Pre-hospital non-invasive ventilation for acute respiratory failure: a systematic review and cost-effectiveness evaluation

Electronic databases

Studies were identified by searching the following electronic databases and research registers:

• MEDLINE In-Process & Other Non-Indexed Citations and MEDLINE (via OvidSP) from 1948 to August 2013

• EMBASE (via OvidSP) from 1980 to August 2013

• Cumulative Index to Nursing and Allied Health Literature (via EBSCOhost) from 1982 to August 2013

• Cochrane Database of Systematic Reviews (via Wiley Online) from 1996 to August 2013

• Cochrane Central Register of Controlled Trials (via Wiley Online) from 1898 to August 2013

• Health Technology Assessment Database (via Wiley Online) from 1995 to August 2013

• Database of Abstracts of Review of Effects (via Wiley Online) from 1995 to August 2013

• Bioscience Information Service (BIOSIS) Previews (via ISI Web of Knowledge) from 1969 to August 2013

• Science Citation Index Expanded (via Web of Science) from 1899 to August 2013

• Conference Proceedings Citation Index – Science (via Web of Science) from 1990 to August 2013

• UK Clinical Research Network Portfolio Database [National Institute for Health Research (NIHR)] from 2001 to October 2012

• National Research Register Archive (NIHR) from 2000 to September 2007

• Current Controlled Trials from 2000 to October 2012

• ClinicalTrials.gov (USA National Institutes of Health) from 2000 to October 2012.

Sensitive keyword strategies were developed using free text and, where available, thesaurus terms using Boolean operators and database-specific syntax to search the electronic databases. Synonyms relating to the setting (e.g. pre-hospital) were combined with terms for NIV. No language or date restrictions were used on any database. All resources were initially searched from inception to October 2012. With the exception of the four research registers, updated searches to August 2013 were conducted on the remaining electronic databases. An example of the MEDLINE search strategy is provided in Appendix 1.

Other resources

To identify additional published, unpublished and ongoing studies, the reference lists of all relevant studies were checked and a citation search of relevant articles (using the Web of Science, Science Citation Index Expanded and Conference Proceedings Citation Index – Science) was undertaken to identify articles that cite the relevant articles. In addition, systematic keyword searches of the internet were undertaken using the Google search engine (Google Inc., Mountain View, CA, USA) and key experts in the field were contacted.

All identified citations from the electronic searches and other resources were imported into and managed using the Reference Manager bibliographic software version 12.0 (Thomson Reuters, Philadelphia, PA, USA).

Study design

All randomised (individual or cluster) or quasi-randomised controlled trials that evaluated pre-hospital NIV (as part of acute treatment by the emergency care system) in patients with acute respiratory failure were included. Non-randomised observational studies were not included in the formal systematic review but were retained and reported descriptively as additional evidence and, if appropriate, used to develop the economic model. In addition, all trials in progress (identified via trial registers) were recorded but not included in the analysis.

The following publication types were excluded from the review: animal models; pre-clinical and biological studies; narrative reviews, editorials, opinions; non-English-language papers; and reports published as meeting abstracts only, where insufficient methodological details are reported to allow critical appraisal of study quality.

Population

All studies of adults (defined as > 18 years of age) presenting to the emergency services with acute (hypoxaemic or hypercapnic) respiratory failure from any cause or no specified cause were included.

Interventions

Pre-hospital NIV (defined as ventilatory support provided before arrival at hospital and delivered to a spontaneously breathing individual without airway intervention) requiring BiPAP or CPAP interventions were included. Head-to-head studies that compared different applications of NIV (e.g. CPAP vs. BiPAP) or different interfaces (e.g. NIV with a face mask vs. NIV with a helmet) were excluded.

Relevant comparators

The relevant comparator was considered to be usual care. This consisted of any alternative treatment to pre-hospital NIV, including standard oxygen therapy, standard medical therapy, delayed NIV or in-hospital NIV.

Outcomes

The outcomes of the review included the need for intubation, mortality (within 30 days), measures of breathlessness or respiratory function and patient-relevant outcomes.

Number of studies identified/included

The literature searches identified 2284 citations. Of these, eight RCTs46–53 and two quasi-randomised trials54,55 met the inclusion criteria. A flow chart describing the process of identifying relevant literature can be found in Figure 2.

FIGURE 2.

Study flow chart (adapted): clinical effectiveness review. 39

Number and type of studies excluded

A total of 55 full-text articles were excluded, as they did not meet all the prespecified inclusion criteria. The majority of the articles were excluded primarily on the basis of inappropriate study design (non-RCT), inappropriate setting (in-hospital NIV) or unsuitable publication type (reviews, commentaries or editorials). One of the excluded studies (the VeNIS BPCO trial),56 which was identified on a trials register, was a planned open-label RCT that was designed to evaluate the effectiveness of pre-hospital NIV compared with standard medical treatment in reducing intubation rates during acute respiratory failure in people with COPD. With an estimated enrolment of 398 adult patients from France, the study was due for completion (final data collection) in December 2014. However, at the time of writing, the study was not yet open for participant recruitment. A full list of excluded studies with reasons for exclusion is presented in Appendix 5.

Description of included studies (design and patient characteristics)

The design and patient characteristics of the 10 included studies46–55 that evaluated the effectiveness of pre-hospital NIV for adults with acute respiratory failure are summarised in Tables 1–3.

All studies were published between 2000 and 2012. Studies were undertaken in a variety of countries and settings including Australia,46 Europe (France,47,48,50 Germany51,52,55 and Spain49) and North America (Canada53 and the USA54). The duration of follow-up was not reported in two studies;46,54 however, the length of follow-up in the remaining studies ranged from 30 days48,51 to hospital discharge or death. 47,49,50,52,53,55 Of the 10 studies, only one study55 recieved funding from one or more commercial sponsors. The design of the included studies required the continuation of management of respiratory failure in the hospital setting and in-hospital management was generally at the discretion of the treating physicians.

