The head injury transportation straight to Neurosurgery (HITS-NS) randomised trial - a feasibility study

Cluster randomised design and cluster eligibility and randomisation

Consideration was given to number of possible designs for randomising patients to early neurosurgery facilitated by acute hospital bypass. After extensive consultation and previous peer review comments, the investigators identified unit of service cluster randomisation as the most efficient and effective design that would not hamper emergency services in the pursuit of time-based targets. This design would also not overwhelm local NCs. Eligible clusters were ambulance stations within the North East Ambulance Service (NEAS) or the Lancashire and South Cumbria division of the North West Ambulance Service (NWAS), which would regularly attend to patients meeting the eligibility criteria below. There were 74 eligible clusters in total within the two participating ASs. The ScHARR (School of Health and Related Research) trial statistician at the University of Sheffield randomised these ambulance stations to intervention (early neurosurgery) or control (usual care) in a 1 : 1 ratio, and stratified that randomisation using a matched pair design. Each matched pair of cluster ambulance stations was equivalent in terms of AS, distance from neurosurgical centre, distance from nearest acute hospital ED and number of full-time acute ambulance patient transportation vehicles. Each cluster ambulance station remained within its randomised trial arm for the 12 months’ duration of patient recruitment.

Eligibility criteria

These are articulated below for the individual patients recruited to the study from the grant application. These criteria underwent several modifications through substantial ethical amendments, which are consequently described later in the text (see Table 1). The final separate NEAS and NWAS protocols that were approved through substantial amendments – as a merged protocol [prepared at the request of the Data Monitoring and Ethics Committee (DMEC)] – are provided in Appendix 6.

Inclusion criteria

Patients injured nearest an acute general hospital ED (NSAH) but not more than a 1-hour land ambulance journey from a NC and thought to be aged ≥ 16 years, when assessed at scene by ambulance personnel, with both:

• signs of significant TBI, such as a reduced consciousness level (GCS score of < 13) and external signs of head injury, and

• no overt signs of ABC compromise.

Important changes from original eligibility criteria for patients

New trauma bypass protocols came into effect in NEAS and NWAS, just before the HITS-NS study commenced (Table 1). Therefore, the study eligibility criteria had to be amended to fit in with their bypass protocols if confusion and chaos in the ASs were to be avoided. Essentially, compared with the original study criteria, NWAS instituted a lower respiratory rate exclusion criteria (< 10 instead of < 12 breaths per minute) and NEAS allowed level 2 emergency medical technicians to assess patients for bypass and included patients with a higher GCS score (≤ 13 instead of ≤ 12).

Finally, the Trial Steering Group (TSG), prior to patient recruitment commencing, asked that the airway exclusion criteria be extended to include any patient for whom a supraglottic airway had been provided at the scene. These amendments to the original eligibility criteria were all approved in a series of substantial amendments by the study Research Ethics Committee (REC).

Patient identification

Eligible patients were identified both through NEAS and NWAS paramedics in the clusters making direct contact with the study research paramedics – when they had identified a patient as meeting the study inclusion criteria – and through daily screening of the AS patient report forms using electronic databases. The latter was highly resource intensive for research paramedics, as it entailed screening large numbers of patients daily who had been ‘trauma pre-alerts’ to receiving hospitals or for whom the word head injury had been written somewhere on the ambulance patient report form (PRF), and sometimes there were delays in the ambulance stations scanning the paper PRFs on to the service electronic data bases, which could result in delays in patient identification. Once eligible patients had been identified from the screening or direct contact, the research paramedic would contact the hospital in question to confirm the identification details (it is not uncommon for head injury patients with impaired consciousness levels to be solely ‘unknown males/females’ throughout their contact with the AS if there are no relatives/friends at the scene of injury) and determine the patient’s current location (inpatient/died/discharged/transferred to another hospital) in order to facilitate the approach for consent. In this way it was assumed that all eligible patients, whether they had been identified by the cluster trial paramedics or not, would be included in the trial. In the end it was the study research paramedics who determined finally whether or not the patient was eligible for recruitment. The screening standard operating procedure (SOP) developed by the Trial Management Group (TMG) and approved by TSG is given in Appendix 2.

Ethical basis and practicalities of obtaining informed consent from participants whenever possible or action for which fully informed consent was not possible

It was not possible to obtain informed consent for the trial from HITS-NS patients at the scene of injury. Patients did not have capacity owing to the HITS-NS inclusion criteria being a reduced consciousness level. The time frames involved did not allow patients’ next of kin, if available, or the nominated consultee for the AS, sufficient time for consideration. Consent for research in this situation is covered by the MCA Section 32. 9 Hence we obtained ethics approval to enrol patients into the trial at scene with later consent for follow-up and inclusion of data from the North Wales REC 10 (10/WNo03/30), which specialises in ethics approval for studies involving adults who lack capacity. Consent to follow-up and inclusion of data were obtained by the HITS-NS research paramedics – or occasionally by a Comprehensive Local Research Network (CLRN)-funded research nurse in one of the participating hospitals – from patients who recovered capacity. When capacity was not recovered, advice was sought from a consultee who was, in general, either the next of kin or the AS nominated consultee for the MCA as appropriate prior to hospital discharge.

It became clear early on in patient recruitment that significant numbers of study patients had relatively minor injuries and were being discharged from hospital within 24 hours of arrival before they could be approached face to face for consent by research paramedics. With the approval of a REC we amended the protocol and patient information sheet to allow these ‘early discharge’ patients to be written to for consent and to reply by sending back a slip in a stamped addressed envelope as agreement to discuss participation by telephone. The patient information sheet and consent form were also sent in the mailing. On receipt of the slip, the research paramedic telephoned the patient to explain the study and obtain consent, where appropriate; patients were encouraged to then sign and return the consent form. Patients were also able to return the consent form by post without discussion if they felt happy to do so. Further approval was gained to send text reminders when mobile phone details were available for patients to respond. At our request the REC was happy for us to retain anonymised data, including 30-day mortality on all eligible patients, counting those who did not respond to the postal invitations to participate further but had not explicitly refused consent. From the outset, we had REC approval to retain anonymised 30-day mortality data on all patients who died within a week of admission to hospital without approaching distressed relatives.

This proposed pathway for obtaining consent is consistent with the DH guidance in relation to research in the emergency setting and Good Clinical Practice in research on adults who lack capacity. 23 The consent SOP developed by the TMG and approved by TSG is given in Appendix 3.

Setting

The study was conducted in two regional AS NHS Trusts – the Lancashire and South Cumbria trauma network subdivision of the NWAS and the NEAS – covering three neuroscience hospital centres (SNCs) and 11 non-specialist ‘acute’ hospitals with type 1 (accept ‘999’ ambulances and supported by full resuscitation facilities) EDs (NSAHs).

The ASs, SNCs and NSAHs covered by the trial are shown below (Table 2). Each AS covers a mixed urban and rural population. TARN data suggest that the trauma cases, (and their outcomes) received by the hospitals in the proposed regions for this study are similar to those nationally represented on TARN. The participating ASs involved received 0.5 out of 3 million emergency ‘999’ ambulance calls in England (excluding London) in 2006–7 and generated similar performance data to that of the rest of England. 24 At the time of commencing recruitment, the London Ambulance Service was already operating early neurosurgery technology (no outcome data available) as part of their trauma system. Therefore, there was a lesser requirement for the trial population to be representative of that within London. The TMG felt that this evidence suggested that they were representative of the population to which early neurosurgery made available by NSAH bypass would apply.

For the purposes of the study, the trial NCs were also identified as research sites, as patients frequently consented therein; the acute hosptials were designated as participant identification centres, where patients were followed up for consent and after discharge.

Time

The HITS-NS trial was not studying a new patient intervention; the new technology under scrutiny was the timing of neurosurgery in patients injured nearest an acute hospital ED compared with the time to any interventions that may be required to stabilise the injured patient’s ABC. Time zero was the time at which paramedics left the scene of the incident with the injured patient. The time frames were measured identically in all clusters/arms of the trial.

Neurosurgery

Neurosurgery included any craniotomy for evacuation of intracranial haematoma, debridement of open fractures and insertion of ICP monitor. Time to neurosurgery was from time zero to the time that the patient arrived in theatre, for whichever of these procedures came first. It was envisaged that this would occur early (within 4 hours of time zero) in patients presenting to the intervention clusters and later after secondary transfer in the control clusters, but was measured identically in both.

Airway, breathing and circulation stabilisation

The interventions that stabilise the injured patients’ ABC that fall outside the scope of paramedic practice include endotracheal intubation (ETI) facilitated by drugs, decompression of tension pneumothorax (if present) and surgery/interventional radiology to control internal haemorrhage, as dictated by the patient’s injuries and physiological status. It was envisaged that most HITS-NS patients would require ETI; the other interventions being less frequently required. The time to each of these interventions was to be recorded, the time to ABC stabilisation being from time zero to whichever ABC intervention procedure was first commenced. It was measured identically in all clusters/trial arms and it was thought likely, but not necessarily, to be a given that this would occur up to 30 minutes earlier from time zero in the control (usual care) group. Paramedics were trained to exclude patients with signs of imminently requiring these interventions from the study.

Paramedic training

This was delivered to a large number of staff. In NWAS training was delivered to > 350 paramedics at 28 stations, and in NEAS to > 500 paramedics and advanced technicians at 46 stations. A paramedic training strategy was developed by the trial manager, Dr Wanda Russell, and was enhanced by additional material developed by the research paramedics (local trial co-ordinators) to ensure optimal delivery of the strategy within the two ASs. The content of the training covered the following dimensions.

The objectives were:

• to introduce the HITS-NS trial

• to inform paramedics of their role in HITS-NS trial

• to identify sources of key information about the trial and further points of contact regarding any aspect of the conduct of the trial relevant to the role of the paramedic.

The structure of the training programme was as follows:

• ‘Why?’ The background to HITS-NS.

• ‘What?’ The research objectives of the trial.

• ‘How?’ The design of the trial – inclusion criteria, recruitment of patients, sample size, interventions, consent process and trial data.

• ‘Who?’ The roles and responsibilities of the chief investigator and the local principal investigators (PIs), research sites (ASs, neurocentres), participant information centres (PICs), research paramedics/local co-ordinators, trial manager, other collaborators and partners.

• ‘Where?’ Geographical locations.

• ‘When?’ Anticipated launch dates and timeline for the trial.

Initially, training in NEAS was delivered on a divisional basis, with the AS stations in the two NEAS divisions potentially involved in bypass to the James Cook University Hospital (JCUH) being targeted for training first. A more extensive training roll-out then followed; however, the roll-out of the new Major Trauma Bypass (MTB) protocol in NEAS required that HITS-NS training was redesigned to show how the HITS-NS trial would run alongside the MTB protocol, and a revised training package was delivered to all paramedics. Training was launched across all HITS-NS stations at the same time in NWAS, with training strategies, including face-to-face training, cascading down via team leaders, self-directed learning packs and online training via the HITS-NS website. Uptake of training among paramedics at all NEAS and NWAS ambulance stations was monitored and logged using processes appropriate to each region. This involved, for example, notification of completed training by individual paramedics via an e-mail sent to the local trial co-ordinator or by the return of a slip certifying completed training (see Appendix 4 for training materials).

Initially, recruitment in NEAS was launched using a phased approach, commencing in January 2012. There were 46 ambulance stations – with 23 randomised to the intervention arm and 23 to the control arm of the trial – involved in recruitment to HITS-NS, with caution initially being observed about allowing intervention stations to go live until a minimum of 70% paramedics at a given station had returned signed forms stating that they had completed their HITS-NS training. This resulted in a gradual roll-out of stations. However, from 2 April all 46 stations were formally launched alongside the launch of the new trauma bypass protocol in the NEAS region. The new trauma bypass roll-out involved a mail-out of the bypass protocol to all paramedic staff in NEAS, and mandatory HITS-NS training material for paramedics was included in this mail-out. Therefore, the NEAS approach to training sign-off was adopted in that the posting of the mail-out to individual paramedics provided HITS-NS training, which did not require a signed form to confirm completed training. This process, therefore, facilitated the launch of the trial across the region.

Similarly, in NWAS, an initial strategy required that a minimum of 70% of paramedics returned HITS-NS training confirmation to qualify individual stations as ready for launch. This was achieved with face-to-face training in the 28 cluster ambulance stations. However, as in NEAS, training was reinforced with the inclusion of a mandatory training pack in the roll-out of the new trauma bypass protocol, and from this point the launch of all stations was considered complete.

The trauma bypass protocols for NEAS and NWAS have had differences that have impacted on the inclusion/exclusion criteria for HITS-NS, necessitating protocol amendments to cater for both regions, which required submission of substantial amendments to the REC, with approvals being granted.

