A randomised controlled trial and cost-effectiveness analysis of high-frequency oscillatory ventilation against conventional artificial ventilation for adults with acute respiratory distress syndrome. The OSCAR (OSCillation in ARDS) study

History of artificial ventilation

An artificial ventilator is essentially a device that replaces or augments the function of the inspiratory muscles, providing energy to ensure a flow of gas into the alveoli during inspiration. Exhalation during artificial ventilation is usually a passive process, so when this inspiratory assistance is removed the inspired gas is expelled as the lung and chest wall recoil to their original volume. In the earliest reports of artificial ventilation, the respiratory muscles of another person, as expired air resuscitation, provided this energy. Baker25 has traced references to expired air resuscitation in the newborn as far back as 1472, and in adults there is a report of an asphyxiated miner being revived with mouth-to-mouth resuscitation in 1744. In the eighteenth century, artificial ventilation became the accepted first-line treatment for drowning victims, although bellows replaced mouth-to-mouth resuscitation. 26 Automatic artificial ventilators which did not require a human as a power source were proposed by Fell 150 years later27 and made commercially available by Dräger in 1907. 28 These were still resuscitation devices as the Dräger company at that time made mine rescue apparatus not medical devices.

The introduction of artificial ventilators into anaesthetic practice proceeded slowly until surgical advances required surgeons to undertake thoracotomies. Without artificial ventilation during a thoracotomy, lung collapse and mediastinal movement made surgery difficult and anaesthesia hazardous. Mortality was markedly reduced with artificial ventilation. A further boost to the development of artificial ventilators occurred in 1952 when a catastrophic poliomyelitis epidemic struck Denmark. Although the combined use of tracheostomy and artificial ventilation reduced the mortality, especially in the patients with bulbar palsy, the artificial ventilation had to be provided entirely by hand and required 1400 university students working shifts. The fear of another epidemic expedited research into powered mechanical ventilators, leading to the development of the first modern ventilator, the Engström, in 1952. 29 Since the advent of microprocessors and computer-controlled gas valves, artificial ventilators have become increasingly sophisticated, though evidence of the effectiveness of any single ventilation mode or ventilator is often lacking.

During both spontaneous breathing and during artificial ventilation, tidal volumes (breaths) have to be greater than the volume of the trachea and conducting airways (the anatomical dead space). Tidal volumes less than the anatomical dead space move gas in and out of these airways but do not ventilate the alveoli, and so no gas exchange with blood in the pulmonary capillaries takes place. Anatomical dead space is usually about 2 ml/kg in adults, tidal volumes are usually set at about 10 ml/kg in anaesthetic practice and 6–8 ml/kg ideal body weight in adults artificially ventilated for acute lung conditions.

However, it has been known for many years that this ‘convective’ model of ventilation does not apply in all circumstances. As early as 1915, Henderson et al. 30 noted that panting dogs were able to eliminate carbon dioxide, even though the volume of each breath was less than their anatomical dead space. In 1954, Briscoe et al. 31 reported that in humans the anatomical dead space appears to be reduced at low tidal volumes, allowing more gas exchange than would be predicted using a convective model of ventilation. However, the absolute amount of carbon dioxide eliminated per breath is very small, so high respiratory frequencies are needed to clear metabolic carbon dioxide production.

The first description of high-frequency ventilation in a clinical setting is attributed to either Lunkenheimer et al. in 197232 or Jonzon et al. in 1971,33 both of whom used the technique to minimise the cyclical effects of intermittent positive-pressure ventilation on the cardiovascular system. Subsequent research into high-frequency ventilation initially concentrated on three techniques to deliver the breaths, HFOV, high-frequency positive-pressure ventilation (HFPPV) and high-frequency jet ventilation (HFJV). External HFOV (EHFOV) using either a cuirass ventilator34 or a pneumatic vest35 was also introduced but was mostly used as an adjunct to physiotherapy and as a research tool rather than a mode of ventilation for critically ill patients.

It became apparent that both HFJV and HFPPV probably had no special properties and conformed to the conventional, convective model of gas exchange. 36 However, it also became clear that carbon dioxide clearance could be achieved with HFOV in animals37 and humans38 with tidal volumes that were half the anatomical dead space or less. There are many theories to explain this phenomenon. All of them reject the simple anatomical (series) dead space concept, and assume there is no sharp cut-off between dead space and alveolar gas, and some form of mixing takes place. The most likely mechanism, termed ‘convective streaming’, is that the interaction of the gas-airway wall friction and the asymmetrical inspiratory-expiratory flow profiles lead to an inward movement of gas in the core of the large airways and an outward movement near the wall. 39 These theories have been extensively reviewed. 40,41

As tidal volumes during HFOV are very small, the peak pressures generated in the alveoli during artificial ventilation are correspondingly modest. Thus HFOV would seem an ideal technique to ventilate patients at risk from pressure-induced lung damage (‘barotrauma’) such as infants with the (infant) respiratory distress syndrome (RDS). This was the rationale behind the early trials of HFOV in infants.

Trials of high-frequency oscillatory ventilation in infants

In pre-term infants with immature lungs, (infant) RDS is a major cause of immediate mortality. In survivors there is also considerable long-term morbidity from bronchopulmonary dysplasia, a condition caused by the combination of high intrapulmonary pressures generated by artificial ventilators, and high concentrations of inspired oxygen. As case reports began to appear in the literature suggesting HFOV might benefit these patients, the National Institutes of Health (NIH) in the USA first convened a conference42 and then commissioned a randomised controlled study. The HiFi study, published in 1989,43 recruited 673 pre-term infants who were randomly assigned to either HFOV using a piston-driven ventilator or to conventional mechanical ventilation. This study showed no survival benefit or difference in the incidence of bronchopulmonary dysplasia in the HFOV group.

Since the HiFi study there have been have been a further 17 RCTs of HFOV in infants. These are described in a series of Cochrane reviews44–47 and independent48 reviews last updated in 2013. In general, though many studies showed more deaths in the conventional ventilation groups, either as individual studies or combined in a meta-analysis, no statistically significant difference could be detected. However, a repeated theme in both the commentaries in the meta-analyses, and in opinion pieces published alongside the trials,49 is that the negative results may be partially due to errors in trial design.

Trials of high-frequency oscillatory ventilation in adults

As part of the preparation for the OSCAR study, we undertook a systematic review of HFOV in patients with ARDS and ALI in July 2006 to update the 2004 Cochrane review. 14 Among the 319 papers identified, there were only two adequate-quality RCTs of HFOV in adults.

The first and largest RCT was published in 2002. 17 Recruitment took place between October 1997 and December 2000 in 13 university-affiliated medical centres in the USA and Canada. A total of 148 patients were recruited. The entry criteria were the ARDS consensus criteria,50 a PEEP of ≥ 10 cmH₂O, and a predicted 6-month survival of 50% or greater. The investigators initially set the HFOV ventilator to a frequency of 5 Hz (breaths per second) and a mean pressure 5 cmH₂O above the mean airway pressure on conventional ventilation, and the amplitude to ‘visible chest wall movement’. The conventional ventilation group were treated with pressure-controlled ventilation to a maximum tidal volume of 10 ml/kg. The HFOV group had better oxygen exchange as measured by P : F ratios, most notably in the first 24 hours of HFOV, though when corrected for the difference in mean airway pressure (which increases oxygen exchange) using the oxygenation index (OI) this difference disappeared. Although more deaths were seen in the control group, this was not statistically significant (see Figure 1, below).

The second study was published in 2005. 51 Recruitment took place in four university-affiliated hospitals in London, Cardiff, Mainz and Paris between October 1997 and March 2001. The entry criteria and HFOV management were virtually identical to the 2002 study. A total of 61 patients were recruited. This study showed a beneficial effect of HFOV on oxygenation, even when airway pressures were taken into account. There was an excess of deaths in the HFOV group which was not statistically significant.

The results for 30-day mortality from both studies have been combined in the forest plot in Figure 1. There is no statistically significant benefit for HFOV seen.

FIGURE 1.

Forest plot showing 30-day mortality in the two methodologically sound trials of HFOV in adult patients with ARDS identified at the start of the OSCAR study. OR, odds ratio.

The report from Bollen et al. 51 contained a post-hoc analysis showing how the treatment effect on mortality varied with the severity of the initial lung damage as determined by the OI. As the OI (and hence the severity of the lung injury) worsened the odd ratios for survival increasingly favoured HFOV. The numbers in each OI band were very small, and there were corrections added to remove other known causes of mortality. There was a clear stepwise increase in treatment effect with increasing OI and hence disease severity, suggesting that HFOV might be more effective in patients with worse ARDS, either because these patients are more prone to secondary lung damage or simply because they have a higher mortality.

