Does home oxygen therapy (HOT) in addition to standard care reduce disease severity and improve symptoms in people with chronic heart failure? A randomised trial of home oxygen therapy for patients with chronic heart failure

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 06/80/01. The contractual start date was in January 2010. The draft report began editorial review in November 2014 and was accepted for publication in May 2015. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Caroline Fairhurst, Sarah Cockayne, Sara Rodgers and David Torgerson all report other grants from the National Institute for Health Research Health Technology Assessment programme.

Copyright statement

© Queen’s Printer and Controller of HMSO 2015. This work was produced by Clark et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Heart failure

Heart failure is a clinical syndrome that arises when the heart fails to pump in a manner adequate to meet the body’s needs. It is increasingly common and affects between 1% and 2% of the UK population. Its incidence and prevalence rise markedly with age. 1 The most common cause of heart failure is myocardial infarction and, as more people survive acute myocardial infarction with modern therapy, so the population of patients with damaged heart muscles grows. 2

Heart failure is the single most common cause for admission to hospital in England and Wales and, following admission to hospital, there is a 25% chance of readmission or death within 12 weeks. By 1 month after an index admission, 15% of patients have died, either as an inpatient or during the days following discharge. 3 The prognosis of heart failure is bleak if it is not well treated. However, one of the greatest success stories of modern medicine is the dramatic improvement in prognosis for patients with chronic heart failure (CHF). Good medical management largely consists of medicines designed to block the adverse consequences of neuroendocrine activation such as beta-blockers, angiotensin-converting enzyme inhibitors and mineralocorticoid receptor antagonists. In selected patients, treatment with cardiac resynchronisation therapy also improves prognosis and, taken together, these treatments approximately double life expectancy. 4

Chronic heart failure has been recognised for many years as having the greatest symptomatic burden of any chronic medical condition. 5 The cardinal symptoms of heart failure are breathlessness and fatigue, particularly on exertion. Worsening breathlessness is part of the cause of most admissions to hospital with heart failure, although many patients also complain of ankle swelling due to fluid retention. Drug therapy is very successful in controlling symptoms, and can induce ‘remission’ in a number of patients; that is, their symptoms can remit almost entirely. However, the clinical course for most patients with heart failure tends to be one of gradual decline, often punctuated with episodes of severe deterioration resulting in hospitalisation.

Symptom severity is most commonly measured using the New York Heart Association (NYHA) classification of symptoms (Table 1).

Towards the end of their lives, many patients with CHF become very symptomatic, with limiting breathlessness on minimal exertion (class III) and even at rest (class IV). Once they have reached this stage, although patients need continued treatment with drugs known to improve prognosis, the emphasis of treatment becomes the relief of symptoms, that is, palliative care. However, although drug treatment with diuretics (which relieve fluid congestion), other drugs and pacing devices may relieve symptoms, for many the last few months and even years of life can be miserable, with persisting severe breathlessness on minimal exertion and episodic hospitalisation.

Although there is some evidence that opioids may relieve breathlessness in patients with chronic airways disease and cancer,6–8 the evidence is mixed in heart failure,9–11 and there is no specific intervention for the intractable breathlessness of severe CHF. Another frequently encountered group with severe breathlessness is those patients with chronic airways disease. In patients with chronic airways disease who also have hypoxia, there is reasonably robust evidence that long-term oxygen therapy (LTOT) can improve prognosis as well as symptoms (see Pulmonary disease and oxygen). By extension from these data, physicians often use home oxygen therapy (HOT) for patients with severely symptomatic heart failure. There is, however, no evidence that LTOT is helpful in CHF, either for the relief of symptoms or to improve prognosis.

Pulmonary disease and oxygen

For many years, patients with chronic airways disease or chronic obstructive pulmonary disease (COPD) have been treated with LTOT, particularly if they have evidence of hypoxia at rest. Treatment is given for at least 15 hours a day (including overnight). The evidence for the benefit of oxygen therapy comes from randomised clinical trials; the Medical Research Council’s (MRC’s) oxygen trial12 and the Nocturnal Oxygen Therapy (NOT) trial13 are perhaps the best known. A Cochrane review of oxygen therapy for patients with chronic airways disease suggests that ‘long[-]term home oxygen therapy improved survival in . . . COPD patients with severe hypoxaemia (PaO₂ [partial pressure of arterial oxygen] less than 55 mmHg (8.0 kPa))’. 14 In the MRC’s oxygen trial,12 treating five patients with severe hypoxaemic COPD with LTOT saved one life over the 5-year study period. 14 The prognostic benefits were not apparent until after more than 1 year of therapy.

Although it does not affect prognosis in people with more modest hypoxaemia, LTOT does appear to help by reducing the severity of breathlessness. 15 There is no evidence that NOT alone (in other words, giving oxygen only at night) improves prognosis. 14 The authors of the systematic review and meta-analysis observed significant heterogeneity in most of their analyses and pointed out that most studies were either single blinded or not blinded at all. They therefore recommended an individual approach to care until data from large randomised controlled trials (RCTs) are available.

Data on the effects of oxygen therapy on quality of life (QoL) in patients with chronic airways disease are not clear-cut. Although the MRC reported that symptoms improved, few data were given. In patients with moderate hypoxaemia, the Cochrane meta-analysis reported a reasonably robust improvement in breathlessness equivalent to a reduction of 0.78 cm on a 10-cm visual analogue scale. 15

It is difficult to estimate adherence to LTOT in people with airways disease. Using the oxygen delivery system for 15 hours per day is clearly burdensome, and most studies suggest that adherence to this demand is less than 50%. 16 The summary figure of 45–70% is commonly quoted,17 but the studies from which the figures are derived are now quite old (see Chapter 9 for discussion). The only recent study suggests that adherence remains poor. 18

Although the original studies demonstrated a positive relationship between benefit and duration of oxygen, the mechanism is not clear. People who used oxygen for a longer period of time would have been more likely to have prevented worse desaturations during sleep or exertion, and the same benefit could have been achieved by supplemental oxygen at night only, or during exertion. Furthermore, if the prevention of exertion-induced desaturation helped exercise tolerance, then increased physical activity and reconditioning over time could have been the mechanism of improved symptoms and prognosis. 19 However, given that the only studies to show an improvement in prognosis had a target of oxygen use for long periods of time during the 24 hours, 15 hours per day remains the recommended prescription for prognostic benefit.

Heart disease and oxygen

Oxygen is commonly prescribed for patients with heart disease. There is a widespread perception that oxygen therapy can do no harm and may possibly be helpful and, thus, patients are often given high concentrations of inspired oxygen immediately following acute myocardial infarction or when they are admitted with acute pulmonary oedema. It is also commonly used during an admission for CHF.

Patients with severe (or even end-stage) CHF can appear very similar to patients with severe chronic airways disease. They are breathless at rest or on minimal exertion despite maximal medical therapy. A consequence is that HOT is often prescribed for severely breathless patients with heart failure, even in the absence of hypoxia, particularly towards the end of life.

There is only very limited evidence for the use of oxygen in heart disease, and much of the evidence suggests that oxygen might be harmful. In a study of patients with acute myocardial infarction, oxygen was given at as near to 100% as possible. Oxygen therapy was associated with a fall in cardiac output and stroke volume, together with a rise in heart rate. 20

The failing heart requires a higher filling pressure. The higher the filling pressure, the worse the cardiac function; hence an intervention causing a rise in filling pressure is deleterious. In a study of 10 patients with CHF,21 100% oxygen caused a rise in cardiac filling pressure, a fall in cardiac output and an increase in systemic vascular resistance (SVR). The SVR represents the load against which the heart has to work: the higher the SVR, the greater the work required of the heart. In another study of 12 patients with CHF,22 high-dose oxygen was associated with an increase in left ventricular (LV) filling pressure.

