The UK resuscitative endovascular balloon occlusion of the aorta in trauma patients with life-threatening torso haemorrhage: the (UK-REBOA) multicentre RCT

Inclusion criteria

Patients were eligible if they met the following criteria:

1. aged, or believed to be aged, 16 years or older

2. with confirmed or suspected life-threatening torso haemorrhage

3. which was thought to be amenable to adjunctive treatment with REBOA.

These criteria were chosen because they reflect the global assessment that expert clinicians intuitively perform when evaluating severely injured patients, and the pressured clinical setting in which this research has to be conducted.

Exclusion criteria

Women known or thought to be pregnant at presentation and patients with who that were deemed clinically unsurvivable were excluded.

Primary outcome

The primary clinical outcome was 90-day mortality (defined as death within 90 days of injury, before or after discharge from hospital). This outcome was intended to capture any potential late harmful effects of REBOA.

The primary economic outcome was lifetime incremental cost per quality-adjusted life-year (QALY) gained, modelled over a lifetime horizon, from a health and personal social services perspective.

Secondary outcome

Secondary clinical outcomes included 3-, 6- and 24-hour mortality, in-hospital mortality, 6- month mortality, length of stay (in hospital and ICU), 24-hour blood product use, need for haemorrhage control procedure (operation or angioembolisation), time to commencement of haemorrhage control procedure, complications/safety data and functional outcome [measured using the extended Glasgow Outcome Scale (GOS-E) at discharge].

Secondary economic outcomes included 6-month costs from a health service and personal social services perspective, as well as quality of life [measured using EuroQol Group’s 5-dimension health status 5-level questionnaire (EQ-5D-5L)] at 6 months; and incremental cost per QALY gained at 6 months.

To note, the clinical outcomes were chosen prior to the publication of a core outcome set for patients undergoing REBOA,34 and prior to the publication of recommendations regarding the choice of outcomes for haemorrhage control trials. 35

Trauma Audit and Research Network National Trauma Registry data

Data on the treatment of trauma patients are routinely collected by TARN, the national trauma registry for England, to which all MTCs are required to submit data. TARN collects demographic, injury, treatment and outcome data [including the GOS-E, and – through a third-party provider – patient-reported outcome measures, including EQ-5D-5L]. Data collected by TARN directly are reported to be very complete and of high quality. 36

NHS digital data

In addition to drawing on TARN data, the trial also linked to NHS England’s Hospital Episode Statistics (HES) data to obtain information on hospital resource use and to Office of National Statistics (ONS) data for medium-term (6-month) mortality.

Mortality

Survival status and, where applicable, date and time of death were recorded in both the TARN and ONS data. However, in order to minimise delays in reporting, we also obtained death data directly from sites.

EuroQol Group’s 5-dimension health status 5-level questionnaire

EuroQol Group’s 5-dimension health status 5-level questionnaire data were also collected. These were initially to be collected directly from the TARN registry. Following the first TARN linkage, it became clear that the EQ-5D-5L results collected by the third-party provider contracted by TARN were incomplete. We therefore asked sites to collect EQ-5D-5L data prior to discharge, and subsequently, at approximately 6 months after randomisation, by telephone.

General rules for statistical analysis

The trial analysis followed a statistical analysis plan (see additional files www.fundingawards.nihr.ac.uk/award/14/199/09; accessed June 2024), which was agreed in advance by the Trial Steering Committee (TSC). The main analysis was based on the intention-to-treat (ITT) principle (i.e. analysed as randomised). There were two planned interim analyses for survival (see Interim analyses) and a final analysis on all outcomes after follow-up was complete. The interim analyses were timed to occur when one-third and then two-thirds of the expected number of patients had been recruited (in line with the recruitment projections) and completed the 90-day follow-up. We wanted to ensure that, should the intervention be deemed beneficial or indeed harmful at an early stage, the number of patients unnecessarily exposed in the trial would be minimised (especially given the concerns raised in one of the previous studies of REBOA of potential for harm). Baseline and follow-up data were summarised using appropriate descriptive statistics and graphical summaries. Treatment effects are presented with 95% credible intervals (CrIs) for the primary and secondary outcomes. Unless stated, all analyses were carried out using Stata 17. 37

Analysis of primary clinical outcome

The number of eligible patients was known to be small, and we therefore adopted a Bayesian inferential framework for this trial, which has been described in detail in another publication. 38 The primary end point was the log odds ratio (OR) of 90-day mortality after MTC treatment with REBOA, compared to MTC treatment alone:

where p_R and p_s are the proportions of patients who died, to 90 days, after SC plus REBOA and SC, respectively.

Bayesian designs permit the inclusion of prior information about δ. The final analysis of the trial used a Bayesian logistic regression with 200,000 iterations allowing for 10,000 iteration burn-in and checking for convergence using autocorrelation and trace plots. We used a range of prior probability distributions, to contextualise the trial’s findings. This approach has been used in a number of recent studies. 39,40 A minimally informative prior was on the log OR of N(0, 1.28²) which rules out extreme ORs, and a non-informative prior on the intercept of N(0, 10²). The enthusiastic priors were obtained through elicitation and are described in Chapter 3. We also present a Kaplan–Meier survival curve.

Analysis of secondary outcomes

Secondary outcomes were also analysed using a Bayesian approach with 200,000 iterations allowing for 10,000 iteration burn-in and checking for convergence using autocorrelation and trace plots. For 3-, 6- and 24-hour mortality, in-hospital mortality, 6-month mortality, need for haemorrhage control procedure and complications/safety, logistic regression was used using the same minimally informative prior as the primary outcome on the log OR and a non-informative prior on the intercept. For length of stay and time to commencement of haemorrhage control procedure, linear regression using non-informative priors was used. GOS-E was analysed using ordered logistic regression and 24-hour blood product use was analysed using negative binomial regression both with non-informative priors.

Sensitivity analysis

Interim analyses

We had planned two interim analyses, after 40 and 80 randomised participants, and a final analysis after the expected maximum of 120 randomised participants. This analysis was based on survival and not mortality.

The stopping rules included:

Harm: Defined by the probability that the 90-day survival OR fell below 1 (i.e. REBOA is harmful) at the first or second interim analysis, was 90% or greater. More formally, our Bayesian futility criterion at each stage was P (δ < 0 | y) ≥ 0.9, where δ is the log OR and y is the observed data.

Success: REBOA would be declared ‘successful’ if the probability that the 90-day survival OR exceeded 1 at the final analysis was 95% or greater, so our Bayesian success criterion was defined as P (δ > 0 | y) ≥ 0.95. Our calculations are based on an estimated control group (standard MTC treatment alone) with a 90-day survival rate of 66.5%. 16

In short, the trial would stop if the posterior probability for harm was 90% or greater, or the posterior probability for benefit was 95% or greater at either interim analysis.

Methods in analysis to handle classifying and analysing protocol non-adherence

We recognised that a number of patients who were randomised to REBOA might not proceed to have full balloon occlusion, for a variety of clinical reasons. These patients are not ‘cross-overs’, but sit on a spectrum of how far a patient progresses down a REBOA-strategy pathway, depending on intercurrent events. There are three main types of intercurrent events:

1. technical failure (inability to achieve arterial access/insert the device)

2. patients improved as a result of other resuscitative measures, and REBOA no longer indicated

3. patients deteriorated, and REBOA no longer possible.

We classified patients, in line with clinical scenarios encountered, as follows:

1. R0 REBOA deemed inappropriate, decided against.

2. R1/C1 Arterial access not attempted (patient improved).

3. R1/C2 Arterial access not attempted (patient deteriorated).

4. R2 Arterial access attempted, but unsuccessful.

5. R3/C1 Arterial access achieved, no balloon insertion (patient improved).

6. R4/C1 Catheter inserted, but balloon not inflated (patient improved).

7. R5 Catheter inserted, balloon inflated.

Classifying patients in this way allowed us to consider the impact of these intercurrent events. 41

We conducted three analyses to accommodate for the intercurrent events, which answered slightly different questions. The first (the main analysis) relates to effectiveness, whereas the second and third relate to efficacy and safety.

QUESTION 1: ‘Does a strategy that includes REBOA (in addition to standard MTC care) reduce the mortality of exsanguinating trauma patients, ignoring all intercurrent events (such as REBOA not being deployed due to clinical improvement, deterioration, or technical failure)?’

This is the ITT analysis, and is relatively straightforward. It is the ‘policy question’ (that healthcare policy-makers want answered) and evaluated the effectiveness (the principal aim of the trial) in a pragmatic fashion. The problem is that, with many patients who were randomised to REBOA not progressing to full occlusion given clinical changes in the patients, the estimate of the treatment effect was conservative (due to potential dilution of treatment effect).