The original systematic reviews of HFOV by Wunsch et al. 14,15 reviewed as part of planning the OSCAR study predated the publication of the Bollen et al. 51 study, though the subsequent 2008 revision made no mention of the study. The systematic review published in 2010 while the OSCAR study was under way18 contained eight studies including the Bollen et al. 51 and Derdak et al. 17 studies, one study involving prone positioning which had been excluded from our original planning review, two studies undertaken in children and two studies published only as abstracts. The only fully reported study of 49 patients published since OSCAR started had no mortality data. 52

Types of high-frequency oscillatory ventilators available

Although there is a wide range of high-frequency oscillatory ventilators available for neonatal use [e.g. SensorMedics 3100A^® (SensorMedics Corporation, Yorba Linda, CA), Stefan SHF3000 (Fritz Stephan GmbH, Gackenbach, Germany), Hummingbird V (Metran Co. Ltd, Saitama, Japan), Dufour OHF1 (Dufour, Villeneuve d'Ascq, France)], there are only two commercially available positive-pressure HFOVs suitable for adults. The ventilator used in all of the studies in adults up to the start of the OSCAR study was the ‘SensorMedics Model 3100B^® high-frequency oscillatory ventilator’ (SensorMedics Corporation, Yorba Linda, CA), manufactured by SensorMedics Corporation in California and distributed in the UK by Viasys Healthcare. This ventilator was approved by the US Food and Drug Administration (FDA) in 2001 for ventilation of selected patients over 35 kg in weight with acute respiratory failure. The second adult HFOV ventilator available is the ‘Novalung’ Metran R100^® high-frequency ventilator (Metran Co. Ltd, Saitama, Japan) (also called the ‘Vision Alpha’) manufactured in Saitama, Japan by Metran Co. Ltd and distributed in the UK by Inspiration Healthcare, which has a long history of use in Japan but, at the time the OSCAR study was being planned, had only recently been CE (Conformité Européenne) marked for European use. A negative pressure external (cuirass) HFOV for adults [‘Hayek Oscillator’, United Hayek Industries (Medical) Ltd, London, UK] is also marketed in the UK and elsewhere.

Current users of high-frequency oscillatory ventilation in the UK

Viasys Healthcare provided us with details of all 3100B ventilators ever sold in the UK at the planning stage of the OSCAR study. A total of 38 ventilators have been sold by 2007 to 25 adult ICUs in England, Wales and the Isle of Man. Eighteen units had one ventilator; the remainder had two or three devices. No further data were obtained during the OSCAR study.

‘Substantial uncertainty’ within units already using high-frequency oscillatory ventilation

To run a RCT of HFOV in ICUs that already own one or more HFOV ventilators would require all clinicians caring for the patients to be substantially uncertain about which ventilation was best for their patients. As patients in the trial would be randomised in a 1 : 1 ratio to conventional positive-pressure ventilation or HFOV, up to 50% of the patients who the clinicians would treat with HFOV under their current protocols or guidelines would be randomised to conventional positive-pressure ventilation. In essence, the clinicians would have to withhold their standard treatment from half of the patients in the trial. The nature of medical care for patients in ICUs, where each consultant works for a block of time before handing onto a colleague, means that ‘substantial uncertainty’ would have to be present in the whole team to ensure the allotted trial treatment was continued throughout a patient’s ICU stay.

Lack of ‘substantial uncertainty’ would expose the study to a considerable risk of bias. Clinicians might elect not to enter the more severely ill patients in the study and treat them with HFOV outside of the trial. This would mean the trial population would not be representative of the UK patients with ARDS and would reduce the generalisability of the results. Crossover from the control to the treatment group might also occur, limiting the ability of the study to show an effect.

We tried to find evidence in the literature to determine whether or not these problems had occurred in previous trials. In the clinical trial of HFOV reported by Bollen et al. in 2005,51 61 patients were recruited from four major European ICUs in 41 months, or about one patient per centre per 3 months. This would be about 0.4% of total admissions. All of these ICUs had prior experience with HFOV. The inclusion criteria for the study were the standard consensus criteria for ARDS50 which include patients with a P : F ratio of < 26.7 kPa (200 mmHg). The mean P : F ratio in the treatment group was 12.6 kPa and in the control group was 16.0 kPa. No Consolidated Standards of Reporting Trials (CONSORT) diagram was published. From published data, we know that about 8% of admissions to general ICUs meet the consensus criteria for ARDS during their ICU stay. Therefore, both the low recruitment rate and the severity of the respiratory failure would suggest there was considerable case selection taking place, though whether or not patients received HFOV outside the study is not known. The trial was stopped prematurely because of ‘poor recruitment’ attributed, in the paper, to lack of local trial-dedicated staff. Crossovers (18%) occurred by protocol in this study.

The other clinical trial reported by Derdak et al. in 200217 took place in 13 university-affiliated medical centres in the USA and Canada over 38 months, a recruitment rate of one patient per centre per 3.6 months. There are no data on whether or not these centres had prior experience with HFOV but seven of the sites’ clinical leads had published on HFOV prior to the study. The inclusion criteria for the study were the standard consensus criteria for ARDS. The mean P : F ratio in the treatment group at enrolment was 15.0 kPa and in the control group 14.6 kPa. Again, no CONSORT diagram was published. Crossovers were 4/75 (5.3%) from HFOV to control, and 9/73 (12.3%) from control to HFOV. Again, these data suggest marked case selection was taking place, though again it is not known if HFOV was used outside of the trial.

We discussed the OSCAR trial with the clinical leads or senior clinicians in five UK ICUs where HFOV was used. Although all initially expressed interest in the study, when we explained that the study would require withholding HFOV in some patients, four of the clinicians suggested they could not take part in a trial under these circumstances.

Finally, we reviewed the experience gained by the Chief Investigator (DY) as a member of the management group for the PAC-Man study. 54,55 The clinicians in ICUs were asked to withhold a pulmonary artery catheter in half of the patients they would normally have used one in. Although the study was successful, a considerable effort was required in the early stages of the trial to generate equipoise. The trial probably only succeeded because it was a trial of a monitoring device, not a treatment, and nearly two-thirds of the ICUs used another monitoring method to generate at least some of the information a pulmonary artery catheter would have given them.

We concluded there are considerable risks to running the trial in centres which already used HFOVs. We believed the major risk was that HFOV would continue outside the trial as a rescue therapy and so the patients in the study would be an unrepresentative sample. The final trial design was a compromise; most ICUs were HFOV-naïve at study commencement, but two centres with HFOV experience did initially join the study and provided a pool of experienced health-care staff.

Identification of other trials of high-frequency oscillatory ventilation and trials competing for the same patient population

When the OSCAR trial was starting we were aware that Professor Meade and Dr Ferguson from the Canadian Critical Care Trials Group (CCCTG) had received funding for a national pilot study of HFOV [the Oscillation for Acute Respiratory Distress Syndrome Treated Early (OSCILLATE) study]. The OSCAR team contacted Professor Meade and agreed a common core data set to facilitate future meta-analyses should the pilot study progress to a full study. The OSCILLATE study did proceed to a full study and ceased recruitment just after the OSCAR study closed to recruitment. The OSCILLATE results56 are covered in Chapter 6.

The National Heart, Lung, and Blood Institute (NHLBI) in the USA funded a phase II study of HFOV using surrogate outcome measures (inflammatory cytokine concentrations in plasma) as a prelude to a full clinical trial (ClinicalTrials.gov identifier NCT00399581). This started in 2006 and was completed in 2009. The Chief Investigator was Dr Roy Brower. The results have not yet been published at the time of writing.

Identification of other trials in the UK competing for the same population

None of the ICUs we contacted as potential trial sites identified any studies of patients with ARDS that would compete for patients.

Three of the OSCAR investigators were on the management group of the Beta Agonist Lung Injury TrIal-2 (BALTI-2) study, a trial of intravenous salbutamol in ARDS. As the OSCAR study commenced, this was in a pilot phase in the West Midlands. It subsequently went to a full study57 and required the OSCAR study team and the BALTI-2 study team to communicate to prevent competition for patients as the entry criteria for the two studies were near identical.

There was a single-centre RCT of the effects of simvastatin in patients with ARDS under way in the Royal Victoria Hospital, Belfast as OSCAR started. This study became multicentre (HARP-2 study,58 ISRCTN88244364) but used different study centres.