Promoting the trial and raising awareness around trial recruitment

Much work was done to raise awareness of the HITS-NS trial in the trial neurocentres and acute hospitals. Meetings were held with research nurses, consultants and research and development (R&D) staff at all three neurocentres (in NEAS and NWAS), at which staff from PICs also attended. The purpose of these meetings was to ensure that contacts in PICs and neurocentres could be established to assist in the process of tracking patients who were recruited into HITS-NS. The study was given good support and systems were put into place that allowed the local trial co-ordinators to make contact with designated individuals at PICs/neurocentres who were able to advise about the location and condition of HITS-NS patients. The research nurses received briefing about the consent-taking process in order to assist in this activity.

Trial site files, including delegation logs for completion, were established at all research sites and PICs.

Awareness of the HITS-NS trial was reinforced in the trial neurocentres and acute hospitals prior to the launch of the trauma bypass protocols, and meetings were held with clinical staff to discuss how HITS-NS could be conducted alongside the new trauma bypass protocols. 25 Close liaison continued with colleagues throughout the duration of the trial to ensure, in particular, that patient consenting processes and data collection ran smoothly.

The trial also received coverage by local press in the NEAS region and a press release was issued in NWAS.

A HITS-NS website was developed to inform the public of the trial and also to include comprehensive training reference material accessible to paramedics only (www.hits-ns.tarn.ac.uk).

Promotional materials – including small tins of a variety of sweets and pens, all with HITS-NS logos – were also distributed to paramedics in NWAS and NEAS.

Development of trial standard operating procedures

Standard operating procedures were successfully developed for the recruitment and consent processes, for data collection and management, for stopping the trial early and for the reporting of SAEs at the outset of the trial. These were subsequently reviewed and modified to include recommendations from the TMG and the TSG and based on observations and developments arising from piloting the procedures where relevant.

A comprehensive screening procedure for identifying potential HITS-NS patients and a screening log for identifying eligible patients were also developed as part of the recruitment SOP. The screening process for identifying potential HITS-NS patients also required close collaboration between the research paramedics and staff, in the informatics departments of NEAS and NWAS, to identify possible appropriate and efficient mechanisms involving data downloads of call data and electronic patient report forms. This particular activity required considerable planning and extensive testing (see Appendices 2 and 3).

In relation to data collection and management, the local trial co-ordinators both completed TARN data entry training, and meetings with TARN data co-ordinators took place both in the North East and the North West, as the TARN data co-ordinators were to assist in prioritising data entry on to the TARN database for HITS-NS patients. The data collection and management SOP included processes to allow the local trial co-ordinators to complete data entry for non-TARN patients using the TARN-based CRF, and also for non-TARN data variables for all patients. The trial manager worked with a TARN analyst to set up a data download system for HITS-NS patients, and the data entry system was also thoroughly tested prior to becoming operational (see Appendix 5).

Implementation challenges

It is important to note, given the feasibility nature of the trial, that there were a number of challenges to be managed during the roll-out (and subsequent implementation) of the study. The key challenges experienced can be summarised as follows.

• Paramedic training for recruitment to HITS-NS:

  • This had been delivered to paramedic staff already pressured by a growing volume of new protocols and procedures in their work.

  • Slow uptake due to the dependency on goodwill among paramedics who are restricted by work rotas and unavailability of cover to allow for training during working hours.

  • Setting up effective systems to allow for monitoring training uptake was extremely difficult prior to the launch of mandatory training alongside the roll-out of the MTB protocols in NEAS and NWAS.

  • A confusion had arisen from misunderstood randomisation instructions from the trial statistician, which led to a number of ambulance stations being initially wrongly identified as intervention/control stations, with about 30 paramedics receiving the wrong training – this situation was promptly rectified as soon as this confusion came to light.

• Challenges within the ASs:

  • The HITS-NS trial was conducted alongside a new MTB Protocol launched in April 2012 (which was applied to the whole trauma population attended by ASs). The protocols identify specific trauma patients outside HITS-NS inclusion criteria who were to bypass into the newly designated MTCs. In the study areas, the MTCs were the SNCs.

  • Previously limited involvement in complex research trials.

  • Wide geographical areas included in the study.

• Diversity in research and governance procedures:

  • Individual organisations have different processes.

  • Time delays in obtaining research site approvals/letters of access for the research paramedic and participant identification centre permissions for the conduct of the trial.

• Delays in staff recruitment.

Close collaboration with partners, and the continuous monitoring of these challenges by the research team and the TMG and steering group, allowed strategies and processes to be refined to ensure that the trial could continue in accordance with the study timeline and planned activities.

Clusters recruiting by ambulance station and study arm

Figures 2 and 3 describe the flow of patients through the study, and it is interesting and instructive to compare this with the study planned CONSORT diagram in Appendix 6 (the full protocol).

FIGURE 2.

Cluster randomisation within ASs.

FIGURE 3.

Study CONSORT diagram. GOS, Glasgow Outcome Score. a, Median/IQR number of patients recruited by each cluster.

Using the matched pairs of randomised ambulance station clusters described above (see Cluster randomised design and cluster eligibility and randomisation) (37 matched pairs), eligible ambulance station clusters were randomised to either control or intervention. Forty-six clusters from NEAS and 28 from NWAS were randomised. No clusters were lost to follow-up. Of 46 NEAS clusters, 43 recruited patients, whereas fewer than half (13/28) of the NWAS clusters recruited patients.

In the intervention arm, therefore, 29 (78%) out of a possible 37 clusters recruited patients with a median cluster size of 6 patients (IQR 3–8 patients); in the control arm, 27 (73%) of 37 possible clusters recruited patients with a median cluster size of 3 patients (IQR 1–7 patients).

The matched pairing of clusters prior to randomisation was an attempt to secure equal numbers of patients in the two arms of the study, as the clusters were matched on numbers of ambulance vehicles at the station (‘busyness’) and distances from neuroscience and acute hospitals. However, as exact numbers of ‘HITS-NS eligible’ patients were not known prior to study commencement, this was clearly an estimate. In the end there was a preponderance of patients in the intervention arm (169) compared with those in the control arm (124), whereas, ideally, 146 patients would have been recruited into each arm of the study.

Screening and responses to consent requests in consort

Large numbers of patients (> 81,000 in both ASs; see Figure 2) were screened for eligibility to cover periods where the study co-ordinators were ‘off shift’ and, therefore, unable to take call and text alerts identifying study patients, or when paramedics had either not recognised or not had time to flag up a HITS-NS patient. As described in Chapter 3 (see Ethical basis and practicalities of obtaining informed consent from participants whenever possible or action for which fully informed consent was not possible) one of the unexpected observations early in the study was that a large numbers of eligible patients turned out not to have injuries that were serious enough to detain them in hospital for > 24 hours. This often meant that by the time the screening process had identified them as eligible and their location had been confirmed by the study co-ordinator, they had left hospital.

A substantial amendment to the original protocol (see Appendix 6) gave the investigators permission to approach patients by mailing the information sheets and consent forms (as personal approaches in hospital were not possible), followed by text reminders. The majority of patients did not respond to these mailings and texts; however, the REC allowed retention of anonymised data on these patients (non-responses to consent request) up to 30 days post injury. A small number of patients (six intervention and five control) declined to consent after face-to-face or mailed approaches; three intervention patients had no NHS number (and no registered GP), which made it impossible to meaningfully identify their 30-day mortality outcomes. Six patients (one intervention and five control) had no meaningful identifiable data on their ambulance PRF (not uncommon in unconscious patients arriving at the ED – often described as ‘unknown male’) and could not be further identified using ED records. One identified control patient was discharged early to ‘no fixed abode’ and therefore could not be approached.

Availability of patients for 30-day and 6-month follow-up in consort

As a consequence of the responses to consent requests, 159 out of 169 (94%) eligible intervention patients and 113 out of 124 (91%) patients had data that were sufficient to analyse their 30-day outcomes.

Twenty-four patients died early during the initial hospital admission. A priori REC approval to retain the anonymised data and 30-day outcomes of those who died prior to 7 days post admission without approaching distressed families for consent had been granted. Out of those who formally consented, some patients were not available on the telephone numbers supplied to provide 6-month outcome data. A minor REC amendment was granted, which allowed the outcome questionnaires to be mailed after no telephone contact had occurred: four intervention and two control patients responded to mailed 6-month outcome questionnaires, leaving (from 46 patients who formally consented) seven intervention patients and seven control patients who were not available for follow-up and were thought to be alive. Four patients (three intervention and one control) were known to have died between providing consent and approach for 6-month follow-up. At the time of writing this gave 33 out of 169 (20%) and 24 out of 124 (19%) of intervention and control patients, respectively, who had data that were sufficient to analyse their 6-month outcomes.

Recruitment period

The initial planned recruitment, as per the application, was to appoint the trial manager in July 2010 and commence recruitment in January 2011. In the end – owing to responding to HTA board questions, contract and subcontract negotiations and the time taken to recruit and train trial staff – the trial manager started in post in May 2011. Recruitment started in the NEAS in January 2012. The first 3 months of the study were spent training paramedics in the NEAS and piloting the study procedures of screening, identifying and locating patients and approaches for consent. Four study patients were recruited between 2 January 2012 and 31 March 2012. Recruitment started in full in the NEAS on 1 April 2012. It took slightly longer to get the NWAS study co-ordinator in post, so recruitment did not start in full until mid-April 2012. Recruitment ran in both ASs until 31 March 2013, as planned for the feasibility study. There was no withdrawal of clusters or research sites during this period, although one of the NHS trusts that acted as a PIC withdrew in mid-December 2012 (see Complaints and serious adverse events). This did not impact significantly on recruitment.

Table 3 compares the basic clinical characteristics of the control and intervention patients in an intention-to-treat analysis. The characteristics compared are those of which paramedics would have been aware at the scene of injury. In both groups, approximately two-thirds of patients were male, with the median age in the mid-forties.

The median GCS score was ‘12’ in both trial arms, the median oxygen saturation was 97% and the median systolic blood pressure was 136 mmHg. In either group, the proportion of patients with normal pupillary responses was 95%. The proportion injured by road traffic collision (RTC) and low energy fall was similar in both groups, at 7–8% and 60%, respectively. The estimated distance to NC was similar (26 minutes vs. 28 minutes for control and intervention, respectively). The average difference with 95% confidence limits is given in the fourth column of the table and indicates that there were no significant differences between control and intervention groups in terms of the characteristics described in Table 3.

The left of each ‘control/intervention’ cell in Table 3 contains the number of patients in each group for which a measurement of the row specified variable was recorded. The following statements describe the circumstances surrounding missingness:

• Two control patients who were subsequently not identified by hospital records had no age recorded on the ambulance PRF, although gender had been recorded.

• Two intervention group patients did not have the exact GCS recorded on the PRF; it was merely recorded that the GCS score was < 13/14, thereby fulfilling the study inclusion criteria.

• For patients in either group with missing physiological variables/mechanism of injury, these were not recorded on the PRF.

• Similarly, the estimated time to nearest SNC could not be calculated as no incident postcode was recorded on the PRF for seven intervention patients and 11 control patients.

Table 4 presents an intention-to-treat comparison of the characteristics of patients in both groups in terms of factors that paramedics could not have known precisely of at the scene of injury. The time from leaving the scene to arrival at the first hospital was between 15 and 20 minutes in both groups. The median ISS was ‘1’ in both groups. The proportion of patients with significant extracranial injury was low in both groups, at 4% for intervention and control groups, respectively, as was the proportion of patients subsequently shown to have TBI in both groups (21.6% and 30.7%). Within the TBI subset of each trial arm, fewer than one-third of patients received any neurosurgical intervention in theatre (11.4% and 31.4%): 15 patients in total. A further 23 patients with TBI went to an ICU without neurosurgical interventions in theatre (10 intervention; 13 control). For these TBI ICU patients not requiring theatre, the ICU care was provided in the NCs in 86% of cases (20/23 patients). Four patients with TBI had ICP monitors inserted in intensive care; one enrolled patient, who did not have TBI but spontaneous subarachnoid haemorrhage, also had an ICP monitor inserted in the ICU.

Interventions directed at stabilising the ABC within 6 hours of leaving the scene were needed in fewer than one-fifth of patients in each trial arm (13.6% and 17.7%). As might be expected, a higher proportion of patients were transferred for further care in the control arm (15.8% vs. 4.9%). Transfers for further care occurred in the intervention arm owing to repatriation to NSAH (n = 4) when no TBI was present or to a SNC in three cases of non-compliance in patients with TBI. In this group of three, only one neurosurgical intervention of ICP monitoring occurred. The 30-day mortality was similar in both groups, being close to 9% (9.4% and 8.8%, respectively). There were no significant differences in the characteristics presented in Table 2 other than the injury severity being, on average, two points higher in the control group and – as expected – a higher rate of secondary transfer to neuroscience in this control group.