As all studies included patients with acute respiratory failure without immediate need for intubation (i.e. spontaneously breathing patients), there was wide variation in terms of underlying conditions resulting in respiratory failure. Six studies46–48,50,54,55 included a selected population of patients with ACPO. Of these, two studies47,50 documented the exclusion of patients with COPD. Conversely, one study52 included patients with acute respiratory failure due to COPD. One study53 enrolled a diverse population, including patients with chronic heart failure, COPD, asthma, pneumonia and acute coronary syndrome with respiratory failure. The sample sizes of the included studies ranged from 23 patients55 to 207 patients,47 with the mean age of participants ranging from 68 years53 to 80 years. 46,47 The percentage of male participants ranged from 41%47 to 56%. 46

Continuous positive airway pressure was the most commonly used mode of NIV in the intervention arm of the included studies. While one study, that of Plaisance et al. ,50 compared the effectiveness of early CPAP (where patients had CPAP for the first 15 minutes after study inclusion, followed by CPAP with medical treatment for 15 minutes) with late CPAP (where patients received conventional medical treatment with supplementary oxygen for the first 15 minutes of study inclusion, followed by the addition of CPAP for another 15 minutes), five studies46–48,52,53 provided CPAP to patients in the intervention group while patients in the control group received conventional medical treatment only. Of the four studies49,51,54,55 that assessed the use of BiPAP versus standard usual care, two were quasi-randomised trials. 54,55 In the study by Roessler et al. ,51 NIV was initially started with CPAP; however, this was quickly changed to BiPAP, if CPAP was tolerated (22 of 24 patients in the intervention group).

The NIV intervention in the included studies, where reported, was provided either by paramedics49,53,54 or by an emergency physician. 47,48,50–52,55 However, in the study by Thompson et al. 53 the response to out-of-hospital emergency calls was the responsibility of an advanced life support team of paramedics, with ongoing online support provided by a physician, remotely. One study46 had no information relating to medical personnel administering pre-hospital NIV.

Although two studies46,49 did not provide details relating to the NIV interface or pressure support levels, eight studies47,48,50–55 used a face mask as the interface of choice for the administration of NIV. In RCTs evaluating CPAP, pressure levels were fixed at 7.5 cmH₂O50 or 10 cmH₂O. 48,53 Applied pressure levels used in other studies47,52 were determined by a titration method based on patient’s response to treatment and degree of comfort. In these studies, the pressure support levels ranged from 5 cmH₂O to 30 cmH₂052 and from 7.5 cmH₂O to 10 cmH₂O. 47 One study provided no details on pressure levels. 46 In studies evaluating BiPAP, a pressure of 12 cmH₂O was applied in the study by Weitz et al. 55 [positive end-expiratory pressure (PEEP), 5 cmH₂O; fraction of inspired oxygen (FiO₂), 0.6], with pressure support titration to achieve a tidal volume of 7 ml/kg body weight or more in treated patients. Similarly, Roessler et al. 51 adjusted pressure support levels from 5 cmH₂O to 20 cmH₂O, depending on comfort (PEEP, 5–15 cmH₂O). In contrast, BiPAP with a fixed airway pressure support (EPAP 7 cmH₂O; IPAP 19 cmH₂O) was administered in a study by Mas et al. 49 No information was available relating to pressure support levels in the study by Craven et al. 54 However, this was the only study that reported that NIV was applied following transfer of the patient into an ambulance equipped with the ventilation system.

Descriptions of the treatment schemes in the control groups were varied and included terminology such as ‘usual treatment’,54 ‘standard therapy’,49 ‘usual care’,46–48,53 ‘standard oxygen therapy’,52 ‘standard medical treatment’51,55 or delayed NIV. 50 As all control groups received conventional medical treatment together with supplementary oxygen for the management of acute respiratory failure or its underlying cause, treatment in the control groups is considered as usual care (standard oxygen therapy) throughout this report. The use of NIV in the control group varied between studies. In the study by Plaisance et al. ,50 which compared early CPAP with late CPAP, NIV use in the control group was regarded as mandatory. Two studies, those of Roessler et al. 51 and Thompson et al. ,53 reported that patients in the control groups were managed with NIV following admission (n = 4/25 and n = 4/35, respectively). NIV use in the control group of these studies was, therefore, regarded as allowed. Although no patients used NIV, Frontin et al. 48 reported that their study allowed the use of NIV in the control group. On the other hand, patients in the study conducted by Ducros et al. 47 presenting with intubation criteria could not receive any type of NIV support. These patients were intubated in the first instance. In this study NIV use in the control group was considered to be prohibited. For the remaining studies,46,49,52,54,55 the use of NIV in the control group was unclear.

Reporting of dosing regimens and number of patients in study groups receiving co-treatments varied across studies. However, the commonest interventions were diuretics46–48,51,53–55 and nitrates. 46,47,53–55

Quality characteristics

The overall methodological quality of the 10 included studies is summarised in Figure 3 and Table 4. Generally, six studies performed well,47–51,53 receiving a positive assessment on at least six out of nine methodological quality items.

FIGURE 3.

Methodological quality graph: review authors’ judgements about each methodological quality item as percentages across all included studies.

Of the eight RCTs,46–53 only six studies reported the method of randomisation. 47–51,53 In seven studies,47–53 similar methods for concealing treatment allocation were used: a randomly generated sequence of treatment allocation concealed in sealed envelopes. Two studies were considered to be quasi-randomised trials. 54,55 In Weitz et al. 55 the study design was described by the authors as a prospective, randomised trial. However, the method of randomisation in this study was based on date of birth. This method of assignment (systematic allocation) is not considered as strictly random. 57 In Craven et al. 54 the study was described by the authors as a prospective, sequential, parallel controlled trial. However, no details were provided on the method of randomisation. Moreover, in this study, 10 emergency service units were divided into five matched pairs (based on similar patient demographics). Five units (one of each matched pair) were then equipped with a BiPAP ventilation system and five without. A convenience sample of adults presumed to have chronic heart failure was given BiPAP by the emergency team during transport and was compared with a control group that received usual care.

The potential sources of bias most frequently identified in studies concerned lack of blinding of outcome assessment and lack of adequate power to detect differences in the primary outcome. Lack of blinding may influence intubation rate (although in a pragmatic trial this may be acceptable) but is unlikely to influence mortality. Many of the studies had small sample sizes (< 100 patients)46,49,51–55 so it is likely they had inadequate statistical power to detect between-group differences, even if they were present. All of the included studies were conducted outside the UK, making generalisability of the findings to the UK setting uncertain.

Effects of interventions

Network meta-analysis using aggregate data

A NMA was undertaken to compare the effectiveness of pre-hospital NIV for adults with acute respiratory failure in terms of mortality and intubation. 58 Figure 4 presents the network of evidence. A total of 10 studies46–55 comparing BiPAP or CPAP against standard care provided information on at least one of the outcomes being analysed, although two studies did not provided information on intubation. 46,55 A summary of all the trials (data) included in the base-case NMA is presented in Appendix 6.

FIGURE 4.

Network diagram of different pre-hospital NIV interventions vs. standard oxygen therapy (usual care) for acute respiratory failure.

