Prehospital randomised assessment of a mechanical compression device in out-of-hospital cardiac arrest (PARAMEDIC): a pragmatic, cluster randomised trial and economic evaluation

Definition

Cardiac arrest is defined as the cessation of cardiac mechanical activity, as confirmed by the absence of signs of circulation. 1 The majority of cardiac arrests outside a hospital occur as a result of cardiac causes (e.g. ischaemic heart disease, myocardial infarction, arrhythmia). Other causes of cardiac arrest include trauma, submersion, drug overdose, asphyxia, exsanguination or other medical causes (e.g. stroke, pulmonary embolus). 1,2

There are three different mechanisms through which cardiac arrest occurs – the development of an arrhythmia that leads to loss of cardiac output [ventricular fibrillation (VF) or ventricular tachycardia (VT)], insufficient cardiac contraction to generate a cardiac output, pulseless electrical activity (PEA) and a failure of the electrical conduction system of the heart (asystole). 3

The manifestations of cardiac arrest are dramatic: within seconds of it occurring the blood supply to the brain and vital organs ceases. The victim loses consciousness and the process of cell death commences. There is a narrow window of opportunity (minutes) during which, if the heart can be restarted, the victim may be successfully resuscitated. The longer the victim remains in cardiac arrest, the worse the outcome and if attempts at restarting the heart are either delayed or unsuccessful then death will occur.

Chain of Survival

The Chain of Survival (Figure 1) describes a series of steps that need to be in place to optimise the chances of survival from out-of-hospital cardiac arrest (OHCA). 4

FIGURE 1.

The Chain of Survival. Reprinted from Jerry Nolan, Jasmeet Soar, Harald Eikeland. The Chain of Survival. Resuscitation (2006), 71;270–1 with permission from the Resuscitation Council (UK) and Laerdal Medical.

Early access

The first link in the chain is early access, which highlights the importance of identifying a patient at risk of cardiac arrest (e.g. someone suffering from an acute myocardial infarction) or someone who has sustained a cardiac arrest (identified by the loss of consciousness and absence of normal breathing) and getting a trained advanced life support (ALS) team to them as rapidly as possible.

High-quality cardiopulmonary resuscitation

The second link in the Chain of Survival is early cardiopulmonary resuscitation (CPR). CPR is the combination of chest compressions and ventilations and is optimally started by those who are initially at the scene of the collapse. This is known as bystander CPR. Bystander CPR increases the odds of survival by 1.23 [95% confidence interval (CI) 0.71 to 2.11] in the studies with the highest baseline survival rates and by 5.01 (95% CI 2.57 to 9.78) in the studies with the lowest baseline rates. 5 When the emergency services arrive on scene they will take over CPR. Current resuscitation guidelines highlight the importance of high-quality CPR for ensuring optimal outcomes from cardiac arrest. 6 High-quality CPR is defined as CPR that ensures that an adequate chest compression depth is achieved (5–6 cm), the compression rate is 100–120 per minute, interruptions are minimised and the chest is allowed to recoil between chest compressions.

Evidence supporting the importance of high-quality CPR is observational: there are no randomised trials evaluating different compression parameters. Nevertheless, high-quality CPR appears to be important to outcomes. 7 Experimental studies show a linear increase in cardiac output and coronary perfusion pressure with increasing compression depths. 8,9 Observational studies in humans found improved shock success10 and better return of spontaneous circulation (ROSC) rates and long-term survival with deeper chest compressions. 11,12 Faster chest compression rates (> 100 minutes) are associated with improved survival13–15 and ensuring that the chest is allowed to recoil between sequential chest compressions also appears to be important. 16

Interruptions in CPR are harmful. 17 A particularly critical time to minimise interruptions to CPR is around the time of attempted defibrillation. Prolonged pre-shock and peri-shock interruptions in CPR reduce the chances of shock success10 and survival. 18

Early defibrillation

Approximately one-quarter of OHCA in the UK occurs as a result of an arrhythmia: either VF or VT. These rhythms are referred to as shockable rhythms, as the arrhythmias may be terminated and cardiac function restored by the successful delivery of defibrillator shocks. The time from the onset of VF/VT to the delivery of a shock is critical to shock success and chances of survival. For every 60–90 seconds that a shock is delayed, the chance of survival falls by approximately 10%. 19

If a defibrillator is immediately available at the scene of a cardiac arrest, defibrillation should be attempted without delay. Where there is a delay in initiating CPR, there is a theoretical rationale that providing CPR before a shock improves coronary perfusion and thereby the chances of achieving sustained ROSC. 20 This concept was evaluated by the Resuscitation Outcomes Consortium in a cluster randomised trial that compared early analysis [30–60 seconds of emergency medical services (EMS)-administered CPR before initial rhythm analysis] with later analysis (180 seconds of CPR, before the initial electrocardiographic analysis). 21 The primary outcome was survival to hospital discharge with satisfactory functional status [a modified Rankin Scale (mRS) score of ≤ 3, on a scale of 0–6, with higher scores indicating greater disability]. The study21 enrolled 9933 patients (5290 to early analysis and 4643 to late analysis), but found no difference in outcomes (cluster-adjusted difference of –0.2%, 95% CI –1.1% to 0.7%).

Post hoc analyses found that for ambulance services with baseline VF survival of < 20%, ‘analyse late’ compared with ‘analyse early’ was associated with a lower chance of favourable functional survival (3.8% vs. 5.5%; OR 0.67, 95% CI 0.50 to 0.90). Conversely, in ambulance services with VF survival of > 20%, ‘analyse late’ was associated with a higher likelihood of favourable functional survival than ‘analyse early’ (7.5% vs. 6.1%; OR 1.22, 95% CI 0.98 to 1.52). 22

In the UK, the Joint Royal College Ambulance Liaison Committee (JRCALC) recommended that defibrillation should not be delayed to allow for a set period of predefibrillation CPR. In practical terms, this means that when an ambulance crew arrive at the scene of a cardiac arrest they will start CPR while the defibrillator/monitor is attached. Once attached, rhythm analysis and, if indicated, defibrillation should take place without further delay.

Post-resuscitation care

The return of a spontaneous circulation marks the start of the post-resuscitation care phase of treatment. 23 Unless the arrest has been relatively brief, most patients who achieve a ROSC will have an obtunded consciousness level, necessitating admission to intensive care. The focus of the post-resuscitation care phase of treatment is upon stabilising cardiac function to prevent a further arrest and minimising the consequences of the cardiac arrest on neurological outcome. This involves the use of targeted temperature management, avoidance of hyperglycaemia and cardiac reperfusion treatments. Most post-resuscitation care treatments are initiated following arrival in the emergency department and in the intensive care unit (ICU).

Incidence and burden of disease

Data from NHS England indicate that UK NHS ambulance services attend approximately 60,000 cardiac arrests each year; of those arrests attended, resuscitation is attempted in just less than half (28,000 cases). 24 Approximately 25% achieve an initial ROSC. However, only approximately one-third of those who achieve a ROSC survive to go home from hospital; thus, the overall survival to discharge rate is approximately 8%. The burden of disease is high, with an estimated 460,000 potential years of life lost, 270,000 of which are working years of life lost.