Time from leaving the scene to arriving in hospital was missing for 24 intervention and 22 control patients as a result of the time of leaving scene not being recorded on the ambulance PRF. There was little difference between the trial arms in this time interval but non-compliance in the intervention group will have influenced this finding.

Injury details (ISS, % extracranial injury, % TBI) were available for all 272 patient when 30-day mortality was available but were unable to be determined in patients who were not identified in hospital (n = 6: 1 intervention and 5 control) and in 4 out of 11 patients who had refused to consent (6 intervention and 5 control), although the patient information form had advised non-consenters that their anonymised injury details could still be retained. Study hospitals were at times reluctant to allow the study co-ordinators access to health records in these circumstances.

The availability of 30-day mortality outcomes has been described in the text following the CONSORT diagram (see Availability of patients for 30-day and 6-month follow-up in consort).

Feasibility outcomes for stream A: recruitment rate versus target

Figure 4 gives the recruitment rate as a percentage of target by month of study during the study recruitment period, by overall study recruitment rate and by recruitment rate in the two ASs. May 2012 to March 2013 is shown as the recruitment period, as the NWAS was not fully recruiting until the end of April 2012, when all the NWAS clusters were trained. The provisional power calculation for the full study suggested that 700 patients would be needed per annum from these research site ASs. However, a feasibility target of 50% (350 patients), with recruitment rising in both sites, was deemed by the TMG, in the grant application, to be an acceptable feasibility target. This gave a monthly recruitment target of 30 patients per month. Hence the 50% target is outlined in black in the figure.

FIGURE 4.

Recruitment rate vs. target for feasibility study.

It had been anticipated by the TMG that (because the NWAS study catchment area was larger in terms of cluster numbers and population served) two-thirds of the recruitment would have occurred in NEAS and one-third would have occured in NWAS, giving a 20 : 10 NEAS/NWAS split in terms of patient numbers for monthly target recruitment of 50%. However, part of the target was that the recruitment rate would be increasing to > 50% towards the end of the feasibility period.

Figure 4 shows that the study was unable to recruit to the 50% target in the majority of months of the study, and that there was no increase in recruitment towards the end of the 12-month recruitment period (light green line). Although recruitment in NEAS was often close to the 50% target and sometimes exceeded it (dark green line), NWAS recruitment was usually about 20% of target (blue line). The higher GCS study inclusion criteria within NEAS need to be borne in mind when interpreting these findings.

Feasibility outcomes for stream A: proportion of study patients with traumatic brain injury on computed tomography head scan

In order to demonstrate the potential effect of early neurosurgery, the feasibility study target in the application was for 80% of enrolled study patients to be subsequently shown to have TBI on a CT scan. Figure 5 shows that the proportion of recruited patients with TBI – in patients whose injury details were known – fell far below this.

FIGURE 5.

Proportion of study patients with TBI vs. study target: all included patients with injury details.

A total of 25% (n = 70; 95% CI 21% to 31%) of 276 patients with known injury details were shown to have a TBI on CT head scan in the overall study sample. The proportion was similar to this in NEAS, at 21% (n = 52; 95% CI 16% to 26%) but significantly higher in NWAS at 55% (n = 18; 95% CI 40% to 70%). However, as mentioned previously, the GCS cut-off for study inclusion in NWAS was one point lower (< 13 vs. < 14) than in NEAS.

As mentioned previously (see Important changes from original eligibility criteria for patients) the application’s study GCS inclusion criteria in NEAS were by necessity changed from the original application (in order to make the study possible alongside the NEAS MTB criteria) from < 13 to < 14. Consequently, an a priori analysis of TBI prevalence, including only those patients who met the original GCS inclusion criteria (GCS score of < 13), was prespecified by the HITS-NS TMG. This is shown in Figure 6.

FIGURE 6.

Proportion of study patients with TBI vs. study target: all included patients with original inclusion criteria of GCS score of < 13.

Figure 6 shows that, in the 180 patients who met the original inclusion criteria, the proportion of patients with TBI was 30% in the overall sample (n = 60; 95% CI 22% to 38%), with the proportion in NEAS being 28% (n = 42; 95% CI 19% to 36%). The NWAS results being guided by the original protocol are unchanged.

Feasibility outcomes for stream A: proportion of study patients when paramedics complied with study allocation

The prespecified rate of compliance with study allocation was 90% in order to demonstrate feasibility. Figure 7 shows that the percentage overall compliance rate was significantly lower than this, at 62% (183/293; 95% CI 57% to 67%). Compliance was significantly higher at 81% (100/124; 95% CI 72% to 91%) in the control group compared with 49% (83/169; 95% CI 41% to 57%) in the intervention group.

FIGURE 7.

Rate of compliance with study allocation overall and by study arm.

Figure 8 breaks this analysis down further by AS. Overall, the NWAS demonstrated excellent compliance with study allocation at 90% (33/37; 95% CI 78% to 95%), the rates being similar (at 91%) in the control group of 20 patients compared with the intervention group of 17 patients for whom compliance was 89%. The picture was somewhat different in the NEAS, with an overall compliance rate of 59% (150/256; 95% CI 50% to 66%), with a significant difference between the control group (for which compliance almost met the study target at 79% of 104 patients) compared with a much lower rate of 45% in the 152 intervention patients.

FIGURE 8.

Rate of compliance with study allocation by study arm and AS.

Feasibility outcomes for stream A: selection bias arising from non-compliance

Table 5 illustrates selection bias arising as a result of non-compliance with study allocation. For the factors of which the paramedics were aware, this occurred only with respect to estimated time to neuroscience centre (ETNC); in the control group, when compliance occurred, the ETNC was on average 2.6 minutes longer (95% CI 1.5 to 12 minutes) than when there was non-compliance; in the intervention group the converse occurred – with compliance the ETNC was 3.5 minutes (95% CI 0.6 to 6.3 minutes), on average, closer than when there was non-compliance.

Table 6 presents the same analysis in terms of factors of which paramedics could not have been directly aware but perhaps may have estimated at the scene of injury. This analysis indicates that in the intervention group compliance was more likely in patients with an ISS that was on average 5.2% higher (95% CI 2.8% to 7.7%) and a TBI prevalence that was on average 23.5% higher (95% CI 1.3% to 35.6%) than when non-compliance occurred. No such selection bias arose as a result of the small amounts of non-compliance in the control group or occurred in terms of subsequent mortality rate.

Prespecified analysis of robustness of screening

In May 2013, 2 weeks after the end of patient enrolment, a search of TARN submissions for TBIs revealed that 323 adults with TBIs presented in the catchment areas of the trial hospitals in the NEAS during the recruitment period and did not appear to have been entered into the study. A search for their ambulance patient report forms by the NEAS study co-ordinator – matching by anonymised information, such as age and date of injury – enabled exclusion of those who had one of the trial NCs as their nearest hospital to the scene of injury. This left 184 patients (Table 7) and revealed that only five could have been eligible for the study but were missed by screening. The reason for non-inclusion of the other patients was ineligibility due to too high a recorded scene GCS score (14 or 15) (n = 62), not having a head injury that was evident to the paramedics at the scene (n = 17), not having being transferred to hospital by a HITS-NS-trained land ambulance crew [helicopter = 18, transfer in = 24, Yorkshire Ambulance Service attended or not HITS-NS trained crew (n = 10) or not having used an ambulance to get to hospital (n = 23)].

Patient satisfaction and nested qualitative cohort

The objectives were:

• to determine the acceptability of the intervention (early neurosurgery) and control (usual care) pathways to patients, families and staff

• to explore service providers’ views of their experiences of their involvement in the HITS-NS trial

• to explore patients’ experiences of taking part in a feasibility study to compare different treatment pathways for patients with head injury.

Patient satisfaction

A patient satisfaction questionnaire was developed for use in the HITS-NS feasibility study to record patient satisfaction relating to the patients’ views of the time immediately after the initial incident and the arrival into hospital only. The initial design was created after completing a literature review on patient satisfaction to find out what components are important for determining patient satisfaction, as it has been shown that this can be used to evaluate the quality of care. The final version was developed after input from the HITS-NS TMG to get feedback and make improvements, and was then piloted with former brain-injured patients. The final version of the questionnaire was administered over the telephone during patient follow-up interviews, alongside administration of the EQ-5D and GOSE, in accordance with the 6-month follow-up SOP developed by the research team. The questionnaire explores 10 dimensions of patient satisfaction represented by statements, and patients were asked to indicate their extent of agreement with each statement by choosing a response from a 5-point Likert scale. The questionnaire proved to be straightforward in its administration and respondents seemed able to provide responses with ease.

Results

Patient follow-up was conducted between November 2012 and November 2013. Although the intention had been to conduct all follow-up interviews at approximately 6 months, this was not always possible because of a number of challenges in other areas of study activity (especially, for example, as a result of the workload challenges faced by the local study co-ordinator for NEAS), and also because of delays in receiving consent and the subsequent difficulties experienced in making contact with consented patients to complete the interviews.

A total of 28 patients and/or their carers took part in follow-up interviews at between 6 months and 17 months after recruitment into the study. Four additional patients who had consented to take part had died before their interviews could take place. Of the 28 patients who were followed up, 22 were able to give responses to the patient satisfaction questionnaire themselves, 3 were able to respond with assistance and additional comments provided by a carer, and for 3 the responses were given by a carer on the patient’s behalf. However, for one patient, who was in a nursing home, the carer was unable to provide answers to the satisfaction questionnaire, as he or she was unaware of the patient’s experiences during the period immediately following admission into hospital and a family member was unavailable to assist. The age range of respondents was between 20 and 92 years, and 19 of the respondents were female. Eleven (40%) of the patients had TBI.

Figure 9 shows the distribution of timings for the interviews.

FIGURE 9.

Timings of follow-up interviews.

Figure 10a–j shows the distributions of responses given to the 10 questions included in the questionnaire.

FIGURE 10a.

Responses to ‘the medical treatment I received was excellent’ (n = 27).

FIGURE 10b.

Responses to ‘I was satisfied with the first hospital where I was treated’ (n = 27).

FIGURE 10c.

Responses to ‘the surroundings and facilities at the hospital were excellent’ (n = 27).

FIGURE 10d.

Responses to ‘I felt well informed of my care throughout my stay in hospital’ (n = 26).

FIGURE 10e.

Responses to ‘the doctors and nurses were polite and friendly’ (n = 27).

FIGURE 10f.

Responses to ‘the doctors and nurses showed their concern for me’ (n = 27).

FIGURE 10g.

Responses to ‘I felt my care was explained to me clearly’ (n = 27).

FIGURE 10h.

Responses to ‘the doctors and nurses answered all my questions’ (n = 27).

FIGURE 10i.

Responses to ‘the information sheets I received were useful and informative’ (n = 26).

FIGURE 10j.

Responses to ‘overall, I was satisfied with the care I received from doctors, nurses and paramedics’ (n = 27).

Paramedic focus group

Four of the eight participating paramedics indicated that they had identified HITS-NS patients and all of the participants confirmed that they had received HITS-NS training in some form.

The questions explored and the key views given (with any direct quotations presented) are as follows.

Summary paramedic focus groups

The themes that emerged from the focus group clearly demonstrated that paramedics considered there to be substantial value in participating in a trial such as HITS-NS from the perspective of determining the most appropriate care for patients, and also for professional development of individual paramedic staff and to promote the AS as an important clinical environment for undertaking research. The HITS-NS trial was well received in general, although one of the main areas of activity that precipitated suggestions for improvement was training. There were no other major issues relating to other aspects of the trial (such as ethics and consent) and its conduct.

Complaints and serious adverse events

There were no patient complaints about the conduct of the study from HITS-NS patients.

One non-study patient complained because he or she was written to in error by the NEAS study paramedic for consent when his or her address was erroneously supplied to the paramedic by the receiving hospital. The complaining patient had the same name as a HITS-NS patient. The chief investigator and the hospital trust in question carried out an investigation and responded in writing to the patient, who appeared satisfied that his or her personal information was safe and had not been disseminated to third parties without consent. To minimise any future similar errors, the need was emphasised to double check that the patient in question was the one admitted on the date in question with a head injury, when two patients with the same name were on the hospital database. There were no further episodes of this nature or patient complaints.

One of the trial acute hospitals – the Royal Lancaster Infirmary – withdrew its participation in the study at the end of December 2012, as it no longer felt able to resuscitate and investigate head injury patients with a reduced GCS score who were arriving by ambulance as a result of local trauma network reconfiguration. Screening of patients injured nearest that hospital continued in NWAS, and the trial co-ordinator felt that one possible patient attended by a control crew might have been eligible.

There were three SAEs during the conduct of the study. All of these occurred when NEAS intervention cluster paramedic crews bypassed patients who were ineligible as their respiratory rate was higher (> 30 breaths per minute) or lower (< 10 breaths per minute) than the range in the study eligibility criteria. All were investigated by the local PI, and it was deemed that no patient harm had resulted from bypassing these ineligible patients.

