Therapeutic hypothermia to reduce intracranial pressure after traumatic brain injury: the Eurotherm3235 RCT

Blood–brain barrier

Alterations in blood–brain barrier (BBB) permeability after acute injury result in water, electrolytes, blood-borne substances and potential neurotoxic agents passing across the vascular system and into the brain parenchyma. Many studies have demonstrated the importance of brain and body temperature on the microvascular consequences of cerebral ischaemia and trauma. One study that assessed the effects of intra-ischaemic brain temperature (mild hypothermia) on the BBB found a reduction in extravasation of the protein tracer horseradish peroxidase. 13 Brain water content is significantly reduced with hypothermia after focal cerebral ischaemia. 14,15 This has been assessed in imaging studies, with magnetic resonance imaging finding that reductions in the apparent diffusion coefficient of water (cellular oedema) are also reduced by hypothermia. 16

In models of post-traumatic injury, hypothermia has also been shown to reduce BBB permeability. Hypothermia may be attenuating BBB permeability by altering matrix metalloproteinases, which are critical extracellular enzymes that can disrupt the BBB. 17 Given that BBB permeability, formation of vasogenic oedema and the extravasation of circulating inflammatory cells can adversely affect post-injury outcome, the effects of hypothermia are an important underlying mechanism for the beneficial effects of hypothermia.

Inflammation and oedema

The inflammatory response after TBI is significantly attenuated by hypothermia in laboratory and clinical studies. As well as attenuating the increase in BBB permeability and leucocyte margination, the endogenous inflammatory response of the central nervous system (CNS) is also reduced by hypothermia. Astrocytes and microglia respond to CNS injury by proliferating around the injury areas and releasing pro-inflammatory communication molecules as an endogenous repair mechanism. Hypothermia significantly attenuates the activation of both astrocytes and microglia. 18–21 Combination therapy with the anti-inflammatory cytokine interleukin 10 (IL-10) and hypothermia therapy was attempted in both TBI and focal cerebral ischaemia. 22 Synergistic effects were seen in focal cerebral ischaemia but not in TBI, suggesting that the cellular biology of inflammation in these two major CNS injuries has an influence on the effect of subsequent hypothermia.

Another major aspect of the inflammatory response to CNS injury is the release of reactive oxygen species by astrocytes and microglia. Hypothermia reduces increases in tissue levels of superoxide, nitric oxide and the hydroxyl radical. 23,24 The level of the superoxide dismutase, the enzyme responsible for scavenging superoxide, is increased by hypothermia,25,26 and the level of the enzyme responsible for synthesising nitric oxide, nitric oxide synthase, is attenuated by hypothermia. 27

Metabolism

By exploiting local measures of glucose metabolism using 2-deoxyglucose techniques, researchers have shown that moderate hypothermia (30 °C) results in reduced glucose utilisation compared with normothermia. 28 Metabolic effects of mild hypothermia have also been demonstrated using nuclear magnetic resonance spectroscopy. 29 Therefore, hypothermia appears to lower metabolic and energy demands, which has potentially beneficial effects on cytoplasmic adenosine triphosphate (ATP) and the maintenance of normal transmembrane ion and neurotransmitter gradients. The magnitude of preservation of ATP levels depends on both the temperature reduction and the severity of the injury. Therefore, an important mechanism for the neuroprotective effects of hypothermia is a reduction or delay in metabolic demand during and after an acute CNS injury.

Excitotoxicity

The effects of moderate hypothermia on glutamate excitotoxicity were reported using microdialysis to assay extracellular/extravascular concentrations of neurotransmitters after global ischaemia. A middle cerebral artery occlusion model is considered a reasonable model for haemorrhagic contusion. Busto et al. 30 showed that intra-ischaemic hypothermia (33 °C and 30 °C) attenuated the rise in extracellular levels of glutamate and dopamine after global cerebral ischaemia. These studies have been replicated in a variety of models of ischaemia, indicating that one of the major mechanisms by which temperature affects neuronal vulnerability is through reducing excitotoxicity following cerebral ischaemia. 23,31,32 Delayed pharmacological treatments that reduce excitotoxicity further improve outcome in combination with hypothermia;33 this may be a promising strategy for further studies. The glutamatergic receptors alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) and N-methyl-D-aspartate (NMDA) are also modulated by hypothermia. Expression of hippocampal glutamate receptors is decreased after transient global ischaemia and this is completely blocked by intra-ischaemic hypothermia. 34

Other neurotransmitters are also modulated by hypothermia. Lyeth et al. 35 demonstrated that hypothermia (30 °C) reduced elevations in cerebrospinal levels of acetylcholine after TBI. Conversely, hypothermia delayed decreases in dopamine, noradrenaline and serotonin after global cerebral ischaemia. 36 However, other studies have demonstrated that hypothermia (32 °C) can improve outcome after CNS injury without attenuating extracellular levels of glutamate and aspartate. 12,29,37 Although the neurotransmitter response in various injury models may be temperature dependent, attenuating other injury cascades may be more important in delivering possible beneficial effects of hypothermia.

Intracellular calcium-dependent signalling

There are pronounced changes in calcium-dependent intracellular signalling pathways after CNS injury. Normal neuronal activity is mediated by signalling through protein kinases and several of these have been documented to be disrupted by TBI and cerebral ischaemia. Temporary cerebral ischaemia inhibits the activity of calcium/calmodulin-dependent protein kinase II (CaMKII), a key protein kinase that mediates synaptic strength, and this is attenuated by hypothermia. 40 Protein kinase C (PKC) translocates to the membrane after cerebral ischaemia and undergoes inhibition; hypothermia rescues the inhibition of PKC activity and its translocation to the membrane. 41 Recently, various transcription factors that participate in normal neuronal functioning have been shown to be sensitive to temperature. The immediate early gene c-Fos, which regulates key genetic responses of neurons, is activated by hypothermia after transient global ischaemia. 18,42,43

These studies suggest that temperature may have profound effects on events associated with neuronal injury as well as the normal processing of neuronal signals throughout brain circuits.

Neuronal cell death

Although neuronal necrosis is commonly seen in most CNS injury models, evidence for apoptotic cell death has also been documented using various histochemical and molecular techniques. As with necrosis, apoptotic cell death appears to be sensitive to post-injury hypothermic treatment strategies. Recent studies have indicated that apoptotic cell death is another important target by which temperature may affect long-term outcome in various models of CNS injury. Various gene families [genes with a similar sequence of deoxyribonucleic acid (DNA) nucleotides] have been shown to be sensitive to post-injury temperature manipulations in models of ischaemia and trauma. 44 The ability of post-injury temperature to affect the acute and more delayed genetic response to injury is important because these genes may play an important role in determining the cellular response(s) that results in secondary injury. Using high-throughput screening and bioinformatics, genomic studies are ongoing in many laboratories; these contemporary technologies will help to determine how hypothermia may protect, and potentially repair, CNS tissues after injury.

Induction of hypothermia

The most widely accepted use of hypothermia is after cardiac arrest. Two RCTs in this patient group found significant neurological improvements in patients whose initial cardiac rhythm was ventricular fibrillation or ventricular tachycardia and who were treated with hypothermia many hours after injury. 81,82 Subsequent data from a large study of patients treated after myocardial infarction suggest that infarct size was reduced in patients who were cooled to < 35 °C before coronary intervention;83 therefore, these studies suggest that faster cooling rates may be beneficial to patient outcome.

Methods of cooling can be broadly divided into two different techniques: surface cooling and core cooling. 84 The above study used surface cooling devices alone and found that a large number of patients did not reach the target temperature quickly enough before the start of the coronary intervention to demonstrate a benefit. 83 Despite advancing technology in surface cooling devices, and the introduction of endovascular catheters for core cooling, average periods of 2–3 hours are still required to reach temperatures of 32–34 °C. 84 The currently available surface cooling devices are also relatively large and cumbersome. This, coupled with the need for staff with specialist knowledge of the management of induced hypothermia, may prevent its use outside the intensive care unit (ICU). 84

One study has examined the feasibility, speed and complication rates of infusing refrigerated fluids intravenously to quickly induce hypothermia in patients with various neurological injuries. 84 The results showed that a 1500-ml infusion of 0.9% saline in patients without cardiogenic shock, administered over 30 minutes, reduced core temperature from 36.9 ± 1.9 °C to 34.6 ± 1.5 °C at 30 minutes and to 32.9 ± 0.9 °C at 60 minutes. Continuous monitoring of arterial blood pressure, heart rhythm, central venous pressure, arterial blood gasses and serum levels of electrolytes, platelets and white blood cells showed no significant adverse events (AEs). 84

When hypothermia develops, the body will immediately try to counteract the temperature drop to maintain homeostasis. 85 One of the key mechanisms of heat production is shivering; this can lead to an increased oxygen consumption of 40–100%, which may be detrimental in this patient population. Sedation drugs are known to increase peripheral blood flow, which, in turn, will increase the transfer of heat from the core to the peripheries, thus reducing the core temperature. 85 Therefore, shivering may be counteracted by the administration of sedatives, anaesthetic agents, opiates and/or paralysing agents. 85

It should be noted, however, that the capacity and effectiveness of the mechanisms of controlling body temperature decrease with age. Younger patients will therefore react earlier, and with greater intensity, than older patients. For this reason, induction of hypothermia in younger patients often requires high doses of sedation drugs to neutralise the counter-regulatory mechanisms. 85

Together with TH therapy, all patients in the intervention group continued to be treated with stage 1 and 2 therapies as required to reduce intracranial hypertension. 86,87 If raised ICP became resistant to these therapies, even when the depth of hypothermia was increased, then care was escalated to include stage 3 interventions. If this was required, TH treatment could be terminated for patients allocated to the treatment group and the patients rewarmed using the rewarming guideline. The reason for treatment escalation was then documented on the daily data collection form.

Meta-analyses

Six meta-analyses were published between 2002 and 2008. 86–91 These included various numbers of trials with varying quality of randomisation and blinding procedures. All found a trend towards positive effects of hypothermia on neurological outcome, although statistical significance was reached in only two reviews (improved neurological outcome: RR 0.78, 95% CI 0.63 to 0.98;86 RR 0.68, 95% CI 0.52 to 0.8987).