Identification of data to inform estimates of the recruitment rate

A systematic review of all epidemiological studies of ALI and ARDS undertaken after the consensus criteria were formulated in 1994 was recently published. 7 The European and Australasian studies using consensus criteria to define ARDS cited in this review, along with additional studies identified from a systematic literature search undertaken in August 2006 by ourselves, are summarised in Table 1.

From these studies, it would appear the incidence of ARDS in ICUs was about 6–8% of all admissions when OSCAR started.

Three estimates of the incidence of ARDS in UK ICUs were available. The Scottish Intensive Care Society Audit Group (SICSAG) published data from 23 of the 26 ICUs in Scotland for an audit run between May and December 1999. 5 They recorded patients meeting the diagnostic criteria for ARDS (including chest radiographs) on a daily basis. The results are in Table 1 (Hughes et al. 5).

Two other (unpublished) estimates of the number of cases of ARDS in UK ICUs were available. In both data sets, the diagnosis of ARDS is based on the P : F ratio only and did not include chest radiograph information or any clinician ‘filtering’.

The Intensive Care National Audit & Research Centre (ICNARC) reviewed 261,193 admissions to UK ICUs over a 10-year period to 2005 and found an incidence of ARDS, defined solely on the P : F ratio in the first 24 hours of ICU admission, of 49.3%. These data are unpublished.

We undertook a similar study using data on admissions to the adult ICU at the John Radcliffe Hospital, Oxford, for calendar year 2005. Of 973 admissions, 850 had simultaneous arterial blood gas analyses and FiO₂ records which allowed P : F ratios to be calculated. The incidence of ARDS, defined on P : F ratio only at any point in the patient’s stay, was 78.9%. As the incidence was so high, we also searched the discharge summaries for any mention of ARDS. Only 2.5% of the patients had both a P : F ratio of < 26.7 kPa and any mention of ARDS in the discharge summary.

The true incidence of ARDS in ICU patients was almost certainly greater than the 2.5% we identified by retrospectively searching the Oxford database of discharge summaries because of errors of omission. However, it is also very clear that estimates of the incidence of ARDS based on the incidence of P : F ratios of < 26.7 kPa grossly overestimate the true incidence of ARDS.

As a result, the ICNARC and Oxford data on the incidence of ARDS were not used to inform recruitment rates or sample size calculations. The erroneously high incidence of ARDS identified in these databases presumably results from the loose definitions of ARDS used by ICNARC and at Oxford which did not include chest radiograph data, and because at least 50% of patients who meet the ARDS oxygenation criteria (P : F ratio of < 26.7 kPa) only have a very transient reduction in the P : F ratio which rapidly improves. 61

Selection of entry criteria, acute lung injury and acute respiratory distress syndrome or just acute respiratory distress syndrome?

As discussed above, acute hypoxaemic respiratory failure was divided into two severity bands by the consensus conference held in 1994. The less severe band was termed Acute Lung Injury or ALI, and includes patients with a P : F ratio of between 26.7 and 40.0 kPa. The more severe band, where the P : F ratio was < 26.7 kPa, was termed acute respiratory distress syndrome or ARDS.

There is often confusion over the terms ALI and ARDS. In some literature, the term ALI is incorrectly used to encompass all patients with a P : F ratio < 40 kPa, and the term ARDS is used to describe a subset of these with a P : F ratio of < 26.7 kPa. We did not use this convention and kept to the definitions published after the consensus conference in which ALI and ARDS are two discrete bands of severity of acute hypoxaemic respiratory failure with no overlap.

The commissioning brief from the HTA requested a study of HFOV in patients with ALI or ARDS. We elected to use ARDS only as an entry criterion for the trial. The reasons for this are as follows.

ALI represents a group of patients who will require between 30% and 45% inspired oxygen to maintain a normal PaO₂ of 12 kPa. This degree of hypoxaemic respiratory failure would normally be managed with simple face mask oxygen as the patients do not require artificial ventilation. It follows from this that patients with ALI who are on artificial ventilators are probably ventilated as a result of non-pulmonary pathology which would not be improved by HFOV, and so would reduce the chance of seeing an effect of any ventilatory strategy if included in a clinical trial. Examples of such patients would be those with neurological conditions such as head injury, meningitis or similar. In a large study of 5183 mechanically ventilated patients in Europe and North America, patients with ALI had the same mortality as patients with no ALI at all. 62

The two RCTs of HFOV17,51 reviewed at the planning stage of OSCAR only recruited patients with ARDS. In the Bollen et al. study,51 a post-hoc analysis revealed that there might have been a treatment effect seen at the more severe end of the spectrum of ARDS. There was no treatment effect seen at the milder end of the ARDS severity spectrum. This suggests any benefit of HFOV would not be seen in patients with ALI.

As ALI is a relatively mild pulmonary insult and does not require artificial ventilation, most patients with this condition are managed on the general wards. In the Europe-wide epidemiological study of ALI and ARDS,3 only 62 out of 6522 ICU admissions (0.9%) had ALI against 6.1% with ARDS. More than half of the patients with ALI rapidly progressed to ARDS, leaving only 0.4% of admissions who had ALI alone. Only two-thirds of these patients with ALI alone were ventilated. By not including ALI patients, we are probably only excluding 0.1–0.2% of all admissions, many of whom will be ventilated for non-pulmonary reasons and could probably not benefit from HFOV. For all of these reasons, we believed it was inappropriate to undertake a study of HFOV that included patients with ALI.

As noted earlier, while the OSCAR study was running, a different set of definitions for ALI and ARDS were developed at a consensus conference. The ‘Berlin Definition’21 will have no effect on the interpretation of the results. The ‘mild’ category is the group of patients previously categorised as ALI, otherwise the definitions are largely unchanged from those first proposed in 1994.

Identification of data to inform the choice of measures used for long-term follow-up

The primary outcome for OSCAR was mortality, as specified in the commissioning brief. The brief also requested longer-term outcome measures. These were included in the OSCAR study, and are reported in this monograph for most, but not all, patients as data collection was still continuing at the monograph submission date.

The HTA commissioned a systematic review into outcome measures for adult critical care in 1998. The results were reported in 2000 both as a monograph and a paper. 63,64 At that time, the three measures of health-related QoL that had been most commonly used in follow-up studies of critically ill patients were the Sickness Impact Profile/Functional Limitations Profile (SIP/FLP), the Perceived Quality of Life (PQOL) and the Nottingham Health Profile (NHP). In addition, the Short Form questionnaire-36 items (SF-36) was increasingly being used.

At the time the review was undertaken, the European Quality of Life-5 Dimensions (EQ-5D) measure was not used in to any extent in critical care research and so did not feature in the reports, even though it was first developed in 1990. 65 However, it has since rapidly gained popularity in critical care research, to the extent that in 2004 a European consensus conference suggested that EQ-5D or SF-36 were the two preferred measures for health-related QoL in survivors of critical illness. 66

The SF-36 is a feasible and reliable instrument with sufficient discriminatory power to detect changes in the health-related QoL of ICU patients with different levels of chronic health and varied severity of their acute illness. 67 SF-36 contains 36 items to measure eight QoL domains: physical functioning, role limitations due to physical problems, bodily pain, general health perceptions, energy/vitality, social functioning, role limitations due to emotional problems, and mental health. 68

European Quality of Life-5 Dimensions is also a general health-related QoL measure that has also proven to be a useful tool in a mixed critical care population. 69 The EQ-5D comprises two parts: the EQ-5D self-classifier, a self-reported description of health problems according to a 5-dimensional classification (i.e. mobility, self-care, usual activities, pain/discomfort and anxiety/depression); and the European Quality of Life Visual Analogue Scale (EQ-VAS), a self-rated health status using a visual analogue scale, similar to a thermometer, to record perceptions of participants own current overall health. The scale is graduated from 0 (the worst imaginable health state) to 100 (the best imaginable state). 65

As part of the background work for the Intensive Care Outcome Network (ICON) studies (studies of long-term ICU survival and QoL being run by the OSCAR Chief Investigator), a systematic review of all of the ICU outcome studies that have used either EQ-5D, SF-36 or both was undertaken. Numerically there are more studies that use the SF-36, though there are eight high-quality studies using the EQ-5D. A similar systematic review was published in 2005 but this only identified five of the eight studies using EQ-5D. 70

A direct comparison of the EQ-5D and SF-36 as measures of health-related QoL in ICU survivors was undertaken in Sheffield in 2004. 71 The report came out strongly in favour of the EQ-5D, because it was simpler, had less floor and ceiling effects and so greater discrimination, and, if response rates were poor, follow-up using face-to-face or telephone interviews was easier.