In contrast, in a study of patients with CHF given lower doses of oxygen (50%), exercise capacity increased and patients were less breathless and had a lower level of ventilation during exercise than during exercise with room air. 23 Findings from studies are inconsistent; another study has suggested that supplementary oxygen has little effect on exercise performance. 24 There is little evidence of the effect of oxygen when given to patients with heart failure at the much more modest levels used for treating chronic airways disease.

There is no evidence on whether or not the low-dose oxygen delivered by home oxygen concentrators is safe in patients with heart failure. There is no evidence about the effects of low-dose oxygen on haemodynamics in patients with severe heart failure.

The equivocal findings are perhaps not surprising. Oxygen is perhaps likely to be helpful only to people who are hypoxic (i.e. have low levels of arterial oxygen). Where it has been measured, oxygen has been found to be normal, or even high, during exercise in patients with CHF. 25 When patients with CHF are found to be hypoxic, there is usually an alternative explanation, such as coincident lung disease or congenital heart disease. 26 Thus, although patients with CHF may resemble patients with chronic airways disease clinically, they are much less likely to be hypoxic, and so might be thought, a priori, to be less likely to gain benefit from oxygen treatment.

Home oxygen therapy for breathlessness

There is surprisingly little evidence that oxygen is effective in treating breathlessness per se. A large observational study of patients with a variety of causes of breathlessness suggested that oxygen therapy was of no benefit,31 and apart from the review and a meta-analysis in people with COPD discussed in Pulmonary disease and oxygen,15 other systematic reviews identified no evidence that supplemental oxygen helped in the relief of breathlessness in patients with heart failure or patients with breathlessness from other causes in the absence of hypoxia. 32–35

A subsequent RCT compared LTOT via concentrator with a sham concentrator for refractory breathlessness due to a mixture of aetiologies (mainly COPD or cancer). Although breathlessness improved over 7 days in both groups, neither was superior. The authors suggested that it was simply airflow over the nasal mucosa, and not oxygen, that might have been the therapeutic intervention. 36

Despite the absence of any evidence in favour of using oxygen, HOT is commonly prescribed for severely symptomatic patients with end-stage heart disease: indeed, HOT in the UK is most commonly prescribed for conditions other than COPD (Department of Health, 2011, unpublished data).

The National Service Framework for coronary disease specifically recommends considering the potential benefit from ‘palliative care services and palliation aids (e.g. home oxygen)’. A Canadian study demonstrated that nearly 30% of oxygen prescribing costs were a result of palliative oxygen prescription (i.e. the patients did not meet the formal criteria for oxygen prescription for respiratory disease). 37 In another survey, 77% palliative physicians gave ‘intractable dyspnoea’ as the most common reason for prescription of home oxygen. 38 Fifty per cent of heart failure patients have breathlessness that markedly limits exercise in their last year of life,35 and in an Australian study between 10% and 20% of patients receiving palliative care were treated with HOT. 39 A small UK study found that 25% of patients receiving HOT had heart failure as their underlying pathology. 40 Finally, it is important to note that data from the Department of Health suggest that between 24% and 43% of the home oxygen prescribed to approximately 85,000 patients in England is not used or leads to no clinical benefit (Department of Health, 2011, unpublished data).

The home oxygen therapy trial rationale

Home oxygen therapy is potentially burdensome for patients and their carers. The concentrator has to be fitted to the home, usually requiring some drilling through walls. The device consumes electricity, although the costs are met by the NHS. The patient is encumbered to a degree by the oxygen. It is delivered through a nasal cannula, usually via a long length of piping, which limits movement. For some, oxygen cylinders to supply oxygen when the patient leaves the house are needed. The oxygen can leave the nose and throat feeling very dry. In addition, the stream of oxygen-enriched air is a potential fire hazard, particularly for patients who continue to smoke.

Home oxygen therapy is thus expensive to the health service and burdensome to the patient, and there is little evidence of its effectiveness. In the absence of any evidence-based guidance on the use of oxygen therapy for patients with CHF despite its widespread use, there is a need for a trial of HOT to find out if patients gain any benefit from its use.

Therefore, the HOT trial was designed to address the question of whether or not there is a role for HOT in patients with severely symptomatic heart failure, in terms of its effect on the symptoms for which it is usually prescribed (e.g. breathlessness), QoL and prognosis. We also planned to conduct a cost-effectiveness analysis.

The study consisted of three parts. The main study was designed to measure the impact of HOT on QoL in severely symptomatic patients. A qualitative substudy was designed to assess the burden on patients and their carers, and an acute oxygen substudy was designed to study whether or not there was any effect on haemodynamics of oxygen given in the same concentration as used by concentrators at home.

Trial structure and protocol

The trial was designed in response to a call from the NHS research and development Health Technology Assessment (HTA) programme (HTA number 06/80; see Appendix 1). The grant for the trial was originally awarded in March 2008.

Stage 1: the North East Oxygen Network trial

The HOT trial was originally named the North East Oxygen Network (NEON) trial. The original conception of the trial was an attempt to provide evidence from a double-blind trial of the effectiveness of HOT in patients with severely symptomatic CHF. Four centres in the UK were involved: Hull, Leeds, Darlington and Leicester.

In the absence of any data from adequately powered studies, it was impossible to know what duration of oxygen therapy might be useful. Whether long-term therapy, following the example of successful treatment for patients with chronic airways disease, or shorter-term overnight oxygen therapy alone was the better treatment was unknown.

The study was thus designed in two phases. In the first phase (Figure 1), patients were to be randomised to receive either NOT overnight, assumed to be around 8 hours per night, or LTOT, aiming for at least 15 hours per day. A second randomisation would allocate patients within the two arms to receive either a true or a sham oxygen concentrator. The design of the second phase depended on the results of the first; whichever of NOT or LTOT appeared the better in phase 1 would then be formally tested in the second phase.

FIGURE 1.

The original design for the NEON trial.

There was discussion among the research group about the optimum design. Some felt that an open pragmatic trial was best, as this would estimate treatment effectiveness rather than efficacy, which is an important question for the NHS. On the other hand, a placebo or sham trial would estimate the true treatment efficacy of oxygen and ensure blinded assessment of outcomes. Indeed, there is some evidence to suggest that the symptom of breathlessness is partially relieved by air blowing over the face. 36,41 Consequently, the original trial design was to use sham concentrators for the control treatment. HOT was to be delivered in the trial using oxygen concentrators rather than oxygen cylinders. A concentrator uses room air and removes a fraction of the nitrogen catalytically, so that the resulting inhaled gas is room air with some nitrogen removed; the concentration of inspired oxygen thus increases from 20.9% to approximately 28%. The sham machines would have delivered only room air.

The aim of the pilot was to recruit 120 patients, 30 in each of the four potential arms (see Figure 1). The results of the initial pilot study were then to inform the second phase study. Whichever of the two oxygen delivery schedules was the more successful would be used in a three-way comparison of oxygen therapy, sham therapy and open-label best medical therapy (BMT).