Functional survival after cardiac arrest is generally good, with the majority of those surviving doing so with a favourable neurological outcome. 25 Survivors may experience post-arrest problems, including anxiety, depression, post-traumatic stress and difficulties with cognitive function. 26

Despite the annual death toll exceeding that of dementia, stroke or lung cancer, there has been relatively little investment in research into this lethal condition. This has created a relatively weak evidence base compared with other diseases (e.g. there are 50-fold more trials per 10,000 deaths from myocardial infarction than deaths from cardiac arrest). A review of the National Institute for Health Research (NIHR) cardiovascular portfolio identified only 4 of 624 studies related to cardiac arrest. Until recently, the pattern was similar in the USA. 27

Existing evidence

At the time of initiating the PARAMEDIC (prehospital randomised assessment of a mechanical compression device in OHCA) trial, there were no large published randomised controlled trials (RCTs) evaluating the LUCAS-2 (Lund University Cardiopulmonary Assistance System-2; Jolife AB, Lund, Sweden) device. A systematic review of the literature in 201228 identified 16 studies investigating the LUCAS-2 device. Four of the studies were animal studies and 12 were human studies. Of the 12 human studies, one was a RCT and 11 were observational studies using either a cohort or before-or-after design (Figure 2).

FIGURE 2.

Forest plot from Gates et al. ’s28 systematic review of studies examining the LUCAS-2 device. M–H, Mantel–Haenszel; RR, risk ratio. Reproduced from Heart, Gates S, Smith JL, Ong GJ, Brace SJ, Perkins GD, 98, 908–13, 2012 with permission from BMJ Publishing Group Ltd.

The main finding of this review was that the existing evidence about the use of the LUCAS device is inconclusive. The animal studies tended to provide evidence that the LUCAS-2 device improved physiological end points, although the results were not consistent across studies.

Studies involving humans similarly lack consistency in the direction of benefit compared with harm. We chose not to perform any meta-analyses because of observed heterogeneity, varying study design and the high risk of bias in most of the included studies.

We concluded that the evidence base is insufficient for making any recommendations about the routine use of the LUCAS-2 device in clinical practice.

This conclusion is similar to the International Liaison Committee for Resuscitation Consensus on Science and Treatment’s recommendation,20 which advised that:

. . . there are insufficient data to support or refute the use of LUCAS-2 CPR instead of manual CPR. It may be reasonable to consider LUCAS-2 CPR to maintain continuous chest compression while undergoing computed tomography (CT) scan or similar diagnostic studies, when provision of manual CPR would be difficult.

Deakin et al. 20

Rationale for intervention

Because of the problems with manual chest compression, several mechanical devices have been proposed. These have potential advantages: they are able to provide compressions of a standard depth and frequency for long periods without interruption or fatigue and they free emergency medical personnel to attend to other tasks.

The LUCAS-2 is a mechanical device that provides automatic chest compressions. It delivers sternal compression at a constant rate, to a fixed depth, by a piston with the added feature of a suction cup that helps the chest return back to the normal position. It compresses 100 times per minute to a depth of 4–5 cm. It is easy to apply, stable in use, relatively light in weight (7.8 kg) and well adapted to use during patient movement on a stretcher and during ambulance transportation. The device is Conformité Européenne (CE) marked and has been on the market in Europe since 2002.

Detailed descriptions of the device and experimental data from animal studies showing increased cardiac output and cortical cerebral flow compared with manual standardised CPR have been published. 35,38

The LUCAS-2 device was introduced into a small number of ambulance services in the UK several years ago, despite the absence of evidence of its effectiveness from randomised trials. 39 It was subsequently withdrawn from routine use by several of the services because of lack of evidence about safety and efficacy and is now used only under restricted conditions. In the absence of evidence of clinical effectiveness or cost-effectiveness and the presence of some concerns regarding safety, the JRCALC, in discussion with the Department of Health, identified the need for large-scale clinical trials to evaluate the device. 40 Until such studies are completed, no further new purchases of the device are recommended by the JRCALC. A briefing note commissioned by the National Institute for Health and Care Excellence (NICE) concluded that ‘there is therefore an urgent need to evaluate this technology to discover whether it is effective and cost-effective in improving survival after cardiac arrest’. The need for a definitive trial is reinforced in the International Liaison Committee for Resuscitation’s analysis of knowledge gaps in resuscitation. 41

Primary objective

To evaluate the effect on mortality of using the LUCAS-2 device rather than manual chest compression during resuscitation by ambulance clinicians (paramedics, technicians, emergency care assistants, etc.) after OHCA at 30 days after the event.

Secondary objectives

To evaluate the effects of LUCAS-2 treatment on survival to 12 months, the cognitive and neurological outcomes of survivors and the cost-effectiveness of the LUCAS-2 device.

Selection of trial sites

The NHS organisations that delivered the trial were ambulance trusts. Initially (at the time of the funding application), we anticipated that three trusts would participate covering the West Midlands, Wales and Scotland. However, the Scottish Ambulance Service withdrew from the trial before the start of recruitment because its legal team was not happy with the research contract. The North East Ambulance Service and South Central Ambulance Service subsequently joined the trial.

Primary outcome

Survival to 30 days post cardiac arrest.

Secondary outcomes

• Survived event (sustained ROSC, with spontaneous circulation until admission and transfer of care to medical staff at the receiving hospital).

• Survival to hospital discharge (the point at which the patient is discharged from the hospital acute care unit regardless of neurological status, outcome or destination).

• Survival to 3 and 12 months.

• Health-related quality of life (HRQL) at 3 and 12 months [Short Form questionnaire-12 items (SF-12)43 and EuroQol-5 Dimensions (EQ-5D)]. 44

• Neurologically intact survival to 3 months [survival with a Cerebral Performance Category (CPC)45 score of 1 or 2].

• Cognitive outcome at 12 months [as measured via the Mini Mental State Examination (MMSE)]. 46

• Anxiety and depression at 12 months [as measured via the Hospital Anxiety and Depression Scale (HADS)]. 47

• Post-traumatic stress at 12 months [as measured via the post-traumatic stress disorder (PTSD) Civilian Checklist (PCL-C)]. 48

• Hospital length of stay.

• Intensive care length of stay (see Appendix 1).

The outcomes defined by the Utstein convention1 for reporting outcomes from cardiac arrest are reported, as well as long-term follow-up at 12 months. We did not measure the incidence of injuries resulting from CPR, for three reasons: first, they are of little importance unless they result in differences in more substantive outcomes such as survival or duration of hospitalisation; second, they are difficult to measure and classify and may not be detected reliably; and, third, organising injury data collection from a large number of hospitals was felt to add significant organisational complexity to the trial, for little benefit.

The CPC score is a 5-point scale for describing the neurological outcome after cardiac arrest and is recommended by the Utstein guidelines. 1 There is a generally accepted split into good neurological outcome (CPC score of 1 or 2) and poor outcome (CPC score of 3–5). The definitions of the categories are:

• CPC 1 Good cerebral performance: conscious, alert, able to work.

• CPC 2 Moderate cerebral disability: conscious, sufficient cerebral function for independent activities of daily life. Able to work in sheltered environment.

• CPC 3 Severe cerebral disability: conscious, dependent on others for daily support because of impaired brain function. Ranges from ambulatory state to severe dementia or paralysis.

• CPC 4 Coma or vegetative state: any degree of coma without the presence of all of the brain death criteria.