Craven et al. 54 performed a cluster randomised trial. However, we had no information on the intracluster correlation coefficient and we did not adjust the effective sample size of this study. Consequently, this study may contribute more information to the analysis than is appropriate. A sensitivity analysis was performed by excluding data from Plaisance et al. 50 because the control group received delayed pre-hospital CPAP rather than in-hospital CPAP, and by excluding data from Craven et al. 54 and Weitz et al. 55 because neither was genuinely randomised.

Mortality

Data were available from 10 studies,46–55 including six comparing CPAP46–48,50,52,53 with usual care and four49,51,54,55 comparing BiPAP with usual care. However, there were no deaths in the Schmidbauer et al. 52 study, so this study provides no information about treatment effect. A summary of the results from the NMA is presented in Table 5.

TABLE 5 - Mortality in pre-hospital NIV patients with acute respiratory failure: posterior results for the odds of death relative to usual care (standard oxygen therapy) (random effects)

Usual care is defined as standard oxygen therapy with conventional medical treatment.

Treatment	Random-effects mean		Predictive distribution		Probability most effective
	OR	95% CrI	OR	95% CrI
NIV
BiPAP	1.94	0.65 to 6.14	1.93	0.50 to 7.98	0.008
CPAP	0.41	0.20 to 0.77	0.41	0.14 to 1.16	0.989
Usual carea
Reference	Reference	Reference	Reference	Reference	0.004
Between-study standard deviation	0.29	0.02 to 0.85	–	–	–

The NMA model fitted the data reasonably well, with a residual deviance close to the total number of (non-zero) data points included in the analysis. The total residual deviance was 18.82, which compared favourably with the 18 non-zero data points being analysed. The between-study standard deviation was estimated to be 0.29 (95% CrI 0.02 to 0.85). This suggests that there was mild heterogeneity between studies in the intervention effects, although with some uncertainty about the true variability in intervention effects between studies.

There was evidence to suggest that CPAP is the most effective of the three interventions (probability = 0.989). The effect of CPAP relative to usual care was statistically significant at a conventional 5% level (OR 0.41, 95% CrI 0.20 to 0.77). The heterogeneity in the effect of CPAP between studies meant that the effect relative to usual care in a randomly chosen study varies according to the characteristics of the study (OR 0.41, 95% CrI 0.14 to 1.16). There was considerable uncertainty associated with the effect of BiPAP relative to usual care (OR 1.94, 95% CrI 0.65 to 6.14). The heterogeneity in the effect of BiPAP between studies meant that the effect relative to usual care in a randomly chosen study varies according to the characteristics of the study (OR 1.93, 95% CrI 0.50 to 7.98).

A sensitivity analysis was performed excluding the studies by Plaisance et al. ,50 Craven et al. 54 and Weitz et al. 55 A summary of the results from the NMA is presented in Table 6. There was little impact on the heterogeneity in intervention effects between studies. As before, the intervention that exhibited the greatest effect was CPAP (OR 0.45, 95% CrI 0.21 to 0.93), although the heterogeneity in the effect of NIV between studies means that the intervention effects in a randomly chosen study varies substantially depending on the characteristics of the study (OR 0.46, 95% CrI 0.13 to 1.41). The effect of BiPAP relative to standard care remained uncertain (OR 1.95, 95% CrI 0.43 to 9.46).

TABLE 6 - Mortality in pre-hospital NIV patients with acute respiratory failure: posterior results for the odds of death relative to usual care (standard oxygen therapy) (random effects) – sensitivity analysis excluding the studies of Plaisance et al.,50 Craven et al.54 and Weitz et al.55

Usual care is defined as standard oxygen therapy with conventional medical treatment.

Treatment	Random-effects mean		Predictive distribution		Probability most effective
	OR	95% CrI	OR	95% CrI
NIV
BiPAP	1.95	0.43 to 9.46	1.95	0.33 to 11.47	0.039
CPAP	0.45	0.21 to 0.93	0.46	0.13 to 1.41	0.949
Usual carea
Reference	Reference	Reference	Reference	Reference	0.012
Between-study standard deviation	0.32	0.02 to 0.92	–	–	–

Intubation rates

Data were available from eight studies,47–54 including five47,48,50,52,53 comparing CPAP with usual care and three studies49,51,54 comparing BiPAP with usual care. The analysis assumes that the lack of intubation data from the studies of Weitz et al. 55 and Austin and Wills46 are not related to the effects of the interventions in these studies. A summary of the results from the NMA is presented in Table 7.

TABLE 7 - Intubation rates in pre-hospital NIV patients with acute respiratory failure: posterior results for the odds of intubation relative to usual care (standard oxygen therapy) (random effects)

Usual care is defined as standard oxygen therapy with conventional medical treatment.

Treatment	Random-effects mean		Predictive distribution		Probability most effective
	OR	95% CrI	OR	95% CrI
NIV
BiPAP	0.40	0.14 to 1.16	0.40	0.12 to 1.39	0.361
CPAP	0.32	0.17 to 0.62	0.32	0.13 to 0.82	0.639
Usual carea
Reference	Reference	Reference	Reference	Reference	0.000
Between-study standard deviation	0.21	0.01 to 0.73	–	–	–

The NMA model fitted the data well, with a residual deviance, 16.05, close to the total number of data points, 16, included in the analysis. The between-study standard deviation was estimated to be 0.21 (95% CrI 0.01 to 0.73). This indicated that there was mild heterogeneity between studies in the intervention effects, although with some uncertainty about the true variability in intervention effects between studies.

Both patients treated with BiPAP and those treated with CPAP were less likely to require intubation than those receiving usual care, although there was evidence to suggest that CPAP was the more effective intervention (probability = 0.639). The effect of CPAP relative to usual care was statistically significant at a conventional 5% level (OR 0.32, 95% CrI 0.17 to 0.62). The heterogeneity in the effect of CPAP between studies meant that the effect relative to usual care in a randomly chosen study varies according to the characteristics of the study (OR 0.32; 95% CrI 0.13 to 0.82). There was considerable uncertainty associated with the effect of BiPAP relative to usual care (OR 0.40, 95% CrI 0.14 to 1.16).