The totality of the REBOA arm tells us what happens in real-life clinical practice, but the interpretation of the results is complex. In order to address the issue of these intercurrent events, we conducted two principal stratum/complier average causal effect (CACE) analyses. These analyses are preferable to a traditional per-protocol analysis, which wastes data and is subject to selection bias. 42 The analysis used a two-staged residual inclusion estimator approach with non-informative priors. For safety, we also did an as-treated analysis using non-informative priors. CACE assumes that the patients in the SC arm, had they been offered REBOA, would have had the same proportion of patients who would not have received REBOA (because of intercurrent events). This is a reasonable assumption, since an equal number of patients in the SC arm would be expected to improve/deteriorate or be difficult to cannulate.

We debated the use of the term ‘compliance’. Although widely established in the statistical/methodological literature, it does not translate well to the circumstances observed in REBOA. Firstly, patients in the UK-REBOA trial were not ‘non-compliant’. Decisions regarding whether REBOA was still indicated, or not, were made by doctors. However, doctors were also not ‘non-compliant’ since the decision not to proceed with insertion was not arbitrary but forced on the provider by intercurrent events.

We believe that better terms to indicate the extent of REBOA treatment received are ‘strategy’ or ‘pathway’. However, since the term ‘compliance’ is established in the CACE analysis literature, we have retained it for the presentation of the CACE analyses.

QUESTION 2: ‘Does a strategy that includes REBOA (in addition to standard MTC care) reduce the mortality of exsanguinating trauma patients; when there is no technical failure, and when patients’ clinical condition did not change (improve or deteriorate)?’

Patients in the non-R5 categories were not excluded from the analysis. CACE analysis simply assumes that there would have been an equal proportion of these patients in the SC arm.

QUESTION 3: ‘Does a strategy that includes REBOA (in addition to standard MTC care) reduce the mortality of exsanguinating trauma patients; when there is no technical failure?’

For the purpose of this analysis, we defined ‘compliance’ (with the caveats regarding the terminology noted above) as patients who were classified as anything other than R2 and ‘non-compliance’ as all patients classified as R2 (arterial access attempted but unsuccessful).

Initial training

We designed a custom intervention implementation and training package, which was delivered as part of the trial site set-up, to facilitate the introduction of REBOA. The aim of the training package was two fold: firstly, to teach REBOA, and secondly, to introduce clinicians to the trial. The instruction was largely based on experience at the Royal London Hospital, as well as the Basic Endovascular Skills for Trauma and Endovascular Skills for Trauma and Resuscitative Surgery courses. Training was initially spread out over 2 days, but after delivering four of the courses, and following feedback from hospitals, we decided to compress the training into a single day. The training was delivered by two senior clinicians, and comprised a small number of didactic tutorials (indications, team organisation, imaging, ethics, post-REBOA management), followed by small group work, focusing on equipment familiarisation, individual skills training and team training. The tutorials were intended to provide background, recognising the diverse clinical backgrounds of the participants. Scenario-based team training in a simulated resuscitation room was utilised to develop decision-making regarding the incorporation of REBOA into standard resuscitative care, as well as the practical process of trial randomisation.

Development of a local service delivery and training framework, for ongoing skill development and training of new staff

Recognising the importance of ongoing and reminder training, we worked with sites to develop a sustainable, local service delivery and training framework. This involved the designation of ‘super-users’ and ‘training leads’ who organised regular refresher training, and initial training for new staff.

Reminder training sessions

The nature of the reminder training session was left to sites, but typically included discussion regarding clinical decision-making, application of the inclusion criteria, ethical considerations, post-REBOA management of patients, as well as simulations using a mannequin (provided by the trial).

Project Management Group

The study was led by CHaRT, a UK Clinical Research Collaboration registered Clinical Trials Unit in HSRU at the University of Aberdeen. The Project Management Group (PMG) consisted of the two co-Chief Investigators (co-CIs), a Senior Trial Manager, a Trial Manager and a Data Coordinator.

Trial Steering Committee

The trial was overseen by an independent TSC, which included a chairperson, a clinician, a statistician and two patient/public representatives. The TSC met at least annually. The TSC adhered to a charter that they agreed and signed at the start of the trial.

Data Monitoring Committee

The trial was monitored by an independent Data Monitoring Committee (DMC) who also oversaw the interim analyses. The DMC met at least annually, and reported to the TSC. The DMC adhered to a charter that they agreed and signed at the start of the trial.

Expected complications

Death and a number of expected complications (including some which result in life-threatening illness, permanent impairment of structure or function, additional medical or surgical intervention, or prolonged hospital stay) were pre-specified outcomes and therefore not reported as serious adverse events (SAEs) or serious adverse device effects (SADEs). Only unexpected SAEs/SADEs were to be reported to the sponsor.

Adverse events related to REBOA

The following AEs could be expected to occur as a result of using REBOA.

• Access-related adverse device effects (ADEs): External haemorrhage at insertion site requiring treatment other than simple pressure, pseudoaneurysm, arteriovenous fistula, dissection of artery, extremity ischaemia, stenosis of artery, distal embolism, air embolism, infection requiring surgical intervention, need for patch angioplasty (surgical repair), need for arterial bypass, need for amputation.

• Other ADEs: Balloon rupture, aortic rupture, side branch cannulation.

Adverse events related to standard treatment

The following AEs could be expected to occur as a result of standard aortic occlusion by means of a thoracotomy or laparotomy:

• AEs related to external thoracic aortic occlusion: Descending thoracic aortic injury, lung injury/bronchopleural fistula, cardiac injury, oesophageal injury, empyema, wound infection requiring surgical intervention, sternal non-union, rib fractures, extremity ischaemia, distal embolism, infection requiring antibiotics only, infection requiring surgical intervention.

• AEs related to external abdominal aortic occlusion: Abdominal aortic injury, wound infection requiring surgical intervention, extremity ischaemia, distal embolism, infection requiring antibiotics only, infection requiring surgical intervention.

Adverse events common to both treatments

• AEs related to impaired organ perfusion: Acute kidney injury requiring renal replacement therapy, mesenteric ischaemia requiring surgical intervention, paraplegia (permanent), paraplegia (temporary), acute respiratory distress syndrome, stroke (embolic or hypoperfusion-related), multiorgan failure.

Adverse event/device effect reporting

The principal investigator (PI) at each site, or their delegated investigator, was responsible for recording and reporting of AEs/ADEs observed during the study period on a trial-specific AE and SAE/SADE eCRF. The PI attempted, if possible, to establish a diagnosis based on the participant’s signs and symptoms. When a diagnosis for the reported signs or symptoms was known, the PI reported the diagnosis as an AE/ADE, rather than reporting the individual symptoms.

Serious adverse event/device effect reporting

All events meeting the study definition of a SAE or SADE were to be entered onto the SAE/SADE eCRF and submitted to the central trial office within 24 hours of the PI becoming aware of the event. The PI at the site was instructed not to wait until all information about the event was available before notifying the trial office of an SAE/SADE. Information not available at the time of the initial report was documented on a follow-up SAE/SADE eCRF. Follow-up information was sought and submitted as it became available. The follow-up information described whether the event had resolved or persisted, if and how it was treated and whether the patient continued on the study or had been withdrawn from treatment. Once received, seriousness, causality and expectedness were confirmed by the Cheif Investigator (CI or delegated clinical lead).

Unanticipated serious adverse device effects

Unanticipated serious adverse device effects (USADEs) were defined as SAEs that were deemed to be related to the study device or any of the research procedures and were unanticipated. USADEs were to be notified to the sponsor and Research Ethics Committee (REC) within 15 days of the trial office becoming aware of the event.

Assessment of seriousness

The PI or designee made an assessment of seriousness. As stated above, death and a number of expected complications (including some that result in life-threatening illness, permanent impairment of structure or function, additional medical or surgical intervention, or prolonged hospital stay) were pre-specified outcomes and were therefore not reported as SAEs/SADEs.

Assessment of causality

The PI or designee was instructed to make an assessment of the causality (i.e. relationship to trial device). Events that were possibly, probably or definitely related to the device were defined and reported as related to the device. Events that were assessed as possibly related or unrelated were defined as not being related. This was determined as follows: (1) Definitely: There was clear evidence to suggest a causal relationship, and other possible contributing factors could be ruled out. (2) Probably: There was evidence to suggest a causal relationship, and the influence of other factors is unlikely. (3) Possibly: There was some evidence to suggest a causal relationship (e.g. the event occurred within a reasonable time after using the device). However, the influence of other factors may have contributed to the event (e.g. the patient’s clinical condition, other concomitant events). (4) Unlikely: There was little evidence to suggest there is a causal relationship (e.g. the event did not occur within a reasonable time after administration of the trial intervention). There was another reasonable explanation for the event (e.g. the patient’s clinical condition, other concomitant treatments). (5) Not related: There was no evidence of any causal relationship. (6) Not assessable: Unable to assess the information available.