The most recent meta-analysis91 included eight trials that studied comparable patient groups at baseline. Hypothermia was shown to reduce mortality by 20%, although this finding was not statistically significant (RR 0.80, 95% CI 0.59 to 1.09). Subgroup analysis showed that this effect was most significant when hypothermia was maintained for > 48 hours (RR 0.51, 95% CI 0.33 to 0.79). Hypothermia was also associated with a non-significant increase of 25% in neurological outcome when measured by the Glasgow Outcome Scale (GOS) at 6 months (RR 1.25, 95% CI 0.96 to 1.62). Despite not reaching statistical significance, the results showed an increased likelihood of improved neurological outcome when cooling was maintained for > 48 hours (RR 1.91, 95% CI 1.28 to 2.85). Another key finding of this meta-analysis was that hypothermia was only of significant benefit to those patients who had not received barbiturate therapy (RR 0.58, 95% CI 0.40 to 0.85).

A criticism of these analyses is that most failed to take account of important differences in patient groups, such as the presence or absence of intracranial hypertension, and differences in treatment protocols, with the exception of the use of hypothermia. Only one differentiated between studies that enrolled patients with normal ICP and those that enrolled patients with intracranial hypertension; this study found no neurological improvement associated with hypothermia. 90 Only two assessed the effects of treatment duration and speed of rewarming,86,87 concluding that cooling for > 48 hours and rewarming rates of 24 hours, or 1 °C per 4 hours, were both key factors in reducing mortality (RR 0.70, 95% CI 0.56 to 0.87) and improving neurological outcome (RR 0.79, 95% CI 0.63 to 0.98).

In summary, the evidence from previous research shows that induced hypothermia may be effective in patients with severe TBI and intracranial hypertension, provided that the treatment is continued for long enough (between 48 hours and 5 days) and that patients are rewarmed slowly (1 °C per 4 hours). Experience with cooling also appears to be important if complications, which may outweigh the benefits of hypothermia, are to be avoided.

Hypothesis

Patients treated with TH (32–35 °C) will have reduced morbidity and mortality rates compared with those receiving standard care alone after TBI.

Research questions

• Does TH (32–35 °C) reduce morbidity and mortality rates at 6 months after TBI as assessed by the Glasgow Outcome Scale – Extended (GOSE) questionnaire?

• Does TH (32–35 °C) reduce intracranial hypertension?

• Is TH a cost-effective treatment to improve outcome after TBI?

Trial design

This was a pragmatic, multicentre RCT to examine the effects of hypothermia (32–35 °C) on outcome after TBI. Participants were randomised to receive either standard care alone or standard care with the addition of TH (Figure 1).

FIGURE 1.

Study flow chart. EEG, electroencephalogram.

The experience from previous hypothermia trials underscores the potential difficulties in using TH treatment for TBI. 10 For this reason, and to reduce intercentre variance, only centres experienced in the care of TBI patients and the use of hypothermia in this patient group were initiated.

At the time of designing the trial protocol in 2008, previous studies had found that TH lasting for at least 48 hours showed a trend towards reduced mortality and improved neurological function after TBI. 91 Hypothermia (32–35 °C) was therefore continued for at least 48 hours in the trial and for as long as was necessary to reduce and maintain ICP at < 20 mmHg. This ICP threshold was used because the Brain Trauma Foundation guidelines86 define intracranial hypertension as an ICP of > 20 mmHg. When ICP was stable, patients were then slowly rewarmed at a rate of 0.25 °C per hour (1 °C per 4 hours).

There are three recognised stages of TBI management (Figure 2). 86,92

FIGURE 2.

Three stages of ICP management after TBI. CSF, cerebrospinal fluid; EEG, electroencephalogram; GCS, Glasgow Coma Scale.

Participants in the hypothermia group continued to be treated with stage 1 and 2 therapies as required to reduce intracranial hypertension. 86,87 If raised ICP was resistant to these therapies, despite increasing the depth of hypothermia, care could be escalated to include stage 3 interventions.

During the preparation of the protocol, evidence suggested that patients were three times more likely to develop complications if hypothermia treatment was continued in conjunction with stage 3 interventions, that is, surgery or continuous infusion of barbiturates (the drugs used to treat ICP). Thus, we advised that the cooling intervention was withdrawn and the patient was rewarmed using the rewarming guideline as soon as it was safe to do so when the patient’s care was escalated. The reason for treatment escalation was then documented on the daily data collection form.

The primary end point of the Eurotherm3235 Trial was the outcome 6 months after TBI using the GOSE questionnaire. Participants were sent the GOSE questionnaire with a covering letter by post 6 months after randomisation by the co-ordinating centre.

Important changes to methods after trial commencement

The inclusion criteria were changed in January 2012, based on the pilot phase findings,93 to remove the upper age limit (previously 65 years) and increase the time from injury from 72 hours to 10 days. These changes provided important information on older patients and allowed patients with evolving brain swelling up to 10 days from injury to be included.

Inclusion criteria

• Believed to be legal age to consent to take part in research.

• Primary closed TBI.

• Raised ICP of > 20 mmHg for ≥ 5 minutes after first-line treatments with no obvious reversible cause (e.g. patient position, coughing, inadequate sedation).

• ≤ 10 days from the initial head injury.

• Cooling device or technique available for > 48 hours.

• Core temperature of ≥ 36 °C (at the time of randomisation).

• An abnormal CT scan of the brain, defined as one that shows haematoma, contusion, swelling, herniation or compressed basal cisterns.

Exclusion criteria

• Patient already receiving TH treatment.

• Administration of barbiturate infusion prior to randomisation.

• Unlikely to survive for the next 24 hours in the opinion of the ICU consultant or consultant neurosurgeon treating the patient.

• Temperature of ≤ 34 °C at hospital admission.

• Pregnancy.

All female patients of childbearing age who met the inclusion criteria underwent a urine pregnancy test. This was performed by the investigator or research nurse in the ICU as part of the screening for eligibility procedure.

The Eurotherm3235 Trial enrolled patients after TBI who had an ICP of > 20 mmHg for ≥ 5 minutes, despite first-line treatments, with no obvious reversible cause, for example patient position, coughing or inadequate sedation. Although early cooling after injury is considered to be beneficial, this is offset by failure to show benefit from hypothermia in the absence of raised ICP. Enrolment to the Eurotherm3235 Trial was therefore allowed up to 10 days following injury.

The trial recruited incapacitated adult patients who were admitted to the neurological ICU. It was not possible to obtain consent from the patients before enrolment to the study as they were sedated and attached to a ventilator. Information booklets were therefore made available in the waiting area of the ICU and the patient’s legal surrogate decision-maker, their next of kin/nearest relative/welfare guardian (in accordance with the laws of each country) was informed about the study by the ICU team soon after their relative was admitted. If they wanted further information about the trial or wished for their relative to participate, then the relative/welfare guardian contacted the researcher.

During this discussion, all stages of the trial were explained and the researcher explained that the patient may not have been eligible at that time, as they must have developed brain swelling before entering the study. Brain swelling can cause further brain damage if not treated quickly, so it was desirable that consent was obtained before any brain swelling occurred so that treatment could be allocated and started very soon after meeting the eligibility criteria. Patients who were consented early entered the study only if they developed brain swelling and met all the inclusion criteria for the trial at that time.

In accordance with the Good Clinical Practice guidelines,94 the patient’s nearest relative or welfare guardian was given the opportunity to ask questions and was given a copy of the trial information sheet to read and keep.

In some cases, the patient’s nearest relative/welfare guardian was not present when the patient was admitted to the ICU. In usual circumstances, the health-care team contacted them to update them on the patient’s condition soon after the patient was admitted. During this call, basic information was provided about the trial and they were asked if they would like more information. If further information was requested, the local investigator or research nurse obtained telephone consent and followed this up with written consent as soon as possible afterwards.

Premature withdrawal

Participation in any research trial is voluntary; therefore, the participant or their legal representative was able to withdraw from the trial at any point. In this case, it was clearly documented on the premature withdrawal form whether or not any previously collected data could still be used in the analysis and which part of the trial the participant was being withdrawn from, using the following options:

1. withdraw entirely – the hypothermia intervention will be safely terminated, no further data will be collected and previously collected data will not be used in the analysis

2. withdraw entirely – the intervention will be safely terminated and no further data will be collected, but previously data collected may be used in the analysis

3. withdraw from the intervention but be willing to be followed up

4. withdraw from being followed up only.

Trial site initiation visits

As soon as hospital approval was given for a site to participate in the trial, the trial manager contacted the PI and research team to arrange a site initiation visit. In the UK, the trial manager visited each site in person to conduct this visit. The visit lasted approximately 2 hours, during which time the staff were taught the inclusion and exclusion criteria, the consenting process, how to randomise a patient, the trial interventions, how to complete the data collection forms on the internet-based electronic database, how to complete a serious adverse event (SAE) form and how to report a SAE. Staff were also given access to the test database to practise completing data input. Follow-up procedures were also discussed and staff were told about the importance of maintaining contact details for the participants and their nearest relative after discharge from the ICU. The procedure for withdrawing a participant from the trial if required was also discussed.

Site initiation visits for the non-UK sites

Owing to the variability of access to the non-UK sites, we developed an alternative method for site initiation in these centres. When it was feasible in terms of travel time, cost and language barrier, the trial manager visited each of the non-UK centres to carry out the trial initiation visit. When this was not feasible, the trial manager arranged a web conference with the PI and research team at each hospital. This meeting lasted for at least 2 hours and was conducted in English.

The countries that we visited were:

• Belgium – as the process of ethics approval was centralised for Belgian centres, we arranged a single site initiation meeting in Brussels and invited all centres to attend. Two additional centres joined the trial after this meeting. In this case, the trial manager arranged one meeting in one of the hospitals and both PIs and research staff from the two hospitals attended.

• Germany and Italy – because of the language barrier in two sites, we felt that the staff at these sites would benefit more from a face-to-face meeting rather than a teleconference.

• Republic of Ireland – the trial manager visited the site in Ireland as it was very similar to centres in the UK and was very easy to travel to.

• Portugal – a trial initiation meeting was possible at this site because the chief investigator was invited to attend the Portuguese Intensive Care Society meeting and so carried out the visit while attending the local conference.