Long-term secondary stream A outcomes

As indicated in the CONSORT diagram, low numbers of enrolled patients responded to approaches to consent for study follow-up. The values below (Table 8) relate to patients who consented and were available for the 6-month follow-up interviews (13 control patients, 15 intervention patients) and those known to have died (18 intervention patients and 11 control patients were known to be deceased). There are no significant differences, but the low response rate – biased heavily towards those known to be deceased (n = 29) or with severe injury (n = 11) – prevents further meaningful comment.

Objective 1: determine the feasibility of conducting a study of early neurosurgery using a cluster randomised trial of pre-hospital bypass

Stream A has shown that cluster randomised pre-hospital trials, with ambulance station as unit of cluster, can be used for health technology/complex intervention evaluation – compliance issues can be addressed if sufficient resources are dedicated to face-to-face training (see Feasibility outcomes for stream A: proportion of study patients where paramedics complied with study allocation).

However, it is not plausible to evaluate the impact of early neurosurgery using this design because of the unexpected case mix in the study cohort. The low rate (25%) of TBI and 7% rate of neurosurgery in included patients were unexpected but critical findings in determining feasibility. The identification of this case mix in those patients with head injury who are eligible for ‘bypass’ is an important new finding from the study. The effect of early neurosurgery is diluted in this cohort and a trial using these (or any prospective) inclusion criteria would be unfeasibly large (> 15,000 patients).

Objective 2: establish the acceptability of both pathways to patients, families and staff

Only a limited evaluation of this was possible because of the low rate of response to invitation to consent for follow-up after enrolment and the fact that patients who did consent to follow-up were (non-representative as they were) in the more severely injured group (who were in hospital for a sufficient time period to allow face-to-face consent). The qualitative work on a cohort of nine patients can be considered to be exploratory.

Notwithstanding this, there was no clear signal of patient/family preference for either pathway, with high levels of satisfaction expressed for either (see Patient satisfaction and nested qualitative cohort) and equal numbers of control and intervention patients having being interviewed.

The selection bias analysis suggests that paramedic preferences are influenced by proximity of hospital and perceptions about likely severity of injury (see Patient satisfaction and nested qualitative cohort).

Objective 3: estimate the magnitude of effect associated with early neurosurgery

It was not possible to do this given that the study recruited only 20 patients who required neurosurgery and most were in the control group.

Objective 4: determine accuracy of identification of isolated traumatic brain injury

In reality this was the rate of TBI in patients who met the study criteria, as paramedics were not given a ‘free hand’ with which to recruit. This lack of free hand reflects the pragmatic implementation of trauma triage tools within the new NHS England trauma systems. This rate of TBI varies between 23% and 55%, dependent on the GCS inclusion criteria and AS. The study had to vary the GCS inclusion criteria by AS in order to make the trial consistent with the varying MTB criteria that were introduced (during study recruitment) within both ASs in early 2012. Within the original study inclusion criteria of GCS score of < 13, the prevalence of TBI was 30%, with the observed values in the two ASs being 28% and 55% (see Feasibility outcomes for stream A: proportion of study patients with traumatic brain injury on computed tomography head scan).

Objective 7: identify major barriers to conducting a cluster randomised trial of early neurosurgery

The overwhelming major barrier is the low prevalence of TBI and need for neurosurgery (as described above) in those who are eligible for bypass according to current major trauma triage criteria that the HITS-NS study inclusion criteria typify.

The analysis of TBI prevalence (see Feasibility outcomes for stream A: proportion of study patients with traumatic brain injury on computed tomography head scan) indicates that, although the NEAS amended study inclusion criteria to the upper GCS score of < 14 (vs. < 13 in the original study protocol) recruited an additional 108 patients to the study, only an additional six (6%) had TBI, with one requiring neurosurgery but four requiring ICU.

With the TARN data analysis of TBIs, there were 62 patients who presented to HITS-NS NSAHs – in the NEAS area – during the study period, who had a scene GCS score recorded as 14 or 15 compared with 52 who met the study inclusion criteria of GCS score of < 14 (see Prespecified analysis of robustness of screening). Over this period, many thousands of patients with head injury and GCS scores of 14–15 presented to NEAS and did not subsequently have TBI (65,000 were screened by the HITS study co-ordinator), so although it would appear that the majority of patients with TBI occur in this high GCS cohort, the prevalence in this cohort of patients with a stable ABC is probably < 1%. It is not practical to bypass all of these patients to the ED of the NCs – without overwhelming them – for no clear patient benefit in the majority of cases.

A further barrier is that compliance was an issue in NEAS intervention clusters stations, but, in future, studies could be overcome by more resources for face-to-face training to achieve NWAS levels of compliance (where the lower numbers of clusters meant that face-to-face training was possible). The preference for face-to-face training was also confirmed by the paramedic focus group.

The low rate of response to invitation to consent – typical of mild head injury population in other studies – could be overcome by the ‘opt out of consent’ approach that we initially requested32 rather than ‘patients opt in’ which the REC insisted on. With an ‘opt-out’ approach, enrolled discharged patients are written to and informed that they will be telephoned to discuss the study unless they send back a slip or text back to say ‘do not contact me as I do not wish to hear more about the study’. The REC insisted that the slip read ‘yes please contact me’ instead, with no option to telephone non-respondents included. However, the REC was flexible in allowing us to record anonymised 30-day outcomes in this group, as long as the patients had been invited and had not refused to consent.

Objective 8: contribute to existing evidence about pre-hospital randomised trials

This has been a successful feasibility evaluation suggesting that the cluster randomised approach at ambulance station level works but that compliance needs face-to-face training.

Recruiting relevant patients with specific injuries has been identified in previous studies as a challenge. This is particularly the case in major trauma studies for two reasons. First, major civilian trauma is an ‘occult’ disease. The major vector is blunt trauma from road traffic collisions and falls, as opposed to more obvious wounding in civilian trauma. Second, as NHS ambulance patients’ records are often not computerised and difficult to link to hospital records (NHS numbers and reliable patient identifiers are often not available to crews during a brief interaction with an unconscious patient), it is hard to reliably predict the numbers of patients with given injuries who will be eligible for given ‘pre-hospital scene based’ study inclusion criteria. Hence there is a need for further pre-hospital major trauma studies to establish feasibility.

HITS-NS was viewed positively by paramedics and awakened enthusiasm for more pre-hospital research.

Trauma Audit and Research Network eligibility criteria

A patient of any age arrives at hospital alive after injury and at least one of the following subsequently occurs:

1. in-hospital death within 93 days during first admission

2. at least 3 days of inpatient care is required

3. level 2 or 3 care is required

4. hospital transfer for specialist care is required.

Most of the recruited HITS-NS study cohort did not meet TARN eligibility criteria, as none of these four criteria subsequently occurred. The predominant injury was mild head injury, with normal CT scan and discharge from hospital within 24 hours of presentation; this was true even in the subgroup with GCS score of < 13, in which > 65% had no brain injury (see Feasibility outcomes for stream A: proportion of study patients with traumatic brain injury on computed tomography head scan).

The NICE high-risk criteria33 for adult patients with head injury are any of:

• GCS score of < 13 at any time

• signs of open, depressed or basal skull fracture

• focal neurological deficit

• post-traumatic seizure

• failure to reach GCS score of 15 within 2 hours of injury

• more than one episode of vomiting post head injury.

The top two of these criteria were in the study eligibility criteria (see Eligibility criteria). Seizure as ‘unstable ABC’ would have been an exclusion criterion for which the patient would not have bypassed. Paramedics are not trained in formal neurological examination beyond the GCS, pupillary responses and FAST scores in stroke; the last two criteria require a period of observation post injury which is not applicable at the scene, so it was not possible to incorporate most of the NICE high-risk criteria into the inclusion criteria for HITS-NS.

The proportion of NICE high-risk patients with TBI/TBI requiring neurosurgery in the literature has not been well studied in the NHS, where studies have quoted rates of TBI for any of the NICE CT indications, which include medium-risk features such as age > 65 years with loss of consciousness or amnesia. 34 Rates of TBI for ‘NICE positive’ head injury patients vary between 10% and 30% but would be expected to be considerably higher in the ‘high-risk group’. Hence the TARN figure was adopted by investigators and 80% was the feasibility target.

The variation in rates of TBI by AS (even with GCS score of < 13 ; see Feasibility outcomes for stream A: proportion of study patients with traumatic brain injury on computed tomography head scan) can be seen as a limitation; however, head injury case mix presenting to EDs has been shown to vary quite significantly in other studies. 2,34 Both rates were too low to meet the feasibility target, particularly with regard to need for neurosurgery. The differences are likely to be caused by variation in the proportion of other patients who present with other causes of reduced GCS score in the context of a fall/assault with external signs of head injury. The major other causes of TBI are mainly intoxication or elderly ‘collapse’.

Compliance in the intervention clusters from the NEAS was limited and was a result of the same training resource (one study co-ordinator) being available to each AS for pragmatic reasons. In reality this made impossible the face-to-face training that achieved the almost-90% compliance in NWAS in NEAS – given the numbers of paramedics and large geographical area, the electronic media and cascade from team leaders were relied on instead. In the end, if rates of TBI had been higher, the investigators did not see this as a barrier to a full trial, as it could have been addressed by increasing resources for training therein.

The REC insistence on an active opt-in for being approached for telephone consent in patients with mild head injury who were discharged early limited meaningful follow-up of the study cohort beyond 30 days post injury. This has been highlighted in the literature as a persistent problem in this mild head injury group. 35 Again, the investigators feel that this could have been addressed in a full trial by persuading the REC with the pilot evidence to allow an opt-out approach that has been used successfully elsewhere. 32 With this approach the patient is written to and advised that a telephone call to discuss the study consent will follow unless a slip declining this is sent back within a fortnight; there have been no patient complaints using this approach.

The cluster randomised approach delivered balanced cohorts in terms of patient baseline characteristics. The median ISS was 2.5 higher in the control group; however, although statistically significant, on a median ISS of 1 this is not a clinically significant difference. 36 There was greater recruitment to the intervention cohort; however, it is unlikely that this reflects post randomisation bias, as the final arbiter of recruitment was the study co-ordinator rather than the scene paramedic. The analysis vis-à-vis the TARN database (see Prespecified analysis of robustness of screening) suggests that 90% of all eligible patients were approached for consent during the study period. It was difficult to predict exact recruitment rates to the study clusters prior to commencement from existing AS data and this is always likely to be the case as screening is very labour intensive (200 patients a day) and median numbers recruited per cluster over 12 months were in single figures (see Prespecified analysis of robustness of screening).

Relationship to other studies

The HITS-NS trial can be compared with other relevant literature in three respects: first, in terms of the feasibility of conducting pre-hospital controlled trials in general (as no trials of bypass have previously been conducted); second, with regard with to observational studies of trauma bypass; and, third, to the latest literature on trauma triage with reference to the sensitivity for detecting brain injury/need for neurosurgery.

The systematic review of pre-hospital trauma trials conducted from a feasibility perspective (see Appendix 1) highlights the same challenges that were encountered during the conduct of HITS-NS, supporting the feasibility approach particularly with regard to recruiting sufficient numbers of patients with the injury profile of interest over a given time. In blunt civilian trauma, the precise nature of a patient’s injuries are never ‘known’ with certainty at the scene – particularly when consciousness is impaired – and, in general, it is difficult for ambulance staff to collate sufficient patient identifiers at scene to permit anonymised data linkage (to hospital records with definitive injury details) to form the basis for feasibility using the pre-hospital population as a denominator.