Assessment of expectedness

The PI or designee made an assessment of expectedness for each SAE/SADE regardless of the causal relationship to the trial device.

Follow-up procedures

All AEs/ADEs assessed by the PI or designee as possibly, probably or definitely related to the device and all SAEs/SADEs that occurred during this time were to be followed until they were resolved or were clearly determined to be due to a patient’s stable or chronic condition or intercurrent illness(es). The CRF was updated with the date and time of resolution or confirmation that the event was due to the patient’s illness as soon as this information became available.

Recording and reporting of urgent safety measures

If the PI, designee or a member of study staff became aware of information that necessitated an immediate change in study procedure to protect clinical trial participants from any immediate hazard, they were instructed to report the urgent safety measure (USM) immediately to the trial office. The trial office would then report any USM immediately to the sponsor, and liaise with the sponsor and site to implement immediate procedures to eliminate any hazard. The trial office would also report immediately by telephone to the REC that had approved the study and follow this up with an e-mail written notice within 3 days of becoming aware of the USM. The e-mail notice would state the reason for the USM and the plan for further action. The PI or designee was to respond to queries from the trial office immediately to ensure the adherence to these reporting requirements.

Videoconferencing platform

We used the Zoom platform (Zoom Video Communications, San Jose, CA, USA). The process and the group discussions were moderated by the CIs of the UK-REBOA trial.

Framework

We used the Sheffield Elicitation Framework methodology, as described by O’Hagan. 48 We adhered to good practice recommendations for eliciting expert opinion47,48 including preparation of the participants for the elicitation workshop, use of an elicitation protocol approved by a REC, provision of feedback to experts and an opportunity to revise elicited responses. 48

Participants

We invited 20 subject matter experts to participate. All participants were from the UK, to reflect the setting of the trial. Invitees included the grant holders (with the exception of those involved with the design and conduct of the elicitation itself), as well as site PIs. We reasoned that these individuals would have both knowledge of the published evidence for using REBOA, and personal experience. Participants included emergency medicine physicians (n = 12); pre-hospital care doctors (n = 3); surgeons (n = 4) and intensivists (n = 1).

Quantities of interest

The quantities of interest chosen to inform the analysis of the UK-REBOA trial were those specified by the protocol: 90-day mortality (the primary outcome of the trial), 6-hour mortality, in-hospital mortality and 24-hour mortality (secondary outcomes). For each of these time points, we elicited experts’ opinions regarding treatment with REBOA (in addition to SC), and without REBOA (SC alone).

Information provided in preparation for the elicitation

We provided participants in advance with an overview of the elicitation process and the concept of subjective probabilities, as well as an evidence dossier which included reference to known studies of REBOA. We provided no commentary on the studies in the evidence dossier so as not to introduce any bias into the process. The list of included studies is included in the evidence dossier in Appendix 3. We also asked participants to provide us with any other published studies or abstracts of which they were aware. The dossier was distributed by e-mail the week prior to the elicitation.

Phases

On the day, the elicitation exercise was split into seven phases:43

1. Presentation of background information on the UK-REBOA trial. We did not present evidence relating to the intervention at this point to avoid bias by ‘anchoring’ the participants.

2. Introduction to Bayesian principles, focusing on the distinction between probability under frequentist and Bayesian paradigms, and emphasising that Bayesian probability represents the subjective level of uncertainty of an event happening and can vary among individuals.

3. Introduction to quantities of interest and their parameters: lower and upper bounds and most likely value of mortality, at different time points, in patients treated with REBOA (in addition to SC) or SC alone.

4. Elicitation training exercise. We worked an example with the participants, using the same online tool used for the actual elicitation, to increase familiarity with the process.

5. Elicitation, first round: Participants’ beliefs for the quantities of interest were elicited using the online elicitation tool. We calculated prior distributions for each participant’s elicited beliefs, and then graphed and presented deidentified individual responses.

6. Group discussion. Participants were then encouraged to discuss their choices. We emphasised that the purpose of the discussion was not to come to a consensus but rather to calibrate individual opinions, and to resolve any questions relating to process.

7. Elicitation, second round: The second round was designed to allow participants to revise and calibrate their beliefs, and therefore used the same questions as the first. Participants were provided with an individual code, to allow first- and second-round responses to be compared. The results were, once again, presented as deidentified individual responses.

The elicitation was supported by a biostatistician, who explained the concepts and was available throughout the day to answer questions.

Data collection

We created an interactive online graphical tool, based on previous work by Mason et al. ,49 and our own work,43 using R software and the Shiny package. 50,51 The purpose of the tool was to allow participants to use it online, while also following instructions/conversation on Zoom. It was also designed to be user-friendly and intuitive.

Participants were provided with individual log-ins, so that responses could be tracked. For each quantity of interest, participants were first asked to provide their ‘most likely’ (median) estimate for a given quantity of interest, using a slider. Participants were then asked to quantify their certainty by selecting lower and upper plausible values, again using a slider. Once participants had selected their values, they clicked a button which then displayed their selection as a probability density graph. An example screenshot is included in Appendix 4, Figure 17. The choices could then be amended, with corresponding changes to the graphical output.

Individual responses were electronically submitted, and analysed in real time. The results (individual as well as pooled) were then displayed, again using Zoom’s screen-sharing function, for discussion.

Derivation of prior probability distributions

To obtain the expert-elicited prior distribution for the model, we adopted the following strategy. We aggregated individual expert-elicited beta distributions into a single pooled distribution (at each time point for each of the intervention and control groups) considering equal weight for each expert. We then sampled 500,000 observations for each pooled distribution and calculated the OR of the sampled data of the intervention to the control group at a given time point. The log-transformed OR of the distribution was incorporated as the prior distribution in analysis models (see Chapter 2 for details).

Mathematical aggregation of experts’ judgement and parameterising the prior distribution

The following section outlines the approaches used to combine the expert knowledge into a single prior distribution, parameterise the pooled prior distribution for incorporating in a logistic regression model with Bayesian inferential framework in the context of a randomised control trial setting.

We describe here the strategy to aggregate K experts’ judgement of the intervention and control arms using equal pooling and implementation of the algorithm in an RCT setting.

1. Capture individual expert judgement as a beta distribution and obtain the parameters of individual beta distribution for a given scenario (say, 6-hour mortality), that is one each for the intervention (I) and control (C) groups at a given time point

1. Obtain the linear pool of the beta distributions of all experts (f_g), considering equal weight for each expert, for each of the intervention and control group.

2. Sample (n = 500,000) from the corresponding linear pool of the distribution for each of the intervention and control arm.

1. Calculate the OR of mortality for the intervention to control arms.

1. Calculate the logarithm of OR.

1. Summarise the parameters of the distribution of log OR assuming a normal distribution with mean μ and variance (σ²). The derived distribution (with mean and variance as hyperparameters) represents the prior distribution of the regression coefficient (intervention vs. control) of the logistic regression model. The prior distribution is defined as:

Process

In total, the elicitation took 6 hours to complete. We encountered no significant technical difficulties with the videoconferencing platform (such as not having or being unable to use Zoom; disconnections; or loss of video or audio feeds).

Despite a relatively large number of participants and additional observers, we found that moderating the session was straightforward. Furthermore, we found that the group discussions resulted in meaningful deliberation and interaction, without being dominated by a small number of individuals.

We also encountered no major technical difficulties (crashes, inability to submit data, inability to enter data, etc.) with the app, and all participants were able to submit their data.

Prior probability distributions

As expected, there was a convergence of elicited distributions in round two. The derived prior probability distributions are summarised in Table 3 and were used in the treatment effect estimation models in Chapter 6. The prior distributions are presented visually in Appendix 4, Figures 18–25.

Sampling and recruitment

Recruitment in Phase 1 was purposive and targeted staff who had a role in the recruitment or randomisation of a patient in the first six sites to randomise a patient into the UK-REBOA trial. Site staff across active centres were sent an e-mail invitation (on behalf of the Clinical Co-CI) to participate in the interview study along with a participant information leaflet (PIL) and asked to contact the process evaluation team if interested. On contact with the process evaluation team, staff were provided with the opportunity to discuss the qualitative interview study further and book a mutually convenient time for a telephone interview. Two attempts were made to engage with potential participants. Sampling was informed by the key principles of information power, because the aim of the process evaluation was focused, the sample was specific (site staff involved in recruitment), rich narratives were provided during the interviews and no cross-case analysis was conducted. 55

Data collection

Qualitative data were collected through telephone interviews conducted with site staff who were recruiting patients to the UK-REBOA trial. A topic guide (developed by the process evaluation team and Co-CIs) was used to direct questions and aimed to elicit site staff’s thoughts, comments, involvement and experience with the trial. Interviews took place between April and June 2018. Interviews were conducted by two members of the research team and were audio-recorded and transcribed verbatim.