A sensitivity analysis was performed excluding the studies of Plaisance et al. ,50 Craven et al. 54 and Weitz et al. 55 A summary of the results from the NMA is presented in Table 8. There was a small increase in the between-trial standard deviation, although this is likely to be a consequence of there being fewer studies rather than a genuine increase in heterogeneity in intervention effects between studies. As before, the intervention that exhibited the greatest effect was CPAP (OR 0.34, 95% CrI 0.15 to 0.77), although the heterogeneity in the effect of NIV between studies means that the intervention effects in a randomly chosen study varies depending on the characteristics of the study (OR 0.34, 95% CrI 0.11 to 1.09). The effect of BiPAP relative to standard care remained uncertain (OR 0.53, 95% CrI 0.11 to 2.28).

TABLE 8 - Intubation rates in pre-hospital NIV patients with acute respiratory failure: posterior results for the odds of intubation relative to usual care (standard oxygen therapy) (random effects) – sensitivity analysis excluding the studies of Plaisance et al.,50 Craven et al.54 and Weitz et al.55

Usual care is defined as standard oxygen therapy with conventional medical treatment.

Treatment	Random-effects mean		Predictive distribution		Probability most effective
	OR	95% CrI	OR	95% CrI
NIV
BiPAP	0.53	0.11 to 2.28	0.53	0.09 to 2.87	0.306
CPAP	0.34	0.15 to 0.77	0.34	0.11 to 1.09	0.692
Usual carea
Reference	Reference	Reference	Reference	Reference	0.002
Between-study standard deviation	0.27	0.02 to 0.86	–	–	–

Electronic databases

Studies were identified by searching the following electronic databases and research registers:

• MEDLINE(R) In-Process & Other Non-Indexed Citations and MEDLINE(R) (via OvidSP) from 1948 to August 2013

• EMBASE (via OvidSP) from 1980 to August 2013

• Cumulative Index to Nursing and Allied Health Literature (via EBSCOhost) from 1982 to August 2013

• Cochrane Database of Systematic Reviews (via Wiley Online) from 1996 to August 2013

• Cochrane Central Register of Controlled Trials (via Wiley Online) from 1898 to August 2013

• Health Technology Assessment Database (via Wiley Online) from 1995 to August 2013

• Database of Abstracts of Review of Effects (via Wiley Online) from 1995 to August 2013

• NHS Economic Evaluation Database (via Wiley Online) from 1995 to August 2013

• BIOSIS Previews (via ISI Web of Knowledge) from 1969 to August 2013

• Science Citation Index Expanded (via Web of Science) from 1899 to August 2013

• Conference Proceedings Index – Science (via Web of Science) from 1990 to August 2013

• EconLit (via OvidSP) from 1961 to August 2013

• UK Clinical Research Network Portfolio Database (NIHR) from 2001 to October 2012

• National Research Register Archive (NIHR) from 2000 to September 2007

• Current Controlled Trials from 2000 to October 2012

• ClinicalTrials.gov (US National Institutes of Health) from 2000 to October 2012.

The keyword strategies developed in the review of clinical effectiveness (see Chapter 3, Methods for reviewing effectiveness, Identification of studies) were used with a sensitive economic evaluation (where applicable) or quality-of-life search filter aimed at restricting search results to economic and cost-related studies (used in the searches of MEDLINE, Cumulative Index to Nursing and Allied Health Literature and EMBASE). All resources were initially searched from inception to October 2012. With the exception of the four research registers, updated searches to August 2013 were conducted on the remaining electronic databases. An example of the MEDLINE search strategy is provided in Appendix 7.

Other resources

To identify additional published, unpublished and ongoing studies, the reference lists of all relevant studies (including existing systematic reviews) were checked and a citation search of relevant articles (using the Web of Science Citation Index Expanded and Conference Proceedings Citation Index – Science) was carried out to identify articles that cite the relevant articles. In addition, systematic keyword searches of the World Wide Web were undertaken using the Google search engine and key experts in the field were contacted.

All identified citations from the electronic searches and other resources were imported into and managed using the Reference Manager bibliographic software (version 12.0).

Model structure

A decision-analytic model was developed to estimate the costs and health outcomes associated with pre-hospital CPAP and standard care in a hypothetical cohort of patients with acute respiratory failure.

Population

The population consisted of a hypothetical cohort of patients with acute respiratory failure of any cause or no specified cause in a given ambulance service setting. Although this cohort can include patients with ACPO/heart failure, COPD or pneumonia, for the purpose of modelling they are treated as a single group. Using an incidence rate, the annual number of instances of patients with acute respiratory failure for a given ambulance service can be estimated based on the population served.

Intervention

There are multiple alternative specifications of the pre-hospital NIV approaches reported in the studies included within the systematic review; the interventions were classified and specified as reported in Chapter 3. These include (a) early CPAP provided by medical responders, (b) CPAP provided by paramedics, (c) CPAP provided by medical responders, (d) BiPAP provided by medical responders and (e) BiPAP provided by emergency physicians.

The clinical expert group decided that CPAP provided by paramedics was the approach most likely to be feasible in the UK, based on knowledge of currently available equipment, training requirements, anticipated costs, existing guidelines and the practicality of use in NHS ambulances. These factors all suggested that the model of pre-hospital NIV delivery most likely to be used in the NHS was CPAP delivered by paramedics. The meta-analysis would generate an estimate of treatment effect for the model by assuming a general effect for CPAP, regardless of whether it was provided by paramedics or physicians.

Comparator

Again, there are multiple alternative specifications of standard care reported in the studies included within the systematic review, which include (a) standard care without any NIV, (b) standard care with in-hospital NIV and (c) delayed pre-hospital NIV.

However, for the purposes of the economic model, in-hospital NIV is chosen as standard care. This assumption was deemed sensible by the clinical expert group, as any patient who received pre-hospital CPAP appropriately would have received in-hospital NIV if pre-hospital CPAP were not available.

Outcomes

The main outcomes in the model were QALYs, which are accrued by patients who survive their acute event (i.e. survive to 30 days).

Model perspective

The model took a lifetime horizon and the economic perspective of the model was the NHS in England and Wales.

Discount rate

Both the costs and QALYs were discounted at an annual discount rate of 3.5%, as recommended by NICE. 85

The key modelling methods together with the evidence sources and assumptions used to populate the model are discussed in detail in the following sections, Short-term outcomes, Effect of pre-hospital continuous positive airway pressure, Long-term health outcomes and Costs.

Mortality risk

The primary outcome measure, mortality, was typically recorded as a binary outcome and results presented as the response rate at 30 days. This 30-day mortality includes deaths in the ambulance and in the hospital.

