Background
Chronic obstructive pulmonary disease (COPD) is a major health problem associated with disability, healthcare utilization, and premature death [
1]. Clinical trials have established that current pharmacologic approaches can improve lung function and reduce exacerbations, but their efficacy in modifying the disease course is a subject of debate [
2]. In recent years, large long-term clinical trials evaluated pharmacologic intervention on mortality rates (as primary or secondary outcomes) in patients with COPD [
3,
4]. Although these studies have suggested that current inhaled therapies have some effects on mortality [
3,
5], no study has shown unequivocally that current pharmacotherapy reduced mortality in patients with COPD [
6].
One possible explanation for the limited effect seen in long-term clinical trials on mortality [
3,
5] could be an actually limited efficacy of current drugs on this outcome. Alternatively, these results may have been related to the design of clinical trials that assessed mortality in patients with COPD, and the heterogeneity of baseline mortality risk among those patients recruited in these clinical trials may have played an important role: it is well established that most trial outcomes (e.g. mortality) occur in a relatively small number of high-risk patients, while most patients are at much lower risk [
7,
8]. A clinical trial enrolling a large group of patients with COPD at low risk of outcome may find the therapy useless overall, but miss detecting its efficacy in high-risk patients [
7]. These considerations suggest the importance of risk stratification in clinical trials [
7]. However, stratification of mortality risks was not performed in the recent Towards a Revolution in COPD Health (TORCH) [
3] and Understanding Potential Long-term Impacts on Function with Tiotropium (UPLIFT®) [
4] clinical trials, which may have limited their ability to detect important differences in mortality.
In the present study, we explored the possibility that the population recruited in the UPLIFT® study, a 4-year trial of tiotropium use in patients with COPD, was composed of patients with large variations in baseline mortality risk. Our strategy was to examine mortality risk heterogeneity using a cluster analysis of variables selected for their previously reported association with increased mortality, and obtained at recruitment in the study. Our goals were to identify clusters of patients with different mortality risks at baseline and to examine the impact of risk stratification on the evaluation of tiotropium effect on mortality, as well as to an outcome associated with increased mortality, e.g. exacerbations. Some of the results of these studies have been previously reported in abstract form [
9].
Discussion
In the present study, we reasoned that classification of patients with COPD using a cluster analysis of multiple clinical variables previously associated with increased mortality may identify clusters of patients with variations in baseline risks of mortality. We found that risks and causes of mortality were markedly different among patients included in the UPLIFT® study. Cluster 2, which accounted for 41% of all patients, contained patients at low risk of mortality and very low risk of respiratory mortality; these patients were also at low risk of exacerbations and severe exacerbations. Compared with this low-risk cluster, risks of all-cause and respiratory mortality were 2.6- and six-fold higher in cluster 3, respectively, and were also increased in clusters 1 and 4, although to a lesser extent. Tiotropium reduced exacerbations in all clusters, and in particular reduced severe exacerbations in high-risk patients (cluster 3). Tiotropium was further suggested to reduce all-cause and respiratory mortality in high-risk patients (cluster 3), but not in other clusters. These findings underscore the potential impact of baseline risk stratification on the interpretation of the results of large clinical trials in patients with COPD.
The present analysis revealed that more than 40% of patients (cluster 2) included in this study had low risk of all-cause mortality and very low risk of respiratory mortality at the time of inclusion in the study. Patients in clusters 1 and 4 (who also represented more than 40% of all patients included) had higher baseline risk for all-cause mortality, but were also at low risk of respiratory mortality. Thus, more than 80% of patients included in the UPLIFT® study were at low baseline risk of respiratory mortality, which is in line with the risk of dying in the general population of those with COPD. When generating the hypothesis that tiotropium could reduce respiratory mortality, a reasonable hypothesis given its effect on FEV
1 and exacerbations [
4], the inclusion of this large number of low-risk patients may have resulted in missing the potential benefit of tiotropium in high-risk patients [
7,
8]. In support of this latter hypothesis, our analysis strongly suggested that tiotropium had a rather large impact on all-cause and respiratory mortality in high-risk patients (cluster 3), although the analysis was likely under-powered to show statistical significance.
To the best of our knowledge, no other study has evaluated the impact of heterogeneity in baseline risks and causes of mortality on the results of pharmacologic clinical trials in patients with COPD. In the past, analyses of potential efficacy of pharmacotherapy on mortality in patients with COPD have relied on conventional (and most often
post-hoc) subgroup analyses of individual variables to examine variability in treatment effects among subgroups [
3,
5]. However, conventional subgroup analyses are inadequate to detect large and clinically important differences in treatment effect among patients when multiple factors determine risk [
8]. This is typically the case for mortality risk in patients with COPD, which is determined by multiple factors including the level of FEV
1 impairment, age, patient-centered outcomes (e.g. dyspnea, HRQoL), and comorbidities (e.g. malnutrition) [
12,
13]. For example, although cluster 3 contained 81% of GOLD Stage 3 and 4 patients (representing 54% and 22% of all GOLD Stage 3 and 4 patients, respectively), GOLD stage
per se was not a significant covariate in the effect of tiotropium on all-cause mortality in a previously published analysis of the UPLIFT® study [
5]. These data suggest that multi-dimensional assessment of baseline mortality risk is more appropriate to provide stratification of mortality risk in patients with COPD.