Data analysis

The approach to analysis was systematic and interpretive, applying an inductive thematic analysis using the Framework approach. 56 One researcher re-read the interview transcripts and generated codes in NVivo (QSR International, Warrington, UK) (used to facilitate data management and initial coding)57 which described relevant features of the data prior to collating into themes. Themes summarised the semantic content of interviewee responses and represented salient issues that were articulated by multiple participants. 58 Following review/refinement of themes, a thematic framework was developed by three members of the team which described the content of all themes and provided illustrative quotes to facilitate data analysis. The thematic coding framework was informed by both a priori questions and issues identified as emerging from the data. A double coder checked the themes and accurately described the content of participants’ responses in a sample of interview transcripts. Any coding discrepancies identified during this process were discussed to reach consensus.

Solution development

Themes from the analysis were tabulated and identified as barriers and/or facilitators to trial delivery. The trial PMG met with the researchers who conducted the qualitative work to discuss potential solutions to the issues identified in the interviews. Proposed solutions were considered in relation to acceptability and deliverability.

Design overview

As the process evaluation developed, it was recognised that many of the challenges within the trial were dependent on people’s behaviour, that is clinicians performing actions (such as randomising a patient or delivering the intervention) that may not be part of their routine practice. There is now a growing body of evidence that suggests behavioural science has the potential to add value to exploring and providing solutions for challenges in the conduct of clinical trials. 59 Within Phase 2 of this process evaluation, we applied the Theoretical Domains Framework (TDF) as a method to help inform data collection and analysis. The TDF is an established framework that categorises behaviour into 14 domains that inhibit or enable behaviour (knowledge, skills, social/professional role and identity, beliefs about capabilities, beliefs about consequences, optimism, reinforcement, intentions, goals, memory/attention/decision processes, environmental context and resources, social influences, emotion and behavioural regulation).

Recent studies have highlighted the utility of the TDF to identify behavioural processes in clinical trials where the performance could be improved. 60–63 The TDF was identified as an ideal framework to support components of this process evaluation as it provides an opportunity to examine behaviours which need to change in order to improve the conduct of a trial, and represents the first step in the process of developing behaviour change interventions. 64 Interventions can be developed to address trial process challenges through mapping barriers and facilitators onto behavioural change techniques (BCTs) via established methods in the behavioural science literature. 65 BCTs are defined as the smallest active ingredient of an intervention such as feedback on behaviour or goal setting, and they can be used alone or in combination with other BCTs. 64 We aimed to develop and implement potential solutions (containing BCTs) to minimise the barriers and maximise the facilitators to trial recruitment and intervention delivery identified from interviews with site staff. 64

Sampling and recruitment

Individuals invited to participate in Phase 2 were from sites which had either recruited a number of patients into the trial, experienced notable difficulties with recruitment, had recently randomised a patient to the trial and/or reported a missed opportunity to recruit an eligible patient. Staff in various roles who were involved in recruitment were invited to take part. E-mail invites were distributed as per previous description for Phase 1.

Data collection

Qualitative interviews were conducted via Microsoft Teams. Interviews were conducted by one member of the research team in October 2020. The topic guide was informed by the TDF, focused on recruitment and intervention delivery and the issues previously identified as important in Phase 1 – that is deployment and insertion of the REBOA catheter. The topic guide was developed and refined by two members of the process evaluation team.

Non-participant observation was conducted during the on-site training for a new recruiting centre at site setup. Detailed notes considering critical conduct problems and behaviours related to trial delivery were collected during this session and considered alongside training materials delivered and provided to sites. Trial training and support materials provided to site staff were compiled and coded using the TDF and BCT Taxonomy v1. 66 These were collected to identify areas where the process evaluation team could help to improve trial processes via adaptation of existing training and support materials (see Stage B subsection under Data analysis).

Data analysis

Sample characteristics

Forty-nine interview invitations were distributed to eligible site staff. Seventeen participants from eight sites were interviewed across both phases (Phase 1 n = 13, Phase 2 n = 5; one participant was interviewed in both phases), with the majority identifying their role as Trauma Consultants (n = 9, 53%) ( Table 4). One of these participants was interviewed in both Phase 1 and 2 as they provided initial perspectives on early process problems and later experiences of more established trial process problems. Taken together, the interviews lasted an average of 37 minutes, ranging between approximately 22 minutes and 1 hour.

Phase 1 findings: Identifying initial difficulties associated with set-up and activation of trial processes

Seven primary themes were identified across the interviews, which could be further organised into barriers or facilitators of trial delivery. The seven primary themes and whether they were reported as a barrier or a facilitator, or both, are summarised in Table 5. Each of these identified themes will be presented in turn with examples.

Phase 1: Proposed solutions based on interview diagnostics from early adopter sites

Table 6 describes potential solutions, developed in collaboration with the PMG, some of which were implemented immediately, and others were combined into solutions within Phase 2 using a behavioural approach. In addition, findings from this Phase were also shared at an Investigators Meeting of recruiting centres in June 2019. Opportunities for discussion and suggestion of solutions was encouraged.

Phase 2 findings: Using a behavioural approach to explore the barriers and facilitators of REBOA recruitment and intervention delivery

Impact of COVID-19 on recruitment

Recruitment had also been previously paused by sponsor on 18 March 2020 due to COVID-19. Sites were able to reopen to recruitment from July 2020; however, not all sites were able to reopen. The number of patients recruited, by sites, is shown in Appendix 7, Table 23 and Figure 26 show recruitment over time. The three highest recruiting sites were Leeds, Royal London Hospital and Birmingham.

Intention-to-treat-analysis

Of the 46 participants allocated to SC + REBOA strategy, 25 (54%) died within 90 days. Of the 43 SC participants for whom primary outcome data are available (1 participant decided against continued participation), 18 (42%) died (Table 10). Using the minimally informative prior, the OR for 90-day mortality was 1.58 (95% CrI 0.72 to 3.52). The posterior probability of an OR > 1 (i.e. that REBOA was harmful) was 86.9% (Figure 8).

FIGURE 8.

Posterior probability of mortality (OR > 1).

When using the elicited enthusiastic prior, the OR for 90-day mortality was 1.40 (95% CrI 0.66 to 2.96) with a posterior probability of an OR > 1 of 81.0%. The adjusted analyses for covariates (see Appendix 8, Table 24) and centre (see Appendix 8, Table 25) showed similar results.

Survival curves

Figure 9 shows the survival curves. There were more early deaths (within hours) in the SC + REBOA group, but deaths in this group also continued to 10 days.

FIGURE 9.

Survival curves.

Mortality at other time points

Mortality at 3, 6, and 24 hours, in-hospital and at 6 months is also shown in Table 10. For in-hospital and 6 months, the results were of the same order as 90-day mortality. For 3-, 6- and 24-hour mortality, however, there was an increased level of mortality in the SC + REBOA arm compared to SC with the greatest difference at 3 hours – 11 (24%) deaths in SC + REBOA compared with 2 (5%) deaths in SC, OR 4.25 95% CrI (1.33 to 15.99). For the adjusted analysis at other time points, the results were similar (see Appendix 8, Table 26–29). Appendix 8, Table 30 shows the learning curve analysis of excluding the first participant randomised to SC + REBOA from each site.

Cause of death

The causes of death, at different time points, are shown in Table 11. The cause of death was listed as ‘unknown’ when a clinician could not determine the cause of death. Typically, this occurred when a patient died very early, before any imaging or operations could be performed, and/or if the patient had injuries that resulted in bleeding, as well as other major injuries, such as a traumatic brain injury. Furthermore, for some patients, a coroner’s report (and therefore cause of death) was still pending at the time of analysis. (These reports can take months to years to be returned to hospitals.)

Table 11 shows that death due to haemorrhage was more common in the SC + REBOA strategy. This difference is apparent at all points, but is accounted for by deaths that occurred early (within 24 hours), after which time there were no further bleeding-related deaths in the SC + REBOA group. Although there were some early deaths due to traumatic brain injury, the overall number of deaths due to traumatic brain injury was similar in both groups.

Haemorrhage control procedures

Table 12 shows the proportion of participants that underwent operations, operations that involved haemorrhage control and the time from randomisation to commencement of such procedures. There were 14 (30%) participants in SC + REBOA and 19 (43%) in SC that underwent a haemorrhage control procedure (OR 0.60, 95% CrI 0.26 to 1.37). The mean time from admission to haemorrhage control procedure (minutes) was 42 [standard deviation (SD) 121] in SC + REBOA and 28 (SD 41) in SC (MD 14.41 95% CrI –22.80 to 52.20). The majority of the participants had a haemorrhage control laparotomy [7/14 (50%) in SC + REBOA and 12/19 (63%) in SC], with one participant having two haemorrhage control procedures.