Much of the literature around paramedic compliance was pessimistic, particularly from the NHS; however, the good rates of compliance with face-to-face HITS-NS training (≈90%), positively enforced by the paramedic focus group, suggest an approach that can work using a cluster randomised design. Training has been identified as key in other trauma trials in which compliance was high. 37

Follow-up rates and consent to follow-up, particularly when large numbers of patients with mild injuries are included, have been flagged as a challenge – as was the case for HITS-NS, although opt-out approaches to consent (which were not permitted by the HITS-NS REC) have worked elsewhere. 32

Two systematic reviews of trauma bypass have been conducted, one with a focus on head injury. These have been observational studies using hospitalised patients with TBI as a denominator rather than the population that the paramedic deals with at the scene. Hence they are of questionable external validity; furthermore, they tend to include patients whose nearest hospital was a NC and/or patients without any evidence of bypass having taken place and, therefore, lack internal validity. Notwithstanding this, the meta-analysis did not suggest benefit from ‘bypass’ in the TBI population. 14

The HITS-NS inclusion criteria – using the analysis in Prespecified analysis of robustness of screening – could be said to have detected up to 52 out of 114 (46%) patients with TBI presenting to non-specialist hospitals in the NEAS during the study period; the main reason for this ‘undertriage’ is the numbers of patients with a higher GCS score (14–15 in NEAS) at scene who are subsequently shown to have brain injury. This is particularly true in the elderly, in whom it is thought that cerebral atrophy allows early intracranial bleeding post head injury to not raise ICP and hence lower GCS. A recent study suggested that > 50% of all patients aged ≥ 65 years with TBI have a scene GCS recorded as 15. 38 Another group in which this is true is patients with extradural haematomas, who are, in general, young. In this group, there is rarely a primary brain injury but bleeding occurs in the vessels outside the brain within the skull, raising ICP and risking significant secondary brain injury if not treated. Many of these patients are observed post detection on CT brain scan, but in those who require operation the median GCS score on presentation to the ED is 14. 39

A further recent paper showed similar rates of undertriage or lack of sensitivity from the study inclusion criteria for patients on the TARN database with TBI; raising the GCS threshold increases sensitivity but reduces specificity, as the vast majority of head injury patients with GCS score of > 12 at scene do not have a brain injury. However, as this study used a pre-selected trauma registry rather than scene population the specificity is likely to have been markedly overestimated. 40 The lack of current trauma triage criteria, which are both sensitive and specific enough for optimum system functioning, has been highlighted elsewhere, but mainly with regard to all major trauma rather than head injury per se; however, in the NHS, 75% of all major trauma victims (defined as an ISS of > 15) have a brain injury. 18

It is known that the NICE high-risk criteria are 95% sensitive for detecting need for neurosurgery33 but, as indicated above, they are not applicable at the scene of injury, as they need observation for 2 hours post injury to be fully applied as well as practitioners who are trained in full neurological examination. Their specificity in this pre-hospital population is unknown.

Study design

We performed a cost–utility economic evaluation using a probabilistic decision analysis model to synthesise available evidence and compare alternative management strategies for patients with suspected TBI injured nearest to a NSAH. To maximise internal validity we followed expert recommendations, consensus modelling guidelines and NICE technical guidance. 57–60 Our base-case analysis followed NICE reference case methodology, taking the perspective of the UK NHS DH (the primary decision-maker) and employing a lifetime horizon.

Model uncertainty and sensitivity analyses

Uncertainty can arise in decision models because of variability, heterogeneity, parameter uncertainty and structural uncertainty. 56 Each of these factors needs to be addressed to explore important determinants of cost-effectiveness, afford decision-makers’ confidence in results, and to indicate the precision of cost-effectiveness estimates.

As a cohort model was implemented with examination of mean values, an examination of variability (or ‘first-order’ uncertainty) is extraneous. Furthermore, as head injury management pathways are population-level interventions, they will be implemented for any patients presenting with suspected significant head injury, making examination of heterogeneity superfluous. ‘Second-order’ parameter uncertainty, that is uncertainty surrounding the true value of a model input within the specified probability distribution, was fully explored in the PSA.

To examine ‘third-order’ parameter uncertainty (i.e. the correct statistical form has been specified for a probability distribution) we performed a number of univariate, threshold and scenario sensitivity analyses on parameter distributions that were thought to be important or uncertain. As influential relative effectiveness inputs were based on uncertain probability distributions elicited from expert opinion, and incremental costs were estimated from small HITS-NS samples, best- and worse-case analyses for parameters related to the pre-hospital triage and bypass strategy were initially performed. Model inputs were set to extreme favourable or unfavourable values and the model re-run to explore the most optimistic and pessimistic estimates for this strategy. An additional threshold analysis was implemented in which incremental costs, relative effectiveness and compliance were varied across their plausible range to determine the parameter levels at which the HITS-NS intervention may become cost-effective. Further univariate sensitivity analyses varied the individual probability distributions of certain model parameters, including population subgroups, baseline outcome probabilities, relative effectiveness, utilities, inpatient costs and post-discharge costs.

Structural uncertainty in methodological choices (varying the discount rates for costs and effects), choice of comparators (including theoretical bypass and no-transfer strategies) and perspective on outcomes (non-health effects of bypass) was also examined. Sensitivity analyses examining relative effectiveness and incremental costs were specified a priori. Other sensitivity analyses were informed by model structuring, evidence synthesis and emerging results. All sensitivity analyses treated unexamined model inputs probabilistically. The sensitivity analyses are summarised in Table 10.

Model implementation

The decision analysis model was programmed in Microsoft Excel^® 2013 (Microsoft Corporation, Redmond, WA, USA) using Visual Basic macros. An Excel add-in was utilised to sample from Dirichlet distributions. 79 Bootstrapping to derive incremental NMB 95% CI, using bias-corrected CI and evaluating 3000 replications, was performed in Stata version 12.1 (StataCorp, College Station, TX, USA). Internal testing was performed throughout model development to ensure that mathematical calculations accurately represented model specifications and were correctly implemented in Visual Basic. 59 Debugging techniques included null and extreme input values, setting equal values across comparators, fixed distributions and code breaks with line-by-line checking of macros. The model was also independently verified by a second modeller.

Background

To avoid potential opportunity costs, health-care systems should typically choose interventions with the highest expected NMB, regardless of statistical uncertainty. 80,81 This approach will maximise health gains from available resources and result in correct decision-making based on current evidence. However, there is often considerable uncertainty in model inputs and the optimal management strategy could differ if the true parameter values were known with certainty. Any errors in the adoption decision will consequently result in forfeited health benefits and wasted resources.

Economic models use probability distributions to reflect the uncertainty of a decision problem and can simulate the entire spectrum of cost-effectiveness results that are possible given potential realisations of parameter values. The probability of making the wrong adoption decision and the resulting effects on costs and QALYs will therefore be available from implementing a PSA; the opportunity losses surrounding a decision can consequently be calculated for each patient. This value is termed the individual expected value of perfect information (EVPI); measuring the expected cost of current uncertainty for each patient by accounting for both the probability that a decision based on existing evidence is wrong and for the magnitude of the consequences of making the wrong decision. 82 A rational decision-maker should be willing to pay for additional perfect evidence up to this level of expected opportunity loss for the total population of patients who may benefit from additional research evidence over the lifespan of the technology, a value known as the population EVPI. 82

Population EVPI places an upper bound on any investment in future research, but does not indicate what type of evidence would be valuable. By focusing attention on eliminating the potential opportunity costs arising from uncertainty in particular model inputs, expected value of partial perfect information (EVPPI) can indicate research into which parameters would be most valuable. 59,83 Identification of variables for which precise estimates would be most valuable will indicate where research funds should be focused and may suggest appropriate research designs. For example, a large EVPPI value for an estimate of relative effectiveness would suggest that a future RCT is potentially useful, whereas a high EVPPI for the prevalence of a disease subgroup could indicate an epidemiological cohort study. Individual EVPPI, defined analogously to EVPI, is the expected NMB given perfect information about the parameter of interest minus the expected NMB with current information. Population EVPPI is similarly derived by multiplying the individual EVPPI by the population of patients that would potentially benefit from additional information. Extension of this technique to groups of parameters is possible using the same principles. 59

Comparing the costs of future research studies against population EVPI and EVPPI provides a necessary condition for future research: study costs must not exceed these values to be cost-effective. However, they do not provide a sufficient condition that any future research study will be helpful and cost-effective, and make the implausible assumption that all uncertainty can be removed from a parameter value. EVSI extends the value of information methodological framework to establish the expected value of conducting studies with different designs and sample sizes. 84,85 Individual EVSI is mathematically defined as the expected difference between the value of the optimal decision, based on a sample of data, informative for certain model parameters, minus the value of information based on existing evidence. 84 Population EVSI is computed equivalently to population EVPI and EVPPI, but should account for the fact that patients recruited into a study are ‘used up’ and cannot benefit from a study’s results, and that the length of a study may reduce the time period for which an intervention is applicable.

The expected benefits of a given study sample (the population EVSI) can be compared with the expected costs of collecting these data, with the difference denoting the expected net benefit of sampling (ENBS). ENBS measures the societal reward for conducting additional research, with ENBS > £0 demonstrating that the marginal benefits of gathering further evidence exceeds the marginal costs, and higher ENBS values representing more efficient study designs. Optimal study designs can then be identified by choosing the largest ENBS from different options with varying sample sizes, follow-up periods and study end points. 59

Expected value of information methodology

Individual EVPI was directly calculated directly from the model PSA output using the standard formula. 59

A state-of-the-art regression-based approach was used for calculation of EVSI and EVPPI. 86–88

Briefly, the method developed by Strong and Oakley86–88 rearranges the statistical formulas for computation of individual EVPPI and EVSI, reframing estimation into a regression problem. The expected NMB for each PSA simulation can be considered a function of the examined uncertain parameters in EVPPI, or simulated data in EVSI. The PSA output can therefore be treated as ‘noisy data’ through which this functional form can be characterised, thus allowing calculation of the individual EVPPI or EVSI at a particular willingness-to-pay threshold using fitted values from regression models. As NMB would not be expected to have a linear relationship with inputs of interest, a non-parametric regression method provides the flexibility to accurately model the correct shape of the association. The theoretical basis of this method is described in detail in the literature86–88 and examples of its use in HTAs are also available. 89,90 In common with previous applications of this method, generalised additive models were used for non-parametric regression,91 implemented in R (The R Foundation for Statistical Computing, Vienna, Austria) using the ‘mcgv package’ and a ‘n = 5000’ PSA sample. 92–94

Individual EVPPI was initially calculated for each model input separately. Groups of parameters were then chosen to match types of research studies that could be conducted in order to logically inform future research prioritisation. Study designs considered included observational studies examining the prevalence of different patient subgroups, utility values for GOS states post-discharge health-care costs, and an experimental study investigating the relative effectiveness of bypass.

The individual EVSI analysis examined the value of conducting a definitive HITS-NS trial, of varying size, examining relative effectiveness between bypass and selective transfer. Regression models in the EVSI analysis require a summary statistic to represent the information gained from additional research. For each patient subgroup the numbers of patients expected to have favourable (good recovery or moderate disability) and unfavourable outcome (death or severe disability) were simulated from the prior distribution in the PSA output, using a binominal distribution. This process accounted for the relative numbers of patients expected in each subgroup by multiplying the overall trial size by the relevant subgroup proportion. Log-odds ratios were subsequently calculated for each patient subgroup as the summary statistic. For small trials, subgroups with low prevalence could have very few patients and a continuity correction was applied, where necessary, to allow calculation of odds ratios. Log-odds ratios for each patient subgroup were included in the final regression model as linear predictors without interaction terms.

Similarity between patients treated by the same ambulance station in a cluster randomised trial could lead to correlated outcomes and between-cluster variability. Cluster sampling consequently provides less information than individually randomised trials. 95 Effective sample sizes used to calculate EVSI were therefore multiplied by the design effect to determine the actual required number of patients to account for the increase in variance resulting from the HITS–NS cluster design. 96 Standard formulas were used to calculate the design effect, using estimates of intracluster correlation coefficient and average cluster size. 97 As HITS-NS was pair-randomised it was not possible to deduce an intracluster coefficient directly from trial data, and a plausible value was taken from a previous pre-hospital trial in the UK. 98 Average cluster size was based on the recruitment observed in HITS-NS. Sample size requirements for a cluster trial are also influenced by variation in cluster size and method of randomisation. However, to simplify calculations we considered only simple randomisation, assuming equal cluster size.

Calculation of population values for each expected value of information metric requires assumptions about the incidence of suspected significant TBI with stable pre-hospital physiology, the predicted time for which pre-hospital triage and bypass is likely to be a viable management option and discount rate. Applicable incidence estimates were unavailable from the literature, and HITS-NS pilot data were used in conjunction with routine statistics to calculate a credible range of values. The total number of eligible trial patients recruited over 1 year in each region was divided by the total AS catchment area adult population, and then averaged across the two regions. Annual incidence of suspected significant TBI was then derived using estimates of the national adult population in England, using data from the 2011 Census. 70 The technology lifespan was arbitrarily defined as 10 years, based on previously published HTAs. 99 A standard discount rate of 3.5% was applied. 60 As the variables influencing the size of population that may benefit from future research are uncertain, we conducted two scenario sensitivity analyses to provide tenable upper and lower bounds. Disease incidence, technology lifespan and discount rate were set sequentially to plausible high and low values, informed by differential recruitment rates between the two HITS-NS trial regions and the literature. Table 11 summarises the values for population expected value of information calculations in base-case and sensitivity analyses.

Calculation of population EVSI must also account for the fact that patients who are enrolled in a study will not benefit from the information generated by the research. Furthermore, as sample sizes increase, accordingly longer trials will reduce the length of time new information will be useful for. We therefore assumed that any trial could feasibly be conducted in up to 8 out of the 11 English NHS ASs, with each AS having 46 ambulance stations available for randomisation. Based on the expected incidence of suspected significant patients with TBI in England and average cluster size, we hypothesised an upper limit on patients who could be recruited in 1 year. Sample sizes exceeding this number would take additional years to enrol the necessary patients. It was also assumed that a further 2 years were required for analysis, reporting and dissemination of results and implementation of findings. Other scenarios were explored in optimistic and pessimistic sensitivity analyses and an analysis based on assumptions regarding a definitive trial detailed in the original HITS-NS protocol. Table 12 summarises the base-case population EVSI assumptions and presents alternative values used in sensitivity analyses.