The investigators negotiated with Air Products Healthcare (Air Products and Chemicals, Inc., Allentown, PA, USA), which was able to manufacture sham machines which delivered room air. At this stage, recruiting sites were thus restricted to those whose contract for home oxygen supplies was with Air Products Healthcare. Plans were developed for the two-stage randomisation.

The original grant application to the HTA programme was with this study design.

Problems with the North East Oxygen Network trial

Trial management

The University of Hull initially agreed to sponsor the trial; however, after a period of some months it decided not to sponsor the study because of the potential issue with indemnity which could arise if a sham oxygen concentrator machine was accidentally given to a patient outside the trial. In addition, a full-time trial manager had not been included in the original submission to the HTA programme.

These problems were rectified by the Hull and East Yorkshire NHS Hospitals Trust agreeing to sponsor the study and through the establishment of a trial management function.

Sham machines

A crucial component of the trial, as originally conceived, was the use of sham machines to deliver room air. However, although it was relatively straightforward for Air Products Healthcare to deliver the sham machines, their use led to two insuperable logistical problems.

As the sham machines had to look as similar to the real devices as possible, it was felt by the company and the sponsor to be a substantial problem that the sham machines might not be kept out of the general pool of machines intended for use in delivering HOT to non-trial patients.

In addition, as the sham machines had to have their alarm functions disabled, there was a risk that they would not detect electrical faults that could potentially be a fire hazard. Specific trial insurance for their use was required, but was too costly to allow the trial to proceed.

Eventually, the trial management group accepted that the problems being raised made the trial as originally conceived impossible to run. Following a HTA site monitoring visit by Professor Jenny Hewison and Dr Vaughan Thomas on 24 September 2010, the group redesigned the study and approached the HTA programme to approve an alternative trial design.

Stage 2: the three-arm home oxygen therapy trial

The redesigned trial was named the HOT trial. We chose a pragmatic study design avoiding blinding and did not include a sham oxygen arm. In turn, this meant that more oxygen supply companies could be used and the number of recruiting centres could be expanded. An additional benefit was to avoid the need for special insurance related to dummy devices. Furthermore, the answer to a pragmatic trial is arguably of more relevance to patients and clinicians than the original study design. The original research question, however, remained unchanged: to assess the effects of HOT on QoL in patients with CHF (Figure 2).

FIGURE 2.

The original design for the three-arm HOT trial.

Patients were randomised to receive BMT, BMT plus NOT or BMT plus LTOT for 15 hours per day. The primary end point was QoL at 12 months as measured by the Minnesota Living with Heart Failure (MLwHF) questionnaire.

The trial started recruitment in April 2012. However, recruitment to the trial was much slower than expected and recruitment targets were not met. There were two major reasons.

First, it is a particular challenge to recruit patients to studies of palliative care,42–44 and the patients being recruited into the HOT study were necessarily very unwell and reaching the end of their lives. Some centres found it difficult to approach such patients and, in some centres, the patients being recruited were predominantly cared for by other groups of health-care workers in their area, commonly in the areas of palliative care, elderly care and primary care. Simply put, frail patients characterised by instability needed to be sufficiently fit and stable to participate in the trial.

The second, and more intractable, problem was the time it took to recruit centres to the trial. Multiple individual site approvals were needed, as for any clinical study, but the peculiar problem for HOT was the need for approval of HOT. Many sites took the view that this should be an NHS excess treatment cost. Hospital trust research and development departments naturally needed approval to prescribe HOT from the relevant primary care providers, leading in many instances to prolonged delays. During the trial set-up period of HOT, there were major upheavals in NHS structures. The replacement of primary care trusts with clinical commissioning groups made it extremely challenging to identify who was responsible for HOT. In some instances, clinical commissioning groups refused permission for the study to go ahead, and sometimes it proved impossible to find the responsible party with whom to negotiate. The median length of time taken from a centre expressing interest in the study to final approval was 9 months.

Stage 3: the two-arm home oxygen therapy trial

The trial management group realised that, with the difficulties in recruiting both patients and centres, it would prove impossible to complete the study within a reasonable time frame and budget. We again approached the HTA programme to ask permission for a modestly redesigned study. We removed the NOT arm, reducing the study from a three- to a two-armed trial comparing BMT with BMT plus LTOT (15 hours per day). LTOT was chosen over NOT as being the only form of oxygen therapy shown to improve prognosis, albeit in patients with COPD. The simplification of the study design allowed the sample size to be greatly reduced (Figure 3).

FIGURE 3.

The final design for the two-arm HOT trial.

In addition, we brought forward the timing of the primary end point from 12 to 6 months. This had the effect of reducing the time required to follow up patients and increasing the time available for analysing the data and writing the report for the HTA. After discussion with the HTA programme, it was agreed that the cost-effectiveness analysis for the trial was not to be included.

Approvals obtained

The Northern and Yorkshire Research Ethics Committee [the Medical Research and Ethics Committee (MREC)] approved the study originally known as the NEON trial on 24 August 2009. Further approvals were received on 18 May 2011 and 2 April 2013 for changes to the design of the study that were implemented to improve recruitment.

The Medicines and Healthcare products Regulatory Agency reviewed the trial protocol and on 28 June 2008 confirmed that oxygen concentrators were medical devices and not medicines. The study, therefore, did not fall under the Clinical Trials Directives and so a clinical trial authorisation was not required.

The details of MREC and research and development department approvals are provided in Appendix 2.

Patient study group

To make the findings of the study as widely applicable as possible, the inclusion criteria were very broad, with few exclusions.

Randomisation

Allocations were centrally generated by the York Trials Unit. Patients were initially randomised into the trial on a 1 : 1 : 1 basis using block randomisation, with randomly permuted block sizes of three, six and nine. Subsequently, after the NOT arm was dropped, 1 : 1 allocation was used, with randomly permuted block sizes of four, six and eight. Patients were randomised by a member of the research team at the recruiting site using the secure, telephone-based randomisation service at the York Trials Unit.

Primary outcome

The primary end point was the total score from the MLwHF questionnaire at 6 months from baseline. As the patient group is highly symptomatic and has a limited prognosis, the 6-month primary outcome is highly clinically meaningful.

The MLwHF questionnaire is a validated, disease-specific QoL instrument widely used in heart failure research to assess both symptom severity and response to treatment. 45,46 It consists of 21 questions focusing on the impact of heart failure on QoL. Patients are asked to rate the extent to which their heart failure has prevented them from living as they wanted during the past month using questions rated on a scale of 0 (no effect) to 5 (very much). The questionnaire is scored by summing the responses to all 21 questions, thus resulting in a score from 0 to 105, with a higher score reflecting poorer QoL. The MLwHF questionnaire-validated QoL score, version 2, is easy to complete and has been shown to be especially effective in older patients with comorbidities. 47

An improvement in the score of 5 is sometimes taken to be a minimum clinically important difference,48 but others have suggested that a change of 1 standard error (SE) around the mean score is needed (around 6 or 7, depending on the population studied). 49 There are no studies of LTOT to help guide us. The MIRACLE (Multicenter InSync RAndomized CLinical Evaluation) trial of biventricular pacing was powered for a 13-point improvement in MLwHF questionnaire score and found a score difference of 9 between treatment groups. 50 The CARE-HF (CArdiac REsynchronization-Heart Failure) study of biventricular pacing found a score improvement of 10.6 with intervention. 51 In the absence of data on which to base study size, we took a MLwHF questionnaire score of 10 as an arbitrary indicator of the minimum improvement necessary to justify the cost and inconvenience of oxygen therapy for patients.