Length of stay

Table 12 shows length of ICU and hospital stay for all patients. The table also shows length of stay for only those patients who survived to ICU (i.e. excluding early deaths). For ICU the median time (days) spent for SC + REBOA was 2 (25th percentile 0, 75th percentile 9) and 5 (25th percentile 1, 75th percentile 28) for SC. For length of stay in hospital (days), the median time was 8 (25th percentile 1, 75th percentile 34) in SC + REBOA and 19 (25th percentile 1, 75th percentile 63) in SC. Due to death being a competing event, the competing risk analysis showed, for those in the SC + REBOA arm, a reduced length of stay (subhazard ratio of 0.75, 95% confidence interval 0.42 to 1.34) (see Appendix 8, Figure 27). In order to better account for the competing risk of death, we also calculated hospital-free and ICU-free days, as shown in Table 12.

Blood product use

Table 12 shows blood product and tranexamic acid use in the first 24 hours following injury. Overall, there was little difference between SC + REBOA and SC. For red cell concentrate, the median number of units was 7 (25th percentile 4, 75th percentile 12) in SC + REBOA and 9 (25th percentile 4, 75th percentile 17) in SC.

Functional outcome

For the functional outcome, GOS-E, the mean score was 2 (SD 2) in SC + REBOA and 3 (SD 2) in SC (proportional OR 0.58 95% CrI 0.26 to 1.25) (see Table 12). The GOS-E ranges from one (death) to five (good recovery). Table 12 also shows these categories with majority of participants being under category one [death; 24/46 (52%) in SC + REBOA and 17/43 (39%) in SC].

Complier average causal effect analysis 1

Question: ‘Does a strategy that includes REBOA (in addition to standard MTC care) reduce the mortality of exsanguinating trauma patients; when there is no technical failure, and when patients’ clinical condition did not change (improve or deteriorate)?'

As noted in Chapter 2, CACE ‘compliance’ (with the caveats regarding the terminology noted in Chapter 2) in the SC + REBOA arm was defined as patients who were classified as R5 (catheter inserted, balloon inflated) and ‘non-compliance’ as all others (i.e. any patient in whom the balloon was not inflated, whether due to technical failure or changes in the patient’s condition). In the SC group, the two patients who had REBOA were classified as ‘non-compliance’ and all other patients were regarded as ‘compliers’. Appendix 8, Table 31 shows the baseline characteristics of these groups. As expected, there are differences between ‘compliers’ and ‘non-compliers’. Table 14 shows the results of the CACE analysis, for all mortality time points. This shows that, even when inability to cannulate and changes in patients’ clinical condition are taken into consideration, the use of REBOA was associated with increased odds of mortality.

Complier average causal effect analysis 2

Question: ‘Does a strategy that includes REBOA (in addition to standard MTC care) reduce the mortality of exsanguinating trauma patients; when there is no technical failure?'

For the purpose of this analysis, in the SC + REBOA arm, we defined ‘compliance’ for the CACE analysis (with the caveats regarding the terminology noted in Chapter 2) as patients who were classified as anything other than R2 (arterial access attempted, but unsuccessful) and ‘non-compliance’ as all patients classified as R2. In the SC group, the two patients who had REBOA were classified as ‘non-compliance’ and all other patients were regarded as ‘compliers’. Appendix 8, Table 32 shows the baseline characteristics of these groups. As expected, there are differences between ‘compliers’ and ‘non-compliers’. Table 15 shows the results of the CACE analysis, for all mortality time points which showed that, even when the patients in whom cannulation was not possible is taken into account, the use of REBOA was associated with increased odds of mortality.

As-treated (safety) analysis

The results of the as-treated (safety) analysis are shown in Appendix 8, Tables 33 and 34.

Resource use and costs – index hospitalisation

Resuscitative endovascular balloon occlusion of the aorta refers to the insertion of a balloon, usually through the femoral artery. REBOA intends to obstruct blood flow upon inflation. The pragmatic trial design allowed any REBOA device to be used. It is assumed that there are no additional staff resources required to administer REBOA and that the skills to deliver REBOA would already be available within the MTC team and would be incorporated into their workload.

Typically, in major trauma, a multidisciplinary trauma team will assemble prior to the patient’s arrival and will meet with the ambulance crew. We have therefore developed an assumed staff mix of the trauma team which was combined with Personal Social Services Research Unit (PSSRU) cost per working hour to calculate the cost per hour of trauma team staff. 69 This was added to the cost of overheads of the ED sourced from Public Health Scotland (PHS), based on data from Scottish MTCs. 70 We applied the cost of the trauma team and ED overheads from arrival (randomisation) through to the patient’s transfer to an operating theatre, death or exit from the ED, whichever happens first. The volume of blood products required for transfusion was sourced from TARN data linkage for trial participants, and unit costs were obtained from the literature and uplifted to 2020–1 prices. 71

The complexity of the treatment required for these injuries means that standard NHS reference costs may under estimate the true opportunity cost of treatment, particularly as such patients require the use of multidisciplinary trauma teams, large blood transfusions, multiple operative procedures and lengthy stays in critical care and on hospital wards. We therefore costed individual components of resource and summed these component costs to generate a total cost for the whole initial hospitalisation admission period. Total NHS resource use for the index hospitalisation was obtained from patient-level data in TARN and the key resource use variables for costing include time of arrival, time of ED departure, time of first operation, time of death/discharge, number and type of operative procedures and volume of blood transfusions that were required.

Hospital resource use was reported and costed per unit of activity (hour, minute, day) using national average unit costs reported by PHS. 70 Scottish unit costs were used because they provide a greater level of detail in costs than the published English costing sources and were therefore more appropriate for a component costing approach. These included the direct and indirect costs for the entirety of the participants’ stay in hospital (in theatre, in the ward and ICU). For operative costs, we applied national average unit costs based on the primary specialty in which the procedure falls. Duration of each operative procedure was not available from TARN. We therefore categorised each procedure as likely to be short (up to 2 hours), medium (2–4 hours) or long (4–6 hours). This categorisation was based on clinical expert judgement of the trial CI. While there is inevitably variability in the duration of each operative procedure, the approach allows an allocation of costs that broadly reflects duration and specialty of different procedures.

The cost of time in types 1 (e.g. general ward), 2 (e.g. HDU) and 3 (ICU/critical care) wards were calculated using the duration of stay in each department and the average direct and allocated cost per day sourced from PHS. 70

Details of all unit costs applied for the index hospitalisation are summarised in Table 16.

Resource use and costs – discharge to 6 months’ follow-up

Secondary care contacts and episodes of care that were commenced between the date of discharge from the index hospitalisation through 6 months post randomisation were sourced, where available, through linkage of patient records to the HES database. The HES database provides information on a variety of secondary care contacts including inpatient admitted patient care, critical care admissions, outpatient consultations and use of accident and emergency (A&E) services and includes secondary care rehabilitation service usage. The database provides the Healthcare Resource Groups (HRGs) and speciality codes that enable mapping of each secondary care contact to the appropriate NHS unit cost. Unit costs are obtained from NHS reference costs for 2020–1.

We had pre-planned a sensitivity analysis that would apply NHS reference costs to the index hospitalisation stay based on HES data from NHS Digital. However, this analysis was not able to be conducted because data linkage, where it was possible to identify the index hospitalisation, was only possible for 39/90 (43%) patients randomised to the study. NHS Digital agreed to provide data linkage only for participants who provided patient or consultee consent (56/90) and of those 56 with consent, data linkage was only available for 39 participants. This was despite the process of obtaining consent being approved by ethics and being conducted in accordance with the trial protocol. Of the remaining 51 patients who were not linked, 37 died during their index hospitalisation and so zero costs were imputed. In total, follow-up costs were available for 76 trial participants. Data availability following the NHS Digital data linkage process is summarised in Figure 10.

FIGURE 10.

NHS Digital data linkage.

All costs are reported from a UK NHS perspective in Great British pounds (GBP; year 2020–1). Healthcare resource use is reported descriptively for each arm of the study, as n (%) for categorical data such as number of procedures and mean (SD) for continuous data (such as length of stay). Total per-participant costs (resource use × unit costs) are reported as mean (SD) for each arm of the trial from index hospitalisation to 6-month follow-up. Incremental costs for SC + REBOA compared to SC alone are estimated using Bayesian generalised linear regression models, with non-informative priors. The most appropriate distributional family and link function for cost data was determined to be a gamma family based on a Parks test, with an identity link. The gamma model accounts for the non-normality of cost data (i.e. a small proportion of participants with lengthy hospital stays and very high NHS costs). Regression models were adjusted for age and gender covariates.