Population ENBS was calculated by subtracting the costs of performing a future trial for a given sample size from the population EVSI. The cost of running a proposed cluster RCT was assumed to be £1000 per recruit in the base-case analysis (Clinical Trials Unit, ScHARR, University of Sheffield, 8 November 2013, personal communication). Other eventualities, including an estimate examining fixed and variable costs based on the HITS-NS pilot study grant, were examined in sensitivity analyses (Table 13).

Introduction

The results of the decision analysis model are presented in four ways. First, the mean lifetime costs and QALYs of alternative head injury management strategies are compared deterministically for the base-case model. Second, to account for parameter imprecision and the uncertainty of expert opinion informing model inputs, results are recalculated in a Monte Carlo PSA. Third, the influences of parameter uncertainty and methodological and structural assumptions inherent in the model’s design are explored in a series of sensitivity analyses. Finally, following this evaluation of the cost-effectiveness of competing management strategies, expected value of information techniques are used to examine the value of further research in this area.

Deterministic base-case results

Table 14 presents the mean expected costs and QALYs accrued by each strategy over a lifetime horizon. Selective secondary transfer was less expensive than the competing management pathways but resulted in fewer QALYs than routine transfer or bypass strategies. The no-transfer strategy provided the fewest QALYs at the second highest cost.

Average cost-effectiveness ratios for each strategy, compared with a baseline of selective secondary transfer, are shown on a cost-effectiveness plane in Figure 15. When choosing between different interventions, it is necessary to determine the extra cost incurred for each additional QALY gained when switching from one strategy to another. Average cost-effectiveness ratios ignore the alternative treatments available and calculation of cost-effectiveness ratios with the next-best alternative in a fully incremental analysis is required for a valid comparison. The no-transfer strategy, providing fewer QALYs at the second highest cost, was strongly dominated by other management options and, therefore, excluded from further consideration. Each option was then ranked in order of increasing effectiveness and an ICER calculated with the next most effective strategy. The final ICERs between selective secondary transfer and routine transfer and between routine transfer and bypass were £2217 and £27,100 respectively. Therefore, ignoring parameter uncertainty and assuming the model is valid, routine secondary transfer is the optimal strategy for management of patients with suspected significant head injury at the standard NHS willingness-to-pay thresholds of £20,000 but bypass would be considered cost-effective if the threshold was increased to £30,000.

FIGURE 15.

Cost-effectiveness plane presenting deterministic point estimates of average cost-effectiveness against a baseline of selective secondary transfer.

Expected value of perfect information

Reflecting the uncertainty in costs and effects, together with the large potential opportunity losses from making the incorrect adoption decision, individual EVPI was substantial at NICE willingness-to-pay thresholds: £1807 at λ = £20,000 and £2594 at λ = £30,000. Given the relatively large annual population with suspected significant head injury and the long time period over which pre-hospital triage and bypass is likely to be applicable, population EVPI was also correspondingly large in the base-case analysis: £35,589,500 at λ = £20,000, and £51,080,000 at λ = £30,000. These figures represent the maximum that the NHS should be willing to invest in future research to eliminate uncertainty about which management strategy to implement; assuming an infinitely sized study, evaluation of all model inputs and the economic model is correctly specified. Figure 21 displays the base-case population EVPI at relevant willingness-to-pay thresholds. Two inflections in the slope of the EVPI curve are visible, at λ = £2000 and λ = £28,000, corresponding to threshold values at which the comparator with highest expected NMB transitions between selective transfer/routine transfer and routine transfer/bypass. Above λ = £28,000 the EVPI continues to rise as the consequences of decision errors are valued more highly and the probability of erroneously adopting bypass remains high. Estimates for population EVPI remained substantial in pessimistic and optimistic scenario sensitivity analyses, varying from £10.9M to £70.7M at λ = £20,000, and £15.6M to £101.3M at λ = £30,000 (Table 21). Individual and population EVPI values for a wider range of willingness-to-pay thresholds are presented in Appendix 7.

FIGURE 21.

Population EVPI for the base-case probabilistic model.

Expected value of sample information

Expected value of sample information and ENBS of a future definitive HITS-NS trial were considered in base-case and scenario analyses, making optimistic and pessimistic assumptions about the size of the future population who could benefit from the data collection. The characteristics of the theoretical trial are summarised in Chapter 4. As sample size increases, simulated estimates of relative effectiveness will become more precise until uncertainty is completely removed at infinite sample sizes. Individual EVSI will therefore asymptotically approach the individual EVPPI value for bypass relative effectiveness parameters at very high sample sizes (£1403 at λ = £20,000, £2249 at £30,000), as shown in Figure 23. An analogous pattern is observed for population EVSI, assuming that the disease incidence and the ability to recruit patients far exceed sample size.

FIGURE 23.

Individual EVSI for a definitive HITS-NS cluster randomised trial at NICE cost-effectiveness thresholds. WTP, willingness to pay.

As the number of study participants increases, the number of ambulance stations and ASs required to recruit the necessary number of patients will expand, determined by the average cluster size and number of ambulance stations available for randomisation in each AS. At higher levels the sample size will exceed the maximum number of patients recruitable per year, extending the length of the trial, curtailing the time that the study findings are useful and consequently reducing the size of the future population that may benefit from the trial. Additionally, patients enrolled in a study will not be able to benefit from the information generated, as they will have already received treatment. In actuality, population EVSI therefore falls at larger trial sizes, as shown in Figure 24 (calculated under base-case assumptions). Inflection points in the population EVSI curve denote sample sizes requiring trials with additional years of recruitment for the necessary number of patients.

FIGURE 24.

Population EVSI for a definitive HITS-NS cluster randomised trial at NICE cost-effectiveness thresholds accounting for prolonged trial duration at large sample sizes (base-case assumptions). WTP, willingness to pay.

Expected net benefit of sampling

The ENBS for a definitive HITS-NS trial with different samples sizes under base-case assumptions is shown in Table 23 and in Figures 25 and 26 for λ = £20,000 and λ = £30,000, respectively. As relative effectiveness between bypass and selective transfer is very uncertain, and a key determinant of cost-effectiveness, even small trials providing relatively imprecise effect estimates would have a substantial impact on reducing decision uncertainty and therefore have positive ENBS. In the base case, ENBS was maximal with a trial of 1040 patients (520 randomised per arm) recruited across 347 ambulance stations in eight ASs over 1 year (with a further 2 years for trial analysis, reporting and dissemination): ENBS £11.0M at λ = £20,000; ENBS £19.6M at λ = £30,000.

FIGURE 25.

Expected net benefit of sampling for a definitive HITS-NS cluster randomised trial at λ = £20,000 (base-case assumptions).

FIGURE 26.

Expected net benefit of sampling for a definitive HITS-NS cluster randomised trial at λ = £30,000 (base-case assumptions).

The optimal trial design was sensitive to varying assumptions about trial characteristics, disease incidence, technology lifespan and patient recruitment limits. Taking an optimistic view of these factors, the most cost-effective trial design at NICE thresholds would be achieved with a trial of 2052 patients recruited from 10 ASs, in 480 ambulance stations and lasting a total of 2 years. Taking pessimistic assumptions resulted in much lower ENBS, but a future trial was still shown to be cost-effective, with the most favourable trial recruiting 636 patients (5 ASs, 140 clusters, 5 years in total) at both λ = £20,000 and λ = £30,000. The study characteristics for a definitive trial envisaged in the original HITS-NS pilot study protocol resulted in a very similar optimum design of 624 patients (4 ASs, 120 clusters, 4 years in total). Interestingly, careful enumeration of fixed and variable trial expenditure, informed by HITS-NS pilot study funding, gave very similar trial costs to the base-case estimate using a marginal per patient estimate. Table 24 summarises the properties of the optimal trials in these different scenarios, with further detail provided on ENBS and trial designs provided in Appendix 7.

Summary of results

1. Objectives 5 and 6 of the HITS-NS study were addressed in the stream B economic evaluation:

   1. – determine the cost per QALY of early neurosurgery and the degree of uncertainty surrounding this estimate

   2. – calculate the EVSI of a fully powered HITS-NS trial.

2. The base-case probabilistic analysis suggests that routine transfer (transport to the local non-specialist hospital and routine secondary transfer of all patients with acute expanding intracranial haematomas or TBI requiring critical care to regional NCs) may provide the optimal management strategy at a willingness-to-pay threshold of £20,000 (mean ICER £2260). At a higher threshold of £30,000, bypass was the most cost-effective option (mean ICER £27,157). At both thresholds there was considerable decision uncertainty, with a high probability of erroneously adopting a suboptimal strategy (54% and 52%, respectively).

3. Sensitivity analyses demonstrate that this result is critically dependent on parameterisation of incremental costs and relative treatment effects for routine transfer and bypass strategies. These model inputs are predominantly based on unadjusted estimates from HITS-NS data or are informed by expert opinion unsupported by empirical evidence and are therefore at risk of systematic error. Plausible alternative assumptions on incremental costs and effects, life expectancy following injury and discounting rates all resulted in reversal of the adoption decision at λ = £20,000, with bypass identified as the optimal strategy. Base-case results should, therefore, be interpreted with extreme caution.

4. The considerable decision uncertainty and important public health burden of TBI is reflected in the large population EVPI. At λ = £20,000, further research up to a value of £35.6M may be indicated to minimise opportunity costs from making the wrong adoption decision for patients with suspected significant TBI.

5. EVPPI analyses demonstrate that future research would have high value in comparing costs and relative effectiveness between bypass and selective secondary transfer: that is, a definitive HITS-NS trial-based economic evaluation.

6. If feasible, EVSI results suggest that a definitive HITS-NS trial examining comparative effectiveness is potentially cost-effective. Maximal ENBS (£11.M) would be achieved with a trial of 520 patients per arm, randomised across 347 ambulance stations in eight ASs and taking 3 years. At higher sample sizes ENBS falls as recruitment costs increase, and trial duration lengthens secondary to the relatively low incidence of suspected significant TBI and the finite number of ambulance stations available for randomisation.

7. EVI analyses are predicated on the parameterisation of the base-case model and assumptions regarding incidence of suspected significant TBI, technology lifespan and cluster trial characteristics. As these factors are highly uncertain, results should be viewed as exploratory.

Limitations

This economic evaluation followed consensus modelling guidelines and has a number of strengths. 57–60 A formal model structuring process was implemented to derive a valid and clinically convincing model structure. Comprehensive systematic evidence searches ensured that the model was populated with valid evidence where possible. Elicitation of expert opinion, informed using the established SHELF framework103 ensured transparency in model inputs when empirical data was lacking. Decision uncertainty was explored extensively in sensitivity analyses and the potential benefit of future research was evaluated in state-of-the-art EVI analyses. However, there are limitations in the model design and parameterisation which could challenge the internal validity and generalisability of results.

First, the model had a limited scope and several potentially important aspects of managing patients with suspected significant TBI were not examined. The HITS-NS cohort included an appreciable minority (approximately 5%) of non-patients with TBI with a major medical diagnosis, for example non-traumatic subarachnoid haemorrhage or sepsis. As a result of heterogeneity and lack of empirical evidence, we did not evaluate this patient subgroup. If there are important differences in costs and effects arising from bypassing such cases, exclusion of these patients could bias results.

Introduction of bypass or routine transfer strategies could incur expenditure relating to changes in patient treatment pathways, such as staff training or administrative support. Given the lack of information available on the extent of these costs and expert opinion that start-up costs are likely to be insignificant, we did not model this factor. However, if reconfiguration costs are substantial, the reported cost-effectiveness of bypass will be overestimated.

The economic model assumes that specialist neuroscience care is scalable with infinite capacity. Surveys of SNC capacity suggest that many units are operating close to their maximum volume,104,105 and admitting increasing numbers of patients for neurocritical care in routine transfer or bypass strategies may not be possible. The potential impact of overcrowding in EDs or ICUs arising from treatment of additional patients, possibly unlikely to benefit from specialist care, has also not been examined. Similarly, capacity constraints arising from difficulties in repatriating bypassed patients to local hospitals after a period of specialist care were not considered. The complexity of representing these real-world phenomena precluded modelling but, reassuringly, expert opinion suggested that they are unlikely to be important.