• Spain – because of the language barrier in one site, we felt that the local staff would benefit more from a face-to-face meeting rather than a teleconference.

Once a site was initiated and activated on the database system, hospital staff could start to screen and recruit patients. Ongoing support was provided to the sites by the chief investigator and trial managers, either by e-mail or office telephone (available Monday–Friday, 09.00–17.00). We also set up a trial helpline using a separate mobile phone account, which was available out of hours and at weekends. The helpline was held by the chief investigator and trial managers on a rotational basis.

Primary

• Outcome at 6 months using the GOSE questionnaire.

Secondary

• Six-month mortality rate.

• Intracranial pressure control.

• Incidence of pneumonia across both groups.

• Length of stay in the ICU and hospital.

• Modified Oxford Handicap Scale score at 1 month, discharge from the randomising hospital or death, whichever took place first.

• Correlation between the predicted outcome using the MOHS score at hospital discharge and the predicted outcomes using the GOSE score at 6 months post injury.

• Health economics.

Primary end-point data collection

• Outcome at 6 months using the GOSE questionnaire.

To collect this information, a questionnaire was sent to all participants 6 months after injury.

This study recruited a very challenging patient group; patients who have suffered a head injury can have personality changes after the injury, which can make it challenging to follow them up. Therefore, drawing on our previous experience of the follow-up of brain-injured patients who had been admitted to the ICU, we knew that it was unlikely that all patients would be able to complete and return the questionnaire by themselves. Depending on the severity of their injury, many patients would still be in hospital at the 6-month follow-up. For this reason, we designed the follow-up procedures very carefully and included these processes in the initial consenting discussion.

At the same time as sending the questionnaire to participants at 6 months, we also sent a letter to those who gave consent for participants to enter the study. This letter informed them that the questionnaire had been sent to participants and asked them to help with completion of the questionnaire.

This procedure was discussed with those giving consent at the time of providing consent for a patient to enter the study. This procedure was also discussed with patients at the time of obtaining their subsequent consent (if applicable).

The procedures for the follow-up of participants varied between those who were recruited in the UK and those who were recruited in non-UK centres.

Follow-up procedure for UK centres

The first process for follow-up, before we sent anything to participants, was for the trial manager to ask the research nurse at each participant’s centre to check the participant’s vital status by determining whether the participant was still an inpatient or had been transferred to a rehabilitation centre. If the participant had been discharged home, the research nurse was then asked to contact the participant’s GP to find out their vital status.

A GOSE questionnaire was then sent by post from the trial office in Edinburgh to all surviving participants 6 months after injury. At this time, we also sent a letter to those who gave consent for participants to enter the study, asking them to help with completion of the questionnaire.

In the pilot stage of the trial, if we did not receive a response from a participant within 3 weeks, we sent a shorter GOS questionnaire to the participant by post. If this was not returned after 2 weeks, then an independent and blinded researcher based in Edinburgh telephoned the participant to complete the short questionnaire with them over the telephone. The steps taken if a participant could not be contacted at this time varied depending on whether or not the participant had regained capacity and had given subsequent consent themselves:

• If the participant had regained capacity and had given retrospective consent, then the process ended here and the participant was documented as lost to follow-up.

• If the participant had not regained capacity and had not given follow-on consent, then the independent researcher telephoned the person who had given consent and asked them to complete the short questionnaire over the telephone.

This process was streamlined at the beginning of the main phase of the trial in January 2012. At this time, we decided to remove the process of sending the shorter questionnaire to participants and instead moved straight to telephoning participants if the first GOSE questionnaire was not returned after 3 weeks. In this case, the independent/blinded researcher based in Edinburgh telephoned participants if they had been discharged from hospital and completed the longer GOSE questionnaire with them over the telephone. If a participant was still in hospital and had not regained capacity to give follow-on consent, then the independent researcher telephoned the person who gave consent for the participant to enter the study and completed the longer GOSE questionnaire with them over the telephone.

Follow-up procedure for non-UK centres

Owing to the language barrier in the participating non-UK centres, it was not possible for the independent/blinded researcher in Edinburgh to contact the non-UK patients for follow-up. We therefore devised a slightly different process for the non-UK centres depending on their local procedures.

The procedure for follow-up was specifically outlined and discussed by the trial manager at the trial initiation meeting with the PI and research team at each of the non-UK centres.

Each PI was asked to identify someone in their hospital who would be blind to the participants’ intervention and could carry out the follow-up. All centres, except those in Belgium, had an independent person at their hospital who carried out the follow-up over the telephone at 6 months. Belgium had a similar procedure to that in the UK in that a letter was sent to both the participant and the person who gave consent at 6 months and, if this was not returned after 3 weeks, an independent person at the centre contacted the participant by telephone and completed the longer GOSE questionnaire with them over the telephone.

Any changes to trial outcomes after the trial commenced, with reasons

No changes were made to the trial outcomes after commencement of the trial.

Data monitoring

The trial did not include any predefined interim analyses. The trial Data and Safety Monitoring Committee (DSMC) met regularly, however, to discuss the unblinded trial data and monitor the progress of the trial.

This committee was assembled using the Medical Research Council (MRC) guidelines for the development of a Data Monitoring and Ethics Committee (DMEC) and comprised three members:

• an expert clinician in intensive care (chairperson)

• an expert trial statistician

• an expert triallist.

A DSMC charter was developed and agreed by all members at the first DSMC meeting in 2010 (see Appendix 3). This charter included the guidelines for early termination, the quorum for decision-making and the reporting of the DSMC recommendations.

Statistical guidelines for termination: Data and Safety Monitoring Committee charter

The DSMC has the responsibility for deciding whether, while randomisation is in progress, the unblinded results (or the unblinded results for a particular subgroup) should be revealed to the Trial Steering Committee.

The DSMC terms of reference stated that it would do this if, and only if, two conditions were satisfied:

1. The blinded results provide proof beyond reasonable doubt that treatment is, on balance, either definitely harmful or definitely favourable for all, or for a particular category of, patients in terms of the major outcome.

2. The blinded results would, if revealed, be expected to substantially change the prescribing patterns of doctors who are already familiar with any other trial results that exist.

Exact criteria for ‘proof beyond reasonable doubt’ are not, and cannot be, specified by a purely mathematical stopping rule, but they are strongly influenced by such rules. DSMC members expressed sympathy with the stopping rule proposed in Part I of the 1976 report to the MRC Leukaemia Committee,98 whereby an interim analysis of major end points would generally need to involve a difference between treatment and control of at least three standard errors to justify premature disclosure. An interim subgroup analysis would, of course, have to be even more extreme to justify disclosure. This rule has the advantage that the exact number and timing of interim analyses need not be prespecified. In summary, the stopping rules as specified in the CRASH trial protocol99 (as successfully applied in other trials including the MRC International Stroke Trial,100 which randomised 19,436 acute stroke patients) require extreme differences to justify premature disclosure and involve an appropriate combination of mathematical stopping rules and scientific judgement.

The DSMC met a total of nine times between February 2010 and October 2014. Of note is that the data showed a trend towards harm in the hypothermia group during the seventh and eighth DSMC meetings but the results did not become significant (difference of three standard errors) until the data were reviewed during the ninth meeting in October 2014.

At this time, the DSMC came to the following conclusions:

• There is no evidence of benefit for the patient group receiving the treatment to be evaluated.

• At best, a result of futility can be expected if the study is continued.

• There are signs of harm by the new treatment investigated including differences in mortality rate and a smaller number of patients achieving a relatively good outcome in the new treatment group, for those surviving the 6-month period after inclusion.

The DSMC therefore proposed that ‘appropriate action is taken rapidly, as a consequence of the analysis done . . . in order to protect patients to be potentially included in this trial from now on’ (abbreviated excerpt from the DSMC letter to the Trial Steering Committee).

This recommendation was sent to the chairperson of the Trial Steering Committee who convened an urgent Trial Steering Committee meeting. The meeting concluded that recruitment to the trial should be stopped early on the basis of futility and possible harm to participants. Recruitment was therefore stopped on 14 October 2014.

Type of randomisation

Participant allocation to either the hypothermia or the standard-care group by the online randomisation service used a minimisation technique with a random element using the following stratification variables:

• trial centre

• aged < 4–5 years or ≥ 45 years

• post-resuscitation Glasgow Coma Scale (GCS) motor component 1 or 2 or 3–6

• time from injury of < 12 hours or ≥ 12 hours

• pupils – both reacting or one or neither reacting

• allocation concealment mechanism.

It was not possible to conceal the allocated group from local investigators and study teams as it was clinically obvious which participants were receiving hypothermia, for example from the equipment required, participant temperature, blood results and fluid requirements.

To overcome potential bias, outcome data assessment were blinded.

Implementation

The central internet-based randomisation service provided by Lincoln generated the random allocation sequence (using minimisation with a random element) and assigned participants to groups.

Patients at each site were enrolled by either the PI or their deputy (e.g. research nurse or research doctor).

Blinding

It was not possible to blind local investigators, study teams or the participants’ next of kin to group allocation as it was clinically obvious which participants were receiving hypothermia.

Outcome data assessment was blinded as the GOSE questionnaire was, in the first instance, posted to participants by the trial office in Edinburgh. If it was not returned in the post within 3 weeks, then an independent researcher who was blind to treatment allocation telephoned the participant and/or their next of kin to complete the questionnaire with them over the telephone.

There were some similarities between interventions in the hypothermia group and interventions in the standard-care group of this trial. Both groups received stage 1 and stage 2 therapy (see Figure 2) throughout the trial period. It was also possible that participants in the standard-care group received temperature reduction, to a minimum of 36 °C, to treat fever using the same cooling technique as for participants in the hypothermia group. If this was required, the centres followed their local guidelines for fever management; no guidelines were provided as part of the trial.

Overall statistical principles

The trial statistician performed the statistical programming and analysis to produce all summary tables and figures using the statistical package SAS^® (version 9.2 or a more recent version; SAS Institute Inc., Cary, NC, USA; SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ^® indicates USA registration).

In general terms, categorical data are presented using counts and percentages, whereas continuous variables are presented using the mean, median, standard deviation (SD), minimum, maximum, interquartile points at 25% and 75% (Q1 and Q3) and number of participants with an observation (n). Data are split by time point when applicable.