• CPC 5 Brain death.

However, recent studies have demonstrated that this score may be insensitive to some of the more subtle, but nevertheless important, longer-term neurocognitive and functional impairments that are experienced by survivors of cardiac arrest. 49,50 The spectrum of impairment of HRQL following cardiac arrest includes memory and cognitive dysfunction, affective disorders and PTSD. 48 The number of patients who are expected to survive to hospital discharge was anticipated to be in the region of 200–300, which allowed more intensive follow-up. We used four clinical outcome measures: the SF-12 is a standard quality-of-life measure that is short and easy to complete. The PTSD PCL-C48 is a 17-item questionnaire measuring the risk of developing PTSD and has been used in previous studies as a good surrogate for the clinical diagnosis of PTSD, which would require a face-to-face interview by a suitably trained professional. The HADS47 is a 14-item self-administered questionnaire, which has been previously used successfully to measure affective disorders in those surviving cardiac arrest. 51 The MMSE measures cognitive impairment. 52 In addition, the EQ-5D44 was used as a health utility measure for the health economic analysis.

Eligibility for clusters

All vehicles that were in service at each participating ambulance station, eligible to attend patients and could carry the device were included in the trial and randomised to one of the trial arms before the start of recruitment.

To maximise the efficiency of the trial, recruitment was concentrated predominantly in urban areas, where each vehicle would attend a higher number of cardiac arrests per year. This avoided the costs of supporting clusters in rural areas that were able to recruit very few patients, increased the size of clusters and increased the survival rate for the trial population by omitting patients who could not be reached quickly and had a very low chance of survival. These considerations should all have helped to improve power to detect a difference between the LUCAS-2 device and manual compression arms in the trial.

Eligibility for individual patients

Patients were eligible if they met all four of the criteria below:

1. Cardiac arrest in the out-of-hospital environment.

2. First ambulance resource was a trial vehicle.

3. Resuscitation attempt was initiated by the attending ambulance clinicians, in accordance with JRCALC guidelines. 53

4. The patient was known, or believed to be, aged ≥ 18 years.

The exclusion criteria were:

• cardiac arrest caused by trauma

• known or clinically apparent pregnancy.

All patients who fulfilled the eligibility criteria were included in the trial. The JRCALC Recognition of Life Extinction (ROLE) guidelines53 were applied to determine patients for whom a resuscitation attempt was inappropriate. This is the case when there is no chance of survival; when the resuscitation attempt would be futile and distressing for relatives, friends and health-care personnel; and when time and resources would be wasted undertaking such measures. If there was clear evidence that life was extinct (Box 1) or if the patient had a ‘do not attempt resuscitation’ order, ambulance staff were authorised to recognise death and withhold CPR.

The LUCAS-2 device cannot be used if patients are too large or too small: the device fits patients with a sternum height of 17.0–30.3 cm and a chest width of < 45 cm. However, patient size was not an exclusion criterion because it would be impossible to apply correctly to the manual compression group, hence potentially introducing bias. Moreover, it was appropriate to include the small proportion of patients who were too large or too small for the LUCAS-2 device in the trial, in accordance with intention-to-treat (ITT) principles. The trial estimated the impact of the LUCAS-2 device on the survival rate among the whole cardiac arrest population. In one Swedish study,29 only 3 out of 159 patients (1.9%) were found to be too small or too large for the LUCAS device. We therefore anticipated that there would be only a small number of patients for whom the LUCAS-2 device could not be used, especially as the LUCAS-2 device accommodates larger patients than the LUCAS version 1 device that was used in the Swedish study.

Ethical considerations

The occurrence of a cardiac arrest out of hospital is unpredictable. Within seconds of cardiac arrest, a person becomes unconscious and thus incapacitated. It was not therefore possible to obtain prospective consent directly from the research participant.

Treatment (in the form of CPR) must be started immediately in an attempt to save the person’s life. In this setting it was not practical to consult a carer or independent registered medical practitioner without placing the potential participant at risk of harm from delaying treatment.

Conducting research in emergency situations in which a patient lacks capacity is regulated by the Mental Capacity Act (2005)54 for England and Wales. The PARAMEDIC trial was approved in accordance with these requirements by the Coventry Research Ethics Committee (REC) (reference number 09/H1210/69). Ethics approval was also gained from the Scotland A REC, but the Scottish Ambulance Service subsequently withdrew from the study.

Approaching survivors

The nature of the condition meant that the majority (85–90%) of people in the study would not survive. Of those patients admitted to hospital alive, the majority (approximately 80%) would be comatose and admitted to an ICU (and thus remain incapacitated). Following admission to intensive care, approximately half of the people who initially survive die without regaining capacity (on average within 48 hours). The average duration of hospital stay for survivors is 18 days. 55

To avoid unnecessary distress to the relatives of the deceased, the timing of the approach was important and had to balance the need to inform at an early opportunity while determining – as accurately as possible – which patients had died. Pilot work for this trial established that it is not possible for ambulance services to determine with sufficient accuracy which patients have died, so the procedure was revised, based on the procedures of the ICON (Intensive Care Outcome Network) study. 56

The participating ambulance services conducted their own checks on patients’ survival using existing data systems, which differed between services. Where possible, ambulance services consulted the NHS Patient Demographics Service. Other checks carried out by either the ambulance service and/or the Warwick Clinical Trials Unit (WCTU), included contacts with hospitals, general practitioners (GPs), the Intensive Care National Audit and Research Centre (ICNARC) and local registrars of births and deaths (see Identification of cardiac arrests).

If a patient was transported to hospital, his/her clinical and contact details were sent to the study co-ordinating centre at the WCTU. Staff at the co-ordinating centre checked the status of each potential survivor with the Medical Research Information Service (MRIS) approximately 6 weeks after the patient’s cardiac arrest. This timing of the approach was selected to ensure that the majority of deaths had been included in the MRIS database. All of the survivors were flagged on the MRIS database and the trial co-ordinating centre was informed of any subsequent deaths.

After these checks, if someone was still believed to be alive then the co-ordinating centre contacted the person at his/her home address by letter to provide information about the study and the follow-up. If there was no response after 2 weeks, then the co-ordinating centre tried to contact the patient by telephone (if the telephone number was known) or by letter. If the patient wished to take part in the follow-up, then they could contact the WCTU by using the reply slip, telephone or e-mail. This gave the participants an opportunity to discuss the study and, if they were happy to proceed, a 3-month follow-up appointment was made. The consent form was either returned by post or signed at the 3-month follow-up visit. Patients who did not respond were approached again at 10 months post cardiac arrest in an attempt to invite them to participate in a 12-month visit.

In the event that the co-ordinating centre was notified (or had reason to believe) that a patient lacked capacity, an approach was made to his/her GP in order to establish if the patient had capacity to consent. In the event that a patient lacked capacity to consent, we sought the views of a personal consultee in order to establish the patient’s wishes. If a personal consultee could not be identified, a carer (unconnected with the study) determined if the patient would be likely to consent to follow-up.

