Overarching discussion
Two COPD populations were considered in this report, stable and post-hospital, in order to account for underlying risk of future exacerbations. Recent hospitalisation (due to exacerbation) was used as a proxy for a higher risk of recurring exacerbation. This division could be seen as artificial, as it does not take into account overall exacerbation history, but was the only possible approach given the information reported in the included studies. In order to be able to gauge the totality of the evidence, clinical effectiveness results for stable and post-hospital populations were presented, where possible, in one forest plot.
For the stable population, there was evidence available to suggest no difference between NIV and usual care in terms of survival (up to 24 months); however, there was a lack of longer-term controlled data on survival. There was a trend (not statistically significant) towards fewer hospital admissions with NIV where studies had reported this outcome, and possibly for improved QoL, although this was not consistent. Not all available evidence on admissions in a stable population could be incorporated into the meta-analysis (and subsequent economic modelling), and it is possible that, based on all evidence, the effect may have been more equivocal. There was less evidence available for the post-hospital population. A benefit from NIV in terms of survival was demonstrated in non-RCTs but not confirmed by RCTs. Hospital admissions findings were inconsistent across three RCTs in this population and there was a lack of QoL data.
In line with these results, the speculative economic model found that NIV may be cost-effective in a stable population at a willingness-to-pay threshold of £30,000 per QALY gained. This is driven by the trend towards fewer hospital admissions but is associated with large uncertainty. The disparity between effect estimates for the post-hospital population is reflected in the cost-effectiveness estimates which range from NIV being dominated by usual care, to ICERs for NIV of near or below £10,000 per QALY gained. This is the first model-based economic evaluation in domiciliary NIV; however, it remains speculative because of the uncertainty around a potential benefit from NIV for a stable population and because of limited and inconsistent evidence for the post-hospital population. The main drivers of the model were clinical effectiveness estimates, duration of benefit (which is currently unknown) and, to a lesser extent, NIV costs. Speculative modelling indicated that a reduction of 24% and 15% (for stable and post-hospital populations respectively) in rate of hospital admissions (per patient per year) would result in NIV being cost-effective at a threshold of £30,000 per QALY gained. Similarly, a 2.5% utility improvement would also hypothetically make NIV cost-effective in a stable population, while a similar change in the post-hospital population does not much affect the current cost-effectiveness estimates. A 2.5% improvement in utility is, however, difficult to translate into a clinically meaningful measure. Given the limited QoL evidence, no changes in utility were included in the base case (other than those caused by avoided hospital admissions), which is contrary to anecdotal patient reports of improvement in daily living with NIV. Assuming there was an improvement in QoL (of daily living), which has not been adequately captured by studies so far, this may increase cost-effectiveness of NIV.
One potential drawback of the model is that it had to consider the two populations separately, while in a real-life setting there is likely to be much more of a continuum of risk. It is possible that the post-hospital population considered represents only a small proportion of COPD patients at the more severe end of the disease spectrum. There was limited information in the included studies on baseline risk in relation to exacerbation history. The model did, however, account (for both populations) for the fact that baseline risk changes, both over time and dependent on exacerbations. Further uncertainty in the model stemmed from the fact that estimates for clinical parameters for the NIV and usual-care arms were in some cases obtained from different sources, and it is unclear how potential differences in populations may affect model outputs.
The remit for this report specified ‘patients with stable end-stage COPD plus chronic HRF, who have required assisted ventilation (whether invasive or non-invasive) during an exacerbation or who are hypercapnic or acidotic on long term oxygen therapy (LTOT).’ Based on the existing clinical effectiveness evidence, no conclusions could be drawn on whether or not a certain type of patient is more likely to benefit from NIV (e.g. in terms of LTOT use or level of hypercapnia), or whether or not a certain type of NIV is more favourable (e.g. higher pressure), and the economic model has not considered any such differences. Most included populations were hypercapnic and in many studies a majority of patients were on LTOT. Performing subgroup analysis based on reported mean baseline CO2 values would have meant dichotomising trials based on an arbitrary threshold and this was not considered appropriate. Exploratory analyses undertaken across both stable and post-hospital populations suggested a trend towards a correlation between changes in CO2 and hospital admissions. Such a potential correlation was not observed for mortality. However, the analysis is using aggregate data for change in CO2 and also for mean difference in hospital admissions, and a causal association therefore cannot be inferred even if there is potential biological plausibility. Further, this was a post-hoc analysis, which is subject to a number of limitations (see Appendix 7). It does suggest that there needs to be further investigation into the association between CO2 and clinical outcomes such as hospital admissions.