The mortality risk of emergency admissions with respiratory illness was modelled using a large cohort data set of 668 patients presented with ‘respiratory disease’ across four ambulance services over a 4-year period from 1997 to 2001. 6 These were the Royal Berkshire Ambulance Service, the Derbyshire Ambulance Service, the Essex Ambulance Service and the West Midlands Ambulance Service. These ambulance services were representative of the types of environment typically encountered in England and included urban, mixed urban and rural, and very rural areas. Patients with ‘respiratory symptoms’ (n = 59, one death) were excluded, as these symptoms seemed to be less severe. There were 79 deaths in 668 patients, which resulted in a mean mortality rate of 11.8%. This was similar to that reported in a systematic review of 15 trials of NIV in ACPO, which reported an average mortality of 10.7% (42 deaths in 389 patients). 22 Thus, in the economic model a mean mortality rate of 11.8% was used with a beta distribution as shown in Table 15.

Scenarios for cost-effectiveness analysis

The distance from hospital of patients is an important factor when considering the optimal cost-effectiveness strategy because of the increase in mortality with increase in distance to hospital for the patient. More specifically, it is reasonable to assume that NIV is one of the interventions that, being currently available only in hospital, accounts for some of the association between distance and mortality. This effect of distance on mortality of emergency admissions for respiratory illness can be observed in the large cohort of 668 patients presenting with ‘respiratory disease’ across four ambulance services over a 4-year period from 1997 to 2001. 6 This relationship can be observed in the raw data, as shown in Figure 7, which show an increase in mortality with an increase in the distance to hospital. Appendix 8 shows mortality variation with distances, categorised as short (< 10 km), medium (10–20 km) and long (> 20 km), with longer distances associated with higher mortality.

FIGURE 7.

Variation in mortality with distance to hospital.

In addition, the average distance to hospital and the distribution of distance to hospital for a cohort of patients will vary according to the geographical setting of the ambulance service. For example, the distance to hospital is greater, on average, for patients in a rural setting (South Western Ambulance Service) than for patients in an urban setting (West Midlands Ambulance Service). Thus, with regard to pre-hospital CPAP for patients with acute respiratory failure, we tested the model in three different scenarios:

1. General population scenario: average distance to hospital and the distribution of distance to hospital is that of the general population in the UK. This scenario reflects the cost-effectiveness of different strategies at the national level.

2. Rural population scenario: average distance to hospital and the distribution of distance to hospital is that of a typical rural setting in the UK. This scenario reflects the cost-effectiveness of different strategies at the rural level; thus, the services in rural areas are able to decide if this scenario and the results best reflect their local practice.

3. Urban population scenario: average distance to hospital and the distribution of distance to hospital is that of a typical urban setting in the UK. This scenario reflects the cost-effectiveness of different strategies at the urban level; thus, the services in urban areas are able to decide if this scenario and the results best reflect their local practice.

This approach was taken because it was possible that different strategies may have different levels of cost-effectiveness in different settings. The decision to commission pre-hospital CPAP service is typically made at the ambulance service level and thus, the users of the results are able to decide which scenario best reflects their local practice and if pre-hospital CPAP is a cost-effective use of resources in their setting. For example, the cost-effectiveness of pre-hospital CPAP may be different in urban and rural areas, that is, it is possible that it might be more cost-effective in a rural setting, where the average distance to hospital is high, but less cost-effective in an urban setting, where the distance to hospital is not high.

The different scenarios are implemented in the model using three different baseline mortality risks: for the general population scenario, the rural population scenario and the urban population scenario. The mortality risk for the general population scenario is 11.8%, as described earlier in this section. The mortality rate for South Western Ambulance Service is used as the mortality risk for the rural population scenario and the mortality rate for West Midlands Ambulance Service is used as the mortality risk for the urban population scenario. The mortality rates used for the different scenarios are as shown in Table 15.

Risk of intubation

A key secondary outcome measure of interest is the intubation rate, which is recorded as a binary outcome. This risk of intubation for respiratory illness was modelled using the data from 3CPO study,18 a multicentre open prospective RCT of 1069 patients presenting with severe ACPO at 26 emergency departments in the UK. The 3CPO study reported a mean intubation rate of 2.9%. This is similar to the intubation rates of 2.7% reported in the British Thoracic Society’s National Respiratory Audit Programme Annual Report 2011/12. 86 The variance was estimated as 0.003% based on elicitation from the clinical experts. The mean value of 2.9% and the variance was then used to calculate the α and β parameters for a beta distribution and, thus, in the model, intubation rate is represented as a beta distribution with an alpha of 4.45 and a beta of 150.

Effectiveness parameters of pre-hospital continuous positive airway pressure used in the model

The ORs estimated from the NMA, as reported in Chapter 3, Effects of interventions, are used as estimates of effectiveness for pre-hospital CPAP. The probabilities of mortality and intubation for patients given pre-hospital CPAP were estimated by applying the log-ORs to the baseline parameters using the formulae below. If the baseline event rate is P, then µ is estimated as logit(P) = log[P/(1 – P)]. Then the absolute probabilities for the interventions (i.e. BiPAP and CPAP) are estimated as

where d is the log-OR for an intervention relative to standard care estimated from the NMA.

Effect on mortality

Pre-hospital CPAP can reduce the mortality of patients as they receive the NIV earlier in the ambulance than they would have if they had to wait to get to the hospital. The log-OR of pre-hospital CPAP for mortality reduction is estimated from the meta-analysis reported Chapter 3, Effects of interventions.

As shown in Table 15, there is an increase in mortality with increase in distance to hospital, with the mortality of patients > 20 km from the hospital twice as high as those < 10 km from the hospital. This suggests that, despite the constant mortality, the absolute benefit increases as the patient’s distance from the hospital increases, that is, patients farther from the hospital achieve more benefit from pre-hospital CPAP than those who are closer; for example, the absolute effectiveness of pre-hospital CPAP is higher in a rural setting than in an urban setting.

Effect on risk of intubation

The effect of pre-hospital CPAP on risk of intubation is modelled as log-OR estimated from the meta-analysis reported in Chapter 3, Effects of interventions. This log-OR parameter is assumed to be independent of the patient’s distance from the hospital, that is, the same risk reduction is applied to all patients irrespective of how far they are from the hospital.

The reduction in number of intubations leads to a reduction in the mean hospital length of stay because mean hospital length of stay for patients without intubation is 5.84 days, compared with 10.82 days for patients with intubation. Thus, the effect of pre-hospital CPAP on the length of hospitalisation is not modelled as an independent relative risk to avoid double counting; it was assumed that the reduction in the mean hospital length of stay is achieved only by reducing the number of intubations.