There is only one published cost-effectiveness study of a treatment for ARDS. 72 This was a retrospective study using data from a large, long-term ICU outcome study undertaken in the USA (project SUPPORT). The treatment studied was artificial ventilation. Utilities were estimated using time-trade off questions and costs were from a hospital perspective.

The decision to use EQ-5D in the OSCAR study was made on a number of grounds. The EQ-5D serves both as a measure of health-related QoL and as a utility measure for calculating quality-adjusted life-years (QALYs). There is a large (3400) reference population database available, and the ICON study generated data on a population of mixed UK ICU survivors at the same time as the OSCAR trial was running, giving two appropriate reference populations. There is a large 11-centre study of survivors of ARDS planned in Baltimore, USA73 which will use EQ-5D as an outcome measure, allowing transatlantic comparisons.

We had originally planned to use formal, laboratory pulmonary function tests to determine residual respiratory dysfunction in survivors. However, two high-quality studies10,11 suggest this might not be cost-effective. The studies followed survivors of ARDS for up to 2 years. 10,11 At both 1 and 2 years, spirometry and lung volumes were normal. There was a reduction in carbon monoxide diffusing capacity (D_LCO) compared with normal values, but, from the HTA review of outcome measures, this test is known to have poor measurement properties. 64 The 6-minute walking distance was also reduced compared with predicted values at both 1 and 2 years, but the patients attributed this to muscle weakness rather than cardio-pulmonary problems. The best measure of respiratory dysfunction was the physical problem domain of the SF-36. Thus if formal pulmonary function testing were to be used as an outcome measure, unless one of the treatments caused additional harm, spirometry and lung volumes would show no difference between groups (a ceiling effect). D_LCO is probably not a valid measure of lung function after ARDS, and the 6-minute walk acts as a surrogate measure for muscle wasting. Thus the probability of distinguishing between treatment groups is very small, and, given both the burden to patients and the cost of transporting patients to pulmonary function laboratories, using laboratory pulmonary function tests as an outcome measure was abandoned.

High-frequency oscillatory ventilation (experimental group)

In the original protocol, we had stipulated that the intervention would be HFOV, delivered using a SensorMedics 3100B ventilator ventilating at a rate of 5–15 Hz (breaths/second). When the study was designed this was the only commercially available high-frequency oscillatory ventilator for adult patients with CE marking marketed in the UK. Prior to the start of the study (early 2007), we became aware that a second CE-marked HFOV ventilator was due to come onto the UK market in May 2007. This ventilator is manufactured in Japan by Metran Co. Ltd (Kawaguchi, Saitama Prefecture, Japan) and marketed as the ‘R-100’ in the Asia-Pacific region. It is imported into Europe and rebadged as the ‘Novalung Vision Alpha’ by Novalung GmbH (Heilbronn, Germany) and distributed in the UK by Inspiration Healthcare Ltd (Leicester, UK). The SensorMedics 3100B was first released in 1993 and has a long history of use worldwide, but it uses analogue electronics, does not incorporate a conventional ventilator, and is noisier than the Vision Alpha device. The Vision Alpha is a digital device, which incorporates a conventional ventilator, but at the start of the study was little used outside Japan. The SensorMedics 3100B uses an electromechanical actuator to generate the oscillation, and, by modifying the driving signal, the ratio of inspiratory to expiratory time (I : E ratio) can be varied. The Vision Alpha uses a rotating mechanical valve and compressed gas to generate the oscillation and has a fixed 1 : 1 I : E ratio. Both ventilators have a frequency range of 5–15 Hz. After careful discussion, the investigators agreed that the study should proceed using the Vision Alpha ventilator, primarily because the transition from conventional ventilation to oscillatory ventilation and back was far simpler and the control interface was more intuitive which was felt to be important as many of the study sites would not have used HFOV prior to the study. The HTA was informed of the decision in June 2007.

The management of artificial ventilation with HFOV was based on two simple algorithms illustrated graphically in Figure 2, designed to allow arterial oxygen and carbon dioxide tensions (PaO₂ and PaCO₂) to be managed separately. The algorithms were derived from guidelines which had been used successfully at Addenbrooke’s Hospital ICU (Cambridge) and the University Hospitals Birmingham ICU for 5 years. The algorithms are virtually identical to the protocols used in the two published RCTs. 17,51

FIGURE 2.

The algorithms for managing HFOV. ABG, arterial blood gas analysis; BP, blood pressure; ETT, endotracheal tube; PAW¯, mean airway pressure; SpR, specialist registrar.

The algorithm for maintaining arterial oxygenation specified starting at a fractional concentration of inspired oxygen (FiO₂) of 1.0, a frequency of 10 Hz, a mean airway pressure 5 cmH₂O above the plateau or equivalent pressure on conventional ventilation, and a cycle volume of 100 ml (the cycle volume is the volume of gas displaced by the diaphragm, the volume reaching the alveoli is far less due to tubing expansion, gas compression and gas inertia). The starting bias flow was 20 l/minute. The target PaO₂ was 8 kPa or greater. As the patient improved, the inspired oxygen was gradually reduced to 0.4 and then the mean airway pressure to 24 cmH₂O. A ‘recruitment’ strategy was used for patients who did not demonstrate the expected improvement in P : F ratio after starting HFOV. When the patient had been stable on a FiO₂ of 0.4 or less with a PaO₂ in excess of 8 kPa, the patient was converted to conventional ventilation and weaned in the usual way. It is not possible to wean patients from artificial ventilation while on HFOV as the ventilator has no facility for spontaneous or patient-triggered breaths.

The algorithm for controlling arterial carbon dioxide tension (PaCO₂) on HFOV involved a target arterial pH of 7.25–7.40, corresponding to a modest respiratory acidosis (mild permissive hypercapnia) in patients with normal metabolic acid-base status. To reduce PaCO₂, the cycle volume was increased up to the maximum for the frequency (on the Vision Alpha ventilator cycle volume increases with decreasing frequency) and if that was insufficient the frequency was reduced. If the frequency was 5 Hz and the cycle volume maximised but the PaCO₂ was still out of range, the on-call clinician was contacted for advice. Early experience with the ventilator identified a small number of patients in whom the PaCO₂ increased rapidly (> 4 kPa rise) after starting on HFOV, and so the algorithm was modified to include a series of faster settings changes in these patients.

Training using the high-frequency oscillatory ventilation

In the original protocol it was anticipated that training would be carried out in the Netherlands and Germany. Before the study started it became clear this would not be practical. An amendment to the protocol was submitted (see Appendix 1, List 2: Substantial amendment 1, No. 5) which allowed training to be carried out in the UK (see Training in use of the ventilator and algorithms).

Troubleshooting with the high-frequency oscillatory ventilation ventilator

Troubleshooting notes prepared specifically for the study were supplied with the abbreviated operating manual. Clinical assistance was available from one of the principal investigators and technical assistance was available from Inspiration Healthcare by telephone at any time.

Supplying the Vision Alpha ventilators and associated disposables to collaborating centres

Collaborating ICUs were supplied with one Vision Alpha high-frequency oscillatory ventilator and humidifier. Inspiration Healthcare supplied service replacements within 48 hours if a ventilator or humidifier failed.

Centres were specifically told that the ventilator:

• was not to be used until the trial office notified them that the appropriate national and local approvals were in place

• was not to be used for treating patients outside of the OSCAR trial

• would be removed from their ICU if there was evidence of violation of its use.

The Vision Alpha ventilators require a disposable ‘tubing set’ for each patient consisting of heated inspiratory and expiratory hoses, oscillator diaphragm assembly, the ‘wet’ assembly for the humidifier, pressure monitoring tubes and expiratory filter. Each patient also required a non-disposable sterile inspiratory valve, used to isolate the oscillator diaphragm from the breathing circuit during conventional ventilation. Each centre kept a stock of disposable tubing sets and sterile inspiratory valves. Used valves were returned to the study office and then sent to Novalung for cleaning and ethylene oxide sterilising. In 2009, the ventilators were retro-fitted with a ‘vent-protect’ heated filter assembly which prevented contact between the patient’s expiratory gases and the inspiratory valve and removed the requirement to sterilise the valves.