Cluster design

One of the major potential sources of bias in cluster randomised trials is selection bias, which can arise if different patients are selected for inclusion in the two trial arms. This can arise when clinicians or people selecting patients are aware of the allocation of the cluster and they may consciously or unconsciously apply inclusion criteria differently depending on the randomised intervention. There is greater scope for selection bias if a large proportion of potentially eligible patients are not included in the trial. In this trial, paramedics assessing patients for inclusion were aware of the allocations; however, we aimed to identify and include close to 100% of the eligible patients, using a combination of methods for identifying eligible patients (see Eligibility for individual patients). This should avoid most selection bias.

Threshold for resuscitation

As the ambulance clinicians who were delivering the interventions were not blinded, there was a possibility that bias could be introduced by different thresholds for resuscitation between the LUCAS-2 and standard care arms: if they believed strongly that the LUCAS-2 device was effective, some of them might have attempted resuscitation in the LUCAS-2 arm on patients who had no chance of survival, and for whom a resuscitation attempt was therefore inappropriate. This would have resulted in a group of patients with a very low probability of survival being recruited to the LUCAS-2 arm but not to the standard care arm, potentially masking any beneficial effect of the LUCAS-2 device. We used several strategies to prevent this bias from occurring, to detect if it was to happen and to correct it if necessary.

First, the criteria that were used to determine whether or not a resuscitation attempt was appropriate, and hence whether or not the patient was eligible, were as objective as possible. The JRCALC ROLE criteria53 were used by all of the participating ambulance services to determine when a resuscitation attempt is inappropriate, and this continued in the trial (see Eligibility for individual patients). Ambulance clinicians were therefore familiar with the application of these criteria and no change of practice was needed during the trial. However, there remained scope for differential application of the criteria to the two trial arms, so further strategies were devised.

Second, all ambulance clinicians in the trial were trained in the trial procedures57 to ensure that they understood the rationale for the trial and the importance of following the trial procedures. The training included a review of existing evidence so that participating ambulance clinicians understood the current position of equipoise regarding the effectiveness of the LUCAS-2 device and discussion of potential sources of bias in the trial and the importance of applying the inclusion/exclusion criteria rigorously to both arms. Training continued throughout the recruitment period to ensure that any new staff were trained before recruiting and that important messages were continually and correctly reinforced (see Appendix 2).

Third, we instituted a programme of regular monitoring of the characteristics of patients who were recruited to the two trial arms, the number of cardiac arrests in each arm when no resuscitation attempt was made and the proportion of cardiac arrests included in the trial in order to detect any imbalances that may be caused by different thresholds for resuscitation. We also monitored the presenting rhythm, proportion of witnessed and unwitnessed arrests, presence of bystander CPR and time from ‘999’ call to crew arrival (using ambulance computer log data). If a lower threshold for attempting resuscitation in the LUCAS-2 arm existed, then we would find a greater number of recruits and a greater proportion of cardiac arrests with resuscitation attempts, a greater proportion with unfavourable presenting rhythms, a lower proportion of witnessed arrests and with bystander CPR and longer times from ‘999’ call to start of resuscitation in the LUCAS-2 group.

Finally, if necessary, we corrected for any inclusion bias in the statistical analysis of the trial, by adjustment of the analysis to take account of imbalance in factors such as presenting rhythm, time since ‘999’ call and presence of bystander CPR. We expected any potential inclusion bias to affect only the group of patients who were least likely to survive and that it would not affect patients in whom a resuscitation attempt would always be made (e.g. those with presenting rhythms with the highest probability of survival) and therefore a comparison between the LUCAS-2 device and manual compression in the subgroups of patients in whom resuscitation was known to be appropriate would be unaffected.

Intervention arm

The LUCAS-2 device used in the trial was the latest version of the LUCAS-2 device, manufactured by Jolife AB and distributed by Physio-Control UK, Watford, UK.

Patients who were allocated to the intervention arm (LUCAS-2 device) received mechanical chest compressions in place of standard manual chest compressions.

On arrival, after confirming cardiac arrest, manual CPR was commenced while the LUCAS-2 device was prepared and applied. Following this, the initial cardiac rhythm was assessed. If the patient was in VF or VT then a countershock was administered in accordance with JRCALC/ALS guidelines. Operational experience showed that the LUCAS-2 device could be deployed within 20–30 seconds of arrival at the patient’s location. Prior to intubation, compressions were provided using the 30 compressions/two ventilations mode. If the patient was intubated, asynchronous compressions and ventilations were provided, with a ventilation rate of 10 per minute.

Defibrillation was performed using the following sequence: pause LUCAS-2 device, analyse heart rhythm; if shock indicated, restart LUCAS-2 device, charge, deliver shock and continue CPR for 2 minutes. This minimised deleterious pre- and post-shock pauses in compressions. The LUCAS-2 device was used in place of standard chest compressions as long as continued resuscitation was indicated, including resuscitation in the field and during transport to hospital.

The trial intervention ceased after care was handed over to the medical team in the hospital or the patient was declared deceased according to the ROLE criteria. 53

Manual chest compression arm

On arrival, after confirming cardiac arrest, manual CPR was commenced and the initial cardiac rhythm was assessed. If the patient was in VF or pulseless VT, then a countershock was applied. Prior to intubation, compressions were provided using the 30 compressions/two ventilations mode. If the patient was intubated, asynchronous compressions and ventilations were provided, with a ventilation rate of 10 per minute.

Minimising interruptions in chest compressions is critical for optimising the chances that a shock is successful. However, it is currently considered unsafe to perform defibrillation during manual chest compression. Defibrillation was therefore performed using current UK recommendations, which are as follows: stop CPR; analyse heart rhythm, charge defibrillator, deliver shock, restart chest compressions and continue CPR for 2 minutes.

Definitions

Additional terms for device trials

For trials of devices, additional terms are used and are defined as follows.

Events that should be reported

All AEs and SADEs and incidents were reported to the trial co-ordinating centre on the appropriate forms (see Appendix 3).

All of the patients in this trial were in an immediately life-threatening situation; many would not survive and all of those who did were hospitalised. These situations were, therefore, expected, and events leading to any of them were reported as SAE/SADEs only if their cause was clearly separate from the cardiac arrest. Events that were related to cardiac arrest and would be expected in patients undergoing attempted resuscitation (including death and hospitalisation) were not reported.

Therefore, events were reported as SAE/SADEs if they were:

• serious

• and were potentially related to trial participation (i.e. they may have resulted from study treatment such as use of the LUCAS-2 device)

• and were unexpected (i.e. that the event was not an expected occurrence for patients who have had a cardiac arrest).

Examples of events that may be SAE/SADEs were the use of the LUCAS-2 device causing a new injury that endangered the patient, malfunction of the device causing injury to ambulance clinicians and malfunction of the device leading to inadequate chest compression.

Reporting serious adverse events

Events satisfying the criteria given above were reported to the study co-ordinating centre, using the event report form, as soon as they became apparent (see Appendix 3).

The SAE/SADE reports that were received by the co-ordinating centre were reviewed on receipt by the chief investigators, and those that were considered to satisfy the criteria for being related to the device and unexpected were notified to the main REC, the Medicines and Healthcare products Regulation Agency (MHRA) and manufacturer within 15 days of receipt.

The SAE reports were also reviewed by the Data Monitoring Committee (DMC) at its regular meetings. AEs that were not considered to be serious were logged and included in annual progress reports.