Costs of standard care

The cost of standard care was assumed as £0. This simplification was made as the analysis is based on incremental costs, that is, it was assumed that all initial treatment costs are same, regardless of whether or not the patient gets pre-hospital CPAP. The zero costs for standard care relates only to pre-hospital and emergency department treatment and does not include hospitalisation costs, intubation costs or additional lifetime costs for survivors. This was deemed sensible by the expert clinical group as it was assumed that all patients would receive NIV in hospital, irrespective of whether or not patients received pre-hospital CPAP. This does not cause any bias even if there are different mortality rates in pre-hospital CPAP and standard-care patients because deaths typically occur during or after emergency department treatment, so it is reasonable to assume emergency department costs are the same regardless of mortality rate. Thus, the only difference in treatment costs between standard care and pre-hospital CPAP was the additional costs of pre-hospital CPAP.

Costs of pre-hospital continuous positive airway pressure

There are a number of costs involved in providing pre-hospital CPAP, such as initial equipment costs, implementation costs and ongoing maintenance costs. These costs were converted into a cost per patient based on a 5-year depreciation period (i.e. assuming new pre-hospital NIV equipment will be required in 5 years) and sharing the overall costs out among the number of patients that would benefit from the service over this time period.

Hospitalisation costs

A further main outcome included in the model is the cost of stay associated with hospitalisations. The hospitalisation stay is dependent on whether or not the patient needs intubation, because length of stay is longer for patients who undergo intubation. Hubble et al. 80 report that the mean hospital length of stay for patients without intubation is 5.84 days. This is similar to the mean length of stay for patients with respiratory failure in the UK, estimated as the weighted average of 8801 patients associated with DZ27D and DZ27E (respiratory failure without intubation with major complications and comorbidity, and respiratory failure without intubation with intermediate complications and comorbidity) reported in the NHS Reference Costs 2011–2012. 89 Thus, the mean inpatient admission cost for hospitalisations was calculated as the weighted average of the costs of patients with DZ27D and DZ27E, from the NHS Reference Costs 2011–2012. 89 The hospitalisation cost used in the model, with a mean cost of £2400, is represented as a gamma distribution with an α of 80 and a β of 30.

Intubation costs

The cost of intubation was estimated as a one-off cost. Although costs for the DZ27A and DZ27B (respiratory failure with intubation with major complications and comorbidity, and respiratory failure with intubation with intermediate complications and comorbidity) were reported in the NHS Reference Costs 2011–2012,89 they were based on a small sample size of 180 patients across the UK whose average total length of stay is 7 days, that is, an additional 1.2 days for intubation. The clinical expert group deemed this as not representative and suggested using the data from Hubble et al. 80 to estimate the costs of intubation. Hubble et al. 80 report a mean hospital length of stay of 10.82 days for patients with intubation (i.e. approximately an additional 5 days for intubation compared with patients without intubation) and it was assumed that 5 additional hospital days spent by the intubated patients will be in the intensive care unit, based on the suggestions by the clinical expert group.

Thus, the cost of intubation was estimated by multiplying intensive care unit costs of £700 per day80 by the average length of stay for intubation, assumed to be 5 days, which results in a mean cost of intubation of £3500. The cost of intubation used in the model is represented as a gamma distribution with an α of 70 and a β of 50.

Lifetime costs of care

Lifetime costs of survivors were estimated using the annual costs and the discounted life expectancy of patients captured from the 3CPO trial. 18 The 3CPO study reported that the mean cost in months 4–6 was £1341, which resulted in mean annual costs of £5300. In the model, the annual cost is parameterised as a gamma distribution with an α of 53 and a β of 100, after discussions with the clinical expert group. It was assumed that the lifetime costs were the same for all survivors, irrespective of whether they were in the standard care or pre-hospital CPAP arm.

Definitions of cost-effectiveness terms

The ICER measures the relative value of two strategies and is calculated as the mean incremental cost divided by the mean incremental benefits. A strategy is dominated when another strategy accrues more QALYs for less cost. Extended dominance occurs when a combination of two alternative strategies can produce the same QALYs as a chosen strategy but at a lower cost. Strategies that are neither dominated nor extendedly dominated constitute the cost-effectiveness frontier, and the ICER is reported for these strategies compared with the next least effective strategy. The willingness-to-pay threshold is the amount of money that the decision-maker is willing to pay to gain one additional QALY. The usual threshold for decision-making at NICE is considered to be around £20,000–30,000 per QALY. The net monetary benefit is defined as the QALYs multiplied by a value for the QALYs (e.g. £20,000) minus the costs of obtaining them, that is, net monetary benefit = (QALYs × λ) – cost, where λ is the willingness-to-pay threshold. The net monetary benefit approach is simpler to calculate and gives equivalent findings (but requires an explicit assumption regarding the value of λ).

Uncertainty analysis

The results presented in the following section include the effects of accounting for uncertainty in the model parameters (the costs, utilities, risks and ORs for mortality and intubation), characterised as probability distributions. Probabilistic sensitivity analysis is undertaken whereby the model is rerun (1000 times), each time with a different value for the risks, ORs, costs and utilities, which are sampled from the probability distributions. The cost-effectiveness plane shows the incremental costs (y-axis) and incremental QALYs (x-axis) compared with usual care. In this chart, if a model run for a strategy had exactly the same costs and QALYs as usual care then the ‘sample’ for that model run would appear at the origin. Samples plotted to the right of the y-axis have more QALYs than usual care and samples plotted above the x-axis have more costs. Samples plotted to the right of a straight line with slope λ passing through the origin are cost-effective, whereas those plotted to the left are not. The cost-effectiveness acceptability curve (CEAC) shows the proportion of model runs for which each strategy is cost-effective over a range of potential willingness-to-pay thresholds (i.e. λ).

Another measure of uncertainty is the overall EVPI. This calculation is carried out based on the theory that the decision-maker will choose the most cost-effective option but could acquire additional evidence to reduce the uncertainties in the decision, for example, know exactly what the HRs for mortality and hospitalisations are for each treatment. In the EVPI calculation, it can be estimated how often making the decision based on current evidence could be wrong, and also how many QALYs (and costs) would be lost by choosing the strategy that is expected to be most cost-effective given current evidence, when in fact one of the other strategies is truly the most cost-effective. The monetary value lost by making a ‘wrong’ decision to choose a strategy based on current evidence can be estimated by valuing the QALYs using the willingness-to-pay threshold for this possible loss, that is, the net benefit lost on each of the occasions when another strategy would be optimum. This can be multiplied by the number of patients per year and the expected lifetime of the decision to estimate the population EVPI.