Life-year and quality-adjusted life-year outcomes

EuroQol Group’s 5-dimension health status 5-level questionnaire data were available from the TARN data set at two follow-up points. The first is administered through TARN prior to the patient’s discharge from their index hospitalisation for major trauma care. The second is administered as a postal questionnaire through a third-party provider at 6 months post admission. Given that there is no generally accepted valuation set for the EQ-5D-5L, we generated utilities by first cross-walking the raw EQ-5D-5L data to the 3L version and applying UK general population tariffs to generate health state utilities. Due to the severity of injury sustained by participants in this trial, it was assumed that all patients were unconscious at the point of randomisation and were therefore assigned an EQ-5D-3L utility value of −0.402 at baseline. Participants who died during the study follow-up period were assigned a utility value of 0 from the date of death until the end of follow-up. These utility scores were then used to calculate the participant’s QALYs over the observed 6-month period using the area under the curve (AUC) approach. The AUC approach assumes a linear change in utility between the time points measured. Incremental life-years and incremental QALYs were estimated using Bayesian ordinary least squares regression models as the data appear to be normally distributed. Life-year and QALY regressions are adjusted for age and gender covariates.

Costs

Tables 17 and 18 detail the results of the resource use and costs generated for the index hospitalisation using the component costing approach. The main driver of cost is the length of stay in hospital in general and particularly in critical care. Together, total hospital length of stay costs account for approximately 80% of the total index admission costs. Due to the larger number of earlier deaths in the SC + REBOA group, key cost drivers of critical care and hospital length of stay are substantially lower in the SC + REBOA group compared to SC.

Table 19 shows that, over 6-month follow-up, SC + REBOA remains substantially less costly than SC alone, due to the competing risk of mortality. The overall finding was robust to whether models were adjusted for age, sex or ISS score.

Quality-adjusted life-years

EuroQol Group’s 5-dimension data collected within the study are presented descriptively in Figures 11–14, in accordance with EuroQol reporting recommendations. 72 Data were available and are reported for N = 29 (SC + REBOA = 15; SC = 14) survivors at discharge and for N = 20 survivors at 6 months (SC + REBOA = 10; SC = 10). Given imputation of 0 utilities for participants who died, it was possible to derive QALYs for N = 57/90 (63%) participants. Data completeness for EQ-5D-5L was lower than expected. This was driven in part to a large proportion of missing data from TARN’s partner provider at 6 months, but also missed TARN data collection prior to discharge. The available data from TARN were supplemented with additional efforts of the trial office to collect further EQ-5D-5L data through participating sites. Life-year gains (LYGs), utilities and QALYs calculated for each arm of the trial are reported in Table 20.

FIGURE 11.

Proportion of respondents reporting any problems at hospital discharge.

FIGURE 12.

Proportion of respondents reporting any problems at 6 months.

FIGURE 13.

Proportion of respondents reporting severe problems at hospital discharge.

FIGURE 14.

Proportion of respondents reporting severe problems at hospital 6 months.

Model parameters

Patients enter the decision tree phase of the model at the point of randomisation to the trial (i.e. the point of arrival in the ED at hospital). The age and gender characteristics of the cohort were aligned with the baseline trial characteristics, where mean (SD) age was 45.35 (18.00) and the proportion of female was 28/88 (31.8%).

The trial data were used to inform the probability of survival to hospital discharge and further survival to 6 months post randomisation. Intervention costs (including the full-index hospital admission costs) were applied based on ITT costing, using the TARN data described in Chapter 7, while follow-up use of resource usage to 6 months post randomisation was based on the available linked data for survivors from NHS Digital HES. Similarly, EQ-5D utility data collected during the trial, cross-walked from the 5L to 3L and valued using UK general population tariffs, were applied at hospital discharge and 6 months. For the 6-month follow-up data, mortality risk, costs and utilities were assumed to be treatment-specific, capturing any potential effect of REBOA.

The long-term trajectory of patient recovery is uncertain, and it is uncertain whether short-term differences in quality of life would be maintained over a lifetime horizon or converge to being equal among survivors at some future time point. We have therefore taken a conservative approach to modelling the longer-term trajectory of patient recovery. We assume that return to general population quality-of-life norms would converge to the pooled mean utility from the trial in the first model cycle. An alternative assumption, to assume that differences in utilities were accrued indefinitely, albeit based on a small sample are explored in scenario analysis. Scenario analysis also explored the impact of applying treatment pooled costs and utility data for the period between hospital discharge and 6 months, on the grounds that any resource use incurred post discharge would be related to initial injury rather than REBOA.

Long-term outcomes, including mortality, long-term recovery in quality of life (utilities) and ongoing costs related to initial injury were obtained from targeted literature searches with variation across studies tested extensively in scenario analyses. Longer-term utility data were available from two studies. The first study reported utility data for N = 335 trauma survivors in the Netherlands. Mean (SD) utility, based on the Dutch value set was 0.691 (0.299) for patients with an initial ISS ≥ 16 followed up for between 12 and 18 months. 75 These utilities were applied at the end of 18 months. A second, smaller study of N = 56 patients provided 15 years of EQ-5D-3L follow-up data for a cohort of major trauma patients in the Netherlands. Average utility (assumed to be mean), applying the Dutch value set at 15 years to a subgroup initially with ISS > 16 was 0.660. 76 A measure of uncertainty such as standard error (SE) or SD was not reported, so it was assumed that the SD of the sampling distribution was equal to 43% of the mean, applying the same ratio of mean : SD as observed from the 18-month study data. 75 The longer-term data from the literature were consistent with an assumption that most trauma survivors will reach their threshold of recovery by 1–2 years post injury. Beyond 15 years, the minimum of the 15-year utility or general population norms was applied, and general population utility was modelled to reduce over time as the surviving proportion of the cohort age.

For the base-case analysis, long-term mortality risk was assumed to be equal to the UK general population age- and sex-adjusted all-cause mortality (ACM) probability. The assumption is justified on the grounds that excess mortality among trauma patients is mostly expected to occur within the initial hospitalisation period. This is consistent with data observed from other trauma studies, including CRASH-2, where the hazard of mortality reduces dramatically over time. 74 However, it is still feasible to assume that survivors may experience an increased mortality risk later in life due to the long-term sequelae of their initial injury (e.g. due to compromised mobility or related chronic illness). To explore the impact of a potential excess mortality risk on model outcomes, scenario analysis applies an excess mortality risk of 1.5.

Patients who survive major trauma events may be at risk of longer-term disability, lower quality of life and thus would also be expected to consume more healthcare resources compared to the general population average. A prospective cohort study from the Netherlands, among 174 trauma patients with an ISS ≥ 16 found that post-hospital costs, including rehabilitation, were mean (SD), €7770 (€13,640) over 24 months of follow-up, with the highest post-hospital costs incurred for spinal injuries. 77 Inflating from 2017 values to 2021 values and converting to GBP, results in a 6-monthly cycle-specific cost of £1764 [(€7770 * 1.06 * 0.86)/4] per 6-monthly cycle, up to 2 years. 78 Beyond 2 years, excess costs were assumed to be £0 for the remainder of the model time horizon. Scenario analysis explores the impact of assuming £0 excess cost for survivors beyond 6 months and applying the full cycle-specific cost for the duration of the model time horizon.

Base-case analysis (probabilistic)

Table 22 shows the expected value of costs, LYGs and QALYs for each treatment strategy in the base case and for several scenario analyses undertaken. QALYs are reported with and without a 1.7 multiplier for severity weighting.

Substantial cost savings associated with SC + REBOA are due to the higher risk of mortality, occurring earlier in the patient journey compared to SC, meaning that there would be substantial life-year and QALY losses associated with the adoption of SC + REBOA in this setting. Despite the cost savings, the magnitude of QALY loss over a lifetime shows that SC + REBOA is both harmful (lost life-years and QALYs) and would be an inefficient use of scarce healthcare resources, when considering increased valuation of a QALY accrued at the end of life. Parameter uncertainty surrounding base-case results, with QALYs weighted at 1.7, is illustrated on the cost-effectiveness plane in Figure 16.

FIGURE 16.

Scatterplot of the cost-effectiveness plane over a lifetime horizon (QALY weighting = 1.7).

Assuming a threshold value of a QALY = £30,000, but applying the maximum weighting for a QALY (multiplier = 1.7), turquoise dots indicate an inefficient use of resource (71% probability for base-case analysis), whereas dark blue dots indicate that REBOA is the most efficient use of resource (29% probability for the base-case analysis).

Iterations on the cost-effectiveness plane show that REBOA is definitively less costly (probability > 99%), due to the competing risk of mortality, but that it is also substantially less effective in terms of QALYs accrued over a lifetime horizon (probability 91%). The findings are robust to a range of scenario analyses undertaken. The results are also consistent with the findings of the clinical effectiveness.

