley PM. Oral corticosteroids in patients admitted to hospital with exacerbations of chronic obstructive pulmonary disease: a prospective randomised controlled trial. Lancet 1999;354:456–460.|
|17.||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.|
|18.||Vijayasaratha K, Stockley RA. Reported and unreported exacerbations of COPD: analysis by diary cards. Chest 2008;133:34–41.|
|19.||Wouters EF, Postma DS, Fokkens B, Hop WC, Prins J, Kuipers AF, Pasma HR, Hensing CA, Creutzberg EC. Withdrawal of fluticasone propionate from combined salmeterol/fluticasone treatment in patients with COPD causes immediate and sustained disease deterioration: a randomised controlled trial. Thorax 2005;60:480–487.|
|20.||Aaron SD, Fergusson D, Marks GB, Suissa S, Vandemheen KL, Doucette S, Maltais F, Bourbeau JF, Goldstein RS, Balter M, et al.; Canadian Thoracic Society/Canadian Respiratory Clinical Research Consortium. Counting, analysing and reporting exacerbations of COPD in randomised controlled trials. Thorax 2008;63:122–128.|
|21.||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.|