Value of information analysis

The interpretation of population EVPI is that, if one could fund research to eliminate the uncertainty in effectiveness for all of the parameters for each strategy (e.g. by a large or infinitely large clinical trial), then the value of eliminating the uncertainty through such research would be expected to be the population EVPI. This can be thought of as the maximum that the health-care system should be willing to pay for additional evidence to inform the decision in the future and, thus, is an upper bound on the value of conducting further research, that is, if the population EVPI exceeds the expected costs of additional research then it is potentially cost-effective to conduct further research. However, EVPI, defined as the maximum investment a decision-maker would be willing to pay to eliminate all parameter uncertainty from the decision problem, has the limitation that it assumes that all information can be determined with certainty.

Expected value of partial perfect information (EVPPI) is similar to EVPI, but instead of evaluating the uncertainty associated with all parameters it focuses on the uncertainty associated with a subset of one of more parameters, allowing the decision-makers to be able to conclude in which variables further research would be most beneficial. The computational time required for EVPPI is markedly more than for EVPI as the process essentially requires two iterations of probabilistic analyses, as standard probabilistic sensitivity analyses are undertaken for each sampled parameter value for the variable(s) under analysis. If the population EVPPI for a subset of parameters exceeds the expected costs of additional research, then it is potentially cost-effective to conduct further research to estimate those parameters.

Expected value of sample information addresses the limitation that values for the parameters can be ascertained without uncertainty, which effectively assumes an infinite trial size, and seeks to provide an optimal number of patients to study within a future trial. In addition, EVSI also allows the evaluation of marginal returns with an increased sample size formally taken into account (e.g. that an additional 100 patients, when only 500 have been recruited, would be likely to provide more value than when 20,000 have been recruited). Within EVSI the costs of the trial are compared with the benefits achieved in order to find the maximum expected net benefit of sampling, which would correspond with the recommended trial size. If the population EVSI of a proposed trial is greater than the costs of the trial, then it is cost-effective to conduct the trial to address the uncertainty. EVSI, similarly to EVPPI, requires two iterations of probabilistic analyses and, additionally, the updating of prior information with the simulated results of the future trial to form a posterior distribution.

Results of rural scenario

The QALY results suggest that the higher baseline mortality in the rural setting combined with the same relative effectiveness of pre-hospital CPAP would result in a higher estimated QALY gain for pre-hospital CPAP over standard care of 0.117 QALYs (mean QALYs = 1.494 for pre-hospital CPAP compared with 1.377 QALYs for standard care). This is because of the higher baseline mortality risk in the rural scenario, which, combined with same mortality ORs of pre-hospital CPAP, results in more lives saved, which leads to more QALYs gained compared with the base-case scenario. The expected costs over a lifetime also differ, with pre-hospital CPAP having higher costs (£16,724) than usual care (£14,540). Comparing pre-hospital CPAP with standard care, the incremental cost per QALY gained is £2184/0.117 = £18,744 per QALY, which is below the typical NICE threshold of £20,000–30,000 per QALY gained.

This can also be observed in the mean ICER, presented as a blue triangle in the cost-effectiveness plane of Figure 10, as it is just below the £20,000 per QALY threshold line (which is in line with our estimated mean ICER of £18,744 per QALY). The cost-effectiveness plane in Figure 10 shows that the majority of the samples fall to the right of the diagonal line, suggesting that pre-hospital CPAP has a higher chance of being cost-effective than usual care.

FIGURE 10.

Cost-effectiveness plane for the rural population scenario.

The CEAC in Figure 11 shows the proportion of model runs for which each strategy is cost-effective over a range of potential willingness-to-pay thresholds. The percentage of model runs in which pre-hospital CPAP was the most cost-effective strategy (at a £20,000 per QALY threshold) was 58.8%, which again indicates less uncertainty as to which strategy is optimum in terms of net benefit.

FIGURE 11.

Cost-effectiveness acceptability curve for the rural population scenario.

Results of urban scenario

The QALY results suggest that the higher baseline mortality in the urban setting, combined with the same relative effectiveness of pre-hospital CPAP, would result in a lower estimated QALY gain for pre-hospital CPAP over standard care of 0.093 QALYs (mean QALYs = 1.520 for pre-hospital NIV, compared with 1.427 QALYs for standard care). This is because of the lower baseline mortality risk in the urban scenario, which, combined with same mortality ORs of pre-hospital CPAP, results in fewer lives saved, which leads to fewer QALYs gained. The expected costs over a lifetime also differ, with pre-hospital CPAP having higher costs (£16,950) than usual care (£14,971). Comparing pre-hospital CPAP with standard care, the incremental cost per QALY gained is £1979/0.093 = £21,284 per QALY, which is above the typical NICE threshold of £20,000 per QALY gained.

This can also be observed in the mean ICER, presented as a blue triangle in the cost-effectiveness plane of Figure 12, as it is above the £20,000 per QALY threshold line (in line with our estimated mean ICER of £21,284 per QALY). The cost-effectiveness plane in Figure 12, shows that only a minority of the samples fall to the right of the diagonal line, suggesting that the chance of pre-hospital CPAP being cost-effective compared with usual care is lower than in the base-case scenario.

FIGURE 12.

Cost-effectiveness plane for the urban population scenario.

The CEAC in Figure 13 shows the proportion of model runs for which each strategy is cost-effective over a range of potential willingness-to-pay thresholds. The percentage of model runs in which pre-hospital CPAP was the most cost-effective strategy (at a £20,000 per QALY threshold) was 41.5%, which indicates greater uncertainty as to which strategy is optimum in terms of net benefit.

FIGURE 13.

Cost-effectiveness acceptability curve for the urban population scenario.

Results of high-cost scenario

The expected costs over a lifetime are higher in this scenario, with the cost of pre-hospital CPAP being higher, at £17,078, than that of standard care (£14,863), a cost difference of £2215. This is higher than the incremental costs in the base-case scenario, where the cost difference is £2032 (pre-hospital CPAP costs of £16,895, compared with a usual care cost of £14,863). This variation is due to the difference in the costs of pre-hospital CPAP in the high-cost scenario and the base-case scenario. In this high-cost scenario, the incremental cost per QALY gained is £2216/0.099 = £22,368 per QALY, which is above the typical NICE threshold of £20,000 per QALY gained, as shown by the blue triangle in the cost-effectiveness plane in Figure 14. The cost-effectiveness plane in Figure 14 shows that the only a minority of the samples fall to the right of the diagonal line, suggesting that pre-hospital CPAP has a low chance of being cost-effective compared with usual care.