Traumatic brain injury frequently results in chronic disability, leading to productivity losses and a long-term informal care burden for family members. 106 Our narrow perspective, including only direct medical and personal social services costs, will therefore substantially underestimate the societal costs of competing strategies. As the proportion of patients with long-term disability was comparable between bypass and selective/routine transfer strategies, it could be argued that exclusion of these costs is unlikely to qualitatively change the results of the base-case model. Similarly, only direct health effects were assessed, but a sensitivity analysis, including a small utility decrement for patients with mild TBI undergoing unnecessary transport to SNCs, illustrated the potential for a marked decrease in the cost-effectiveness of bypass if such factors are important. However, qualitative interviews with HITS-NS participants and relatives did not reveal any serious concerns regarding non-health-related consequences arising from the bypass strategy.

Furthermore, examination of aero-medical transfers, urban environments and suspected patients with TBI with unstable pre-hospital physiology were outside the remit of the economic evaluation. The model also consciously focused on simulating management within the NHS. Any generalisation of results to other populations beyond the HITS-NS setting and inclusion criteria should therefore be circumspect.

Second, model structuring was restricted by the limited availability of evidence on TBI epidemiology, hospital costs, treatment effectiveness and pathophysiology. Theoretical disease logic and treatment pathway conceptual models emphasised the importance of time to resuscitation and neurosurgery on outcome, but we were unable to represent these factors directly. The final model, comprising strategy level estimates of costs and consequences for relevant patient subgroups, provided a pragmatic structure indirectly accounting for these factors and retaining clinical credibility. However, the ability to examine the influence of different geographical settings, triage compliance and secondary transfer rates was curtailed.

No information was available on the covariance between inpatient costs and short-term outcomes, and our treatment of these variables as independent in the PSA is a further limitation of the model structure. Expert opinion was highly uncertain as to the magnitude and direction of any correlation. Given the large degree of decision uncertainty at NICE willingness-to-pay thresholds, small differences in expected costs and QALYs arising from incorrect specification of this relationship could potentially influence the choice of optimal management strategy.

Within each modelled patient subgroup there will be a considerable diversity of patients with differing characteristics and prognoses. Applying a cohort methodology, with consequent use of mean values, impeded an examination of uncertainty due to heterogeneity. However, competing management strategies are service-level interventions and hence would be applied to the entire population presenting with HITS-NS inclusion criteria. Exploration of heterogeneity, for example the cost-effectiveness in different age groups, may therefore be less relevant.

Third, and most importantly, there was a very limited evidence base available to parameterise model inputs. No level 1 RCT data were identified and systematic reviews highlighted the dearth of valid observational evidence. The shortage of applicable information was compounded by the unique HITS-NS inclusion criteria, which, to our knowledge, have not been previously studied, and the lack of 6-month follow-up data available from the pilot study. To ensure a believable and accurate model, we therefore prescribed a minimum quality standard for including evidence, eliciting expert opinion, when necessary, in preference to ‘cherry-picking’ data or using inappropriate studies at high risk of bias. Moreover, in the event of valid evidence that was not directly applicable because of differing study populations, we transparently modified published estimates using formal ‘external bias adjustment’ techniques. 107 The base-case model consequently provides a framework to accurately represent current beliefs on competing management strategies and synthesise relatively unbiased evidence.

Although inclusion of expert opinion allows full representation of the uncertainty associated with the decision problem, it is possible that elicited probability distributions do not accurately reflect true parameter values. Beliefs about pivotal relative effectiveness estimates were characterised from intensive care consultants working in a NC. These clinicians may not have full insight into outcomes in patients treated in non-specialist centres and could have imperfect knowledge on the effects of operative management of expanding intracranial haematomas compared with neurosurgeons. Strategy-level estimates of outcomes are also dependent on weighing several other competing factors including pre-hospital deterioration, compliance with bypass and secondary transfer rates. It is debatable whether or not clinicians can accurately integrate all of these considerations into credible effect estimates. Further challenges to elicitation include well-recognised difficulties in understanding probability distributions and odds ratios103,108 and cognitive biases inherent in the elicitation process. 103 To maximise the validity of expert opinion we followed an established elicitation framework, used natural frequencies to derive effect estimates and conducted detailed briefing and training exercises. 103 Other, less critical model inputs were elicited and reviewed within the TMG, and this lack of independence could be considered a limitation with the potential for introducing subjectivity into the analysis.

In the absence of any relevant external studies, HITS-NS pilot study data were used to inform incremental cost differences between selective secondary transfer and bypass. The low sample size available resulted in very imprecise results and prevented regression modelling to adjust for case mix differences in all but the mild TBI subgroup. Although patient characteristics were broadly similar between study groups, findings could potentially be biased secondary to confounding. Additionally, complete case analyses were performed and, if cost data were missing at random, or missing not at random, selection bias may have arisen. 109 Missing data levels were low and hence unlikely to influence results, but ideally a principled statistical method for imputing missing data, such as multiple imputation,110 would have been used. Resource use was valued deterministically using NHS reference costs averaged over NHS hospitals, and limited data was available on management intensity for patients admitted to critical care. It is therefore possible that cost differences between specialist and non-specialist hospital care were not fully delineated.

Other notable weaknesses in model parameterisation include GOS utility values (risk of selection bias and valuation in a non-UK population), relative effectiveness data for patients with major extracranial injury (estimate from non-contemporaneous study with likelihood of subsequent improvements in trauma outcomes) and post-discharge costs (empirical evidence source largely based on expert opinion). All of these model inputs were assessed as borderline with respect to the minimum quality standards required for model inclusion and were therefore subjected to sensitivity analyses to illustrate the potential impact on results from alternative assumptions.

In common with established modelling practice, we used an informal Bayesian approach to fitting parameter distributions, based on the evidence available. 56 In practice this will lead to very similar distributions to an analysis using uninformative prior distributions. However, given the paucity of available evidence and low sample sizes informing some model inputs, a formal Bayesian synthesis using subjective priors may have been more appropriate to account for existing beliefs about suspected significant TBI management.

Fourth, an inability to fully evaluate the model is a further study weakness. Detailed model verification was performed and descriptive validity has been expounded in the foregoing discussion. However, the inclusion of theoretical interventions and reliance on expert opinion for parameterisation restricted the possibilities for internal or external model validation, with little opportunity to check that model results match actual management strategy outcomes. To our knowledge there are no other ongoing or planned studies examining this population or decision problem, making prospective validation unlikely. However, the model output appears to have face validity and is not inconsistent with costs and outcomes reported in other broadly related studies. 3,41,111

Finally, there are a number of potential limitations that could affect expected value of information results. These analyses are premised on the implemented decision analysis model and are therefore subject to identical biases arising from model inputs and structure described previously. Additionally, population-level results are heavily dependent on assumptions regarding the lifespan of bypass as a relevant health technology and the unknown true incidence of suspected significant TBI in the UK. Incidence rate estimates differed widely between the two trial regions and, although different scenarios were explored in sensitivity analyses, the actual value of future research is consequently uncertain.

When computing EVSI we calculated trial sample sizes based on an individually randomised trial, inflated by a design effect determined by assumptions on intracluster correlation coefficient and average cluster size. Generating simulated trial results using an explicit multilevel statistical model to directly account for clustering may have offered a more theoretically sound approach, but is likely to be an academic distinction with minimal influence on results. Owing to the paired cluster randomisation method used in HITS-NS, we were unable to calculate an ICC directly from pilot study data, and we therefore had to use plausible estimates from other non-TBI pre-hospital trials. Several real-world aspects of trials, such as restricted randomisation, loss to follow-up, cluster dropout, unequal cluster sizes and non-compliance were also not accounted for, but are unlikely to materially change the findings. Although not a prespecified objective, ideally we would have also examined the EVSI of including incremental costs in a trial-based economic evaluation or calculated ENBS for a trial comparing bypass with routine transfer. Unfortunately, the computational challenges were significant and implementing these analyses was beyond the resource and time constraints of the current study.

Interpretation of findings

The main determinants of cost-effectiveness in the decision analysis model were incremental costs and effects for patients with TBI requiring acute neurosurgery or critical care. Model inputs for these variables were very uncertain and small changes in their values resulted in conflicting adoption decisions. Previous observational studies, consistent with cost data in HITS-NS, suggest large incremental differences in costs between management in specialist and non-specialist centres. 3,41 Disparities of this magnitude are less likely to be explained by confounding, suggesting that our parameterisation may indeed reflect reality. However, ultimately, evidence from a well-conducted randomised trial is necessary to provide a definitive estimate.

The incremental cost difference in patients with TBI requiring critical care may be intuitively explained by more aggressive management, with longer ICU stays and higher rates of neurosurgical interventions in specialist centres. 41,112 However, the reasons for the higher costs observed in the HITS-NS study between bypassed patients with acute neurosurgical lesions compared with those undergoing secondary transfer are much less clear. The finding could be explained by the play of chance arising from the small pilot study sample. Outliers with high treatment costs, possibly secondary to management of associated extracranial injuries, will have much greater influence in such a small sample. Confounding arising from crude analyses may also be responsible. Alternatively, there may be a true difference in costs possibly explained by the much worse prognosis expected in patients undergoing delayed neurosurgery, which increases early mortality and reduces treatment costs. Interestingly, the HITS-NS results were replicated to some extent in adjusted analyses studying similar patients from the Nottingham Head Injury Register (NHIR), lending credence to these findings.

The result from the base-case economic model that bypass may not be cost-effective for suspected patients with TBI at the standard NICE cost-effectiveness threshold may be unexpected, given the strong support for introducing regional trauma networks from professional bodies and expert opinion leaders. 113,114 The conceptual models of pathophysiology and treatment pathways elucidate the potential benefits and hazards of bypassing patients with TBI, and indicate the manifold factors that will interact to determine strategy level cost-effectiveness. Considering this complexity, the plethora of low-quality and poorly applicable evidence that has been cited in support of bypass and lack of familiarity with heath-care costs, it may be unsurprising that subjective expert opinion will differ from an objective and careful dissection of the decision problem. Contrarily, proponents of bypass could conjecture that nuanced clinical opinion based on extensive experience will be superior to an economic model ignoring many subtleties of trauma management.

Expert beliefs on the comparative effectiveness of bypass for patients with TBI requiring critical care were very cautious, indicating a wide range of possible effect estimates consistent with either a beneficial or a harmful effect. This may suggest that the support for regional trauma networks and bypass of general major trauma patients may not extend to patients with TBI, with the potential for deterioration and secondary brain injury.

Expected net benefit of sampling analyses suggest that a future definitive bypass trial would, if feasible, be cost-effective to reduce decision uncertainty, even in conservative sensitivity analyses. Regardless of model limitations, this finding is unsurprising as elicited expert opinion was very uncertain on the effectiveness of alternative management strategies and empirical evidence was entirely lacking. The major areas of uncertainty pertained to relative effectiveness in patients with TBI requiring acute neurosurgery or critical care, and a relatively small trial would provide substantial information on their posterior distributions and impact adoption decisions. As the EVPPI for these parameters is very high, but the prevalence of relevant patient subgroups in the HITS-NS population is low, EVSI continued to increase at very large sample sizes. However, because of a reduction in the number of patients who may benefit from the trial, and time period for which study results would be useful, as study duration increased, ENBS falls at higher sample sizes.

Relationship to previous studies

No previous economic evaluations have been conducted that specifically investigated pre-hospital triage and bypass in patients with suspected significant TBI and stable pre-hospital physiology. However, a number of previous studies have studied patients with TBI using simulation techniques, or have examined bypass in major trauma patients. Dissimilar populations, lack of valid model parameterisation and non-cost–utility approaches limit the inferences that can be drawn from comparisons with these studies.

Stevenson et al. 13 preformed a simulation study examining bypass of patients with TBI in a UK setting. The model, almost entirely based on informal expert opinion, estimated an additional six survivors per million total population per year if a bypass strategy was introduced. This finding is consistent with the increased number of surviving patients observed with bypass within the HITS-NS model. Unfortunately, further insights are not possible, as costs, disability and triage of undifferentiated patients were not considered.

The 2007 NICE head injury guidelines included a cost–utility study examining bypass of patients with TBI compared to a routine transfer strategy1 using a similar decision tree approach to the HITS-NS model. A base-case ICER of –£26,340 in the south-east quadrant of the cost-effectiveness plane was reported, with bypass strongly dominating the secondary transfer approach. This result was consistent in conservative sensitivity analyses and the conspicuous discrepancy with findings from the current model deserves close scrutiny. A clear difference in the NICE model, generating the extremely favourable ICER for bypass, is an assumption that ED and inpatient costs do not differ between each treatment pathway. In common with the Stevenson study,13 the model also only included patients with severe head injury, rather than the wider spectrum of TBI and non-patients with TBI to which bypass will apply. Moreover, the rationale for parameterisation of the model is not transparent, with very heavy reliance on subjective, informal expert opinion and inclusion of evidence rejected from the HITS-NS model because of extremely high risk of bias.