Secondary outcomes

Other indices of QoL, exercise performance and severity of heart failure were also measured as part of the study.

Adverse events

In this study an adverse event was defined as any untoward medical occurrence in a patient which did not necessarily have a causal relationship with the study treatments or procedures.

Health-care professionals were asked to report any adverse event occurring in participants in both groups using either the SAE form or the adverse event form within 24 hours of becoming aware of the event. A SAE was defined as an event that resulted in death, was life-threatening, required hospitalisation or prolongation of existing hospitalisation, resulted in persistent or significant disability or incapacity, consisted of a congenital anomaly or birth defect or was otherwise considered to be medically significant by the investigator. The reporting health-care professional was asked to indicate whether or not, in his or her opinion, the event was related to the treatment, to indicate if it was expected and to grade the intensity of the event. Further follow-up reports were completed if necessary or until the local principal investigator considered the event to be resolved or to have become a chronic ongoing condition.

Trial completion

The flow of participants through the trial is presented in a CONSORT (Consolidated Standards of Reporting Trials) diagram. The numbers of participants withdrawing from treatment and/or the trial are summarised with the reasons where applicable (see Figure 6).

Baseline data

All participant baseline data are summarised descriptively by treatment group. Continuous measures are reported using summary statistics (mean, SD, median, interquartile range, minimum, maximum) and categorical data are reported as counts and percentages. Comparisons were made between trial groups ‘as randomised’ and ‘as analysed’ in the primary analysis. No formal statistical comparisons were undertaken.

Primary analysis

The primary outcome was health-related quality of life (HRQoL) as measured by the MLwHF questionnaire scores at 6 months. A lower MLwHF questionnaire score indicates a better QoL. The MLwHF questionnaire consists of 21 items and a total score is obtained by summing the item scores, where all 21 items have a response. The number of missing MLwHF questionnaire responses was examined at each time point. Multiple imputation using chained equations was used to fill in missing questionnaire items, where there were fewer than four missing item responses, using other items in the questionnaire (http://178.23.156.107:8085/Instruments_files/USERS/mlhf.pdf). Linear regression was used to perform the imputation and five imputed data sets were created. The mean of the imputed values for each patient was used to replace the missing item. Where the mean was less than 0 or greater than 5 (the range of permitted scores for the MLwHF questionnaire) it was replaced with the nearest permitted value.

Our primary analysis compared MLwHF questionnaire scores between the LTOT and BMT treatment groups using a covariance pattern mixed model, where effects of interest and baseline covariates are specified as fixed effects, and the correlation of observations within patients is modelled by a covariance structure. The outcome modelled was total MLwHF questionnaire scores at 3, 6 and 12 months. The model included as fixed effects baseline MLwHF questionnaire score, age, log-NT-proBNP level, creatinine level, treatment group, time and a treatment group–time interaction term. Age, NT-proBNP levels and creatinine levels were all continuous variables as assessed at baseline. NT-proBNP data were found to be significantly positively skewed and so were log transformed.

Different covariance structures for the repeated measurements, which are available as part of Stata version 13, were explored and the most appropriate pattern used for the final model. Diagnostics including Akaike’s information criterion65 were compared for each model (smaller values are preferred).

Participants were naturally included in the model only if they had full data for the baseline covariates and outcome data for at least one post-randomisation time point (3, 6 or 12 months).

Estimates of the adjusted mean difference (AMD) between treatment groups in MLwHF questionnaire scores were extracted from the model for all time points with 95% confidence intervals (CIs) and p-values. The primary end point is the treatment effect estimate at 6 months. Estimates for the other time points serve as secondary outcomes. An overall effect of the intervention across all included time points was not extracted.

Secondary analysis

Baseline and follow-up assessments

After written informed consent had been obtained, baseline data were collected using the nurse and patient baseline questionnaires.

After giving consent and completing baseline assessments, patients were randomised by a member of the research team at the recruiting site using the secure, telephone-based randomisation service at the York Trials Unit.

In the event that the local clinical team felt that a patient needed to be assessed for obstructive sleep apnoea, the patient was not randomised until a clinical decision was made.

Home oxygen therapy

If the patient was randomised to receive HOT, a clinical prescription was completed and sent to the local oxygen supply company holding the standard NHS contract within that particular region. These were Air Liquide UK (Birmingham, UK), Dolby Vivisol (Stirling, UK), BOC Healthcare (Manchester, UK) and Air Products (Crewe, UK). Concentrators were delivered to the patient’s home by the recruiting hospital’s usual oxygen supplier, in accordance with existing NHS agreements. The oxygen supply company typically installed the equipment within 3 days of the prescription being issued. The installing engineer instructed the participant about the safety requirements of using the machines and gave details of how to claim for the electricity costs of running the machine. In accordance with NHS agreements, the concentrators should have been serviced approximately every 6 months.

At the end of an individual patient’s trial participation, oxygen concentrators were removed from the patients’ homes unless they wished to continue treatment.

Best medical therapy

Patients allocated to this arm received the BMT currently available. Patients received the maximally tolerated medical management for their heart failure and reached their target dose of inhibitors of the renin–angiotensin system, a beta-adrenoceptor antagonist and an aldosterone antagonist.

Follow-up

Patients with CHF usually attend clinic every 6 months and follow-up in the study was arranged around standard attendances to prevent patients being unduly burdened with additional hospital visits. At the conclusion of the trial, final clinical data were collected using existing hospital records, including admission and mortality. The data collected are summarised in Table 2, together with the schedule for data collection.

Trial completion

Participants were deemed to have exited the trial when:

• they had been in the trial for 24 months or 6 months if there was insufficient time to follow the patient further

• they wished to withdraw from the trial

• their treating physician or medical researcher withdrew them from the trial

• they were lost to follow-up

• they died.

As well as withdrawing fully from the trial, participants had the option of:

• withdrawing from receiving oxygen (if that had been their allocation)

• withdrawing from the collection of data via patient questionnaires

• withdrawing from the collection of data via nurse questionnaires.

Trial recruitment

The HOT study was stopped early by the funders because of poor patient adherence to the oxygen prescription. Recruitment began in April 2012 and stopped in February 2014. Randomisation to the NOT arm was stopped in April 2013. In total, 139 patients were randomised, 57 to each of the LTOT and BMT arms and 25 to the NOT arm. The overall rate of recruitment is shown in Figure 4.

FIGURE 4.

The overall rate of recruitment in the HOT trial.

The original sample size for the three-arm trial was 450 patients. The aim was to recruit these patients in 12 months; however, recruitment was slower than expected and, by the end of April 2013, 74 patients had been recruited into the trial (25 to each of the LTOT and NOT arms and 24 to the BMT arm). The decision was made to drop the NOT arm and continue the trial with two arms with a revised sample size of 222 patients. The recruitment period was extended to August 2014. When the trial was closed at the end of February 2014, a total of 139 participants had been randomised.

Over the course of the trial, a total of 15 participating sites joined, all in the UK (Table 3). Recruitment was staggered, with sites joining over the course of the trial. At least one trial participant was recruited in 13 out of the 15 sites (Figure 5 and Table 3). The median number of participants recruited per site was 4 (range 1–76). Over half of the participants (n = 76) were recruited from the Hull site, where the chief investigator was based.