All applicable statistical tests are two-sided and have been performed using a 5% significance level, leading to 95% (two-sided) CIs, unless otherwise specified.

Distributional assumptions underlying the statistical analyses are assessed by visual inspection of residual plots. Normality is examined by normal probability plots. If the distributional assumptions for the parametric approach were not satisfied, further data transformation (for achieving normality), or other suitable methods, were considered.

The intention-to-treat (ITT) and per-protocol (PP) populations are used to summarise all data and all analyses unless otherwise specified.

There has been no imputation of the data with regard to missing values or withdrawals for the statistical summaries and statistical analysis unless otherwise specified.

Statistical analysis plan

A detailed statistical analysis plan, setting out full details of the proposed analyses, was finalised before the trial database was locked for analysis.

As a result of the internal pilot phase,93 the sample size for the full trial was reduced from 1800 to 600 participants. Two factors contributed to this decision:

1. The original sample size may have underestimated the possible benefit of hypothermia as, unlike participants in most previous hypothermia trials, participants in the Eurotherm3235 Trial had to have evidence of brain swelling (raised ICP of > 20 mmHg) to be eligible for randomisation.

2. We showed that an enhanced cooling intervention could be delivered, as described by Peterson et al. 91

These data therefore informed the revised power calculation.

Population for analysis

List of analyses

Primary outcome

Using an ordinal analysis of the GOSE scores, together with covariate adjustment (primary efficacy analysis), we were able to increase the statistical efficiency of the analysis101,102 so that a trial involving 600 participants would have power equivalent to that of a trial involving 1000 participants that assessed a binary outcome. We calculated that, with such an analysis, the study would have the equivalent of 80% power to detect a rate of unfavourable outcome (GOSE score of 1–4) that was 9 percentage points lower with hypothermia than with standard care (51% vs. 60%) at the 5% significance level (two-sided).

Data were analysed on an ITT basis. All randomised participants for whom outcome data were available were therefore included.

Participants were analysed depending on their allocated intervention, irrespective of whether or not their actual management complied with the allocated intervention.

The distributions of the 6-month GOSE scores between the two randomised groups (hypothermia vs. standard care) were compared using ordinal regression, adjusting for the following baseline covariates that were included in the minimisation algorithm:

• age (included as a continuous variable using a linear term in the regression model)

• post-resuscitation GCS motor component 1 or 2 compared with 3–6

• time from injury of < 12 hours compared with ≥ 12 hours

• pupils: both reacting compared with one reacting compared with neither reacting (included as an unordered categorical variable in the regression model).

As the covariates specified for the statistical analyses were required to inform the minimisation algorithm, there were no missing data for baseline covariates. A strenuous effort by the trial managers was also made to obtain complete outcome data at 6 months.

For this analysis, the eight-point GOSE was collapsed to six categories by pooling the ‘death with vegetative state’ and ‘lower severe disability’ categories. This was carried out to ensure that the analysis could not give ‘credit’ to an intervention that reduces mortality at the expense of increasing the proportion of severely disabled survivors.

The treatment effect was reported as an estimated adjusted common odds ratio (OR) with its corresponding 95% CI. The treatment groups are ordered in such a way that a common OR of < 1 corresponds to a benefit for hypothermia over standard care.

If there is strong evidence that the treatment effect does not satisfy the proportional odds assumption, which is required for the ordinal analysis, then this in itself will demonstrate that the distribution of the GOSE score differs between the two randomised groups. In such a situation, the distribution of the GOSE score in the two groups will be explored descriptively to give an insight into the nature of the treatment effect.

Secondary outcomes

The following outcome measures were tested for differences between the two treatment groups:

Quality control/validation: summary tables

A random selection of unique analysis and summary tables was quality controlled using manual methods (e.g. comparison of results in the table with results calculated using a calculator, spreadsheet, database output or any alternative summarisation tool).

Quality control/validation: statistical analysis

Quality control/validation of statistical analyses was performed by peer review of program code, log and output. Additionally, the primary outcome analysis was replicated independently by a second statistician.

Participant flow

Of the 2498 participants who were assessed for trial eligibility, 387 were randomly assigned, 386 received the intended treatment (after a withdrawal pre intervention) and 387 were analysed for the primary outcome (Figure 3). The most common reasons for exclusion from the trial were that ICP remained at ≤ 20 mmHg (41% of 2111 exclusions), the participant was unlikely to survive (8%) or the participant was already receiving hypothermia treatment (6%).

FIGURE 3.

The CONSORT flow diagram. Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

Recruitment

A total of 64 centres were initiated in 18 countries worldwide, with a target recruitment at each centre of four participants per year. A total of 387 participants were randomised from 47 of these centres between November 2009 (pilot phase from 10 January 2009 to 15 September 201193) and October 2014, when recruitment was stopped. Of these 387 participants, 53% (n = 205) were recruited in the UK. Follow-up ended 6 months after recruitment stopped, in April 2015.

Figure 4 shows the predicted versus actual recruitment throughout the trial.

FIGURE 4.

Recruitment graph. Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

Why the trial was ended

Recruitment was stopped in October 2014 when the Trial Steering Committee, in concurrence with the original recommendations of the DMEC, concluded that there were signs of harm with the induced hypothermia treatment and, at best, a result of futility would be expected if the trial were to continue. These findings were apparent when the study’s designated primary outcome measure, analysed in accordance with the study’s pre-stated statistical analysis plan, was examined.

Baseline data

On an ITT basis, a total of 195 participants were randomised to hypothermia and 192 participants were randomised to standard care. Table 1 shows that there were no significant differences between groups in baseline characteristics.

Numbers analysed

Table 2 shows the breakdown of recruitment and treatment allocation by centre and Table 3 shows the total number of participants screened and recruited at each centre.

The breakdown of participants recruited in the UK and non-UK centres was 53% and 47%, respectively.

Outcomes and estimation

All analyses were performed with SAS software. Analyses were performed on an ITT basis, incorporating all participants who underwent randomisation and for whom outcome data were available, with participants evaluated depending on their assigned intervention.

For the primary analysis, the distribution of the 6-month GOSE scores between the two groups (hypothermia vs. control) was compared with the use of ordinal logistic regression and with adjustment for the following baseline covariates:

• age (included as a continuous variable with the use of a linear term in the regression model)

• post-resuscitation GCS motor score [1 or 2 (no response or extensor response) vs. 3 to 6 (flexion or better response)]

• time from injury (< 12 hours vs. ≥ 12 hours)

• pupillary response (both reacting vs. one reacting vs. neither reacting; included as an unordered categorical variable in the regression model).

For this analysis, we collapsed the eight-point GOSE to six categories by pooling ‘death with a vegetative state’ and ‘lower severe disability’. This ensured that the analysis would not favour an intervention that reduced mortality at the expense of increasing the proportion of severely disabled survivors. Figure 5 shows the survival curves between groups.

FIGURE 5.

Kaplan–Meier survival curves per group.

Table 4 shows a detailed breakdown of the score category per group for both the full eight-point GOSE, the collapsed six-category GOSE and the dichotomised GOSE.

Table 5 shows the results of the primary analysis and primary subgroup analyses for the ITT population.

The results of the primary analysis were unexpected and showed that participants in the hypothermia group had significantly poorer outcomes at 6 months (p = 0.04) and a higher mortality rate (p = 0.05) than those treated with standard care alone. Furthermore, the primary subgroup analysis showed that age and post-resuscitation GCS motor component were specifically linked to outcome; in this patient population, these findings would be expected.

Secondary analyses

Secondary outcomes of the trial were:

• 6-month mortality rate

• intracranial pressure control

• incidence of pneumonia across both groups

• length of stay in the ICU and hospital

• modified Oxford Handicap Scale score at 1 month, discharge from the randomising hospital or death, whichever took place first

• correlation between the predicted outcome using the MOHS at hospital discharge and the GOSE score at 6 months post injury

• health economics.

The results of the secondary end-point analyses are shown in Table 6.

TABLE 6 - Results of the secondary end-point analyses

Values were calculated with the use of a repeated-measures model adjusted for age, postresuscitation GCS motor score, time from injury, pupillary response, study day and (when available) baseline value. These are post hoc analyses.

An OR or common OR of < 1 corresponds to a benefit for hypothermia over control.

Results were adjusted for age, post-resuscitation GCS motor score, time from injury and pupillary response.

The six grades of the MOHS were collapsed to four categories by pooling dependent, fully dependent and death.

Analysis of secondary outcomes: hypothermia vs. control	Estimate (95% CI)	p-value
Unadjusted HR for death incidence between randomisation and 6 months	1.45 (1.01 to 2.10)	0.047
Physiological measuresa
Adjusted mean difference in ICP on days 1–7 (mmHg)	–0.48 (–2.04 to 1.08)	0.55
Adjusted mean difference in core temperature on days 1–7 (°C)	–2.14 (–2.34 to –1.94)	< 0.001
Adjusted mean difference in MAP on days 1–7 (mmHg)	1.20 (–0.46 to 2.86)	0.16
Adjusted mean difference in CPP on days 1–7 (mmHg)	1.61 (–0.36 to 3.58)	0.11
Adjusted mean difference in squared proportion of ICP measurements of ≤ 20 mmHg on days 1–7	440 (–1.60 to 1000)	0.47
Incidence of patients who had presence of pneumonia on days 3–7	1.04 (0.69 to 1.58)	0.84
Adjusted mean difference in log-transformed length of ICU stay (log-hours)c	0.05 (0.11 to 0.22)	0.54
Adjusted common ORb for MOHS grade at 28 daysc,d	1.65 (0.91 to 3.02)	0.10

The physiological measures detailed in Table 5 are also shown as trends in Figures 6 and 7.

FIGURE 6.

Intracranial pressure control per group. Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

FIGURE 7.

Temperature control per group. Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

Adapted with permission from Andrews et al. 1 Copyright © (2015) Massachusetts Medical Society.

Ancillary analyses

Prespecified subgroups for the primary analysis were defined on the basis of the baseline covariates described (see Table 1), the location of the centre (UK vs. non-UK) and the volume of the centre (≥ 10 vs. < 10 participants). We performed these analyses by including an interaction term between each intervention and the relevant covariate in the ordinal logistic–regression model; a stricter level of statistical significance (p < 0.01) was used because of their exploratory nature.