FIGURE 14.

Cost-effectiveness plane for the high-cost scenario.

The CEAC in Figure 15 shows the proportion of model runs for which each strategy is cost-effective over a range of potential willingness-to-pay thresholds. The percentage of model runs in which pre-hospital CPAP was the most cost-effective strategy (at a £20,000 per QALY threshold) was 35.4%, which indicates increased uncertainty that pre-hospital CPAP is optimum in terms of net benefit. This was also observed in Figure 14, where most of the samples fall on left side of the diagonal line, that is, there is only a one-third chance of pre-hospital CPAP being cost-effective at a threshold of £20,000 per QALY.

FIGURE 15.

Cost-effectiveness acceptability curve for the high-cost scenario.

Results of low-cost scenario

The expected costs over a lifetime are lower than in the base-case scenario, with the cost of pre-hospital CPAP being £16,421, compared with £14,863 for usual care and, comparing pre-hospital CPAP with standard care, the incremental cost per QALY gained is £1558/0.099 = £15,728 per QALY, which is below the typical NICE threshold of £20,000 per QALY gained, suggesting that it is cost-effective. The cost-effectiveness plane for the high-cost scenario is shown in Figure 16.

FIGURE 16.

Cost-effectiveness plane for the low-cost scenario.

The CEAC in Figure 17 shows the proportion of model runs for which each strategy is cost-effective over a range of potential willingness-to-pay thresholds. The percentage of model runs in which pre-hospital CPAP was the most cost-effective strategy (at a £20,000 per QALY threshold) was 79.8%, which indicates greater confidence to state that pre-hospital CPAP is optimum in terms of net benefit.

FIGURE 17.

Cost-effectiveness acceptability curve for the low-cost scenario.

Results of lower-cost scenario

The expected costs over a lifetime for this scenario are even lower, with the costs of pre-hospital CPAP being £15,977, compared with £14,863 for standard care, that is, pre-hospital CPAP incurs additional costs of £1114 compared with standard care. This is much lower than the incremental cost in the base-case scenario, which is £2032 (pre-hospital CPAP costs of £16,895 compared with usual care costs of £14,863), and this variation is a result of the lower costs of pre-hospital CPAP in this scenario than in the base-case scenario.

The incremental cost per QALY gained is £1114/0.099 = £11,248 per QALY, which is much lower than the typical NICE threshold of £20,000 per QALY gained, suggesting that pre-hospital CPAP is cost-effective, as shown in the cost-effectiveness plane in Figure 18. The cost-effectiveness plane in Figure 18 shows that the majority of the samples fall to the right of the diagonal line, suggesting that pre-hospital CPAP has a high chance of being cost-effective compared with usual care.

FIGURE 18.

Cost-effectiveness plane for the lower-cost scenario.

The CEAC in Figure 19 shows the proportion of model runs for which each strategy is cost-effective over a range of potential willingness-to-pay thresholds. The percentage of model runs in which pre-hospital CPAP was the most cost-effective strategy (at a £20,000 per QALY threshold) was 93.8%, which indicates a much greater confidence to state that pre-hospital CPAP is optimum in terms of net benefit.

FIGURE 19.

Cost-effectiveness acceptability curve for the lower-cost scenario.

Expected value of partial perfect information

Partial EVPI provides the value of reducing the uncertainty surrounding particular input parameters in the decision model, and this can be used to identify the parameters for which more precise estimates would be most valuable to focus further research.

The EVPPIs associated with the parameters are illustrated in Figure 21. At the threshold of £20,000 per QALY, EVPPIs associated with ‘baseline mortality’, ‘pre-hospital CPAP mortality effectiveness’ and ‘total costs of pre-hospital CPAP’ are £14.85, £156.12 and £37.54, respectively. Other parameters do not come out of the analysis as substantial, as shown in Figure 21. In addition, the combined EVPPI for the three parameters (i.e. ‘baseline mortality’, ‘pre-hospital CPAP mortality effectiveness’ and ‘total costs of pre-hospital CPAP’) is £174.61, which is close to the overall EVPI; this suggests that most of the uncertainty in the model is due to the uncertainty in these three parameters.

FIGURE 21.

Individual patient EVPPI at £20,000 per QALY.

The high EVPPIs associated with the three parameters above suggest that further experimental research to estimate these parameters will potentially be cost-effective. EVPPI for the whole population can be estimated as EVPPI per patient multiplied by the number of patients affected by the decision over the lifetime of the technology. Assuming an annual incidence of 3.5 per 100,000 population, the undiscounted population EVPPI for the three parameters (i.e. ‘baseline mortality’, ‘pre-hospital CPAP mortality effectiveness’ and ‘total costs of pre-hospital CPAP’) at the threshold of £20,000 per QALY is £180 × 2100 × 5 = £1.83M. If an annual incidence of 40.8 per 100,000 is assumed, then the population EVPPI is £21.3M.

Expected value of sample information

Expected value of sample information seeks to provide an optimal number of patients to study within a future trial by comparing the costs of the trial with the benefits achieved in order to find the recommended trial size. For any given set of parameters, EVSI values are always lower than partial EVPPI values and hence EVSI analysis was conducted only for ‘baseline mortality’, ‘pre-hospital CPAP mortality effectiveness’ and ‘total costs of pre-hospital CPAP’, as the EVPPI values for other parameters are not significant (i.e. further experimental research to estimate other parameters may not be cost-effective). The EVSI values were estimated for a RCT conducted to estimate the ‘baseline mortality’ and ‘pre-hospital CPAP mortality effectiveness’ and the analysis was carried out for different sample sizes in order to estimate the optimal trial size.

It should also be noted that the trial has benefits other than EVSI as, for example, it also allows the estimation of the incidence of cases eligible for pre-hospital NIV, the costs of setting up and running the service, and determining whether or not a large trial would be feasible. This suggests that the EVSI will be higher than that reported in Table 24. If we assume that the trial will also address the uncertainty in the cost of pre-hospital CPAP, then an additional EVSI of £35.56 per patient can be added on top of the original EVSI.

If the population EVSI of a proposed trial at a given sample size is greater than the costs of the trial, then it is cost-effective to conduct the trial to address the uncertainty. For example, it is cost-effective to do a trial with 100 patients in each arm (trial 4) if the overall cost is less than £1.08M or £12.67M, depending on which source of incidence is more believable. If an incidence of 40 patients per 100,000 population is to be believed, then there is a clear argument for conducting a trial, as the expected benefits of the trial outweigh the costs of conducting such a trial.