FIGURE 5.

Participant recruitment by site (two sites did not recruit any patients).

Two sites did not recruit any patients. Significant attempts were made to recruit a patient in Aneurin Bevan University Health Board but no eligible, consenting patients were identified; East Cheshire NHS Trust was granted research and development approval only shortly before recruitment ceased and, therefore, did not have time to recruit a patient.

Figure 6 shows the CONSORT flow diagram of participants through the trial.

FIGURE 6.

The CONSORT flow diagram. a, Based on patients who provided full covariate data for the primary analysis model and primary outcome data at at least one of the 3-, 6- or 12-month time points.

Withdrawals

A greater proportion of patients in the LTOT arm (n = 8, 14%) than the NOT arm (n = 2, 8%) formally withdrew from their allocated trial treatment. One patient in the NOT arm withdrew from completing the postal patient questionnaires and one patient in each of the NOT and BMT arms withdrew from completing the nurse questionnaires. A total of four patients (one LTOT, one BMT and two NOT) requested full trial withdrawal and two further patients (one LTOT, one NOT) were withdrawn by a health professional. The reasons given for each of the change of circumstances are shown in Table 4. There were 25 recorded deaths during the course of the trial (LTOT, 6; BMT, 12; and NOT, 7).

Baseline participant characteristics

Completed baseline questionnaires were received from 139 (100%) randomised participants. Participant baseline characteristics are summarised by treatment group (LTOT and BMT arms only) in Tables 5 and 6.

The majority of patients in the study were male (n = 80, 70%) and the mean age was 72 years (range 38–94 years). The most common cause of the participants’ heart failure was ischaemic or coronary heart disease (n = 96, 84%), and the vast majority of participants were in NHYA class III (n = 108, 95%).

To be eligible for the trial, participants had to have a LVEF < 40% or be graded as at least ‘moderately’ impaired on visual inspection. LVEF was not recorded in eight participants in the LTOT or BMT arm, and all of these participants had either ‘moderate’ or ‘severe’ LV impairment. One participant with a LVEF > 40% was randomised as he or she was visually assessed as having ‘severe’ LV impairment. NT-proBNP level was not an entry criterion to the study, but the very high levels suggest that patients with severe heart failure were recruited.

In general, the two treatment groups were comparable across baseline participant characteristics; however, there was a slight imbalance in NT-proBNP level and proportion of patients taking aldosterone antagonists. The mean NT-proBNP level was higher in the LTOT arm than in the BMT arm, and the proportion of patients taking an aldosterone antagonist was greater in the LTOT arm. NT-proBNP level was pre-specified as a covariate in the primary analysis and so this imbalance was controlled for.

Primary outcome

The primary outcome was MLwHF questionnaire score at 6 months post randomisation. At baseline, a response to one item was missing in two patients and responses to two items were missing in two patients. At 3 months, among those for whom the MLwHF questionnaire was returned, there were no missing data. At 6 months, seven patients had a missing response to one item. At 12 months, one questionnaire was returned not completed, with a note to say that the patient was too unwell to complete the questionnaire; otherwise there were no missing data. At 12, 18 and 24 months, where patients returned a questionnaire, there were complete item data for the MLwHF questionnaire. Imputation of missing items was, therefore, necessary only on baseline and 6-month data. Summaries of the MLwHF questionnaire score (post imputation) by treatment group at each time point are presented in Table 7.

A covariance pattern model was used to compare MLwHF questionnaire score between the LTOT and BMT arms, adjusting for baseline MLwHF questionnaire score, age, log-NT-proBNP level, creatinine level, treatment group, time and a treatment group–time interaction. The model included all patients who provided full data for each of the included covariates, and MLwHF questionnaire outcome data at one or more post-baseline time points up to 12 months, and so was based on 102 out of 114 patients (89.5%): 51 (89.5%) in each group. Baseline characteristics of participants as included in primary analysis model are compared between the treatment arms in Table 8. It does not appear that the loss of the 12 patients for whom covariate or outcome data were missing significantly impacted on the balance between the treatment arms achieved at randomisation.

The assumptions of the linear model were checked visually. The normality of the standardised residuals was assessed via a histogram and Q–Q plot (see Appendix 5), and the homoscedasticity of the errors was checked by plotting the residuals against the fitted values. These plots gave no reason to be concerned about the validity of the assumptions.

In total, 88 participants provided valid MLwHF questionnaire data at 6 months (LTOT n = 45; BMT n = 43); however, five of these participants did not provide data for at least one of the included baseline covariates (MLwHF questionnaire score, age, NT-proBNP level, creatinine level) so they were not included in the model. A further 19 participants (LTOT n = 8; BMT n = 11) were included in the model, as they provided valid MLwHF questionnaire data at 3 and/or 12 months, resulting in an analysed sample of 102 participants. An estimate of the treatment effect at 6 months was extracted from the model. As well as the 85 participants in the model who provided primary outcome data at 6 months, participants who provided data at 3 and/or 12 months but not at 6 months are taken into account when the treatment effect at 6 months is estimated owing to the specification of the covariance pattern between the within-patient repeated measures.

There was no evidence of a difference in MLwHF questionnaire score between the LTOT and BMT groups at 6 months (AMD –0.10, 95% CI –6.88 to 6.69; p = 0.98) (see Table 7); the LTOT group had a slightly lower MLwHF questionnaire score at 6 months, but the difference was not statistically significant. Figure 7 plots the adjusted means obtained from the model by treatment group over time.

FIGURE 7.

Adjusted means for MLwHF questionnaire score by treatment group over time from the primary analysis model.

Secondary outcomes

For each of the following models, assumptions were checked in the same way as for the primary analysis and no significant violations were observed.

Mortality

At 6 months, 15 patients had died (LTOT n = 4, NOT n = 5 and BMT n = 6). A further 10 deaths were reported after 6 months of follow-up (LTOT n = 2, NOT n = 2 and BMT n = 6). Mortality was analysed as a time-to-event outcome. The main analysis compared the BMT and LTOT treatment groups. For each group, the distribution of time from randomisation to death is described using Kaplan–Meier survival curves (Figure 8).

FIGURE 8.

Kaplan–Meier survival curve.

Unadjusted Cox regression gave a HR of 2.03 (95% CI 0.76 to 5.40) for BMT relative to LTOT, but this was not statistically significant (p = 0.16). Adjusting for baseline CCI score, the HR was slightly lower (1.84, 95% CI 0.68 to 4.96; p = 0.30), indicating, again, that the risk of death was higher in the BMT group than in the LTOT group, but the difference was not statistically significant (p = 0.30). The CCI was not a significant predictor in the survival model (p = 0.19)

Additional survival analyses were undertaken comparing the NOT and LTOT arms only including patients randomised up to the time that the NOT arm was dropped from the study.

The number of patients recruited to the LTOT, NOT and BMT arms was 25, 25 and 24, respectively (recruited up to the end of April 2013). There was no significant difference in survival between the groups (Kaplan–Meier, χ² = 2.07, df = 1; p = 0.15). The oxygen arms were combined and compared against the BMT arm including only those patients randomised up to the time that the NOT arm was dropped. Figure 9 shows the survival curve for BMT against LTOT and NOT. There was no significant difference in survival between the groups (Kaplan–Meier, χ² = 1.12, df = 1; p = 0.29). The 75th percentile for the BMT group was 269.0 days (SE 79.3 days) and for the combined LTOT and NOT group it was 655.0 days (SE 0.98 days). Unadjusted Cox regression gave a HR of 1.64 (95% CI 0.65 to 4.17), indicating that the risk of death was higher in the BMT group than in the LTOT and NOT groups, but the difference was not statistically significant (p = 0.30). The model adjusting for baseline CCI score gave very similar estimates and CCI was not a significant predictor of mortality.