The results of this subanalysis showed no statistically significant difference in patient outcome between UK centres and non-UK centres (p = 0.17) and no statistically significant difference in outcome between high-volume recruiting centres and low-volume recruiting centres (p = 0.12).

Modified Oxford Handicap Scale grades were analysed in the same way as GOSE scores, but we collapsed the six grades to four categories by grouping together dependent, fully dependent and death. In the analysis of the between-group difference in mortality, Cox proportional hazards regression was used to estimate the intervention effect.

Other continuous outcomes were tested with an analysis of covariance; for binary outcomes, logistic regression was used. ICP, core temperature, MAP and CPP on days 1–7 were analysed post hoc with the use of a linear model, with study days as repeated measurements with a compound symmetry covariance matrix. All of these analyses used the same covariates as were prespecified for GOSE scores, together with the baseline value of the relevant variable.

Occasionally, participants who have suffered a brain injury have severe brain swelling that does not respond to the usual treatment methods. In these cases, participants require further treatment with either drugs (barbiturates) or surgery (decompressive craniectomy).

A further covariance analysis that was carried out was the incidence of treatment escalation or stage 2 therapy failure across groups. Table 11 shows the proportion of participants in the Eurotherm3235 Trial who developed severe brain swelling and the days on which the therapy failure occurred.

Harms

An AE is any untoward medical occurrence in a clinical trial subject. Many untoward events are expected in patients admitted to the ICU because of the severity of their illness and/or injury.

The treatment of any untoward medical occurrence is part of the standard care for patients admitted to the ICU. Therefore, the decision was made in the design stage of the trial that no AE data would be collected in this study.

A SAE is defined by the National Research Ethics Service in the UK104 as any AE that:

• results in death

• is life-threatening

• necessitates hospitalisation or prolongation of existing hospitalisation

• results in persistent or significant disability or incapacity

• is a congenital abnormality or birth defect.

Death is an expected outcome in approximately 25% of all critically ill patients with severe TBI105 and, therefore, it was not reported as a SAE in this study.

The specific SAE data that we collected were:

• bleeding – defined as a new haemorrhage requiring ≥ 2 units of packed red cells

• cardiovascular instability – defined as a systolic blood pressure of < 90 mmHg for ≥ 30 minutes86 (of note, terminal hypotension data was not collected)

• thermal burns on > 5% of body surface area using the Lund–Browder chart

• cerebral perfusion pressure of < 50 mmHg for ≥ 15 minutes.

Unexpected events considered to be SAEs, that are not described above, could also be collected by investigators by using the ‘other’ option on the SAE form. Table 12 shows all SAEs that were collected throughout the trial.

Reporting of SAEs was low throughout the trial, with an overall incidence of 7.5%. Most of these events were collected in the hypothermia group. We believe that the reason for this is that the four specific SAEs that we outlined were all possible side effects of hypothermia; therefore, these SAEs may have been thought to be physiological effects of the brain injury in the standard-care group rather than being related to the trial, especially as treatment in the standard-care group was not protocolised.

Generalisability

The Trial Steering Committee and trial sponsor accepted the recommendation of the DSMC after its ninth meeting in full and terminated recruitment early. Early stopping of any trial can potentially reduce the external validity of the results; however, the burden of proof required for early stopping for possible harm is considerably lower than that for overwhelming evidence of efficacy. 106 In this case, the GOSE scores collected after the decision to terminate confirmed the result of the trial detailed earlier.

Although the trial was open, we believe that the external validity was strong, as centres:

1. were able to use their own technique for maintenance of hypothermia

2. could rewarm the participant based on their physiological parameters and condition from day 3 onwards

3. could cool participants to normothermia in the control group for fever control

4. could escalate care to stage 3 interventions if/as soon as was required in both groups.

The prescribed induction of hypothermia means that the same technique was used to initiate hypothermia in all participants. We know that compliance with the guideline was good, as Figure 7 shows the rapid reduction in temperature to 35 °C, and treatment compliance showed that 97.1% of core temperature observations were within range in the hypothermia group. This strengthens the internal validity of the trial and its result. The use of a blinded outcome assessor also improved the internal validity of the trial as the same method for scoring was used for all GOSE questionnaires and scoring was carried out by the same researcher throughout the trial. Furthermore, the imputed algorithm in the trial online database was used to ensure that there was no human error/bias in scoring.

The trial involved 387 participants from 47 centres across 18 countries worldwide, and all were able to use their own maintenance technique for hypothermia without a protocolised control group. We therefore believe that the results can be generalised to any global setting. The cooling intervention studied was already being used in clinical practice in each of these centres and was tested in the way that centres normally used it.

Search methods for identification of trials

Searches were not restricted by date or by publication status. The search strategy is recorded in the protocol (see Report Supplementary Material 1). The dates were chosen to be complementary to, and have an overlap of 12 months with, the previous systematic review110 conducted by one of the co-authors; this was to capture any trials that were in the process of publication.

Ethics approval and consent

This study did not require ethics approval or individual participant consent.

Methodological criteria for selection of randomised controlled trials

The inclusion criteria were that trials must (1) be RCTs, (2) investigate adult moderate (i.e. a GCS of 9–12) or severe (i.e. a GCS of ≤ 8) TBI and (3) investigate TBI that was sustained following an acute, closed head injury. The use of TH was the intervention of interest. RCTs that did not have two treatment arms with TH compared with a control group were excluded. For the purpose of this review TH is defined as any intervention with the intention of reducing core body temperature to below the physiological norm (36 °C). The method of temperature reduction was noted during data extraction. Participant outcomes at follow-up were assessed using mortality and poor outcome data. There was a particular emphasis on outcomes recorded on a scale, for example the GOS, the Ranchos Los Amigos Scale or an equivalent scale. Reporting of poor outcomes includes mortality figures, in accordance with the usual reporting methodology.

Data extraction

The output of all searches was exported into Microsoft Word (Microsoft Corporation, Redmond, WA, USA) and two co-authors were involved at each stage of review. The PRISMA statement114 was followed at all three stages of review. First, duplicates from across the databases were removed. Following duplicate removal, the abstracts and titles were assessed against our inclusion and exclusion criteria. Second, abstracts were read in detail against the defined criteria. Third, the remaining articles were read in full and the final decision on inclusion or exclusion was made. A third reviewer was available to resolve any discrepancies in agreement. See Figure 9 for the PRISMA flow diagram.

Statistical analyses

Mortality, poor outcomes (including mortality) and pneumonia data were extracted and the RR and corresponding 95% CIs were calculated. The Mantel–Haenszel approach was used to assess the significance of the overall RR (RR_overall) as an effect estimate using the null hypothesis RR_overall = 1 using the z-test.

As per the Cochrane Handbook,109 both fixed and random effects were generated in the first instance for each analysis.

Statistical heterogeneity was assessed using the chi-squared and I²-index tests to estimate the degree of trial variability. Thresholds for interpretation of heterogeneity were also taken from the Cochrane Handbook, with 0–40% representing possible low heterogeneity, 30–60% representing possible moderate heterogeneity, 50–90% representing possible substantial heterogeneity and 75–100% representing possible high heterogeneity.

Sensitivity analysis

Forest plot analyses were performed for all studies and were performed separately for those deemed at least risk of bias as assessed using a Cochrane-based evaluation.

Risk of bias of included randomised controlled trials

There were two methods used for assessing the quality of included RCTs: the Cochrane Collaboration’s tool for assessing the risk of bias and a modified Jadad score. 109

Using the Cochrane Collaboration’s tool for assessing the risk of bias, each RCT was given a score out of 18. Two RCTs scored ≥ 14 (top quartile) so were considered as having a low risk of bias. Conversely, 18 RCTs scored ≤ 13 and so were considered as having a high risk of bias. See Appendix 7 on assessment of bias. These scores were used for further analysis of risk of bias for the primary outcomes of mortality, poor outcomes and new pneumonia.

Quality of included randomised controlled trials

On the modified Jadad score, those scoring 3 were high quality and those scoring 0, 1 or 2 were low quality. In total, six of the included studies were classed as high quality and 14 studies were classed as low quality. These scores were used for further analysis of quality for the primary outcomes of mortality, poor outcomes and new pneumonia.

Mortality results

Mortality was reported in 20 studies. Figure 9 shows mortality data for 20 studies, totalling 2270 participants. When the results of these 20 RCTs were statistically aggregated, there was a significant reduction in mortality in the TH group (RR 0.84, 95% CI 0.74 to 0.96; p = 0.01). A fixed-effects model was used and supported by a moderate I²-index (I² = 43%).

FIGURE 9.

Forest plot of mortality data for 20 studies. M–H, Mantel–Haenszel.

Mortality results with quality and risk-of-bias analysis

Six of the included studies were classed as high quality and 14 studies were classed as low quality, as per the modified Jadad score assessment of risk of bias (Figure 10). The high-quality studies showed a higher mortality in the TH group (RR 1.22, 95% CI 0.99 to 1.57), but this result was not significant (p = 0.06). The low-quality studies showed a higher mortality in the control group (RR 0.69, 95% CI 0.58 to 0.81; p < 0.00001). There was significant heterogeneity between the two subgroups (I² = 94.2%).

FIGURE 10.

Mortality results with quality and modified Jadad score assessment of risk-of-bias analysis. M–H, Mantel–Haenszel.

Risk-of-bias assessment was also undertaken using the Cochrane Collaboration’s tool and again affected mortality outcomes (Figure 11). Two studies had a low risk of bias and showed a higher mortality in the TH group (RR 1.37, 95% CI 1.04 to 1.79; p = 0.02), whereas the 18 studies having a high risk of bias showed a higher mortality in the control group (RR 0.72, 95% CI 0.62 to 0.84; p < 0.0001).

FIGURE 11.

Forest plot of mortality results with quality and Cochrane Collaboration assessment of risk-of-bias analysis. M–H, Mantel–Haenszel.

Poor outcomes

Twenty-one trials reported on poor outcome involving 2286 participants. There were significantly more poor outcomes in the control group than in the TH group (RR 0.81, 95% CI 0.75 to 0.87; p ≤ 0.00001) (Figure 12), with possible substantial heterogeneity (I² = 72%).