Training in use of the ventilator and algorithms

As HFOV had not been used previously in most of the ICUs in the study, a robust training/mentoring system was needed. Experience in neonatal trials where HFOV was introduced into special care baby units that had not previously used the technique suggested a major investment in training was required. 43,49

Before the study started, the clinicians in these units were trained to operate the HFOV ventilator and follow the treatment algorithms. Training was offered in various forms to suit the collaborating unit. During the first year of the study, a 2-day workshop-based course on the HFOV ventilator and how to manage ventilated patients was organised at each centre. In addition, follow-up ‘drop-in’ sessions were offered, usually on the ICU. These shorter sessions were to suit busy units and allow staff to dip in when possible during clinical shifts to top up their skills.

This training was to be backed up with centralised training programs run in Birmingham and Oxford for staff from each study site, targeting the ICU consultant medical staff, senior nursing staff and the local research nurses co-ordinating the OSCAR trial. These used the teaching suites equipped with patient simulators [Laerdal ‘SimMan^®’ (Laerdal Medical Limited, Orpington, UK)] available at Birmingham and Oxford. The trial had a full-time, clinically trained research fellow in the team for the first year to lead and organise the training both centrally and locally, and two half-time senior nurse trainers for the next 3 years of the study. In addition, a member of the team regularly travelled to collaborating centres to support the use of the HFOV. Inspiration Healthcare also offered local training based on the need at individual centres.

Teaching material was prepared at the trial office and distributed to the ICUs taking part, both electronically and on paper. A newsletter was used to share any problems and solutions related to HFOV. We had planned to use a website to distribute information, but decided targeted e-mails with the information were more effective. The trial website held only contact details, details of regulatory approvals and the protocol.

Medical Devices Regulations 2002

As the trial employed a medical device (the Vision Alpha ventilator) for a purpose for which it was CE marked, approval from the competent authority [the Medicines and Healthcare Regulatory Authority (MHRA)] was not required.

Conventional ventilation (control group)

The control group received conventional positive-pressure ventilation using conventional pressure-controlled artificial ventilation.

Clinical management of patients in the control group (conventional ventilation)

When implementing the control intervention in the OSCAR trial, we suggested, but did not mandate, that the conventional ventilation strategy be based on limited tidal volume, pressure-controlled artificial ventilation using tidal volumes of 6–8 ml/kg, ideal body weight and fixed PEEP/FiO₂ combinations as used in the ARDSnet study,74 the only ventilator mode for patients with ARDS with proven benefit. Most ICUs used pressure-controlled or pressure-supported ventilation modes. Two ICUs used airway pressure release ventilation (pressure-controlled ventilation with a very long inspiratory time and a very short expiratory time) on some of the control patients.

Proposed duration of treatment and weaning

The patients continued on HFOV until they had recovered sufficiently to be weaned from artificial ventilation when their FiO₂ was 0.4 or less, their mean airway pressure was 24 cmH₂O or less, and the local clinician was satisfied that there were no non-pulmonary impediments to weaning. The HFOV ventilation mode does not allow any form of spontaneous (patient-triggered) ventilation which is normally required for weaning, so at this point the patients were placed back on conventional ventilation and weaned according to local protocols using inspiratory pressure support (Figure 3).

FIGURE 3.

Patient treatment and weaning flow chart.

The point at which patients could be fully weaned from conventional artificial ventilation depended on a large number of factors that could not be protocolised.

Setting

In the protocol, we planned to recruit patients in 12 ICUs in the NHS in the UK, which were able to care for Level 3 patients as defined by the Department of Health’s Comprehensive Critical Care: A Review of Adult Critical Care Services. 75 To achieve adequate recruitment, additional centres were added so a total of 30 ICUs had taken part by the end of the OSCAR trial.

Intensive care units

In the protocol, the timeline for the project was approximately 57 months. The anticipated start date for the project was 1 June 2007 and the completion date was 29 February 2012. In this timeline, it was anticipated that the first 6 months of the study would be allocated to ethic approvals, design and production of trial material and on-site training for the staff using ventilators.

Initially, 12 centres were recruited to the study: 10 were allocated the Vision Alpha ventilators which had been leased and two sites had their own ventilators. The John Radcliffe Hospital was the first site to recruit in December 2007. The remaining 11 sites were due to start shortly after this date. However, there were delays as the final Multicentre Research Ethics Committee (MREC) (ethics) approval was not obtained until October 2007, despite submitting to ethics in July 2007. Also, there was a delay in the delivery of the ventilators from Japan and Germany and therefore the ventilators were not available for a December 2007 start.

Recruitment was lower than anticipated and it was recognised very early on in the trial (May 2008) that the required number of patients would not be achieved in the original timescale and that an extension would be required. An application for this was submitted to the HTA. The decision was to award the funding, conditional on a HTA recruitment review visit which was scheduled for 18 December 2008.

Following the visit, a 13-month extension approval was confirmed in June 2009, moving the completion date of the project to the end of March 2013. Ten additional ventilators were purchased and this allowed an additional 10 new sites to collaborate (see Table 3). Allowing for the lead in time for arrival of the ventilators (12 weeks), it was estimated that the start date for recruitment for these new sites would be 1 October 2009. However, owing to a delay in arrival of the ventilators, setting up of new sites was not possible until end/beginning of 2009/2010. There were a further five sites which had their own ventilators and were approached and agreed to recruit patients into OSCAR.

In 2009/2010, some reallocation of the ventilators occurred:

1. Aberdeen Royal Infirmary closed for recruitment when the local Principal Investigator left in May 2009 and the ventilator was relocated to Stirling Royal Infirmary.

2. Royal United Hospital in Bath closed in June 2009 with the ventilator relocated firstly to the John Radcliffe Hospital to support HFOV training and then relocated to Queen Alexandra Hospital in Portsmouth once the new sites had all been set up.

3. University Hospital of Wales closed for recruitment in July 2010 owing to lack of recruitment and the ventilator from this site was sent to the highest recruiting site at the time, Queen Elizabeth Hospital in Birmingham, for use as a second ventilator.

In December 2010, the leasing period for the first 10 ventilators was coming to an end. Negotiations with Novalung led to the lease being extended to 3 March 2011, at no additional cost to the study. Also, there were enough additional funds to extend the lease on seven of these ventilators up to end of recruitment (August 2011). Thus, in the beginning of 2011, it was planned that four of the lowest recruiting sites would be closed so the lease could be extended for the remaining sites.

By the end of August 2011, 637 of the target of 802 patients had been recruited into the trial. It was decided to approach 17 sites (as indicated in Table 3) to see if they were prepared to continue recruiting patients into OSCAR until the end of July 2012 (when the last 12-month follow-up was due for the 637 patients). The sites were seven with their own ventilators and 10 which had a purchased (study) ventilator. A substantial amendment to the protocol was submitted to the MREC, as some documentation needed updating because follow-up might not be complete for all patients and this was approved (see Appendix 1, List 6: substantial amendment 5).

Inclusion criteria for centres

An ICU was considered for collaboration in the trial if it met the following criteria:

1. The number of annual admissions to the ICU suggested patients with ARDS occurred sufficiently frequently (amended from the original protocol as stated in Appendix 1, List 2: substantial amendment 1, No. 1).

2. The ICU had a history of collaborating in research and staff members were keen to be involved.

3. All consultants in the ICU had ‘substantial uncertainty’ about the use of HFOV generally and were prepared to enter patients into a trial comparing HFOV with conventional treatment for patients with ARDS.

4. Consultants were willing to attend HFOV training.

5. The PI was prepared to negotiate the release of all other appropriate staff for HFOV training.

Inclusion criteria for patients

Our aim was to recruit adults (age ≥ 16 years old) admitted to an ICU with ARDS who were predicted to require artificial ventilation for 48 hours or greater. ARDS was defined using the American-European Consensus Committee definition of a P : F ratio of < 26.7 kPa51 from two arterial blood gas analyses 12 hours apart. We exclude patients who weighed < 35 kg as the ventilators were not CE approved for treatment of patients below this weight. Patients with obstructive lung pathologies and conditions in which HFOV might be hazardous were excluded.

Patients were therefore eligible for the trial if they met the following inclusion criteria:

1. age ≥ 16 years

2. weight ≥ 35 kg

3. was receiving artificial ventilation via an endotracheal or tracheostomy tube

4. had acute hypoxaemic respiratory failure as defined by:

   1. – lowest recorded P : F ratio measured between onset of artificial ventilation and time of screening of ≤ 26.7 kPa with a PEEP ≥ 5 cmH₂O

   2. – bilateral infiltrates on chest radiograph

5. was not likely to be extubated by the following evening (predicted by attending clinician)

6. had been mechanically ventilated for < 7 consecutive days (≤ 168 hours) at the point of randomisation.