FIGURE 9.

Kaplan–Meier survival curve for BMT vs. LTOT plus NOT.

The proportional hazards assumption was checked using log–log plots of the estimated survivor function, plotting scaled Schoenfeld residuals and by the Grambsch and Therneau test. Visual inspection of the log–log and Schoenfeld plots indicated potential non-proportionality of the treatment group variable in each model; however, the Grambsch and Therneau test did not indicate evidence of non-proportionality.

Adverse events

The following sections detail the SAEs and non-SAEs for all three groups.

Adherence

The oxygen concentrators were generally installed in the homes of patients allocated to receive HOT within 2 weeks of randomisation (median 5 days). Two participants did not receive a machine as one died and one withdrew from the trial very shortly after randomisation. Eight (14%) patients in the LTOT arm formally withdrew from the trial treatment throughout follow-up and requested that the oxygen concentrator be removed from their home. Here, we present data on patient-reported adherence and data from the oxygen suppliers’ meter readings.

Overnight Embletta sleep study

If an Embletta device was locally accessible, participants were asked to complete an overnight sleep test at baseline and at 6, 12, 18 and 24 months. The Embletta records multiple signals to detect whether or not a patient has sleep apnoea, either obstructive or central. A pair of nasal prongs records air flow and a pulse oximeter worn on the finger records arterial oxygen saturation. Together, these signals detect episodes of reduced breathing (hypopnoea) or arrested breathing (apnoea). A band is worn across the chest and upper abdomen: this channel records respiratory effort. If decreased nasal air flow (and decrease in arterial oxygen saturation) suggests apnoea or hypopnoea, and if respiratory efforts are detected, an obstructive event is recorded. If, on the other hand, no respiratory efforts are detected, a central event is recorded.

The apnoea–hypopnoea index (AHI) is used to indicate the severity of sleep apnoea and represents the number of apnoea and hypopnoea events per hour of sleep. AHI values are categorised as normal (0–4), mild sleep apnoea (5–14), moderate sleep apnoea (15–29) and severe sleep apnoea (30+). Considering that both AHI and oxygen saturation can give an overall assessment of the sleep apnoea severity. The AHI, the number of desaturations per hour and the total sleep time with saturation less than 90% at each time point are summarised by treatment group in Table 34.

The sleep test was conducted in only five sites (Barnet, Darlington, Dundee, Durham and Hull), and only if the patient consented to the test, hence the low number of tests conducted. There was a variety of potential problems with the sleep study. Some patients found the Embletta difficult to set up and use despite training at the hospital; some found that it interfered with sleep and were unwilling to use it again in the event of a poor-quality recording. As there is no currently recognised indication for sleep studies in heart failure patients per se, it is likely that some investigators did not encourage patients to take part in this substudy.

Background

As there is some evidence that oxygen administration to patients with heart failure might be harmful, and that, in particular, oxygen administration might be associated with adverse haemodynamic consequences, we undertook an acute haemodynamic substudy as part of the HOT trial.

The evidence for harm with hyperoxia is limited to small-scale studies, the main ones being by Haque et al. 21 and Mak et al. 22

Haque et al. 21 studied 10 patients with class III and IV CHF who breathed 100% oxygen for 20 minutes. The administration of 100% oxygen reduced cardiac output from 3.7 l/minute to 3.1 l/minute and increased pulmonary capillary wedge pressure (from 25 mmHg to 29 mmHg). Systemic vascular resistance increased from 1628 dyn × s/cm⁵ to 2203 dyn × s/cm⁵, and there was no significant change in pulmonary vascular resistance. In a smaller substudy of only seven patients, similar changes were seen in patients breathing 24% oxygen.

Mak et al. 22 studied 16 patients with slightly milder stable CHF (NYHA class II and III) and 12 subjects with normal LV function. Subjects again received 100% oxygen for 20 minutes. In the patients, LV end-diastolic pressure (equivalent to pulmonary capillary wedge pressure in Haque’s study21) increased from 21 mmHg to 25 mmHg, cardiac output fell from 4.6 l/minute to 4.1 l/minute and SVR increased from 1626 dyn × s/cm⁵ to 1901 dyn × s/cm⁵. There were similar changes to cardiac output and SVR in the control subjects.

Other evidence for the haemodynamics effects of oxygen is limited to very old studies of small numbers of patients (such as a study of six patients20 with acute myocardial infarction, in whom 100% oxygen caused a fall in cardiac output). There is no literature suggesting harm from LTOT in patients with CHF.

The oxygen dose that patients received in the main HOT trial was 2 l/minute, equivalent to 28% oxygen and much lower than the 40% threshold for harm suggested by Haque et al. 21 There is no formal study assessing the effects of this dose of oxygen on haemodynamics in the type of patient included in the HOT study.

We therefore assessed the haemodynamic effects of oxygen delivered for 10 minutes at a similar concentration to that received by patients in the main HOT study (28%), with the aim of establishing at least the short-term safety of low-dose oxygen.

Methods

Results

Demographic data are shown in Table 35. The patients were younger than those in the main study, reflecting the fact that most were undergoing assessment for possible heart transplantation. The patients were well treated and had severe LV systolic dysfunction. Most had ischaemic heart disease and 32% were in atrial fibrillation.

There were no complications during any of the procedures, but one patient experienced an episode of pulmonary oedema 1 hour after catheterisation and was hospitalised overnight.

Table 36 shows the effects of 10 minutes of 28% inspired oxygen on the directly measured intravascular pressures during cardiac catheterisation. The patients had mild pulmonary arterial hypertension and high left heart filling pressures but there was no effect of oxygen on any of the variables measured.

Table 37 shows the effects of 28% inspired oxygen on the oxygen saturation in pulmonary artery and aorta together with the effects on derived haemodynamic variables. The increase in the inspired oxygen led to an increase in pulmonary and systemic arterial saturation. There was a small fall in pulmonary vascular resistance and an increase in cardiac output.

The change in pulmonary vascular resistance correlated with change in mean pulmonary artery pressure rather than with pulmonary capillary wedge pressure, and the change in cardiac output correlated with change in pulmonary artery saturation (Figure 10), suggesting that what small changes to central haemodynamics were seen were driven by changes in the pulmonary rather than the systemic circulation.

FIGURE 10.

Relations between haemodynamic variables. CO, cardiac output; PA, pulmonary artery; PVR, pulmonary vascular resistance.

Background and rationale

Despite a survival benefit in COPD, patient adherence to LTOT is known to be problematic. 17,18 Adherence appears to be related to symptom burden and performance status, and is better in patients who have had training in the use of the concentrator. 18,70–72 Given the lack of data about the use of home oxygen for people with symptomatic heart failure, this substudy explored the experience of patients and their informal carers regarding the use of HOT during participation in the main study to gain a deeper understanding of adherence issues than could be gained from monitoring concentrator use alone.