FIGURE 12.

Forest plot of poor outcome. M–H, Mantel–Haenszel.

Poor outcome results with quality and risk-of-bias analysis

There were six high-quality studies that showed no significant difference in poor outcomes between the control and TH groups (RR 1.03, 95% CI 0.92 to 1.16; p = 0.56). However, when the 15 low-quality studies were analysed together, they showed more poor outcomes in the control group (RR 0.69, 95% CI 0.62 to 0.76; p < 0.00001).

When the two low-risk-of-bias studies were analysed, they showed that poor outcomes were more likely in the TH group (RR 1.16, 95% CI 1.02 to 1.32; p = 0.03). Conversely, when the 19 studies with a high risk of bias were analysed together, they showed significantly more poor outcomes in the control than the TH group (RR 0.70, 95% CI 0.64 to 0.77; p < 0.00001).

New pneumonia results

Thirteen studies reported new pneumonia, accounting for 1128 participants. There was significantly more new pneumonia in the TH group (RR 1.43, 95% CI 1.15 to 1.78; p = 0.001) (Figure 13) when all studies were analysed together.

FIGURE 13.

Forest plot of new pneumonia data. M–H, Mantel–Haenszel.

New pneumonia results with quality and risk-of-bias analysis

There were three high-quality studies that, when analysed together, showed no significance in new pneumonia between the two groups (RR 1.33, 95% CI 0.79 to 2.25; p = 0.28). In contrast, the 10 low-quality studies suggested significantly more new pneumonia in the TH group (RR 1.35, 95% CI 1.07 to 1.69; p = 0.01).

The two low-risk-of-bias studies showed no significant difference in new pneumonia between the two groups (RR 1.41, 95% CI 0.72 to 1.84; p = 0.32), whereas the 11 studies having a high risk of bias suggested significantly more new pneumonia in the TH group (RR 1.47, 95% CI 1.17 to 1.84; p = 0.0008).

Mortality: comparison of control groups and comparison of controlled normothermia temperature ranges

The first subanalysis on mortality specifically assessed the methodology used for temperature management in the control groups. There was no significant difference in mortality between the TH group and the control group with controlled normothermia (RR 0.89, 95% CI 0.74 to 1.08; p = 0.25) with possible low heterogeneity (I² = 25%). However, there was significantly greater mortality in the no temperature control group than in the TH group (RR 0.81, 95% CI 0.68 to 0.97; p = 0.02) with possible substantial heterogeneity (I² = 63%).

The results for those RCTs that implemented controlled normothermia of any degree were analysed in isolation. The groups were split into those that ensured that their control groups were < 38 °C and those that included 38 °C within their normothermia control range. Those studies that stated controlled normothermia without specific temperature ranges were placed in the latter group. Those studies that controlled to < 38 °C showed no significant difference in mortality outcomes between the control group and TH group (RR 1.01, 95% CI 0.82 to 1.32; p = 0.73). Contrastingly, those control groups that included 38 °C within their normothermia range showed significantly more mortality in their control groups (RR 0.61, 95% CI 0.44 to 0.87; p = 0.005).

Early therapeutic hypothermia compared with late therapeutic hypothermia

When mortality was analysed accounting for timing of TH from time of TBI, there were significant differences between the subgroups (Figure 14). For those participants with early hypothermia (< 24 hours from injury), there was significantly greater mortality in the control group than in the TH group [RR 0.76, 95% CI 0.64 to 0.90; p = 0.001 (1542 participants)] with possible low heterogeneity (I² = 2%). Whereas, among those participants with TH induced ≥ 24 hours after injury, there was no significant difference in mortality between the control and TH groups [RR 1.08, 95% CI 0.86 to 1.34; p = 0.51 (641 participants)] with possible substantial heterogeneity (I² = 79%). There was also possible substantial heterogeneity (I² = 83.5%) between the early and late hypothermia subgroups.

FIGURE 14.

Forest plot of mortality analysed accounting for timing of TH from time of TBI. M–H, Mantel–Haenszel.

Mortality: effect of duration of follow-up

There was variation between the RCTs regarding duration of follow-up. Two groups were created that analysed those RCTs with follow-up for 3 months (five studies; 271 participants) compared with those with follow-up for > 6 months (14 studies; 1999 participants) (Figure 15). For those followed up for 3 months, there was no significant difference in mortality between the control and TH groups (RR 0.73, 95% CI 0.47 to 1.17; p = 0.20) with possible low heterogeneity (I² = 0%), whereas studies with follow-up for > 6 months showed significantly greater mortality in the control group (RR 0.85, 95% CI 0.74 to 0.98; p = 0.02), there was possible moderate heterogeneity in this subgroup (I² = 55%). There was no heterogeneity between the two subgroups (I² = 0%).

FIGURE 15.

Forest plot of the effect of duration of follow-up on trial results. M–H, Mantel–Haenszel.

Mortality: date of publication

Those studies that reported mortality data were split into two groups: five were published before 2000 (223 participants) and 15 were published in or after 2000 (2047 participants). Those published before 2000 showed no significant difference in mortality between the groups (RR 0.71, 95% CI 0.49 to 1.04; p = 0.08), whereas those studies published in or after 2000 found higher mortality in the control group than the TH group (RR 0.86, 95% CI 0.74 to 0.99; p = 0.03) (Figure 16).

FIGURE 16.

Forest plot of the effect of date of publication on outcome. M–H, Mantel–Haenszel.

Summary of results

Overall, the results of this systematic review shed doubt on the overall outcome benefits of TH. Low-quality studies are more likely than high-quality studies to show that TH improves mortality. In addition, this review also highlights the importance of temperature management in the control group of TH trials and, in particular, the avoidance of fever.

Explanation of results

Adding to the literature, this review has shown that the control group management is important in assessing the efficacy of TH. When the temperature of the control group is controlled (controlled normothermia), avoiding fever, then TH was no longer beneficial. Furthermore, when controlled normothermia included ‘fever’ temperatures then the control group had a greater mortality. This is significant because it suggests that it is not just the implementation of TH that is important, but also the maintenance of adequate fever control. The importance of fever control has implications for the conclusions suggested by RCTs; for example, the Eurotherm3235 Trial had a control population with values of normothermia and the results showed harm from TH. Studies that do not adequately maintain normothermia in their control groups may be overestimating the benefits of TH, as worse overall outcomes in the control groups may be down to poor fever control rather than lack of TH.

Subgroup heterogeneity appears to be important in a variety of analyses. When looking at the induction of TH, there is a lower mortality in those who had TH implemented within 24 hours; however, cooling after 24 hours conferred no mortality benefit compared with the control treatment. This could suggest that cooling early is most beneficial or that there are differences between the subgroups; this is supported by the possible substantial heterogeneity. The results of the ongoing POLAR (Prophylactic Hypothermia trial to Lessen Traumatic Brain Injury) study123 of early hypothermia for neuroprotection could provide further evidence to support this hypothesis. Those patients with late TH may have more severe injuries with more difficult to control ICP meaning that treatment was less likely to be successful regardless of time of TH onset. The two most recent RCTs both investigated late-onset TH. 1,107

Other studies have shown that subgroup differences are important in relation to surgical intervention; this is not always accounted for during statistical analysis of mortality and poor outcomes. These studies have provided evidence that TH may be more beneficial in those patients who are cooled after surgical haematoma evacuation than in those with diffuse brain injury. This review was unable to investigate this further because diffuse and focal TBI patients were commonly grouped together during data analysis.

Overall, this finding is important as it suggests that there may be certain subgroups of TBI patients who are more likely to benefit from TH. Differences could depend on the type of injury, surgical intervention before TH, early or late cooling and the temperature management within the control group. With all of these possible confounding factors, it is unsurprising that there are conclusion discrepancies among the most recent reviews.

Quality of evidence

Two methods were used to assess quality and the risk of bias: a modified Jadad score was used in conjunction with a domain-based assessment for risk of bias. The two latest RCTs1,107 were of high quality and had a low risk of bias. However, all other studies showed a high risk of bias as assessed by the objective domains. Previously, concern has been expressed that small, single-centre studies are distorting the conclusions. 110 This review highlights the importance of this as, when the larger, higher-quality RCTs were analysed separately, they showed consistently different conclusions to other trials.

Deviation from systematic review protocol

The original primary and secondary outcomes remained as published in the protocol (see Report Supplementary Material 1). However, several additional subgroup analyses were added as secondary outcomes following protocol publication.

Of the original secondary aims, some subgroup analyses could not take place because of a lack of information supplied within RCTs or homogeneity between the studies. These were depth of cooling on outcomes, effect of rate of rewarming, the duration of cooling on outcomes and the use of propofol sedation.

Data extraction was completed as per protocol. There was always agreement between the two co-authors undertaking the data extraction but a reason for exclusion of an article was not always documented.

The Cochrane Collaboration’s tool for assessing the risk of bias109 was used rather than the 1998 Downs and Black checklist. 124 The modified Jadad score for trial quality was added to increase the objectivity for claims of quality and risk of bias.

Limitations of the review

The main limitations of the previous reviews were the inclusion of many single-centre, low-quality RCTs; this has now been partly addressed with the addition of the two latest RCTs. However, as discussed above, underlying subgroup differences may be affecting the overall conclusions.

There was significant variation in the length of follow-up post injury, ranging from 3 months to 2 years. This has also been a factor in other systematic reviews. It is not possible to control this confounder because of the variations in the standalone RCT protocols. In the context of hypothermia after cardiac arrest, it has been suggested that 6 months is the earliest time point that neurological status should be assessed. 125

This review updated some of the original data used in the Crossley et al. 110 paper and these data extraction amendments have been included and analysed in this paper.

Similarities to, and differences from, other reviews

Although this review has several similarities to recent review papers, it also has some significant differences. Crossley et al. 110 were able to demonstrate a significant reduction in mortality and poor outcomes with the use of TH. However, Zhu et al. 111 found no difference in either of these outcomes between the TH and control groups. The latter systematic review is in agreement with other RCTs published since 2015: the Eurotherm3235 Trial found evidence of harm from TH,1 whereas B-HYPO (Brain Hypothermia study)107 found no improvement in outcomes from TH when compared with strict fever control or controlled normothermia; both studies assessed the use of late TH post TBI. Importantly, both of these RCTs avoided fever in their control groups, which does differ from the control groups of previous RCTs. 110,111 Previous systematic reviews reported a general belief that TH was associated with new pneumonia. However, until this review, only Zhu et al. 111 had been able to show significance from this result. This current systematic review has gone some way to explain the varying conclusions on the effectiveness of TH in TBI by analysing the main subgroups.