Identification of cardiac arrests

Data were recorded on all of the cardiac arrests within the trial areas. This allowed assessment of the proportion of cardiac arrests that were enrolled into the trial and helped to ensure that no eligible cardiac arrests were missed. Data were collected by the attending ambulance clinicians, using the routinely completed PRF. Data from the PRFs were then transcribed on to the trial case report forms (CRFs) by the trial research fellows (see Appendix 4). The data collection was retrospective and each ambulance service had its own system, which meant that identifying cardiac arrest cases was challenging. The ambulance services, generally, had more than one system for identifying cardiac arrest cases to ensure that none was missed. One ambulance service had to go through all daily paper PRFs on station; others had electronic PRFs and were able to perform searches through a database. Other methods included linking in with national reporting of cardiac arrests to the Department of Health.

Data forms were collected in a central place at participating ambulance stations and collected by research paramedics on a weekly basis. For ineligible cardiac arrests (no resuscitation attempt, aged < 18 years, pregnant, traumatic aetiology, non-trial vehicle was first on scene), the ambulance service also sent the trial co-ordinating centre details of the arrests for monitoring purposes.

Hospitals did not undertake prospective data collection for trial participants because of the logistical difficulties that this would present. Hospitals were contacted, as necessary, to seek information about whether patients had been discharged or had died in hospital before contacting them for follow-up. Further details about length of stay in the ICU and hospital were also sought from hospitals, ICNARC and Hospital Episode Statistics (HES). Authority was granted by the Confidentiality Advisory Group to collect these data without seeking consent from the next of kin.

Deaths

Data management

All of the data collected during the trial were handled and stored in accordance with the Data Protection Act 1998. 60 Data were, as far as possible, anonymised, but this trial involved the use of identifiable personal data for follow-up. All transfer of data between ambulance services and the study co-ordinating centre used secure methods, such as encrypted e-mail. All of the study data were entered into a study-specific database, which was set up by the programming team at the WCTU at the start of the study. All specifications (i.e. database variables, validation checks, screens) were agreed between the programmer, statistician, chief investigators and trial co-ordinator. All trial documentation and data were archived after completion of the trial and are stored in accordance with the WCTU standard operating procedures61 (see Appendix 7).

Power and sample size

Statistical analysis

We performed ITT analyses to estimate the treatment effect of the LUCAS-2 device and presented results as a point estimate (RR or mean difference), with uncertainty estimated by the 95% CI.

We also used CACE analyses to estimate the effect in cardiac arrests when the protocol was followed. 66 CACE estimates the treatment effect in people who were randomly assigned to the intervention and who actually received it, by comparing compliers in the intervention group with those participants in the control group who would have been compliers if they had been allocated to the intervention group. This analysis retains the advantages of randomisation and avoids introducing bias, hence CACE is preferred to per-protocol analysis. 67 CACE assumes that the probability of non-compliance with the LUCAS-2 device would be the same for people who were actually randomised to the control as for those who were randomised to the LUCAS-2 arm. If allocation is random, this assumption will hold. A second assumption is that outcomes are not affected just by being randomised to the LUCAS-2 or control groups; in other words, there is no systematic difference in outcomes between patients attended by LUCAS-2 device-containing vehicles and those attended by control vehicles, except that caused by the different treatments provided. CACE analysis enables us to estimate the unobserved proportion of the control group who would have been non-compliers if randomised to the LUCAS-2 group. We did two CACE analyses, defining compliers in different ways. In CACE 1, we treated as non-compliant those cases in which the LUCAS-2 device was not used for unknown or trial-related reasons that would not occur in real-life clinical practice (e.g. crew was not trained in trial procedures, crew misunderstood the trial protocol, the device was missing from the vehicle). This analysis omits trial-related non-use and should be a better estimate of the treatment effect in real-world clinical practice than an ITT analysis. In the CACE 2 analysis, we treated as compliant only those cases in which the LUCAS-2 device was actually used and this analysis therefore estimates efficacy, that is, the treatment effect in patients who received LUCAS-2 device treatment.

For ITT analyses, we used logistic regression models to obtain unadjusted and adjusted ORs and 95% CIs. The prespecified covariates used in the adjusted models were age, sex, response time, bystander CPR and initial rhythm. We attempted adjusting for the clustering design using multilevel logistic models using the Statistical Analysis Software (SAS) GLIMMIX procedure with logit link function based on the binomial distribution. Because of the extremely low survival rates in each cluster (vehicle), the multilevel models could not be fitted with the vehicle random effect, as this effect was not estimable. As a result, ordinary logistic regressions were fitted. We also undertook prespecified subgroup analyses by (1) initial rhythm (shockable vs. non-shockable); (2) cardiac arrest witnessed versus not witnessed; (3) type of vehicle (RRV vs. ambulance); (4) bystander CPR versus no bystander CPR; (5) region; (6) aetiology (presumed cardiac or non-cardiac); (7) age; and (8) response time. The analyses by region and type of vehicle were added during recruitment on the recommendation of the TSC. We fitted logistic regression models for the primary outcome measure with the inclusion of an interaction term to examine whether or not the treatment effect differed between the subgroups. Age and response times are continuous variables and we assessed these using multivariate fractional polynomials68 (see Appendix 8).

We did all analyses using SAS version 9.3 (SAS Institute, Marlow, UK). Interim analyses were conducted at least once per year during recruitment and supplied confidentially to the DMC. The DMC considered the results of the interim analysis and made recommendations to the TSC about continuation of recruitment or any modification to the trial that may have been necessary.

Recruitment of clusters

Four hundred and eighteen emergency vehicles (287 double-manned ambulances and 131 single-manned RRVs), at 86 ambulance stations, were included in the trial. One hundred and forty-seven vehicles (100 ambulances and 47 RRVs) were assigned to the LUCAS-2 group and 271 clusters (187 ambulances and 84 RRVs) were randomised to the control group (Table 3 and Figure 4). The overall LUCAS-2-to-control ratio was 1 : 1.8 (1 : 1.87 for ambulances and 1 : 1.79 for RRVs). Seventy-two clusters were terminated early during the recruitment period, for a variety of reasons, including the vehicle being taken out of front-line service, scrapped or transferred to a station that was out of the trial area.

FIGURE 4.

The Consolidated Standards of Reporting Trials (CONSORT) diagram for vehicles. a, A vehicle can be counted twice, as it can deliver/not deliver intervention on different occasions. Seventy-two (17%) of vehicles changed status. CA, cardiac arrest; CAT A, category A call.

Stations were opened for recruitment and vehicles were randomised between April 2010 and February 2013, so the duration of recruitment for each cluster varied between 4 and 38 months. The number of patients recruited by each vehicle was also very variable, ranging from 0 to 41 patients (Figure 5), with a mean cluster size of 13 patients in the LUCAS-2 arm and 11 in the control arm. Some of the zeros are accounted for by five stations with 17 vehicles that never commenced active recruitment. Individual ambulance staff attended on average 4.1 eligible cardiac arrests [standard deviation (SD) 3.6 eligible cardiac arrests] in the control group and 3.0 eligible cardiac arrests (SD 2.3 eligible cardiac arrests) in the LUCAS-2 group over the study period.

FIGURE 5.

Number of vehicles attending different numbers of cardiac arrests.