Nicholl et al. 67 investigated the cost-effectiveness of introducing regional trauma networks in England for major trauma patients using a very simple decision analysis model, reporting an ICER of £1262 and an 80% probability of cost-effectiveness if the cost of implementing a fully effective bypass system was < £34M. 67 The authors highlight the limited evidence available, necessitating a number of ‘heroic assumptions’. These simplifications are not necessarily a problem, as the purpose of a model is to usefully inform a decision question rather than replicate real life. However, the postulation that acute care costs are the same before or after the introduction of trauma networks appears untenable. Furthermore, the similarity between average major trauma patients and the specific HITS-NS population is uncertain, making external validity questionable.

Other patient-level health-economic studies have examined the cost-effectiveness of trauma centre care using observational data. 115–117 These studies compared patients treated in trauma centres (bypassed, directly admitted and transferred in) to untransferred patients cared for in non-specialist hospitals. As they do not examine counterfactual outcomes for patients treated with or without bypass, they have little relevance to the HITS-NS decision problem. Overall, in common with other many other areas of TBI research, there is little cost-effectiveness literature available to inform the HITS-NS economic model.

Implications for clinical practice

Since the inception of the HITS-NS study, trauma care in the NHS has been reconfigured, with the introduction of regional trauma networks. 44,118 Pre-hospital triage with bypass of patients meeting HITS-NS inclusion criteria has now surpassed selective transfer as conventional practice. The relevance of these results to fully implemented trauma systems therefore requires careful consideration.

Above the conventional willingness-to-pay threshold of £20,000, NICE judgements about the acceptability of a health technology will depend on the relative degree of cost-effectiveness, the nature of the intervention and disease and any wider societal costs and benefits. Bypass demonstrated an ICER within NICE’s stated range of borderline cost-effectiveness and results were highly uncertain, even without considering the model limitations or absence of robust evidence. The introduction of trauma networks could be considered an innovative health technology addressing a previously disadvantaged population of trauma patients8 and improved outcomes after head injury could have a major societal impact through increased productivity and a reduced burden on families. It could therefore be contended that in this context the rational course of action would be to avoid any risks and costs from further reorganisation, and persist with the bypass-adoption decision. Decision-makers may also find alternative sensitivity analyses, for which bypass had a favourable ICER of < £20,000, to be more believable than the assumptions inherent in the base-case analysis.

Furthermore, routine transfer could be viewed as a theoretical intervention, unlikely to ever be implemented in the NHS. From this perspective bypass would be the optimal strategy, providing noticeably higher expected net benefit than selective secondary transfer (mean PSA ICER £15,526). However, recent studies have highlighted that a very high proportion of non-surgical patients with severe TBI are now transferred for specialist care [83% in HITS-NS, 73% in RAIN (Risk Adjustment In Neurocritical care)], suggesting that there may be little difference between selective and routine transfer strategies in practice. 41 Likewise, as routine transfer is the standard of care currently recommended in NICE head injury guidelines, excluding it from deliberation seems unreasonable.

Conversely, if the assumptions of the base-case model are considered convincing to decision-makers, it could be asserted that bypass protocols should be modified to exclude patients with suspected significant TBI and stable pre-hospital physiology, although, based on a transparent and thorough exposition of the available evidence, there are several rebuttals to this position. The accuracy and reliability of pre-hospital triage is poorly understood and HITS-NS data suggest that a meaningful number of non-TBI major trauma patients would consequently not be bypassed to trauma centres. Administering multiple triage rules simultaneously or further complicating existing major trauma triage instruments may not be practicable in the pressured pre-hospital environment. Given the financial and administrative investment in trauma system reconfiguration, there is also likely to be sizable clinical and public opposition to further re-organisations of care based on cost-effectiveness results with a high risk of error.

Notwithstanding its potential lack of cost-effectiveness, the adoption of bypass may also be fundamentally irreversible. Ongoing regionalisation of hospital services may result in closure of non-specialist hospitals or degradation in the skills required to manage patients with moderate and severe TBI. 119,120 This may prevent reintroduction of secondary transfer strategies and could have important implications for the substantial numbers of patients with significant TBI who present with GCS levels higher than triage rule inclusion cut points. Recent analysis of TARN registry data and the analysis in stream A, suggests that appreciable numbers of patients with TBI ultimately requiring critical care or neurosurgery will continue to be transported to NSAHs. 40

Implications for future research

The extensive literature reviews conducted for the HITS-NS model demonstrate that the evidence base supporting bypass in TBI is extremely tenuous. As noted by Nicholl et al. 67 in a related economic evaluation ‘it is remarkable how poor the design of relevant studies has been’. Although there are possible theoretical benefits from achieving earlier definitive care through bypass,113 there are potential hazards from prolonged primary transport of patients with TBI, and it could be posited that proof of concept for bypass has not yet been unequivocally proven. This viewpoint would support further research to reduce the probability that an incorrect adoption decision has been implemented, with the concomitant opportunity costs for suspected significant patients with TBI.

However, despite a theoretical demonstration that both necessary (EVPPI exceeds the costs of research) and sufficient (ENBS shows that marginal benefits of sampling exceed the marginal costs) conditions for future research are met, a definitive bypass trial is unlikely to be feasible. Principally, the key HITS-NS feasibility objectives of treatment compliance and incidence of significant TBI were not met. Moreover, prevailing ‘opt-in’ consent requirements would prevent any meaningful follow-up data. Trauma systems may also be a fait accompli, with clinical opinion resisting further experimental research in this area. It is also possible that the relevance of a comparison between bypass and selective transfer will continue to recede over the duration of any trial, secondary to continued regionalisation of emergency and trauma services. 121

Alternative non-randomised research designs to investigate the effectiveness of bypass are likely to be at very high risk of bias. Future cohort or case–control studies of neuroscience care compared with non-neuroscience care are critically limited by confounding, cannot provide valid evidence of comparative effectiveness and would add nothing to the existing weak evidence base. An interrupted time series study, examining outcomes before and after implementation of trauma systems, is a more promising design. However, this would require access to disability outcome information on the entire spectrum of patients to which bypass technology would apply, including patients with mild TBI, and data not available from current routine data collection sources such as Hospital Episode Statistics (HES) or TARN. 18,122 Case submissions to TARN are now linked to best-practice tariff payments to SNCs, incentivising data collection and potentially resulting in differential enrolment of patients between specialist and non-specialist centres. Additionally, the significant numbers of unmatched submissions resulting from interhospital transfers and discrepancy between TARN and HES data suggest the further potential for irresolvable selection bias.

In the contingency that bypass is thought to be cost-effective, its introduction is considered irreversible or collecting further valid evidence on its effectiveness is impossible; however, there are other areas in which future research may be beneficial. Current major trauma triage rules are primarily based on expert opinion, and results from HITS-NS and other recent studies suggest that they may have suboptimal accuracy for identifying patients with TBI requiring specialist care. 40 A cohort study designed to measure the sensitivity and specificity of major trauma triage rules would further examine this hypothesis. If corroborated, future research in this area could include derivation, validation and impact studies to improve over and undertriage rates.

Population EVPPI analyses identified several other groups of variables with a high upper bound on the returns to future research. Cohort studies investigating the incidence of relevant patient subgroups, utility values for GOS health states and post-discharge costs are likely to be cost-effective in reducing uncertainty within the decision analysis model. In contrast with a trial examining comparative effectiveness, such studies would be of shorter duration and require fewer resources. Lack of information on these parameters has been a weakness in previous TBI health-economic models and additional evidence may have considerable value in future HTAs (external to the bypass decision problem).

Pre-hospital controlled trials conducted in the UK NHS

Pre-hospital controlled trials conducted outside the UK NHS

Methods

Search terms (Table 26) were developed with reference to the US National Library of Medicine’s ‘Medical Subject Headings’ database and through contact with a researcher with extensive experience of both conducting and appraising pre-hospital controlled trials (see Contact with Dr Janette Turner, University of Sheffield, below). Search terms of more than one word were enclosed in quotation marks, and where more than one word ending was possible (e.g. injury, injuries, injured) a truncation symbol (*) was used.

Three databases were used to conduct the search: MEDLINE (via the OVID Technologies interface: http://gateway.ovid.com), CINAHL (via the EBSCOhost interface: http://ebscohost.com/academic) and the Cochrane Library (www.thecochranelibrary.com). These were chosen based on the scope of each database covering all the journals it was thought that pre-hospital controlled trials would be published in, as advised by the two above-named researchers. As well as these databases, reference searching of articles identified was also employed, along with communication with above-named authors for any trials that had been missed or any ‘grey literature’ (studies conducted and completed but unpublished).

The MEDLINE database search was conducted on the 24 March 2013 and that of the CINAHL and Cochrane Databases was conducted on 25 March 2013. An ‘autoalert’ search was set up on the MEDLINE database so that any trials which were published following the initial search, up to and including 30 June 2013, could be identified and included. Only the MEDLINE search will be described here.

Each search term was searched for individually in both the ‘title’ and ‘abstract’ fields, using the OVID ‘search fields’ function. Following this, search terms from the same key concept were combined with ‘or’ and the search terms from different key concepts were combined with ‘and’. This gave the following search strategy:

(pre-hospital or ‘out of hospital’ or ambulance or paramedic or ‘EMS’ or EMS) and

‘controlled trial’ and

(trauma or wound or injur*)

A date limit of 15 years was set. This was chosen pragmatically, based on advice from Professor Lecky (see also Contact with Dr Janette Turner, below), to try to maintain ecological validity with current ambulance practice. A date limit of less than 15 years would have missed important pre-hospital trials conducted in the UK. A date limit of greater than 15 years would mean that practices and availability of resources may have been very different from those today, making associations between older and contemporary trials meaningless.

Following the conduct of the above searches, eligibility criteria (Box 1) were applied to the papers identified. These were chosen so that the scope of the trials included in the review would be focused on the pre-hospital care of those with traumatic injuries by land-based ambulance services. It was predicted that this would give a homogenous group of studies from which meaningful comparisons and conclusions could be drawn. The eligibility criteria were first applied to the titles of the papers, then the abstracts and finally the full reports.

The trials finally selected for inclusion in the report were reviewed in terms of five criteria (Table 27) highlighted in previous reports as being key to the success of the conduct of controlled trials,132 and of pre-hospital trials in particular. 128 As this review was concerned with the success of the conduct of the trials, rather than with the efficacy of the health-care interventions they investigated, no assessment, in terms of the success of the interventions under investigation, was made. All trials were reviewed by the author using the criteria specified in Table 4 and the preliminary results were reviewed by Professor Lecky in terms of adequacy and consistency.

Contact with Dr Janette Turner, University of Sheffield

from: Nathan Chapman

to: Janette Kay Turner

date: Fri, Mar 29, 2013 at 2:00 PM

subject: Re: Student of Fiona Lecky – request for advice regarding lit review of pre-hospital RCTs

Hi Janette, I hope you’re well. I have now completed the literature search for my systematic review of pre-hospital controlled trials. I refined my research question following a preliminary review of the literature, and decided to focus on pre-hospital controlled trials regarding trauma published in the last 15-years. The 15-year limit was chosen pragmatically to include important pre-hospital trials conducted in the UK in the late 90s early 00s, but not to go so far back as to lose relevance with up-to-date pre-hospital practice. I searched for the following terms in titles or abstracts:

• (pre-hospital or pre-hospital or ‘out of hospital’ or ambulance or paramedic or ‘EMSs’ or EMS) and (trauma* or wound* or injury or injuries) and ‘controlled trial’

The eligibility criteria were as follows:

• Conducted, at least in part) in the pre-hospital environment

• Controlled trial (randomised or cluster)

• At least 50% of participants suffering trauma

• Involved, at least in part, land based ambulance services (i.e. not helicopter EMS only studies)

• Actual report of trial conduct and results (not just trial protocol)

I also kept any systematic reviews and protocols I identified during the search to identify furthers controlled trials. I identified 15 articles in medline, an additional 2 in Cochrane and no additional articles in EMBASE. Then, using the systematic review and protocols I had found, as well as the document you sent me, I identified 2 further articles, giving me a total of 19. I was wondering if when you have time you could have a quick look over this list and let me know if you think I have missed anything important? Please only do this if you have time though, I know you must be busy. Best wishes, Nathan.

from: Janette Kay Turner

to: Nathan Chapman

date: Fri, Apr 5, 2013 at 4:28 PM

subject: Re: Student of Fiona Lecky – request for advice regarding lit review of pre-hospital RCTs

Hi Nathan I’ve had a scan and i can’t think of anything to be added now you have narrowed it down to trauma. There were a few I thought of but when I checked they were old studies outside your 15-year period. Your search strategy is good and by looking and checking the systematic reviews you should have captured everything. Bit of a sad scenario isn’t it! Let me know if you need any more help. Janette

Critical appraisal tables

Parameter uncertainty sensitivity analyses