Ongoing trials of therapeutic hypothermia for traumatic brain injury

There are two ongoing RCTs that are yet to be published: the POLAR RCT126 and the RCT of Long-term Mild Hypothermia for Severe Traumatic Brain Injury trial (LTH-1). 127 Both author groups were contacted during the production of this systematic review.

The POLAR RCT is a neuroprotection study, the results of which may support the finding of this review that early hypothermia is beneficial.

In the earlier 2014 systematic review,110 the B-HYPO107 investigators kindly provided preliminary data from their RCT. The full RCT has now been published and has been included in this updated systematic review.

Key messages

• High-quality studies show no significant difference in mortality, poor outcomes or new pneumonia.

• When temperature in the control group is controlled normothermia, there is no difference in mortality.

Statistical analysis

For all data collected, descriptive statistics by allocated treatment and dichotomised by GOSE outcome (favourable vs. unfavourable) were generated.

Data

Responses were received from a total of 33 centres out of the 47 centres in the main trial, of which 16 (48%) were UK and 17 (52%) were non-UK sites. A total of 286 anonymised CRFs (UK, n = 163; and non-UK, n = 123) were returned by e-mail or post to the Eurotherm3235 Trial office by 31 July 2016. Data were returned on 143 participants randomised to TH and 143 participants randomised to the control groups. This distribution was similar to the main trial.

Demographics and baseline characteristics

The demographics of this sample were well matched between the TH and control group and were very similar to those of the main trial ITT population. Table 13 shows that the baseline characteristics of the two ‘additional data collection’ groups were similar in all respects. Furthermore, the TH and control groups were well balanced for comorbidities (see Table 13).

The incidence of pre-ICU admission of secondary brain injuries because of either hypoxia [n = 27 (20.9%) in TH vs. n = 17 (13.2%) in the control group; p = 0.136] or hypotension [n = 15 (12.0%) in TH vs. n = 11 (8.6%) in the control group; p = 0.493] was not different between the groups. However, clinical signs of hypoxia were proportionally more frequent in the TH group (n = 12, 44.4%) than in the control group (n = 1, 6.3%), and blood gas evidence of hypoxia was more frequent in the control group [13 (48.2%) in TH vs. 13 (81.3%) in the control group; p = 0.031], although the absolute number of arterial gases identifying hypoxia were identical in both groups (see Table 13).

Additional data collection sample mortality

Hypothermia was associated with a statistically significant increased risk of death at 6 months (Figure 17). In the additional data collection sample, a total of 78 participants died: 46 participants in the TH group (32.2%) and 32 participants in the control group (22.5%) (HR 1.624, 95% CI 1.025 to 2.573; p = 0.0388). These data were consistent with the ITT study population and suggest no reporting bias in this retrospective cohort.

FIGURE 17.

Survival curve between randomisation and 6 months by treatment allocation in the additional data collection cohort.

Intracranial pressure management: medical therapy

The main trial results showed that the mean daily ICP was similar in the two groups. During the first 4 days after randomisation, there were fewer failures of stage II therapy to control ICP in TH than in the control group (57 vs. 84 participants). In this exploratory analysis, the use of medical treatments to control ICP was similar in both groups over days 1–7 after randomisation: 128 participants (92.1%) in the TH group received medical treatments for raised ICP compared with 124 participants (91.9%) in the control group (p = 0.88) (see Table 15). There was no significant difference in the average number of participants treated with hypertonic saline per day [mean 35.0 (SD 6.2) in the TH group vs. mean 39.6 (SD 10.9) in the control group; p = 0.353] or the average number of daily treatments given [mean 1.9 (SD 0.7) in TH group vs. mean 2.5 (SD 0.9) in the control group; p = 0.15]. Similarly, for mannitol, the average number of participants treated (mean 37.0 (SD 9.4) in the TH group vs. [mean 43.3 (SD 16.0) in the control group; p = 0.388)] and the average number of daily treatments [2.0 (2.0–2.0) in the TH group vs. 2.0 (2.0–2.0) in the control group; p = 0.71] were not significantly different. The average number of daily treatments with furosemide was statistically significantly higher in the control group than the TH group [mean 1.2 (SD 0.4) in the TH group vs. mean 1.9 (SD 0.4) in the control group; p = 0.004], but there was no difference in the average number of participants treated per day [mean 13.3 (SD 4.3) in the TH group vs. mean 15.7 (SD 5.6) in the control group; p = 0.384] (Table 15).

Intracranial pressure management: surgical treatment

The surgical management of raised ICP was also similar between groups. The number of participants undergoing decompressive craniectomy before randomisation was 21 (15.4%) in the TH group, compared with 23 (16.3%) in the control group (p = 0.952), and after randomisation was 34 (24.8%) in the TH group, compared with 37 (26.2%) in the control group (p = 0.893) (see Table 15). The proportion of participants undergoing hemicraniectomy was slightly greater in the control group than in the TH group [29 (58.0%) in TH group vs. 37 (64.9%) in the control group], whereas bifrontal craniectomy was more common in the TH group than in the control group [21 (42.0%) in the TH group vs. 20 (35.1%) in the control group]. Only one participant in each group underwent decompressive craniectomy before and after randomisation. However, these distributions were not significantly different between the groups (p = 0.593) (see Table 15).

Follow-up computed tomography

Follow-up CT was performed in significantly more participants in the control group than the TH group: 105 (80.8%) in TH group compared with 120 (90.2%) in the control group (p = 0.045) (Table 16). The distribution of Marshall scores based on the second CT scan was similar in both groups, with no statistically significant differences (p = 0.323). In both the TH and CG groups of the additional collection exercise, type II diffuse injury and high- or mixed-density lesion not evacuated were the most common pathologies observed [26 participants (24.8%) in the TH group vs. 31 participants (26.7%) in the control group, for both] (see Table 16). For both the TH group and the control group, the distribution of Marshall scores with the type of decompressive craniectomy was not significantly different (p = 0.506 in the TH group; p = 0.189 in the control group) (see Table 16). In the TH group, removable lesions were most frequently associated with both hemicraniectomy and bifrontal craniectomy [10 participants (41.7%) had a hemicraniectomy vs. five participants (38.5%) who had a bifrontal procedure]. By contrast, in the control group, removable lesions were most commonly associated with hemicraniectomy (14 participants, 46.7%), but a type II diffuse injury was slightly more common in those receiving bifrontal craniectomy (four participants, 28.6%). For both hemicraniectomy and bifrontal craniectomy, the distribution of Marshall scores by treatment allocation was also not significantly different (p = 0.787 for hemicraniectomy; p = 0.598 for bifrontal craniectomy) (see Table 16).

Sedation and analgesia

Midazolam infusion was used more often in the TH group: the mean number of participants per day was 96.0 (SD 12.8) in the TH group compared with 78.7 (SD 16.0) in the control group (p = 0.046), with the same hourly infusion rate in both groups [median 10.0 (IQR 10.0–10.0) mg/hour in the TH group vs. median 10.0 (IQR 10.0–10.0) mg/hour in the control group; p = 0.71) (Table 16). Alfentanil and fentanyl were the most commonly used opioids in all centres. The number of participants receiving alfentanil each day was higher in the TH group than in the control group [mean 37.1 (SD 4.7) in the TH group vs. mean 27.7 (SD 5.2) in the control group; p = 0.004], but the hourly dose was the same in both groups [median 2.5 (IQR 2.5–2.5) mg/hour in the TH group vs. median 2.5 (IQR 2.5–2.5) mg/hour in the control group; p = 1.0]. In contrast, the number of participants treated per day with fentanyl was not significantly different between groups [mean 56.0 (SD 6.9) in the TH group vs. mean 58.4 (SD 7.7) in the control group; p = 0.545] or the hourly dose [median 250 (IQR 250–250) µg/h in the TH group vs. median 250 (IQR 250–250) µg/h in the control group; p = 0.71] (see Table 17).

The average number of participants sedated with propofol per day was similar in both groups: a total of 89.9 (SD 14.2) participants in the TH group, compared with 79.1 (SD 17.8) participants in the control group (p = 0.236). The mean daily dose of 4615.3 (SD 658.8) mg/day in the TH group compared with 4801.4 (SD 657.0) mg/day in the control group (p = 0.606) (see Table 15). Correcting the daily dose for the participants’ weight did not reveal any between-group difference: 60.0 (SD 8.6) mg/kg/day in the TH group compared with 62.7 (SD 5.4) mg/kg/day in the control group (p = 0.5). After adjustment for age, post resuscitation GCS motor component, time from injury and pupil reactivity, no significant differences in the average daily dose of propofol in each group were found. A potentially toxic level infusion rate of propofol (> 5mg/kg/hour) was used in a few participants per day in both groups [median 3.0 participants (IQR 2.3–5.3) in the TH group and median 4.0 participants (IQR 3.0–4.0) in the control group; p = 0.535].

Atracurium was the predominant neuromuscular blocking agent used in all 33 centres. The number of participants receiving atracurium was significantly greater in the TH group [mean 31.1 (SD 6.8)] than in the control group [mean 22.0 (SD 4.5)] (p = 0.012) after randomisation. However, the median hourly infusion rate was slightly higher in the control group than in the TH group: median 50.0 (IQR 50.0–50.0) mg/hour in the TH group, compared with median 60.0 (IQR 50.8–60.0) mg/hour in the control group (p = 0.017) (see Table 17).

