There are three main results from this systematic review. First, evidence suggests that supplemental oxygen during physical exercise does not provide a clinically relevant benefit. Second, the evidence for any other supplemental intervention is not strong enough to recommend or discourage their use in clinical practice and third, there were major methodological limitations in most trials that may explain some of the inconclusive findings. We discuss each of these results in turn.
] reported in 1956 that oxygen increased exercise performance in patients with COPD. Since then, many investigators assessed the short-term effect of increased oxygen availability during exercise[15
]. Some investigators argue that patients tolerate higher exercise intensities or longer exercise time with supplemental oxygen leading to larger training effects[43
]. Others believe that only without oxygen, an adequate hypoxemic stimulus is provided for peripheral muscles to improve exercise capacity.
The studies by Emtner[25
] and Rooyackers[28
] demonstrated that patients indeed tolerate higher exercise intensities if supplemented by oxygen. Mean differences on the CRQ domain scores, however, showed a slight but clinically not meaningful trend towards a benefit with oxygen supplementation (figure ). The trial by Emtner[25
] was the only one that showed a consistent trend towards a small benefit of oxygen on HRQL and exercise capacity. Across all studies, however, results from exercise testing were contradicting. Supplemental oxygen did prolong exercise duration in constant work rate tests, but led to considerably smaller improvements of functional exercise capacity (figure ). It was hypothesized earlier that those patients with the highest oxygen desaturation during exercise would benefit most from supplemental oxygen[45
]. The trials do not provide sufficient evidence for or against this hypothesis.
There is limited evidence on the safety of oxygen during exercise and on the safety of exercise without oxygen in patients with desaturation. Clinicians may have concerns about training in hypoxemia because of adverse events and will encourage oxygen supplementation in patients with desaturation during exercise. In theory, oxygen carries the risk of CO2
retention in COPD patients. The only trial reporting on CO2
] did not observe significant differences of CO2
levels during exercise tests with oxygen compared with exercise on room air. However, exercise tests may have been too short to assess the effect of CO2
retention. Exercise is a risk indicator for unmasking latent pulmonary hypertension[46
], but supplemental oxygen may reduce this risk by decreasing the sympathetic tone and the respiratory rate allowing for less end-expiratory pressure[47
] did not find any differences in resting mean pulmonary artery pressure between patients with and without oxygen. However, patients stopped exercising when oxygen saturation fell below 90% so that the risk of the exercise program under hypoxemic conditions on the development of pulmonary hypertensions could not be studied.
Several studies found a positive acute effect of oxygen during exercise testing on exercise capacity and a number of physiologic mechanisms for the effects of oxygen have been proposed [48
]. However, these results on the short-term benefit of oxygen during exercise testing seem not to translate into improvements of clinically relevant outcomes during exercise programs. Current data do not suggest benefit from the use of oxygen during exercise to enhance training effects (figure ), but show some benefit in terms of HRQL (figure ) Given the limited methodological quality of trials, any conclusions are vague. The general use of oxygen is only justified, if larger trials of good quality show its benefit on clinically relevant outcomes. The mechanisms of the effects of oxygen during exercise are still insufficiently understood and call for more basic research[15
Assisted ventilation also aims at increasing oxygen availability during exercise, but the trials indicated no additional benefit. An exception may represent overnight non-invasive positive pressure ventilation. This treatment may improve quality of sleep as well as daytime gas levels and respiratory muscle function thereby providing a better milieu (pH, PaO2
) for peripheral muscle function. One trial[32
] found statistically significant improvements of functional exercise capacity and also large improvements of HRQL (mean differences between groups 0.45 to 0.85 in CRQ domain scores, table 4, see Additional file 4
) with additional non-invasive positive pressure ventilation. These results support the hypothesis formulated by authors of a recent meta-analysis showing that nocturnal non-invasive positive pressure ventilation alone has no effect on exercise capacity and HRQL, but may be beneficial as an adjunct to respiratory rehabilitation[51
]. The eight trials that assessed various supplemental interventions during rehabilitation produced inconclusive results that do not allow recommendations for clinical practice yet.