Registry-enabled design

We relied on data routinely collected by TARN, England’s national trauma registry, to characterise the trial population (although most outcomes were collected directly, using a UK-REBOA trial-specific eCRF). Our intent was to make the trial as simple to run as possible, particularly given that many patients presented out of hours, and to avoid duplication of effort. However, there were a number of downstream effects. Firstly, the time from discharge to submission of data to TARN, by sites, was variable, and often prolonged for patients awaiting injury details from post-mortem examinations. This made the characterisation of the trial population more difficult, especially for interim analyses. There were also more missing data than anticipated, requiring queries to sites. (Fortunately, these data had actually been recorded in patients’ health records, but not been transcribed into TARN.) Some of these issues may relate to the large proportion of patients who died, and sometimes died very early on. Lastly, we discovered that a small number of patients had opted out from all national health data collection (https://digital.nhs.uk/services/national-data-opt-out). These patients have all their TARN records deleted, without a record of such a deletion (which also comes under the opt-out). Fortunately, we were able to obtain the necessary data points directly from sites, under research-without-prior-consent rules.

Working with NHS Digital

The data obtained from NHS Digital (required for the health economic analysis) was of high quality, but it took almost the entire duration of the trial to secure the necessary permissions to use it. Data linkage was not agreed to for participants without either participant or consultee consent, limiting the usefulness of the data for the health economic analysis. This was largely related to the research-without-prior-consent framework, which NHS Digital had only limited familiarity with. Multiple changes in case workers, each requiring additional questions to be answered, further compounded the issue. Future research-without-prior-consent trials relying on NHS Digital data would benefit from early discussion.

Initial error in statistical design

Our initial design parameters contained an error in the formulation of the variance in calculations, resulting in an overestimation of the operating characteristics. Following extensive consultation with the funder, and external reviewers, we relaxed the success threshold, and added informative priors, resulting in acceptable probabilities of declaring success if REBOA had indeed been beneficial. 38 As it turns out, the design error – or the revised design – had no impact, given that REBOA turned out likely to be harmful, and the large size of the effect.

Streamlined framework with limited dataset

Executing clinical trials in patients at imminent risk of dying, particularly when they present out of hours, and with little notice, is extremely difficult. Although the UK’s research delivery framework has many advantages, like all healthcare systems, it is not well suited to support such studies. Our reliance on clinicians to enrol patients and collect some data (albeit minimal) in real-time necessitated a limited dataset, focused on answering the trial’s key question. If we had our time again, we would likely have instigated more comprehensive data collection, to capture data such as changes in blood pressure in response to balloon inflation and other physiological data. However, this would have greatly increased the cost of the trial.

Choice of primary outcome

We have already alluded to the issue of choice of primary outcome, and the change in thinking that has taken place over the past 5 years. Our choice of 90-day mortality was based partly on the critical care literature, and concern that early benefit (prevention of early death) might be associated with late harm (increased mortality as a result of acute respiratory distress syndrome, acute kidney injury, etc.).

Experience with the ‘Pragmatic, Randomized Optimal Platelet and Plasma Ratios’ (PROPPR) trial, which compared two types of transfusion strategies in trauma patients,87,88 and several subsequent trials, has resulted in a better understanding of the impact of the time point at which mortality is evaluated. Deaths from haemorrhage occur early. Later deaths are typically due to causes such as traumatic brain injury or multiple organ failure. Although the latter may occur equally in both arms of the study, and the difference thus remains the same, the baseline mortality rate increases, which makes it more difficult to detect differences. A recent consensus statement – which was the product of a conference that involved the National Institute of Health, the Food and Drug Administration, the Department of Defense and researchers – has therefore advocated for using more proximate mortality endpoints in trauma haemorrhage control trials. 35

However, the above debate did not impact on the UK-REBOA trial, because the signal (of harm) is so strong – even at 90 days post injury. However, in keeping with the assertions above, the effect size is even greater at earlier time points.

Setting

The UK-REBOA trial was a study of the use of REBOA in-hospital. The use of REBOA in-hospital may, however, be ‘too late’, especially since patients can be delayed in arrival at the hospital and, when there, they can often be taken to an operating theatre, for definitive control of haemorrhage, very quickly. Potentially the result may have been different had REBOA been considered in the pre-hospital setting, which at the time of the trial design was not being considered in the UK. The pre-hospital environment differs from the in-hospital setting conceptually; it makes sense to obtain haemorrhage control as early as possible after injury, before large volumes of blood have been lost, and inflammatory sequelae are superimposed onto haemorrhagic shock. The findings of the trial should therefore not be extrapolated to the pre-hospital setting.

Staff who facilitated recruitment and data collection

Aintree University Hospital NHS Foundation Trust: Carol Brooks, Shirley Cooper, Calum Edge, Kate Fenlon, John Fletcher, John Gilbert, Melanie Harrison, Tanya Ingram, Raimundas Lunevicius, Gemma Mullen, John Pilavakis, Olga Rutka, Abdo Sattout (PI), Sarah Stevenson, Lilian Wajero, Timothy Wharton.

Barts Health NHS Trust (Royal London Hospital): Sumira Alberali, Ben Bloom, Noemi Caponi, Richard Carden, Arun Castro, Raine Astin-Chamberlain, Michael Carver, Ben Clarke, Helen Curnoni, Anna Dobbie, Neil Durge, Derek Hicks, Shabana Issa, Tony Joy, Brian Kennedy, Charlotte Lindsay, John Mackenna, John Mackenney, Max Marsden, Somayyeh Mossadegh, Anna Morgan, Dan Nevin, Henry Nnajiuba, Helen Oliver, Nimca Omer, Andrea Rosetto, Jennifer Ross, Claire Rourke, William Rush, Samy Sadek (PI), Joanna Shepherd, Rebecca Stoner, Anthony Thaventhiran, Harriet Tucker, Grace Tunesi, Paul Vulliamy, Simon Walsh, Anne Weaver, Jared Wohlgemut.

Hull University Teaching Hospitals NHS Trust: Sina Askari-jirhandeh, Andrew Blackmore, Biju Cherian, Tom Cowlam (PI), Mark Higson, Matthew Hines, Victoria Jowett, Simon Long, Vicki Martinson, Jeremy Osman, Jack Preece, Ben Rayner, Krystyna Simpson, Augustine Smithies, Chris Srinivasan, Paul Stewart, Elizabeth Stone, Dhruba Tamuli, William Townend, Fraser Young.

Imperial College Healthcare NHS Trust (St Mary’s): Zazyneb Al-Saadi, Chris Aylwin (PI), George Bailey, Colin Bicknell, Niamh Bohnacker, Michela Cicchetti, Jo-Anna Conyngham, Andrea D’Mello, Marwa El-Zanfaly, Sarah Finlay, Dan Frith (PI), Shehan Hettiarachty, Sophie Hunter, Nadia Guidozzi, Oliver Jefferson, Rachel Jennings, Moya Kirmond, Athanasia Kountourgioti, Chris Lambert, Graham Lawton, Sonia Fernandez Lopez, Heather McLachlan, Anu Mitra, Noora Mohamoud, Abby Harper Payne, Alberto Rigoni, Alexander Rolls, Cosmo Scurr, Danny Sharpe, Rhys Thomas, Sophie Turner, Lajeesh Vettikkat, Eyston Vaughan-Huxley, Claire Webster, Isabella Wilkes, Rachel Williams, Mark Wilson, Michael Wilson, Henry Wydell, Louise Young, Karolina Zimmerman.

King’s College Hospital NHS Foundation Trust: Duncan Bew (PI), Roger Bloomer, Emma Clarey, Ellie Corcoran, Hannah Cotton, Adam Czapran, Maria Depante, Clare Donegan, Matt Edwards, Tom Hurst, Kevin O’Reilly, Sian Saha, Malcolm Tunnicliff, Umar Wali, Simone Waitt.

Leeds Teaching Hospitals NHS Trust: Hala Ahmed, Kimberley Atha, Rebecca Blythe, Camilla Boyton, Stephen Bush, Paul Curley, Peter Cutting, Rosie Darwood, Nikki Dewhirst, Sally Docherty, Joanne Fletcher, Karen Flood, Yien Fah Fong, James Forsyth, Thushan Gooneratne, Christopher Hammond, A Hassan, Sarah Higgins, Susannah Howard, Craig Irvine, Harriet Jennings, D.H. Jones, Jonathan Jones, Megan Lawrence, Christina Leddie, James Lenton, Simon McPherson, Nonica Maftei, Andrew Mavor, Sundararaj Manou, Rashmi Menon, Samantha Monkman, Jacqueline Moorhead, Frances Mulley, Ahmed Nassef, Stuart Nuttall, Jai Patel, Emily Pawson, Kate Priestly, Sapna Puppala, Sara Rahma, David Russell, Julian Scott, Laura Scott, David Shaw, Venugopal Shankar, Tim Stansfield (PI), Sally Stone, Adene Thornton, Constatinosis Tingerides, Merane Todd, Max Troxler, Venugopal Shanker, Junaid Sultan, Katie Smith, Tom Wallace, Paul Walker, Emma Watson, Alexander Wilkins, Christina Wood, Janet Woods, Amanda Yang, Shahzadi Zeb.