Exclusion criteria for patients prior to trial entry

Patients who were likely not to benefit from HFOV included the following:

1. Patients with left atrial hypertension from any cause, diagnosed clinically or with echocardiography or pulmonary artery catheterisation.

2. Patients who had been mechanically ventilated for more than 7 consecutive days at the point of enrolment.

3. Patients with moderate or severe airway disease expected to cause expiratory airflow limitation.

4. Patients who would have had a lung biopsy or resection during this hospital admission.

5. Patients with any other condition the clinician believed would make receiving HFOV hazardous.

Administrative, practical and ethical exclusions:

1. Patients who had previously enrolled in the OSCAR trial.

2. Patients (or their representative) who refused consent.

3. Patients (or their representative) who did not understand written or verbal information and for whom no interpreter was available.

4. Patients who had been enrolled in another therapeutic trial in the 30 days prior to randomisation.

5. Patients in whom active treatment had been withdrawn or withdrawal was planned.

Where a patient met one of the exclusion criteria, screening was stopped. Most patients were expected to be unable to give informed consent when recruited so patient representatives were used to provide an opinion/assent.

Conventional positive-pressure ventilation (control group)

Patients randomised to the control group received conventional positive-pressure ventilation using conventional pressure-controlled artificial ventilation (as detailed in Chapter 2).

High-frequency oscillatory ventilation

Patients randomised to the experimental group received (artificial) HFOV delivered using a Vision Alpha ventilator (as detailed earlier).

Patient flow

Figure 4 illustrates the flow of patients (screened and randomised) in the OSCAR trial.

FIGURE 4.

The flow of patients (screened and randomised) in the OSCAR trial.

Patient consent

Patients were almost invariably unable to give informed consent owing to alterations in conscious level caused by their illness and therapeutic sedation. As a result, assent from personal or professional consultees (a relative or a nominated health-care professional) was obtained in line with the legal requirements in England and Wales (Mental Capacity Act 200576), and in Scotland [Adults with Incapacity (Scotland) Act 200077]. See Appendix 2 for informed consent process, information and forms.

If a patient or their representative [‘Consultee’ (personal or nominated professional), in England and Wales; ‘Welfare Guardian/Nearest Relative’ in Scotland] refused assent, the patient received usual treatment as defined by the clinician responsible for their care.

Checking eligibility and ventilator availability

Prior to entry, the appropriate items on the eligibility checklist were ticked and the patient was assessed for suitability for entry into the OSCAR study.

Prior to randomisation, it was necessary to check the availability of the HFOV ventilator, in case it was in use treating another trial patient.

Pre randomisation

Patients who satisfied the eligibility criteria had their baseline data collected prior to randomisation.

The procedures for the data collected prior to randomisation are listed in Table 4. The use of antimicrobial drugs and sedatives was recorded. As an aide-memoire and because antimicrobial drugs are sometimes used for purposes other than treating infections (e.g. erythromycin used as a prokinetic agent), a table of antimicrobials was included in the case report form (CRF). Similarly, some drugs are variously used for their sedative or analgesic properties (e.g. morphine) so an explanatory table was included. These are given in Box 1 and Table 5.

The randomisation system

Randomisation was carried out by the Health Services Research Unit in Aberdeen. Randomisation was telephone based and was continuously available. Random allocation was generated by an independent programmer in an equal number assigned to each intervention.

Method of randomisation

We used stratified block randomisation. The randomisation was stratified by site, age of the patient (≤ 55 years and > 55 years) and P : F ratio (≤ 15 kPa and > 15 kPa). The block lengths were random so that prediction of intervention allocation could not be made.

Patients not in the trial

Brief details of patients initially eligible for the trial but not entered into the trial was recorded on a ‘Why not in trial’ log at each collaborating unit to monitor for bias in case selection and provide full reporting in accordance to the CONSORT statement. 78

Immediately post randomisation

Hospital admission date and time and reason for admission were recorded following randomisation. Data on the initial ventilation and the patient’s height (to allow calculation of ideal body weight) were also recorded.

For patients randomised to receive HFOV, the date and time the patient was connected onto the HFOV ventilator was recorded. In addition, the date/time the first FiO₂ was reduced and the cause of the acute hypoxaemic respiratory failure were recorded.

First 24 hours in intensive care unit

The ICU date and time of entry and the details of the patient’s past medical history required to calculate the risk of in-hospital death using the Acute Physiology and Chronic Health Evaluation II (APACHE II) algorithm79 were recorded using definitions slightly modified from those used by the ICNARC Case Mix Programme, a UK-wide ICU audit programme (Box 2). The patient’s primary condition was also recorded using the ICNARC coding method convention. 80

Clinical data were also recorded to calculate the APACHE II score. The procedure for recording the data is listed in Box 3, which is an updated version of the definitions used in the original APACHE II study. 79

Daily data in intensive care unit

Following ICU admission, data were collected every morning, starting at 8 a.m. (or as near as possible) the day after randomisation.

The data collected on all patients were details of organ support, antimicrobial use, sedative use, muscle relaxant use, nitric oxide use and information on prone positioning. For patients on conventional ventilation, exhaled minute volume, total respiratory rate (and hence tidal volume), PEEP and plateau pressures, blood gases and inspired oxygen were recorded. For the patients on HFOV, the frequency, mean airway pressure, measured amplitude, cycle volume, bias flow, completed hours on HFOV and the use of a cuff leak were recorded, along with arterial blood gases and inspired oxygen.

Transition from high-frequency oscillatory ventilation to conventional ventilation for weaning

As noted earlier, there is no facility for spontaneous breaths while a patient is receiving HFOV. As weaning from artificial ventilation back to spontaneous (unassisted) breathing involves a gradual removal of artificial ventilator support with a concomitant increase in spontaneous breathing, this could not be undertaken on the HFOV ventilator. When a patient randomised to HFOV had improved sufficiently (i.e. FiO₂ was ≤ 0.4 or less and the mean airway pressure was ≤ 25 cmH₂O) the patient was switched to conventional ventilation, initially using the conventional ventilation mode on the Vision Alpha ventilator for up to 48 hours and then using the ICU’s normal ventilators. If, during the first 48 hours on conventional ventilation (either Vision Alpha or standard conventional ventilator), the consultant had decided that the patient might benefit from further high-frequency ventilation, the patient was placed back on HFOV and this was recorded accordingly. Figure 3 summarises the patient treatment and weaning flow chart.

The point at which patients could be fully weaned from conventional artificial ventilation depended on a large number of factors that could not be protocolised. The assessments made whilst weaning a patient off HFOV are illustrated in Figure 3.

Hospital stay and after hospital discharge

Patients either died on the ICU or survived and were transferred to a hospital ward. The date, time and vital status at ICU discharge were recorded. No data were collected while in the hospital ward, except the hospital discharge date.

Patients were ‘flagged’ on the Office for National Statistics (ONS) database to ensure reliable collection of the survival data. Lists of survivors to hospital discharge were sent to the ONS regularly where two checks were carried out. The first was list cleaning, which maximised the chances of identifying individual patients on the ONS databases. The second check was to reveal any patients who have died after hospital discharge to ensure follow-up questionnaires were not posted out to deceased patients. For some patients it was necessary to contact the patient’s general practitioner to obtain the patient’s vital status.

Follow-up assessments

This section describes the methods used for patient follow-up to 1 year after randomisation, to obtain data on health-related QoL and health-care costs. As the trial recruitment period was extended into the period planned for follow-up, the follow-up data are incomplete at the time of writing, and so only a limited number of results are reported in Chapter 4.

Patients were followed up at 6 and 12 months after randomisation using self-completed postal questionnaires. Return of questionnaires was tracked carefully by the OSCAR trial office. Patients who died after hospital discharge but prior to the mailing were identified from the ONS returns and removed from the mailing list. The questionnaire included standard instruments to measure health-related QoL and calculate utility indices for the first year after ICU discharge [Short Form questionnaire-12 items (SF-12) and EQ-5D, see Chapter 1]. Questions on social- and health-service use were included to allow a health economic analysis. Freepost envelopes were provided to maximise the response rate.