Our data strongly suggest that such stratification of clinical trials on baseline mortality risk is of utmost importance in studies assessing specifically mortality. Prognostic enrichment strategies, aimed at including subjects at high risk of mortality, may be especially useful in such studies [
18]: because the appropriate sample size necessary to show a reduction in mortality rates will depend on effect size and the event rate in the placebo group, selecting a population at high risk of mortality would increase the likelihood of showing an effect of a drug, if there is one [
18]. Prognostic enrichment may not increase the relative risk reduction (e.g. percent improvement in mortality), but will increase the absolute event counts, allowing for a smaller sample size [
18]. Another advantage of this strategy will be to allow reduction in the duration of the study, presumably lowering study drop-off rates, a common problem in long-term clinical trials, and reducing the cost of the study.
In future studies, prognostic enrichment may rely on validated multivariate risk indices collected at study entry [
12,
13]. The BODE index, which is a predictor of all-cause and of respiratory mortality [
13], may be particularly suited for selecting high-risk patients (and excluding low-risk patients) in clinical trials assessing the effect of therapy on mortality in patients with COPD. However, the BODE index requires a 6-minute walk test, which may be difficult to use in the screening of patients for large-scale multicenter studies. Alternative strategies for recruiting patients at high risk of mortality may eventually be considered. Based on the characteristics of high-risk subjects (cluster 3) in the present study, heavy current or ex-smokers with severe airflow limitation and impaired HRQoL (as measured by a high SGRQ total score) would be candidates for recruitment in these studies. Another possibility would be to recruit subjects who had experienced at least one hospitalization for COPD in recent years, as several studies have shown that 3-year mortality rates in these patients is approximately 50% [
19,
20]. Although the most effective enrichment strategy remains to be established, we suggest that future clinical trials assessing treatments aimed at reducing mortality in patients with COPD should include patients at high risk of mortality.
The present analysis was performed on patients with complete data for five variables, which were previously reported to predict mortality in patients with COPD and available in the UPLIFT® study. The UPLIFT® study was performed in 5993 patients, but 287 (4.8%) patients had missing data for these variables, leading to their exclusion from the analysis. Because these data were missing at random (not shown), as it is often the case in large clinical trials, this is unlikely to affect our conclusions significantly. Important predictors of mortality (e.g. dyspnea [
13] and a history of severe exacerbations [
21]) were not available for inclusion in the cluster analysis at study entry. Further, limited clinical data were available to characterize differences fully among the identified clusters. However, our analysis was validated by showing that these four clusters had marked heterogeneity in future outcomes, including exacerbations, hospitalizations, and all-cause and respiratory mortality. In the present study, risk of exacerbations was also heterogeneous among clusters, but tiotropium reduced exacerbations in all clusters of patients, confirming its positive impact on this outcome [
4]. Consequences of the absence of risk stratification on the results of the UPLIFT® study were less important for exacerbations than for mortality. First, mortality was a relatively rare outcome that occurred in approximately 15% of patients, whereas exacerbations occurred in more than two thirds of patients over 4 years. Second, increases in all-cause and respiratory mortality risks in the highest- versus lowest-risk clusters (up to 2.6- and six-fold, respectively) were more important compared with those in exacerbation risks (fewer than two-fold). However, the impact of risk heterogeneity was also important for severe exacerbations leading to hospitalizations (a relatively rare event with substantial heterogeneity in risk), and tiotropium significantly reduced hospitalizations only in high-risk patients (cluster 3).
Competing interests
In the past 5 years, Pierre-Régis Burgel has received fees for speaking, organising education or research, or consulting from Almirall, Nycomed-Takeda, AstraZeneca, Boehringer Ingelheim, Chiesi, GlaxoSmithKline, Novartis, Pfizer. Jean-Louis Paillasseur was full time employee of Clindatafirst and EFFI-STAT, CROs which received fees from Boehringer Ingelheim, Nycomed and AstraZeneca. Daniel Dusser has received fees for speaking, organising education or research, or consulting from Boehringer Ingelheim, Novartis, Pfizer, Chiesi, Dey Pharma and Nycomed. Nicolas Roche has received (i) fees for speaking, organising education or research, or consulting from Aerocrine, Almirall, Altana Pharma-Nycomed-Takeda, AstraZeneca, Boehringer Ingelheim, Chiesi, GlaxoSmithKline, MEDA, MSD-Chibret, Mundipharma, Novartis, Pfizer, Stallergenes, TEVA; (ii) research grants from Novartis, Nycomed, Boehringer Ingelheim and Pfizer. Yufeng Liu has received consulting fees from Boehringer-Ingelheim. Marc Decramer has received research grants or fees for consulting or speaking from Novartis, Nycomed, Boehringer Ingelheim, GlaxoSmithKline, Altana and AstraZeneca. Dacheng Liu, Armin Furtwaengler, and Norbert Metzdorf are full-time employees of Boehringer Ingelheim Pharma GmbH & Co KG.
Authors’ contributions
PRB, JLP, DD, NR, and MD conceived and designed the paper; DL and YL performed the statistical analysis; PRB, JLP, DD, NR, AF, NM, and MD performed the analysis and interpretation. PRB, JLP, DD, NR, and MD helped to draft the manuscript for important intellectual content. All authors read and approved the final manuscript.