Recruitment of patients

Patients were recruited between 15 April 2010 and 10 June 2013. Recruitment began first in the West Midlands (start date 15 April 2010) and, subsequently, in the other regions (start dates: Wales, September 2011; South Central, October 2011; North East, May 2012). During the recruitment period there were 16.019 cardiac arrests attended by vehicles from participating stations and trial vehicles attended 11,171 emergency incidents. Cardiac arrest was confirmed and resuscitation was attempted in 4689 cases, of which 218 cases were ineligible and excluded. In total, 4471 patients were therefore enrolled in the study (Figure 6).

FIGURE 6.

Patient flow chart. a, Reasons that the LUCAS-2 device was not used: crew not trained, 78; crew error, 168; no device in vehicle, 26; unsuitable patient, 102 [patient too large, 58; patient too small, 22; other reasons (e.g. chest deformity), 22]; device issues, 14; not possible to use device, 140; and reason unknown, 110. Reasons for the LUCAS-2 device use in control arm were crew error.

In the LUCAS-2 arm, 638 cases received manual chest compression and in the control arm 11 cases received LUCAS-2 chest compression (see footnote a to Figure 6).

The majority of patients were recruited in the West Midlands (2723/4471, 60.9%), with smaller proportions in the other three regions. RRVs were the first vehicle in attendance for 1635 out of 4471 (36.6%) recruits, whereas ambulances were the first vehicle on scene for 2836 out of 4471 (63.4%) recruits (Table 4).

Differences in baseline characteristics between the groups were small (Table 5). Slightly more patients in the manual CPR group had cardiac arrest at home [2336/2819 (82.9%) vs. 1336/1652 (80.9%)] and witnessed arrest [1749/2819 (62.0%) vs. 1001/1652 (60.6%)].

Cumulative recruitment through time is shown in Figure 7.

FIGURE 7.

Cumulative recruitment: April 2010 to June 2013. Dotted line represents the original target recruitment; dashed line represents the revised target recruitment; and the green solid line represents actual recruitment.

Treatment

A substantial proportion of the patients who were randomised to the LUCAS-2 group (638/1652, 38%) did not receive mechanical chest compressions. This was expected because in clinical practice there would be occasions in which the LUCAS-2 device would not be used and it is appropriate to include these cardiac arrests in a pragmatic trial. The LUCAS-2 device was used for 985 out of 1652 patients (60%) in the LUCAS-2 arm; in 29 out of 1652 (1.8%) cases the intervention used was unknown. The LUCAS-2 device was not used for trial-related reasons in 272 cases and 256 cases of non-use were classified as being for reasons that would occur in normal clinical practice. The reason for non-use was not known for 110 cases. In the control arm, the LUCAS-2 device was used in 11 out of 2819 cases (0.4%) as a result of crew error.

Treatments used were similar between the trial arms, although the proportion of patients receiving intravenous drugs was slightly higher in the LUCAS-2 group (1366/1652, 82.7%) than in the control group (2255/2819, 80.0%) (Table 6).

Follow-up

The number of patients followed up with visits at 3 and 12 months was limited by lack of response to the initial contact and patients declining participation in the follow-up (Table 7): 146 out of 278 survivors (52.5%) were followed up at 3 months and 143 out of 264 survivors (54.2%) at 12 months. The proportion of patients visited was slightly lower in the LUCAS-2 group than in the control group (49.0% vs. 54.1%), raising the possibility that there may have been differential losses between the trial arms (Table 8).

Outcomes: intention-to-treat analysis

Survival status could not be ascertained for one patient, who was from overseas and returned home < 30 days after the cardiac arrest, but it was known up to 12 months for all of the other participants. In the ITT analysis, 30-day survival was similar in the LUCAS-2 and control groups [LUCAS-2, 104/1652 (6.3%); control, 193/2818 (6.8%); adjusted OR 0.86, 95% CI 0.64 to 1.15].

The analyses did not show a clear advantage to the LUCAS-2 group for any outcome; there was very little effect of the LUCAS-2 device on ROSC (OR 0.99, 95% CI 0.86 to 1.14) and survived event (OR 0.97, 95% CI 0.82 to 1.14), but for survival at 30 days, 3 months and 12 months the point estimates favoured manual chest compression, although the 95% CIs included 1. The number of patients surviving with a favourable neurological outcome (i.e. a CPC score of 1 or 2) was lower in the LUCAS-2 group than in the control group (adjusted OR 0.72, 95% CI 0.52 to 0.99; Table 9).

Complier average causal effect analyses

The CACE 1 analysis included cases for which the LUCAS-2 device was not used in situations when it would not be used in normal clinical practice. Its results were similar to the ITT analysis, which suggests that there is no advantage to the LUCAS-2 device when used as it would be in routine clinical practice. In the CACE 2 analysis, which included only cases for which the LUCAS-2 device was actually used in the intervention group, the effects of the LUCAS-2 device, again, were similar to the ITT analysis, although the estimate for survival with a CPC score of 1 or 2 was slightly more extreme. Therefore, there was not a substantial difference in the treatment effect of the LUCAS-2 device when it was actually used. A concern with the ITT analysis is that because it includes a large number of non-compliant cases, any benefit among patients for whom the device was actually used would be diluted by a lack of effect among cases that were non-compliant, leading to underestimation of the treatment effect. The CACE analyses demonstrate that the lack of advantage to the LUCAS-2 device in the ITT analysis is not due to dilution of the treatment effect (Table 10).

Subgroup analyses

In the subgroup analyses there was no evidence of different treatment effects between the subgroups for 30-day survival according to whether or not the cardiac arrest was witnessed; the type of vehicle (ambulance or RRV); whether or not the patient received bystander CPR; aetiology; and region (Table 11). In the subgroup analysis by initial rhythm, there was a difference in treatment effect between patients with a shockable initial rhythm and those with PEA or asystole. Survival was lower in the LUCAS-2 group than in those with shockable initial rhythms (OR 0.71, 95% CI 0.52 to 0.98), but higher in the PEA/asystole group (OR 1.38, 95% CI 0.80 to 2.36).

The analyses of age and response time using fractional polynomial models did not show any interaction effects of these variables with the treatment effect of the LUCAS-2 device. Therefore, we did not find any evidence that the treatment effect of the LUCAS-2 device differs depending on the age of the patient or ambulance service response time.

Serious adverse events

Seven clinical AEs were reported in the LUCAS-2 group (three events of chest bruising, two of chest laceration and two of blood in mouth). No SAEs were reported. Fifteen device incidents occurred during operational use (four incidents in which alarms sounded, seven in which the device stopped working and four other device incidents). No adverse or SAEs were reported in the control group.

Follow-up at 3 months and 12 months

We considered using multiple imputation (MI) to attempt to correct for the effects of missing data, but we felt that the assumptions of any imputation procedure were unlikely to hold, given the large number of missing data and the high probability that missingness was related to outcome. Analyses therefore used available cases only, with no imputation.

At the 3-month follow-up (Table 12), SF-12 mental and physical scores and EQ-5D quality-of-life scores were slightly lower in the LUCAS-2 group than in the control group.

In the 12-month follow-up (Table 13), all of the results were in the same direction, indicating worse outcomes in the LUCAS-2 group. For some of the outcomes, the 95% CIs excluded zero, whereas for others the data were compatible with a zero or small positive effect.