Newcastle upon Tyne Hospitals NHS Foundation Trust: Dion Arbid, Maite Babio-Galan, Benjamin Brown, Lauren Butler, Matt Cadamy, Tony Calder, Ryan Clark, Stephen Clark, Brian Carroll, Jim Connolly, Peter Coyne, Leigh Dunn, Peter Goode, Nigel Fox, Arti Gulati, Abigail Harrison, Carole Hays, Robert Jarman, Philip Johnstone (PI), Jacques Kerr, Sarah Kirtley, Jenny Lett, Sohom Maitra, James McCaslin, Margaret McNeil, Craig Ord, Rebecca Petini, Sarah Platt, Reuben Saharia, Julia Scott, Bas Sen, Graham Soulsby, Kirsty Tristam, Peter Truran, Jason Urron Kimberley Webster, Tessa Wilkinson, John Williams, John Wright.

North Bristol NHS Trust: Jonathan Benger, Kristina Birch, Julian Blackham, Borislava Borislavova, John Bowditch, Adam Brown, Jules Brown, James Cameron, David Campbell, Ed Carlton, Phil Cowburn, Kate Crewdson, Leilah Dare, Keith Davies, Beverley Faulkner, Emma Gendall, Marianne Gillings, Elizabeth Goff, Scott Grier, Joydeep Grover, Debbie Harris, Timothy Hooper, Vicki Hughes, Dominic Janssen, Richard Jeavons, Claire Jewkes, Kirsten Jones, Ben Jordan, Phillip Kaye, Jason Kendall, Nagaraj Kumar, Simon Laing, David Lockey, Gordon Macfarlane, Alex Manara, Aidan Marsh, Rebecca Maxwell, Stephen Meek, Nicola Morgan, Patrick Morgan, Chris Newell, Delia Parnam-Cope, Simon Odum, Caroline Oliver, Harvey Pynn, Emma Redfern, Stephen Robinson, Kerry Smith, Reston Smith, Susan Smith, Jasmeet Soar, Deanna Stephens, Julian Thompson (PI), Ian Thomas, Matthew Thomas, Rebecca Thorpe, Edward Valentine, Sian Veysey, Ben Walton, Curtis Whittle, Ruth Worner, Paul Younge.

Nottingham University Hospitals NHS Trust: Shafique Ahmad, Lauren Blackburn, Adam Brooks (PI), Kate Brown, Paul Candler, Shalini Chinna, David Clarkson, Craig Douglas, Darren Forward, Ramzi Freij, Angelo La Valle, Chris Lamb, Shane Macsweeney, Chris Mason, Alex Navarro, Rory O’Connor, Matt O’Meara, Bob Winter, Adam Wolverson, Ju Young Um.

Oxford University Hospitals NHS Foundation Trust (Oxford): Chris Acott, Nurul Ali, Susan Anthony, Elaine Armstrong, Tanya Baron, Graham Barker, Neil Barker, Steven Barker, Sally Beer, Charlotte Brown, Rebekah Caseley, Andrew Chadwick, Benjamin Clare-Grey, Rebecca Clark, Sarah Cocker, Jessica Cottrell, Barbara Crowe, Andrea Dale, Colin Davenport, Luciana De Barbosa-Dias, Alberto Sobrino Diaz, Karen Dineen, Iain Edgar, Alexis Espinosa, Aoife Fitzgerald, Maria Garcia, Domonique Georgiou, Ben Griffiths, Conrad Groenwald, Oliver Harris, Jonathan Hughes, Rudi Jarai, Aimee Jeffs, Deon Louw, Joshua Lyons, Rufino Roger Magallano, Grigoris Makris, Sachin Mandalia, Priyadarshini Marathe, Jose Martinez, Syed Masud, Alex Mattin, Maria Fernandez Mendoza, Rocio Fernandez Mendoza, Edward Norris-Cervetto, Alex Novak (PI), John McMaster, Ross Moy, Aman Paul, Elena Perez, Claire Pickering, Victoria Prabhu, Vishakha Prasad, Surabhi Ramsundar, Veronica Sanchez, Ros Simpson, Priyanka Singh, Simon Smith, Enrico Sorrentino, Tim Sparkes, Hannah Thraves, Andrew Wigham, James Winchester.

Sheffield Teaching Hospitals NHS Foundation Trust: Paul Barker, Tom Bircher, Sarah Bird, Trevor Cleveland, Chris Connolly, Ben Cooper, Andreas Crede, Michael Dennison, Ben Edwards, Gordon Fuller, Stephen Goode, John Griffiths, Sherif Hemaya, Anil Hormis, Khurram Iftikhar, Robert Jones, Avril Kuhrt, Daniel Kusuma, Tom Locker, Ben Loryman, George Lye, Edward Mills, Timothy Moll, Christopher Press, Mark Regi, Stuart Reid (PI), Steve Rowe, Karen Robinson, Neil Sambridge, Justin Squires, Neil Strawbridge, Stephen Thomas, Douglas Turner, Erica Wallis, Matthew Wiles, Anna Wilson, Samantha Wilson, Chris Yap.

South Tees Hospital NHS Foundation Trust (Middlesbrough): Kerry Colling, Tracy Ruddick, Chris Smith (PI).

St Georges University Hospitals NHS Foundation Trust: Thomas Breen, Louisa Brown, Diego Campos Ceia, Raj Das, Nadine Davison, Richard Carden, Orla Dooley, Alex Eeles, Kevin Enright, Risolvo Gianluca, Grazia Freni, Sobika Gangeswaran, Will Glazebrook, Mark Haden, Tim Hardiman, Richard Hartopp, Paul Holmes, Hannah Hornsby, Mansoor Husain, Anthony Hudson (PI), Heather Jarman, Jeyanathan Jeysankar, Sarah Jolly, Gabriel Jones, Kamila Kalka, Sarah Krishnamandan, Ana Lisboa, Alvina Lone, Catherine Loughran, William McGuiness, Marco Machado, Ahmed Mahdi, Raul Marques, Leto Malli, Gary Maytham, Phil Moss, Faheem Obaidullah, Teresa Parras, Benjamin Patterson, Alex Pickard, Naomi Pritchard, Michael Putnis, Mehrad Ramzamy, Lakshmi Ratnam, Seyed Renani, Mark Reaveley, Daniel Roberts, Oluwaseun Samuel, Guy Sanders, Ariadna Sanchez, Claire Seel, Narani Sivayoham, Audrey Tan, Lorenzo Tiraboschi, Harriet Tucker, Pallavi Walkar.

University Hospital Birmingham NHS Foundation Trust: Amy Bamford, Colin Bergin, Barry Boland, Matthew Boylan, Ronald Carrera, Amy Clark, Lauren Cooper, Liesl Depsy, Natalie Dooley, Moin Durrani, Karen Ellis, Emma Fellows, Morgan Foster, Stephanie Goundry, Mycroft Halliwell-Ewen, Jennifer Hardy, Samantha Harkett, Damian Keene, Justine Lee, Hadassah Lhlenfeldt, Christopher McGhee, Azam Majeed, Tracy Mason, Ansar Mahmmod (PI), Kalyana Murali, Aoife Neal, Paul Parker, Stephanie Porter, Emma Reeves, Saif Rehman, Angus Selby, Hazel Smith, Elaine Spruce, Jon Vendrew.

University Hospitals Coventry and Warwickshire NHS Trust: Daniel Anotike, James Ashford, Richard Atkins, Chris Bassford, Richard Bowman, Pamela Bremmer, James Chinery, Sasathorn Chutimaworaphan, Laura Conway, Tim Daves, James Davidson, Marie Fogarty, Arun George (PI), Edward Hartley, Duncan Haynes, Michael Helme, Marius Holmes, Aliraza Husain, Karen Jones, Andrew Kelly, Anna Kurek, Lucinda Lacey, Caroline Leech, Thomas Mcarthy, George Madden, Laura May, Katy Moon, Catherine Morgan, Charles Pairaudeau, Mark Pell, Mark Porter, Sathananthan Ratrani, Matt Robbins, Rachel Rose, Magdy Sakr, Robert Simpson, Carly Swann, Chris Turner, Imogen Virgo, Geraldine Ward, Laura Wild, Helen Wilkins, Nichola Williams, Matthew Wyse.

University Hospitals of North Midlands NHS Trust: Felicity Avann, John Beckett, Jon Bingham, Astrophel Castro, James Chinery (PI), David Cooper, Chris Day, Scott Farmery, Richard Fawcett, Kay Finney, Minerva Gellamucho, Richard Hall, Paul Hancock, Joanne Hiden, Ben Hockenhall, Alex James, Matt O’Meara (PI), Mary-Jane Newton, Sam Parry, Ida Ponce, Fatemah Sakinia, Jo Smith, Richard Smith, Resti Varquez.