A letter was sent out to survivors 2 weeks ahead of the first follow-up questionnaire at 6 months and the second questionnaire at 12 months. This letter was added in as a substantial amendment to the protocol (see Appendix 1, List 5: substantial amendment 4) as a reminder to the participant that they would be receiving a questionnaire related to their health in the near future, in an attempt to maximise the response rate. If the questionnaire was not returned a month after it was posted, the vital status of the patient was rechecked using ONS and a reminder letter and another questionnaire were posted. This reminder letter was added in as a substantial amendment to the protocol (see Appendix 1, List 3: substantial amendment 2).

A carer’s questionnaire was also included as part of the follow-up package. This was added in as a substantial amendment to the protocol (see Appendix 1, List 2: substantial amendment 1, No. 9), as part of the study involved learning about the carers and the financial impact on them as a result of providing care to someone who has been treated on an ICU. Collecting the data on carers allowed an insight into the cost of the interventions not just to the NHS but to the wider society.

Primary outcome

The primary outcome was all-cause mortality 30 days after randomisation. This allows comparison with previous studies and is the FDA’s preferred ICU mortality outcome. Most deaths in observational/interventional studies of ARDS occur within 30 days.

Secondary outcomes

Serious adverse events

The reporting guidelines from the National Research Ethics Service (NRES) for safety reporting in research other than clinical trials of investigational medicinal products were used for serious adverse events (SAEs).

A SAE was considered as an untoward and unexpected occurrence that a research participant experienced which: (i) resulted in death; (ii) was life-threatening; (iii) required hospitalisation or prolongation of existing hospitalisation; (iv) resulted in persistent or significant disability or incapacity; (v) consisted of a congenital anomaly or birth defect.

A SAE was recorded on the appropriate trial form by the clinician caring for the patient and reported immediately to the Principal Investigator at that centre. The Principal Investigator then reported the SAE to the Chief Investigator of the OSCAR trial. The Chief Investigator gave an opinion on whether or not the event was:

• ‘related’ (resulted from administration of any of the research procedures), and

• ‘unexpected’ (the type of event was not listed in the protocol as an expected occurrence).

A confirmed, related SAE was submitted to a MREC (England or Scotland) within 15 days of the Chief Investigator becoming aware of the event using the NRES report of SAE form. Events were also reported to the Data Monitoring and Ethics Committee (DMEC).

Expected events

Most known events related to artificial ventilation should occur equally in both groups. Exceptions that were noted that might occur more frequently in the HFOV group were:

1. air trapping

2. secondary effects of air trapping such as reduced carbon dioxide clearance, and impaired venous return to the thorax.

Revised recruitment target

From 1 December 2007 (start of recruitment) to 1 April 2009, the observed rate of recruitment had averaged to 1.01 per centre per month, approximately half of what was expected and thus the trial was severely underrecruiting.

After reviewing various strategies of increasing recruitment with experts in this research field, the DMEC and the Trial Steering Committee (TSC), it was agreed that the effect size based on the clinical relevant difference should be revisited and closely assessed. In addition, the assumptions around control group mortality should be checked by the study statistician reviewing the control group 30-day mortality. It was agreed that this information should be shared with the trial statistician and the HTA only. The sample size was recalculated using a 10% absolute change in mortality, with 80% power and 5% significance level and the control group mortality, giving a total of 802 patients (with 401 per treatment group), allowing for a 3% withdrawal rate. This recalculated sample size was discussed with clinical members of the TSC, who felt this represented a reasonable compromise between an achievable and clinically credible improvement in mortality, and the need to reduce the sample size and hence costs.

Final statistical analyses

The Data Monitoring and Ethics Committee

The DMEC comprised a senior statistician, a senior clinician and a senior triallist (chairperson); see Appendix 4 for membership details.

Standard operating procedures for the DMEC were:

1. During the period of recruitment into the study, interim analyses of the proportion of patients alive at 30 days and analyses of deaths from all causes at 30 days were supplied, in strict confidence, to the chairperson of the DMEC, along with any other analyses that the committee requested.

2. In the light of these analyses, the DMEC advised the chairperson of the Steering Committee if, in their view, the randomised comparisons provided both (i) ‘proof beyond reasonable doubt’ that for all, or some, the treatment was clearly indicated or clearly contra-indicated and (ii) evidence that might reasonably be expected to materially influence future patient management.

3. Data relating to the safety of patients were reviewed by the Chair of the DMEC. The data reviewed were specifically related to:

   1. – procedure related ‘serious, unanticipated adverse events’ (death or serious disability)

   2. – procedure related adverse events/complications

   3. – deaths at 30 days (any cause).

The DMEC met to review progress at 1 year and when a formal interim analysis was due.

The Trial Steering Committee

The trial was guided by a group of respected and experienced critical care personnel and triallists as well as a public and patient representative. Face-to-face meetings were held at regular intervals determined by need but not less than once a year.

Standard operating procedures for the TSC were:

The TSC, in the development of this protocol and throughout the trial, was responsible for:

• major decisions such as a need to change the protocol for any reason

• monitoring and supervising the progress of the trial

• reviewing relevant information from other sources

• considering recommendations from the DMEC

• informing and advising on all aspects of the trial.

Project management groups’ responsibilities

This group was made up of the investigators on the grant application to the HTA programme plus the OSCAR co-ordinating team (see Appendix 3 for membership details). They were responsible for:

• monitoring the progress of the trial and discussing project milestones

• reviewing centre and patient recruitment to the trial

• discussing day-to-day management issues that arise.

Collaborators’ responsibilities

Co-ordination within each participating hospital was through a local collaborator who:

• complied with the protocol at all times

• discussed the trial with medical and nursing staff who treated ICU patients and ensured that they remained aware of the state of the current knowledge, the trial and its procedures (posters and other ‘reminders’ were provided by the trial office)

• ensured that patients in the ICU were considered promptly for the trial

• ensured that the trial CRFs and consent forms were completed in full

• ensured the trial was conducted in accordance with the Research Governance Framework and Good Clinical Practice (GCP) and fulfilled all national and local regulatory requirements

• allowed access to source data for audit and verification.

Co-ordinating centre responsibilities

The trial was co-ordinated by the Intensive Care Society (ICS) Trials Group based at the Kadoorie Centre for Critical Care Research and Education at the John Radcliffe Hospital in Oxford. Administrative support was supplied by the Nuffield Department of Anaesthetics, University of Oxford. Assistance with trial management and the statistical elements of the trial was supplied by the Clinical Trials Unit at Warwick University. Health economic support and analysis was provided by the Academic Unit of Health Economics at the University of Leeds. Randomisation services were provided by the University of Aberdeen.

The co-ordinating centre had the responsibility to:

• assist and facilitate the setting up of centres wishing to collaborate

• organise training in the use of the Vision Alpha high-frequency oscillatory ventilator

• provide study materials and organise a 24-hour randomisation service

• respond to any questions from collaborators about the trial

• give collaborators regular information about the progress of the study

• monitor the collection of data, process and seek missing data

• assure data security and quality

• organise any interim and main analyses

• organise TSC, DMEC and collaborators meetings

• carry out the trial according to the Research Governance Framework and GCP.

Summary of CONSORT flow chart

Figure 6 illustrates the CONSORT diagram78 of the OSCAR trial.

FIGURE 6.

The CONSORT flow chart for the OSCAR trial. ‘Not meeting all inclusion criteria’ are reasons D + E in Table 12; ‘meeting all inclusion criteria but ventilator not available’ are reasons A + B + C in Table 12; ‘meeting all the inclusion criteria but also met one of more of the exclusion criteria’ are reasons F + G + H in Table 12 and ‘meeting all the inclusion criteria but excluded due to the other reasons’ are reasons I to Z in Table 12.

In brief:

• Two thousand seven hundred and sixty-nine patients were screened for the OSCAR trial.

• Of these, 1974 (71.4% of those screened) did not meet the inclusion criteria or met the inclusion criteria and were excluded for another reason.

• The total number randomised was 795 (28.7% of those screened).

• Of these, one patient remained in ICU and six patients were in hospital when the database was closed for the analysis for this report (14 October 2012).

• Of the patients randomised, three withdrew from follow-up (one who had received HFOV and two who had received conventional ventilation). The withdrawal rate was 0.3% of those randomised (the anticipated withdrawal rate used for the sample size calculation was 3%).

• The number of patients who did not receive the randomised intervention was very similar between the two intervention groups.

• Follow-up data have been collected on 215 patients (64% of those approached for follow-up) at 6 months and on 179 patients (60% of those approached for follow-up) at 12 months (as of database closure on 14 October 2012). Follow-up of the remaining patients continues but the results will not be included in this report.

• All randomised patients had the primary outcome data recorded.