All of the follow-up results should be interpreted cautiously because of the low percentage of patients known to be alive who were included in the follow-up. This was caused by refusal of consent and non-response to the invitation to participate in the follow-up and led to high numbers of missing data, with a consequent risk of bias. It is possible that those who declined to take part and those who did not respond may have had worse outcomes than those who responded. Slightly fewer people were followed up in the LUCAS-2 group, which could have been as a result of worse outcomes in that group.

The results of the ‘quality of CPR study’ are shown in Table 14. Data recording problems in one ambulance service meant that only compression depth was analysable from this service, so data for both compression rate and compression fraction were available from only three services.

Compression depth was fairly consistent across services, with the mean varying between 41 and 49.4 mm. Mean compression rate was substantially faster in one service than the other two that had data, and exceeded the guideline recommended rate of 120 compressions per minute. Compression fraction was very consistent across the three services that had data, at between 65% and 66%.

Within-trial analysis

The within-trial analysis aimed to determine the cost-effectiveness of the LUCAS-2 device compared with manual chest compression over the period of the trial (i.e. from cardiac arrest to 12 months’ follow-up). The analysis used QALYs as the main outcome and adopted the perspective of the NHS and PSS. Utility values were derived from patient questionnaires; resource use was obtained from a variety of sources, including the trial CRFs, large data sets (i.e. HES, ICNARC) and self-completed patient questionnaires. Neither costs nor QALYs were discounted given the 12-month time period. The results are reported as incremental cost-effectiveness ratios (ICERs).

Quality-adjusted life-years

Quality-adjusted life-years reflect both duration and quality of life and their estimation requires the production of utility weights for each health state observed in the trial population. HRQL was assessed using the EQ-5D,42 which has been validated for use in the critical care patient group. 80 Surviving patients completed the EQ-5D at 3 and 12 months post cardiac arrest. The EQ-5D responses were converted to health-state utility values using the UK tariff. 81 Utility values were combined with survival information to calculate QALYs for the trial period using an area under the curve (AUC) approach. As patients were unable to complete the measure at baseline, estimates had to be made of their baseline utility level. Following strategies previously used in studies that have dealt with this scenario,82–84 we assumed that patients who experienced a cardiac arrest had a baseline utility value of ‘0’ (which is equivalent to dead). We then assumed a linear transition from ‘0’ to the 3-month utility value and, similarly, from the 3- to the 12-month utility value. A utility weight of zero was assigned to patients who died within 3 months, which may underestimate total QALYs. We explored an alternative assumption in the sensitivity analysis, for which the survival days of these patients were assigned the average 3-month utility estimated in our sample.

Resource use and costs

The costs considered in this analysis included intervention costs (i.e. cost of the LUCAS-2 device and ambulance costs), costs of hospital inpatient stays, accident and emergency (A&E) admissions and outpatient visits and the use of primary care and community-based health and social care services (such as GP and social worker visits).

Resource use data were collected prospectively and retrospectively. Hospital resource utilisation was obtained through linkage with the HES data set. We extracted data from the HES for study participants from cardiac arrest to 12 months after randomisation. The data set records information on the total number of days in hospital, the number of in-person and telephone outpatient visits and the number of A&E admissions. To identify the number of hospital days spent in the ICU (Table 16) we used information from the ICNARC data set. Patients who stayed for < 24 hours in hospital were also identified. Some of these were regular admissions and, based on the profile of their procedural use, were assigned the cost of 1 day in ICU. Another group of patients who stayed for < 24 hours in hospital were recorded in HES as day case or regular day admissions. Hospital costs were obtained by multiplying the number of days or visits of each service by the corresponding unit cost derived from NHS reference costs databases. 85

Following hospital discharge, health-care resource use questionnaires were completed by surviving patients at 3 and 12 months post cardiac arrest. Patients were asked about their use of health and social services during the previous 3 months, including further inpatient and outpatient care and primary and community-based health and social services. For the 6-month period during which non-hospital resource use data were not collected (between 3 and 9 months post cardiac arrest), we used the average resource utilisation between the initial (0–3 months) and final (9–12 months) period. Patients who died within 3 months were assumed to have incurred no community-based health and social care costs. Health-care resource use was multiplied by the relevant unit costs extracted from the national reference costs (Table 17). 86

A microcosting study was undertaken to establish the cost of the LUCAS-2 device and determine the relevant cost per application. This included (1) the cost of purchasing the device and accessories; (2) the cost of fitting the device to the ambulance; (3) maintenance costs; and (4) initial and ongoing staff training costs. The frequency of use observed in the trial was used to estimate the expected number of applications in order to calculate the expected cost per application (Table 18).

Missing data

The primary analysis used MI to handle missing data using baseline characteristics (sex and age) to impute missing follow-up HRQL and resource use information. Unlike more simple imputation approaches, MI reflects both the structural uncertainty related to the parameters of the imputation model and the uncertainty arising from missing data. Practically, to obtain total costs and QALYs at 1 year for each patient, missing data on HRQL and resource use were addressed using chained equations. We used truncated models to reflect the specific distribution of HRQL and resource use and generated 10 data sets. Estimates from each imputed data set were combined following Rubin’s rule. 87 In the sensitivity analysis, we also report results from the complete-case analysis, which included only patients with non-missing utility values.

Cost-effectiveness analysis

The main cost-effectiveness outcome is the 1-year cost per QALY. ICERs were calculated where one intervention was more expensive and more effective or less effective and cheaper than the other. 88 The ICER is calculated by dividing the difference in mean cost between the two arms by the difference in mean QALYs between the two arms:

where C_i and E_i are the cost and effectiveness of the LUCAS-2 device and C_c and E_c are the cost and effectiveness of manual compression, and ΔC and ΔE are the incremental cost and effect of the intervention compared with the comparator. Thus, the ICER represents the cost per QALY gained. ICERs below the NICE willingness-to-pay threshold (λ) of £20,000 are considered to indicate cost-effectiveness. For the main analyses, as effects were observed within 12 months, no discounting of costs or effects was required. ITT analyses were conducted throughout.

Uncertainty and sensitivity analysis

Uncertainty was explored by conducting non-parametric bootstrapping via 1000 resampled analyses. Cost-effectiveness planes (scatterplots of the 1000 bootstrap replications) were created. 89,90 Sensitivity analyses were conducted to determine the impact of assumptions on the cost-effectiveness results. We compared results from complete-case analysis against MI and also carried out analyses using an average group cost for outliers with high costs. Owing to the large number of non-compliers, results of a per-protocol analysis were also reported. We also derived net monetary benefits (NMBs) for each patient using the following:

We then estimated linear regression models to identify predictors of NMB, including treatment arm.

Long-term decision-analytic model

To assess cost-effectiveness over the lifetime horizon, a decision tree combined with a Markov model was constructed91 (Figure 13). The model starts with a decision tree reflecting patients’ risk of death at different time points and patient CPC score at the end of the trial. The end points of the decision tree are the starting point of the lifetime Markov model. We chose to model the intervention impact from baseline application rather than simply extending outcomes and costs from 12 months onwards. The main motivation for this was to enable better capture of uncertainty during the trial period and allow propagation of this through the lifetime horizon. Beyond 1 year post cardiac arrest, costs, HRQL and survival were modelled in two subsets of patients: patients with (1) good neurological outcomes at 1 year (CPC score of 1 or 2) and (2) poor neurological outcomes at 1 year (CPC score of > 2). Relevant model parameters were extracted from the trial data and from the literature (Table 19). The parameters of interest included relative survival rates in these subgroups that were applied to UK reference mortality rates published by the Office for National Statistics (ONS) and annual cost and HRQL data for cardiac arrest survivors with/without good neurological outcomes (from trial data). Annual costs for patients with poor neurological outcomes were obtained from trial data (outpatient visits and community-based health and social care) and residential care costs were added, as these patients are likely to require daily support/institutionalisation. 86

FIGURE 13.