The evidence overall indicates that some patients may be more likely to benefit from NIV than others, but it is difficult to relate this to underlying risk of future exacerbations. It therefore remains uncertain at which level of risk it might be beneficial to commence domiciliary NIV, if indeed such a risk could be determined accurately for an individual patient during the course of their disease. The current recommendations in the UK suggest that domiciliary NIV is considered on health-economic grounds if a patient has had three hospital admissions with acute HRF.40 There may, however, be other, as yet undetermined, patient characteristics which could influence the effectiveness of NIV.
Uncertainty also remains regarding the length of time NIV may provide benefit for. There are at least two RCTs looking at the effect of discontinuing NIV (see Chapter 5, Discontinuation studies), but it was beyond the scope of this report to explore this question.
Research recommendations
Based on the current evidence, it is possible that there are some patients who may derive greater benefit from NIV, but at present it is not possible to define those patients’ characteristics.
A number of currently ongoing studies may go some way to adding to the evidence base, but this may depend to some extent on how the study populations are defined (see Chapter 4, Ongoing studies, for full details). There is at least one ongoing trial (the UK HOT-HMV trial) which includes a population with a higher underlying risk of recurrent events similar to the post-hospital populations in this report. This study had almost finished recruitment, but no results were available to include in this report. Recruitment appears to have been slower than expected in at least some ongoing studies, which may be explained by narrow inclusion criteria.
No further ongoing trials in a stable population were identified, which is perhaps a reflection of the lack of evidence of benefit in this population. Uncertainty remains regarding a potential effect on QoL and long-term effects on survival.
The results from this report will need to be re-examined in the light of any new results from the ongoing trials, particularly in terms of reducing some of the uncertainty in the economic model. As such, recommendations for additional RCTs would be premature.
Should there be a need for additional new RCTs, these could include a sham NIV arm in order to minimise potential bias, as well a higher-pressure and lower-pressure NIV arm, to enable further exploration of the relationship between pressure and effectiveness. However, some argue that sham NIV may lead to an overestimate of the potential benefit of NIV, because of its engendered disbenefits in terms of QoL, in which case a control arm based on usual care only may be more appropriate. A three-arm trial (NIV vs. sham NIV vs. usual care) may be required. In view of the small sample sizes in published and ongoing trials, as well as recruitment issues, broader inclusion criteria could be considered, together with planned subgroup analyses.
Future studies should consider measuring and reporting outcomes in a way that could usefully inform an economic model. So QoL should be measured in way that enables conversion into utilities, exacerbations should be reported with an indication of severity and the relationship between exacerbations, hospitalisations and mortality should be clearly reported. Where continuous data are reported, analysis of covariance should be used for adjusting for baseline imbalances. Alternatively, the proportion of patients achieving a level of clinically important improvement could also be reported.
An appropriately conducted IPD analysis of all study data may help to answer some of the outstanding questions about the type of patient that might benefit most from NIV. A recent Cochrane review110,124 has incorporated IPD analyses but based on a smaller group of studies and without considering hospitalisations or survival. The Cochrane analysis was not aimed at identifying potential effect modifiers. Feasibility of an IPD analysis would depend on the availability of information on potential effect modifiers for all patients and differences in methodological quality between studies. Small patient numbers and potentially insufficiently high event rates in the trials published so far may also preclude analysis of all effect modifiers of potential interest (e.g. age, GOLD severity, level of hypercapnia, use of LTOT, history of exacerbations).