Breathless patients may have mixed feelings about oxygen therapy. 73–75 Positive feelings include having ‘treatment’ and an instinctive welcome of oxygen as it is routinely used during episodes of severe decompensated heart failure by emergency health professionals. The presence of an oxygen concentrator in the home and regular attention from professionals may be reassuring. Patients with prior experience of cylinders may find the concentrator easier to handle with less need for deliveries. However, the presence of the concentrator and associated paraphernalia can cause home disruption, limitations to activity, problems with tubing and anxieties about it ‘going wrong’. The fixed nature of the concentrator in the house may restrict outside activities and hence adherence is lower in patients able to do activities outdoors. 74,75

In general, adherence is lower for treatments given for maintenance or prevention than for those which are directed at an acute symptom with a clear temporally related dose–response relationship. In a RCT of oxygen of sham concentrators for people with advanced disease and symptomatic breathlessness, 50% of patients reported no benefit (for their breathlessness) and did not want further oxygen therapy. 36,76

The sensation of breathlessness correlates poorly with measures of lung function or arterial oxygen saturation and has a complex genesis in heart failure. Two factors may make it difficult to identify the possible role of oxygen therapy for people with advanced heart failure, particularly if they are not hypoxaemic enough to warrant HOT in order to gain a survival benefit. The first is the lack of immediate relief of breathlessness (particularly when patients intuitively believe that they should be helped by oxygen) and the second is the fact that it might be a benefit that is mediated by airflow rather than by the oxygen.

Aims and objectives

This planned substudy aimed to explore the views of patients with CHF using a home oxygen concentrator, and their carers, with regard to benefits and burdens. In addition, we wished to understand their experience of this in the context of participation in the HOT trial.

Methods

Ethics approval was given by Northern and Yorkshire Research Ethics Committee (reference 09/H0903/43).

Data analysis

Thematic framework analysis was used. This is an approach developed for conducting applied qualitative research,77 and involves moving through the stages of familiarisation with the data, identification of a thematic framework, indexing the data using the thematic framework, arranging the data into charts, mapping and interpretation of the data. SN, MJ and Lesley Jones carried out the preliminary coding of interview transcripts, checking each other’s coding for similarity to increase rigor. SN used the NVivo computer software (QSR International, Warrington, UK) to manage the data.

All transcripts were coded by at least two researchers. The coding strategies and emerging themes were then agreed between all three researchers. Once the thematic framework had been drawn up, the findings were then synthesised to provide a summary of patient and carer views with particular regard to the perceived benefits and burdens of HOT using a concentrator, and their experience of participation in the HOT trial. Data from informal carers and patients were analysed together, as we considered the dynamics of the family unit to be integral to the use of oxygen therapy in the home.

Results

Clinical effectiveness

Although there was no difference in MLwHF questionnaire score between the LTOT and BMT arms at 6 months, we did see evidence for an improvement at 3 months. The LTOT group had a statistically significant lower MLwHF questionnaire score (difference –5.47, 95% CI –10.54 to –0.41; p = 0.03). This represents a clinically important difference;48 however, a difference of 10 was used to power the trial. The improvement may reflect a placebo effect and represent the effect of receiving active treatment, and having the oxygen available as a possible acute treatment in times of severe symptoms. That the improvement disappeared at 6 months might be a result of the waning of the placebo effect coupled with the burden of the oxygen usage. However, 3 months seems to be a long time for a placebo effect. Although placebo responses can last up to 12 weeks, it is unusual and would be particularly so in the situation where the patient is expecting an immediate benefit.

There were twice as many deaths in the BMT arm (n = 12) as in the LTOT arm (n = 6), although the impact of oxygen therapy on survival was not statistically significant (p = 0.16). Hazard rates were similar up to around 6 months post randomisation, after which only two more deaths were recorded in the LTOT arm but a further six were reported in the BMT arm.

As we observed a significant difference in the MLwHF questionnaire score at 3 months (favouring LTOT), we hypothesised that LTOT might be keeping sicker patients alive longer and so be artificially pulling the MLWHF score up at 6 months (lower scores indicate a better QoL) relative to BMT. In the BMT arm, the higher mortality of the more symptomatic patients would have the effect that the average QoL of the group would tend to improve as the more severely symptomatic patients died.

To investigate this hypothesis, we repeated the primary analysis model restricted to patients who returned data at 3 and 6 months (and so necessarily were alive at 6 months). However, restricting the analysis to those patients alive at all three time points showed no evidence of such an effect.

Quality of life measures

Patients in the LTOT arm at 3 months reported an improvement in average breathlessness over the past 24 hours on the numerical rating score by a clinically important amount57 (which was no longer apparent at 6 months). The improvement coincided with an improvement in the distress caused by breathlessness, reporting of having coped better with breathlessness, and a higher level of satisfaction with the treatment received for breathlessness. At 6 months, there was no difference in average or worse breathlessness experienced over the last 24 hours, but patients in the LTOT arm continued to have increased median scores for coping and satisfaction with treatment.

There are two potential explanations. First, it seems from other work that airflow (whether or not it is oxygen enriched) can reduce the sensation of breathlessness; therefore, there may be a ‘real’ symptom improvement from using the flow of gas across the nasal mucosae,36 which may not persist to 6 months, particularly in patients who are less well. Second, the improvement in distress due to breathlessness at 3 months may be because of the ‘safety net’ issue seen in the qualitative substudy. In other words, patients may have had increasing self-efficacy, that is the oxygen was something they had in reserve for a bad day. That effect, too, might disappear over time as the underlying condition progresses or the therapy does not lead to hoped-for improvements. However, it is notable that improvements in self-rated ‘coping with breathlessness’ persisted for the first 12 months and ‘satisfaction with treatment of breathlessness’ until 6 months.

Safety

There were no safety concerns with the use of HOT. There were theoretical concerns related to possible mechanical complications from tubing, and at a very early stage there was a suggestion of an increase in admissions because of peripheral vascular disease in the oxygen-treated patients. However, as the study progressed, there were no significant differences in event rates between the two groups. Although more participants in the BMT arm than in the LTOT and NOT arms experienced one or more SAEs, the difference was not statistically significant.

Might oxygen be associated with an increase in survival? Although the difference between the survival curves is visually striking (see Figure 8), the difference in survival between the two groups was not statistically significant. Coupled with the results from the acute oxygen substudy, which did detect a favourable haemodynamic response to oxygen, the evidence is reassuring that HOT, at least when the fraction of inspired oxygen is < 30%, is safe.

However, the study was greatly underpowered to detect a difference in mortality. The premature end to the study reduced the chances of detecting a survival effect even further. It should be borne in mind that the target dose of oxygen at 15 hours per day is an almost arbitrary figure arrived at from analysis of patients in long-term trials of another condition. It remains possible that oxygen used for shorter periods of time might be beneficial for people with heart failure.

Why was the home oxygen therapy trial neutral?

The most likely explanation for the apparent lack of effect of LTOT for heart failure in the HOT trial is that oxygen has no effect on breathlessness in this clinical situation. There are several possible confounding issues.

Substudies

Limitations

We were unable to blind oxygen delivery as the original study design had planned. However, the open trial was a pragmatic one which addressed the important question of treatment effectiveness. The trial was stopped early because of poor adherence to the 15 hours per day prescription. The prescription was based on extrapolation from studies of patients with a different pathology, chronic airways disease, who had severe hypoxia. It may be that shorter periods of exposure might have been effective, in terms of either symptom relief or preventing hospitalisation.