Structure of the decision-analytic model.

A discount rate of 3.5% was applied to costs and effects in the Markov model (see Figure 13). The within-trial analysis was conducted using the statistical software Stata version 13.1 (StataCorp LP, College Station, TX, USA) and the decision-analytic model was built using Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA).

Cost-effectiveness analysis

The ICER is calculated by dividing the difference in mean cost between the two arms by the difference in mean QALYs between the two arms:

where C_i and E_i are the expected cost and effectiveness of the LUCAS-2 device and are the expected cost and effectiveness of manual compression, and ΔC and ΔE are the incremental cost and effect of the intervention compared with the comparator. Thus, the ICER represents the cost per QALY gained. ICERs that are below the NICE willingness-to-pay threshold (λ) of £20,000 are considered to indicate cost-effectiveness. A discount rate of 3.5% was used to discount costs and benefits.

Uncertainty and sensitivity analysis

We first performed several one-way sensitivity analyses by adding and subtracting 20% of the main parameters of the model (i.e. costs, QALY and 1-year mortality) and assessed the subsequent impact on the ICERs. The value of 20% is arbitrary, but was considered likely to represent any uncertainty that might exist in parameter values. A probabilistic sensitivity analysis was then conducted for a more comprehensive account of uncertainty in the model parameters. Cost-effectiveness acceptability curves (CEACs) that show the probability of cost-effectiveness across a range of values for λ were created using the net benefit approach. 94 With this transformation, we avoid problems encountered with ICERs (same sign but opposite quadrants) and CEACs are much simpler to calculate. 95

One-year costs

Using the complete-case analysis, the average cost at 1 year in the LUCAS-2 group was higher than in the manual CPR group, with an incremental cost of £106.70, with hospital costs being the main cost driver (Table 21). Overall, the average cost in each category of costs is higher in the LUCAS-2 group than in the manual CPR group.

We also observed higher costs in the LUCAS-2 arm than in the manual CPR arm in all cost categories in analyses that followed MI. Using the imputed data sets, we computed the total cost incurred in each patient group (the sum of all costs across all patients) that we divided by the number of 1-year survivors in each group (i.e. 177 patients in the manual CPR arm and 89 patients in the LUCAS-2 arm). We obtained a costs per 1-year survivor of £32,192 in the manual CPR arm and of £52,548 in the LUCAS-2 arm.

Quality of life

The mean utility scores in each group were measured at 3 and 12 months (Table 22). At both 3 and 12 months, HRQL was higher in the manual CPR group than in the LUCAS-2 group. An independent-sample t-test indicated that these differences were statistically significant. Changes in HRQL between the 3- and 12-month assessments were not statistically significant. Table 22 also reports the average QALY over 1 year accrued by all patients in both groups based on the AUC calculations. The mean 1-year QALY is small as a result of the high 1-year mortality rate in the sample (> 95%). We observe a small difference in mean QALY over 1 year (0.007), with patients in the manual CPR group having a higher average QALY than patients in the LUCAS-2 group. HRQL of survivors at 12 months was also estimated by neurological outcome status (as measured via the CPC score). We found a significant difference in HRQL between patients with good neurological outcome (CPC score of 1 or 2; mean 0.75) and patients with poor neurological outcome (CPC score of 3 or 4; mean: 0.47).

Cost-effectiveness at 1 year

Table 23 presents the cost-effectiveness results, showing the incremental costs and QALY for each arm of the trial, as well as the corresponding ICER. Results are shown for both the ITT and per-protocol analyses and, in each case, complete-case and MI results are presented. In addition, to obtain an approximation of the CACE, which was conducted for clinical outcomes, we inflate the ITT results using the proportion of compliers in the sample (i.e. 60%).

At 1 year, we found an incremental QALY of –0.0072 and an incremental cost of £106.70, which indicates that the LUCAS-2 device is dominated by manual chest compression, that is, the LUCAS-2 device is more costly and less effective than manual chest compression. When a per-protocol analysis was conducted instead, manual compression still dominated and results from the MI led to the same conclusion. The conclusions remain unchanged when QALYs were derived using SF-12 instead of EQ-5D. Overall, the results suggest that manual chest compression dominated the LUCAS-2 device, with the LUCAS-2 device having higher costs and providing lower QALY benefits than manual CPR. Interpretation, however, should be tempered by the very small between-group differences observed in QALYs and the relatively small differences in costs.

In Figures 14 and 15, we present the results of the 1000 bootstrap replications in the cost-effectiveness plane for both the complete-case analysis and the analysis based on MI. In both cases, the 1000 estimates are spread mainly in the north-west quadrant of the cost-effectiveness plane, meaning that the LUCAS-2 device is more costly and less effective than manual chest compression; however, it is worth noting that QALY losses are minimal. In other words, these results confirm the finding that the LUCAS-2 device is dominated by manual CPR. None of the iterations is below conventional values of the threshold (£20,000 per QALY). It is worth noting that the iterations in the MI analysis are more concentrated in the north-west quadrant (i.e. only a small number of iterations correspond to a decrease in costs). This suggests that the complete-case analysis may underestimate the incremental costs of the LUCAS-2 device. A possible explanation is that data of more costly (e.g. older) patients are more likely to be missing.

FIGURE 14.

Cost-effectiveness plane for the LUCAS-2 device compared with manual chest compression. Complete-case analysis based on 1000 bootstrap replications.

FIGURE 15.

Cost-effectiveness plane for the LUCAS-2 device compared with manual chest compression. MI analysis based on 1000 bootstrap replications.

Net monetary benefits

Linear regression models using age, sex and treatment allocation as covariates and independent variables were run to predict NMBs. Treatment allocation was found to be a significant predictor of NMB; NMB was significantly smaller (more negative) in the LUCAS-2 group.

Long-term cost-effectiveness (Markov model)

The cost-effectiveness estimates were extrapolated over a lifetime time horizon using the Markov model. The lifetime cost-effectiveness results obtained with the model are presented in Table 24. The base-case analysis is based on a cohort of patients aged 60 years, followed over 40 years, which corresponds to the average age of patients who survived at 1 year. Results suggest that the LUCAS-2 device is dominated by manual CPR, with an incremental cost of £2376.40 and an incremental QALY of –0.1286. This finding is robust to a range of sensitivity analyses as shown in Table 24.

Figure 16 shows the results from the probabilistic sensitivity analysis that takes parameter uncertainty into account. The CEAC indicates that the probability that the LUCAS-2 device is cost-effective is only around 20%, irrespective of the value of λ.

FIGURE 16.

Cost-effectiveness acceptability curves for the LUCAS-2 device compared with manual CPR.

Hospital Episode Statistics: inpatient

Hospital Episode Statistics: outpatient

Hospital Episode Statistics: accident and emergency