We thank Mydin et al1 for their interest in our article.2 They contend that the main findings are explained by patient selection and that for many of these patients management with non-invasive ventilation (NIV) is inappropriate and end-of-life care pathways should be introduced instead.

We agree that patient selection is one of the important explanations for the difference in outcomes of observed clinical practice when compared with the randomised controlled trial (RCT) results and repeatedly emphasise this within the discussion. Patient selection alone however is unlikely to explain the poor survival observed as we also demonstrate that patients subject to pre-hospital oxygen poisoning have poorer outcomes and patients treated with NIV often have significant delays in the initiation of treatment contrary to the RCT evidence and guideline recommendations.

We have also found that patients who fit the RCT and guideline criteria for NIV do not in some cases receive this treatment while escalation to invasive mechanical ventilation (IMV) is the exception. The study also describes inadequate documentation of both escalation plans and do not resuscitate orders. So it is quite possible that some of these patients are receiving NIV when instead end-of-life care may be more appropriate, but there are many other important issues that explain the observed outcomes. End of life in chronic obstructive pulmonary disease (COPD) exacerbations is a difficult area of care for which the guidelines are currently vague and where our own data have shown that in large-scale studies all predictors of outcome combined only explain a minority of the variance in outcome.3 Finally studies of patient choice when offered IMV for respiratory failure in COPD have shown patient preference for intervention beyond that considered appropriate by intensivists in many cases.4 In essence, this is an area where prospective research is required to better understand both the wishes of patients and the costs and benefits of interventionist or palliative choices.