The a priori sample size calculation was based on the independent-sample t-test at 6 months, giving a conservative estimate of 200 participants to detect a difference in a MLwHF questionnaire score of 10 points, assuming a SD of 25 points (effect size of 0.4), at 80% power with 5% significance. Allowing for 10% attrition at 6 months, we needed to recruit and randomise 222 participants.

The primary result was the treatment effect at 6 months extracted from an adjusted covariance pattern model. Adjusting for baseline covariates by using an analysis of covariance (ANCOVA) increases the statistical power compared with a t-test for the same number of participants. With an ANCOVA, the sample size required to detect the same difference can be reduced by a factor of (1 – ρ²), where ρ is the correlation between the baseline covariates and the outcome. For example, with a sample size of 200 at 6 months, assuming a moderate correlation of 0.5 between the baseline covariates and MLwHF questionnaire score at 6 months, we would have had 90% power to detect an effect of magnitude 0.4 with an ANCOVA compared with 80% power with the t-test.

An ANCOVA can use data available only for the time point of interest, here 6 months; however, in the covariance pattern model, we could include the 3- and 12-month post-randomisation assessments. In a repeated-measures model, missing data are less problematic than when using a t-test, as observations at each time point influence estimates of treatment effects at every other time point, owing to the specification of a covariance pattern for the within-patient repeated measures. Thus, patients whose observations are limited to the 3- and/or 12-month time points will nevertheless be taken into account when the estimate at 6 months is made. Clearly, such individuals will not influence the estimates as greatly as individuals whose data are complete, but the philosophy is that some data from a patient is better than none. The extra information further increases the power available from the ANCOVA; however, it is beyond our knowledge to quantify this extra benefit and we have found no literature that discusses this exact problem.

Calculating a sample size based on an adjusted repeated-measures analysis is complicated and requires knowledge of parameters that are not known in advance (particularly the correlation between baseline covariates and the outcome). The calculations are sensitive to initial assumptions and so we used a t-test to provide a conservative estimate to minimise the risk of underpowering the trial.

A total of 88 participants provided usable primary outcome data at 6 months; 83 also had complete baseline data for MLwHF questionnaire score, age, NT-proBNP level and creatinine level. To detect a 10-point difference with a SD of 25 at 6 months, the estimated power available in a two-sided test of two independent means is 46% but by ANCOVA (adjusting for the covariates and assuming a correlation between baseline covariates and 6 months MLwHF questionnaire score of 0.5) is 55%.

As a result of the trial being underpowered, the risk of a type II error is increased and any of the results, especially those with an associated p-value greater than 0.05, are inconclusive. The low sample size means it is not possible to distinguish whether the lack of a detected effect is a result of the trial being underpowered or whether there is a true lack of effect. In addition, a lack of adherence with oxygen may have diluted any effect of long-term home oxygen.

The MLwHF questionnaire is the most widely used instrument for the assessment of QoL in patients with heart failure. It asks patients to rate the severity of their symptoms on a 6-point scale (from 0 to 5) but does not assess the impact of those symptoms on the patient’s life directly. It might be that another scale, such as the Kansas City Cardiomyopathy Questionnaire,99 would have been more sensitive to changes in health state. The Kansas City Cardiomyopathy Questionnaire asks patients to record the impact that a symptom has on their life rather than merely the presence or absence of a symptom. It might be that the patients in the HOT study were sufficiently habituated to their symptoms that although they were, perhaps severely, breathless, they were used to it and thus felt that the breathlessness had little impact on their life.

Conclusions

The prevalence of hypoxia in patients with severe heart failure, at rest, following exercise and during an overnight sleep test, is low. There is no evidence that LTOT, although it is safe, improves the symptoms, prognosis or severity of heart failure in patients with severe CHF at 6 months. There is no evidence to support the use of HOT in patients with heart failure but, as the study was terminated early, we cannot exclude the possibility of a type II error.

Recommendations for future research

We suggest that two further studies might be appropriate:

1. a trial of patients with severe heart failure randomised to have emergency oxygen supply in the house, supplied by cylinders rather than oxygen concentrator, powered to detect a reduction in admissions to hospital

2. a study of bed-bound patients with heart failure who are in the last few weeks of life, powered to detect changes in symptom severity.

However, given the problems with the conduct of HOT encountered in the present study, mounting such a trial would face similar problems with both blinding (and the use of a sham device) and adherence.

The HOT investigators are in close contact with the investigators of the OXYGEN-HF trial (ACTRN12609000103268). OXYGEN-HF is a randomised trial of 285 patients comparing the effects of HOT, medical air (placebo) and no treatment (control) on all-cause admissions to hospital at 6 months. Many of the trial’s secondary end points are the same as ours, and the findings of both studies will be of value in a meta-analysis.

The home oxygen therapy trial

Contributions of authors

Professor Andrew L Clark (Professor of Clinical Cardiology) conceived the study and applied for, and received, the original grant. He was principal investigator for the study. He conducted the review of the data for HOT in heart failure and is the principle author of this report.

Miriam Johnson (Professor of Palliative Care) cowrote the initial grant application, trial protocol (main study and qualitative substudy) and ethics application (main study and qualitative substudy), was a member of the trial management group, provided palliative care and symptom outcome assessment expertise, was principal investigator for the qualitative substudy, conducted data collection and analysis for the qualitative substudy and contributed to writing the report.

Caroline Fairhurst (Statistician) conducted the statistical analysis of the main trial, prepared the results for publication, and contributed to the preparation and review of the final report.

David Torgerson (Director of York Trials Unit, Health Sciences) helped design the study, gave trial methodological advice and was a co-applicant.

Sarah Cockayne (Research Fellow, Health Sciences) was a trial co-ordinator on the study, and contributed to the design of the study and to the preparation and review of the final report.

Sara Rodgers (Research Fellow, Health Sciences) was a trial co-ordinator on the study, and contributed to the preparation and review of the final report.

Susan Griffin (Senior Research Fellow, Health Economics) analysed the EQ-5D data, and contributed to the preparation and review of the final report.

Victoria Allgar (Statistician) provided statistical support during the design and conduct of the main trial.

Lesley Jones (Senior Lecturer, Social Sciences) cowrote the qualitative substudy protocol and provided senior qualitative expertise regarding data analysis and report writing.

Samantha Nabb (Health Psychologist, University of Hull) helped to create the interview schedule investigating the patient and carer experience of the oxygen concentrator, interviewed patients and carers and analysed the data.

Ian Harvey (Trial Manager) contributed to the design, development and delivery of the trial, and to the preparation and review of the final report.

Iain Squire (Professor of Cardiovascular Medicine) helped design the study, contributed to the writing of the report and was a co-applicant.

Jerry Murphy (Professor of Cardiovascular Medicine) helped design the study, contributed to the writing of the report and was a co-applicant.

Michael Greenstone (Consultant Chest Physician and Honorary Clinical Senior Lecturer) co-wrote the initial grant application and trial protocol, and contributed to the writing of the report.

Publications

Clark AL, Johnson MJ, Squire I. Does home oxygen benefit people with chronic heart failure? BMJ 2011;342:d234.

Davidson PM, Johnson J. Update on the role of palliative oxygen. Curr Opin Support Palliat Care 2011;5:87–91.

Gadoud A, Johnson MJ. What palliative care clinicians need to know about heart failure. Progress Palliat Care 2014;22:26–31.

Data sharing statement

All available data can be obtained from the corresponding author.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.

NHS R&D Health Technology Assessment programme HTA number 06/80