An important result of this systematic review with implications for future research is the low methodological quality and small sample sizes. For example, the majority of trials did not consider stratification for important prognostic factors such as exercise capacity[52
] for randomization. In some trials there were baseline imbalances between groups, for example in terms of exercise capacity[27
]. The influence of these imbalances on the results was not investigated in any of the trials. Concealment of random allocation and blinding of treatment providers or outcome assessors was also not addressed in most trials.
Sample sizes were small except in three trials[34
]. Pragmatic trials comparing active interventions, as included in this systematic review, are very useful for clinical practice when clinicians are confronted with the choice between interventions[53
]. However, small sample sizes are problematic in pragmatic trials for at least two reasons: First, differences between study groups tend to be smaller in pragmatic trials than in trials comparing an active intervention with placebo or a sham intervention. Figure shows the results and 95% confidence intervals of a trial comparing respiratory rehabilitation with usual care and of a trial comparing respiratory rehabilitation with and without a supplemental intervention with different sample sizes. It illustrates the importance of sufficient sample sizes in pragmatic trials by showing that for pragmatic RCTs in respiratory rehabilitation, in which widely established patient-important outcomes such as HRQL are used, sample sizes of up to 40 per group will produce imprecise results (large confidence intervals). This imprecision hinders interpretation. Another reason for sufficient samples sizes is that in pragmatic trials patient profiles are usually more variable than in explanatory trials reflecting the wide patient spectrum encountered in clinical practice[53
]. The greater variability in patient profiles carries, on one side, a greater risk for confounding and, on the other side, subgroup analyses will be important to assess whether the effects differ between patient subgroups (effect modification). Subgroup analyses based on prognostically important patient characteristics will provide more differentiated evaluations than one mean for the whole study group, but they require sufficient sample for well-balanced intervention groups.
Figure 4 Sample size and interpretation of randomized controlled trials in respiratory rehabilitation. Forest plot with simulated results from two trials with varying sample size, in which the CRQ was used. Boxes with 95% confidence intervals represent point estimates (more ...)
We propose that investigators consider the following aspects in future pragmatic trials on respiratory rehabilitation: First, preliminary sample size considerations should be based on realistic estimates for expected differences between groups, which are typically smaller than in trials without active comparators. To better understand what these sample sizes mean, 95% confidence intervals around the predicted point estimate can be calculated as shown in figure . This approach will help to better foresee the consequence of different sample sizes on interpretation of the data[54
]. Second, COPD patients represent a heterogeneous group and stratification for prognostically important variables should be considered to avoid baseline imbalances that bear on outcomes[55
], as seen in some trials included in this review[27
]. Third, more attention needs to be paid to general requirements for RCTs of high quality like method of randomisation, concealment of random allocation and blinding of those who assess treatment effects.
The strengths of our systematic review include the broad literature search including several databases and extensive hand searching for trials with direct comparisons of interventions that are used in clinical practice. In addition, we contacted authors for additional data and received them from the majority of investigators. This greatly enhanced the informativeness of included studies and thereby of this review. A weakness of this review includes the discussion that is limited to the best-investigated supplements because of the number of interventions included in this review. However, the aim of this review was to analyze current evidence from a meta-epidemiological perspective not giving to much emphasis to single studies. Some may criticize that we did not pool the results from trials on supplemental oxygen during exercise using meta-analysis. However, desaturation or no desaturation during exercise was an important inclusion criterion in four of the five trials and investigators wanted to learn about the effect of supplemental oxygen in these subgroups, in particular. Therefore, we considered the patient profiles of these trials to be too different to provide meaningful pooled estimates. Instead, we provided forest plots (figures and ) to show the individual studies' point estimates and 95% confidence intervals for the CRQ domains and the exercise tests to allow comparisons across studies.
In conclusion, data for most supplemental interventions during respiratory rehabilitation are inconclusive. Oxygen during exercise does not seem to provide a patient-important additional benefit for COPD patients during a respiratory rehabilitation, but methodological shortcomings of the trials on supplemental oxygen do not allow conclusive answers. Future trials should pay careful attention to the methodological trial design and to sufficient sample sizes.