Plots of daily percentage of patients on each type of support

The daily counts of number patients for all percentage plots are shown in Table 39.

Mortality rate at first discharge from intensive care unit/intensive care unit length of stay

The number (and percentage) of patients who died in ICU or were alive at discharge from ICU are summarised in Table 41 by each intervention. Note that there are 794 patients in total with ICU data, as one patient was still in an ICU when the database was analysed.

No statistically significant difference in mortality rates at first discharge from ICU was found between the two interventions (chi-squared test: p = 0.5664). The difference in mortality rates between the two interventions was 2.02% (95% CI –4.85% to 8.86%).

For the survival analysis, patients who died up to first discharge from ICU were uncensored, and all others (including the one patient in ICU) were uncensored. No statistically significant difference in mortality rates at first discharge from ICU was found between the two interventions when adjusting for centre, sex, P : F ratio and APACHE II score (logistic regression: p-value 0.5400). The odds of being alive (as opposed to dying) at first discharge from ICU were 1.102 (95% CI 0.807 to 1.505) when on conventional ventilation compared with HFOV.

Table 42 details the summary statistic for the length of stay in ICU (from randomisation to first discharge). No significant difference was found in the length of stay in ICU between the interventions. Using a linear regression model and adjusting for centre, sex, P : F ratio and APACHE II score, the mean [standard error (SE)] estimates were conventions: 15.8 (1.1) days; HFOV 17.5 (1.1) days and a difference of 1.8 (1.2) days: p-value for difference = 0.1314.

Figure 8 illustrates the Kaplan–Meier curve for the probability of survival in ICU against the number of days in ICU (with the number of patients at risk). The event here was mortality from randomisation to first discharge from ICU. Thus all patients who did not die in ICU and were discharged alive are censored (as their death date was beyond that of discharge from ICU date).

FIGURE 8.

Kaplan–Meier curves displaying the probability of survival in ICU over the time in ICU for both interventions.

From Table 42, it is evident that the number of days in ICU range from 0 to 114 over both intervention groups. From Figure 8, we can see that the probability of survival in ICU appears to increase for patients on HFOV from day 50 (in ICU) to just beyond day 75. However, this is based on a small sample of patients at risk.

Mortality rate at first discharge from hospital/hospital length of stay

The number (and percentage) of patients who died prior to hospital discharge or were alive at hospital discharge are summarised in Table 43 by each intervention. Note that there are 788 patients in total with hospital data – one patient was still in ICU and the other six patients were in hospital and had not been discharged.

No statistically significant difference in mortality rates at first discharge from hospital was found between the two interventions (chi-squared test: p = 0.6187). The difference in mortality rates between the two interventions was 1.77% (95% CI –5.19% to 8.71%).

No statistically significant difference in mortality rates at first discharge from hospital was found between the two interventions when adjusting for centre, sex, P : F ratio and APACHE II score (logistic regression: p-value 0.5276). The odds of being alive (as opposed to dying) at first discharge from ICU were 1.104 (95% CI 0.812 to 1.501) when on conventional ventilation compared with HFOV.

Table 44 details the summary statistics for the length of stay in hospital (from randomisation to first discharge from hospital). No significant difference was found in the length of stay in hospital between the interventions. Using a linear regression model and adjusting for centre, sex, P : F ratio and the APACHE II score, the mean (SE) estimates were conventions: 31.0 (3.0) days; HFOV 32.8 (3.1) days and a difference of 1.9 (3.2) days: p-value for difference = 0.5426.

Figure 9 illustrates the Kaplan–Meier curve for the probability of survival in hospital against the number of days in hospital (with the number of patients at risk).

FIGURE 9.

Kaplan–Meier curves displaying the probability of survival in hospital over the time in hospital for both interventions.

Mortality rate one year after randomisation

The number (and percentage) of patients who were alive/died on year after randomisation are summarised in Table 45 by each intervention. An assumption that all of those who have not reached 12-month follow-up are alive has been made for the survival analysis. The uncensored observations are all of those who died prior to 12 months and the censored are all other subjects.

Allowing for the censoring, no statistically significant difference in mortality rates at 12 months post randomisation were found between the two interventions (log-rank test: p = 0.2419). The difference in mortality rates between the two interventions was 0.38% (95% CI –6.54% to 7.29%).

No statistically significant difference in mortality rates at 12 months post randomisation from hospital was found between the two interventions when adjusting for centre, sex, P : F ratio and APACHE II score (Cox’s regression model: p-value = 0.1781). The odds of being alive (as opposed to dying) at 12 months post randomisation were 1.148 when on conventional ventilation compared with HFOV.

Figure 10 illustrates the Kaplan–Meier curve for the probability of survival up to 12 months post randomisation (with the number of patients at risk).

FIGURE 10.

Kaplan–Meier curves displaying the probability of survival up to 12 months post randomisation for both interventions.

Six-month follow-up

From Table 43, there are 400 patients who are alive at hospital discharge.

From Table 46, there were 15 patients who died from hospital discharge to the 6-month follow-up time point and one patient withdrew from their 6 month follow-up. Also, there are 37 patients who are not due for their 6-month follow-up questionnaire. Thus, 53 patients will not have had their 6-month follow-up.

In total, 347 patients (86.8% of those discharged) were sent the 6-month questionnaire.

In total, we received back 221 questionnaires (63.6% of those sent). Thus, our follow-up rate at 6 months is currently 64%. Of these 221 questionnaires, 6 were returned blank and 215 were returned with data. There are 126 (36.3%) non-responders at 6 months. Table 46 lists the numbers sent the questionnaires and followed up at 6 months by treatment group.

The statistical results from the linear regression models (unadjusted estimates and adjusted for centre, sex, APACHE II score and P : F ratio) are given in Table 47. The outcomes for the 6 months are in Tables 48–50.

Twelve-month follow-up

• Similar to the 6-month flow of patients, the 12-month rates are displayed in Table 51.

• The follow-up rate at 12 months is currently 60%.

Severity of illness

The APACHE II score was used to compute the risk of dying and thus the severity of illness. For the calculation, the diagnostic coefficient was set to 0 (i.e. all the patients were given a respiratory infection diagnosis). A logistic regression model was set up to assess the risk of dying, with the APACHE II score as the independent variable. The cut-off point of a score of 26 for the APACHE II score was chosen, as this divided patients into equal risk groups [patients with a score of < 26 (N = 607) had more than 50% risk of survival, whereas those with a score of ≥ 26 (N = 188) had less than 50% risk of survival]. Thus, those in the latter category were more severely ill. The 50% risk cut off was selected during study design.

There is an indication that the interaction term for the subgroupings of illness and intervention is significant, with patients who are more ill likely to survive (up to 30 days) on the conventional ventilation group and patients who are less ill likely to survive (up to 30 days) on the intervention group (Table 52 and Figure 11).

FIGURE 11.

Subgroup analysis: Kaplan–Meier curve for number of days survived (up to 30 days) with subgroups of severity of illness (n = 795).

The statistical significance of the interaction term (p = 0.0104) indicates the need for caution in interpreting these findings.

The interaction of intervention and the two groups of illness for the number of days in ICU (Table 53 and Figure 12) and number of days in hospital (Table 54 and Figure 13) were not significant.

FIGURE 12.

Subgroup analysis: Kaplan–Meier curve for number of days in ICU with subgroups of severity of illness (n = 794).

FIGURE 13.

Subgroup analysis: Kaplan–Meier curve for number of days up to hospital discharge with subgroups of severity of illness (n = 788).

P : F ratio

The cut-off point for the P : F ratio was based on the median value of 15 kPa (i.e. the value where 50% of the data fall above and below 15 kPa).

There was no indication of an interaction between intervention and the categories of P : F ratio for any of the outcomes (Tables 55–57 and Figures 14–16).

FIGURE 14.

Subgroup analysis: Kaplan–Meier curve for number of days survived (up to 30 days) with subgroups based on P : F ratio (n = 795).

FIGURE 15.

Subgroup analysis: Kaplan–Meier curve for number of days in ICU with subgroups based on P : F ratio (n = 795).

FIGURE 16.

Subgroup analysis: Kaplan–Meier curve for number of days in hospital with subgroups based on P : F ratio (n = 788).

Causes of hypoxaemic respiratory failure

The two categories for the cause of hypoxaemic respiratory failure were pulmonary and extrapulmonary.

There was no indication of an interaction between intervention and the categories of hypoxaemic respiratory failure for any of the outcomes (Tables 58–60 and Figures 17–19).