Cardiovascular support

The use of drugs to support the cardiovascular system was greater in the TH group than in the control group. The average number of participants receiving noradrenaline per day [mean 95.4 (SD 12.9) in the TH group vs. mean 78.1 (SD 18.7) in the control group] and the daily infusion rate [mean 15.3 (SD 3.5) mg/day in the TH group vs. mean 12.2 (SD 2.5) mg/day in the control group] were not significantly different (p = 0.067 and p = 0.078, respectively) (Table 18). Dobutamine was used in more TH participants [mean 7.3 (SD 3.1) participants/day] than control group participants [mean 2.3 (SD 1.0) participants/day] (p = 0.002), although the daily infusion rates were similar [median 292.5 (IQR 221.3–466.9) mg/day in the TH group vs. median 315.0 (IQR 193.9–441.9) mg/day in the control group; p = 0.71]. Similarly, more participants received daily vasopressin in the TH group [mean 5.7 (SD 2.4)] than in the control group [mean 1.9 (SD 0.9)] (p = 0.002), with a daily dose of median 24.9 (IQR 16.7–31.6) UI/day in the TH group compared with median 60.4 (IQR 6.0–778.8) UI/day in the control group (p = 0.71) (see Table 18).

Steroids

The number of participants receiving steroids was greater in the TH group [23 (17.8%)] than in the control group [9 (6.7%)] (p = 0.01) (Table 19). The most frequently used steroid was hydrocortisone: 20 participants, with a median daily dose of 100 mg (range 25–600 mg), used this steroid in the TH group compared with six participants, with a median daily dose of 125 mg (range 50–200 mg), in the control group.

Fluid balance

Daily fluid balance was similar in both groups. The number of participants per day with a positive fluid balance [mean 84.4 participants (SD 19.2) in the TH group vs. mean 87.3 (SD 17.0) in the control group; p = 0.773] or negative fluid balance [mean 32.9 (SD 13.9) in the TH group vs. mean 30.9 (SD 11.3) in the control group; p = 0.773] was similar. Average daily fluid balance was also similar between the groups [mean 781.7 (SD 409.9) ml in TH vs. mean 657.3 (SD 406.6) ml in control group; p = 0.579] (Table 19).

Laboratory results and organ failure

There were no significant differences in the average daily haemoglobin [mean 102.4 (SD 2.8) g/l in the TH group vs. mean 102.8 (SD 4.6) g/l in the control group; p = 0.839], haematocrit [mean 0.3 (SD 0.0) in the TH group vs. mean 0.3 (SD 0.0) in the control group; p = 0.139] or platelet count [mean 163.1 (SD 15.2) × 10⁹/l in the TH group vs. mean 195.5 (SD 36.9) × 10⁹/l in the control group; p = 0.053] between the two groups (Table 20).

Liver function was not significantly different in the two groups. There were no significant differences between the two groups for average daily bilirubin [mean 8.2 (SD 0.7) µmol/l in the TH group vs. mean 8.3 (SD 1.3) µmol/l in the control group; p = 0.832], alanine transaminase (ALT) [median 32.0 (IQR 29.3–41.5) U/l in the TH group vs. median 30.5 (IQR 29.0–44.8) U/l in the control group; p = 1.0] and albumin [median 25.8 (IQR 25.5–27.4) g/l in the TH group vs. median 26.9 (IQR 26.5–28.5) g/l in the control group; p = 0.128] (see Table 20). Average daily APTT (activated partial thromboplastin time) [mean 33.2 (SD 2.1) seconds in the TH group vs. mean 31.2 (SD 1.3) seconds in the control group; p = 0.046] and APTTr (activated partial thromboplastin time ratio) [mean 1.2 (SD 0.1) in the TH group vs. mean 1.1 (SD 0.0) in the control group; p = 0.037] were significantly increased in the TH group.

Renal function was not significantly different in the two groups for average daily serum urea [mean 4.30 (SD 0.83) mmol/l in the TH group vs. mean 4.33 (SD 1.0) mmol/l in the control group; p = 0.959] or creatinine [mean 54.9 (SD 4.9) µmol/l in the TH group vs. mean 57.9 (SD 3.3) µmol/l in the control group; p = 0.205] (see Table 20). Electrolyte concentrations were identical in the TH and control groups: serum sodium was median 145.2 (range 144.2–145.4) mmol/l in the TH group, compared with median 145.0 (range 142.8–145.6) mmol/l in the control group (p = 0.62) and potassium was mean 4.03 (SD 0.14) mmol/l in the TH group, compared with mean 4.03 (SD 0.11) mmol/l in the control group; p = 1].

There was no difference in the number of participants requiring renal replacement therapy (RRT). Seven participants (5.5%) in the TH group required RRT, compared with three participants (2.4%) in the control group (p = 0.33) (see Table 20). Evidence of rhabdomyolysis was reported in the same number of participants in each group [11 (9.1%) in the TH group vs. 11 (8.9%) in the control group; p = 0.87]. Additionally, elevated levels of creatine kinase and/or troponin T were reported in a similar proportion of participants across the two groups [37 (33.9%) in the TH group vs. 30 (28.6%) in the control group; p = 0.484]. The average serum creatinine kinase level was median 753 (IQR 505–1280) U/l in the TH group, compared with median 1013 (IQR 606–1554) U/l in the control group (p = 0.336), and the troponin T level was median 41.0 (IQR 10.0–142.0) ng/l in the TH group, compared with median 15.0 (IQR 3.755–38.3) ng/l in the control group (p = 0.46).

Acid–base estimation

Metabolic acidosis was reported significantly more often in the TH group than in the control group [45 participants (36.9%) in the TH group vs. 27 participants (22.1%) in the control group; p = 0.017], as was the use of an infusion of bicarbonate [17 (13.7%) in the TH group vs. 5 (4.2%) in the control group; p = 0.018]. Average daily lactate concentrations were significantly greater in the TH group than in the control group [mean 1.2 (SD 0.1) mmol/l in the TH group vs. mean 1.1 (SD 0.1) mmol/l in the control group; p = 0.027], although serum bicarbonate concentrations were similar in both groups [mean 23.0 (SD 0.7) mmol/l in the TH group vs. mean 23.6 (SD 0.5) mmol/l in the control group; p = 0.10].

Organ dysfunction

Organ dysfunction was quantified using a modified abbreviated SOFA score. All analyses were made comparing minimal (SOFA score of ≤ 1) with significant (SOFA score of > 1) organ dysfunction. There was no significant difference in the distribution of SOFA scores between the TH group and the control group; 47 participants (32.9%) in the TH group and 55 participants (38.5%) in the control group had a SOFA score of ≤ 1, and 96 participants (67.1%) in the TH group and 88 participants (61.5%) in the control group had a SOFA score of > 1 (p = 0.388) (Table 21). In the TH group, there were significantly more participants who died with a SOFA score of > 1 [39 (84.8%) died vs. 57 (58.8%) survived] than with a SOFA score of ≤ 1 [7 (15.2%) died vs. 40 (41.2%) survived] (p = 0.004). This was not seen in the control group: 21 participants (65.6%) died compared with 66 participants (60.0%) survived with a SOFA score of > 1, and 11 participants (34.4%) died compared with 44 participants (40.0%) survived with a SOFA score of ≤ 1 (p = 0.712). The distribution of SOFA scores was not significantly different for the dichotomised GOSE in either the TH group [73 (71.6%) unfavourable vs. 23 (59.0%) favourable with a SOFA score of > 1, and 29 (28.4%) poor vs. 16 (41.0%) good outcome with a SOFA score of ≤ 1; p = 0.218] or the control group [53 (60.2%) poor vs. 33 (62.3%) good outcome with a SOFA score of > 1 and 35 (39.8%) poor vs. 20 (37.7%) good outcome with a SOFA score of ≤ 1; p = 0.951].

Immunity and infection

Therapeutic hypothermia was associated with a significantly lower average daily leucocyte count [mean 9.9 (SD 0.9) × 10⁹/l in the TH group vs. 11.0 (SD 0.7) × 10⁹/l in the control group; p = 0.031]. Similarly, the neutrophil count was also reduced [mean 7.8 (SD 0.6) x10⁹/l in the TH group vs. mean 9.0 (SD 0.7) ×10⁹/l in the control group; p = 0.005]. However, there was no statistically significant difference between the two groups for average daily C-reactive protein [mean 148.8 (SD 31.9) mg/l in the TH group vs. mean 160.4 (SD 31.9) mg/l in the control group; p = 0.511].

The number of participants suspected of having infection was similar over days 1–7 of data collection [111 (81.6%) in the TH group vs. 106 (78.5%) in the control group; p = 0.627]. Confirmed infections were also similar [98 (79.0%) in the TH group vs. 94 (77.0%) in the control group; p = 0.825] (Table 22). Pneumonia was the most common infection in both groups, although there was no significant difference in the number of episodes between the two groups [84 (58.7%) episodes in the TH group vs. 78 (54.5%) in the control group; p = 0.551]. Endotracheal aspiration was the technique most frequently used to confirm a respiratory infection [35 participants (44.3%) in the TH group vs. 29 participants (40.8%) in the control group; p = 0.754]. There was no significant difference between the two groups in the number of participants who received antibiotics [116 (89.2%) in the TH group vs. 113 (87.6%) in the control group; p = 0.828].

Coenrolment

The number of participants coenrolled in other trials was similar in the two groups: 16 out of 143 participants (11.4%) in the TH group and 12 out of 143 participants (8.6%) in the control group (p = 0.55). The distribution of trials to which participants were coenrolled was not different. Coenrolment was not more frequent in non-survivors than in survivors in either the TH group [5 (11.1%) non-survivors vs. 11 (11.6%) survivors; p = 0.839] or the control group [3 (9.7%) non-survivors vs. 8 (7.4%) survivors; p = 0.972]. Similarly, coenrolment was not more common in those participants with an unfavourable outcome by GOSE in the TH group [11 (11.2%) unfavourable outcome vs. 3 (7.9%) good outcome; p = 0.796] or the control group [7 (8.2%) poor outcome vs. 4 (7.7%) good outcome; p = 0.833].

Method of cooling

The majority of participants were cooled with an automated device and surface cooling; Blanketrol^® (Patient Temperature Management Solutions, Cincinnati, OH, USA) and ARCTIC SUN^® 5000 Temperature Management System (Bard Medical Division, Covington, GA, USA) were the most commonly used devices (58.2% of the TH group sample), although a small percentage of participants were cooled with an intravascular device (ZOLL^®; ZOLL, Pitsburgh, PA, USA) (23.9% of the TH group sample).

Unfavourable outcome at 6 months by GOSE was not associated with the type of cooling device used (p = 0.441).