IJCA-12750; No of Pages 9

International Journal of Cardiology xxx (2010) xxxxxx

Contents lists available at ScienceDirect

International Journal of Cardiology

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / i j c a r d

Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis
Niklas Nielsen a,, Hans Friberg b, Christian Gluud c, Johan Herlitz d, Jrn Wetterslev c
a Department of Clinical Sciences, Section of Anesthesia and Intensive Care, Lund University, S-221 85 Lund, Sweden and Department of Anesthesia and Intensive Care, Helsingborg Hospital, S-251 87 Helsingborg, Sweden b Department of Clinical Sciences, Section of Anesthesia and Intensive Care, Lund University, S-221 85 Lund, Sweden and Department of Emergency Medicine, Lund University Hospital, S-221 85 Lund, Sweden c Copenhagen Trial Unit, Center for Clinical Intervention Research, Rigshospitalet, Copenhagen University Hospital, DK-2100 Copenhagen, Denmark d Department of Cardiology, Sahlgrenska University Hospital, S-413 19 Gothenburg, Sweden

a r t i c l e

i n f o

a b s t r a c t
Background: Guidelines recommend mild induced hypothermia (MIH) to reduce mortality and neurological impairment after out-of-hospital cardiac arrest. Our objective was to systematically evaluate the evidence for MIH taking into consideration the risks of systematic and random error and to GRADE the evidence. Methods: Systematic review with meta-analysis and trial sequential analysis of randomised trials evaluating MIH after cardiac arrest in adults. We searched CENTRAL, MEDLINE, and EMBASE databases until May 2009. Retrieved trials were evaluated with Cochrane methodology. Meta-analytic estimates were calculated with random- and xed-effects models and random errors were evaluated with trial sequential analysis (TSA). Results: Five randomised trials (478 patients) were included. All trials had substantial risk of bias. The relative risk (RR) for death was 0.84 (95% condence interval (CI) 0.70 to 1.01) and for poor neurological outcome 0.78 (95% CI 0.64 to 0.95). For the two trials with least risk of bias the RR for death was 0.92 (95% CI 0.56 to 1.51) and for poor neurological outcome 0.92 (95% condence interval 0.56 to 1.50). TSA indicated lack of rm evidence for a benecial effect. The substantial risk of bias and concerns with directness rated down the quality of the evidence to low. Conclusions: Evidence regarding MIH after out-of-hospital cardiac arrest is still inconclusive and associated with non-negligible risks of systematic and random errors. Using GRADE-methodology, we conclude that the quality of evidence is low. Our ndings demonstrate that clinical equipoise exists and that large welldesigned randomised trials with low risk of bias are needed. 2010 Elsevier Ireland Ltd. All rights reserved.

Article history: Received 29 March 2010 Accepted 4 June 2010 Available online xxxx Keywords: Induced hypothermia Heart arrest Mortality Neurological outcome Intensive care Critical care

1. Introduction Between 40 and 54 patients per 100,000 inhabitants per year suffer out-of-hospital cardiac arrest where cardiopulmonary resuscitation is attempted [1,2]. Return of spontaneous circulation (ROSC) is achieved in 25% to 40% of the patients [3,4]. Survival to hospital discharge [47] and persistent neurological decits vary considerably [4,711] but the mortality and risk of neurological impairment is high. Global brain ischaemia and the reperfusion injury following resuscitation may lead to brain tissue degeneration and loss of neurological function [12]. Avoidance of fever and mild induced hypothermia (MIH) (3336 C) seem to mitigate this damage in the experimental setting [1317].

Corresponding author. Department of Anaesthesia and Intensive Care, Helsingborg Hospital, S. Vallgatan 5, S-251 87 Helsingborg, Sweden. Tel.: +46 42 4061000. E-mail address: niklas.nielsen@telia.com (N. Nielsen). 0167-5273/$ see front matter 2010 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ijcard.2010.06.008

Guidelines from the American Heart Association [18] and the European Resuscitation Council [19] recommend MIH based mainly on two trials [20,21]. These trials and guidelines have focused interest on a previously neglected patient group that thus may have beneted, but criticism has been raised that guidelines were based on small trials with high risks of bias including only subsets of the relevant population [2226]. A recent Cochrane review [27] supports the guidelines, but this review merely considered some bias components, included trials with high risk of bias in the main analysis, did not consider the risks of random errors, and did not GRADE the evidence. Accordingly, the retrieved data may not justify the conclusion, a previously encountered problem with some Cochrane reviews and guidelines [2830]. In many countries MIH has become standard care [3134], while practice varies in others [3537]. The objective of this review was to systematically evaluate the benets and harms of MIH for adult cardiac arrest patients taking into account the risks of systematic errors (bias) and random errors (play of chance). To quantify the estimated effect of MIH, we conducted

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

2 Table 1 Terms for data base search.

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx Trials Registry Platform Search Portal for missed, unreported, or ongoing trials. Trials were included if they were randomised, included human adults resuscitated from outof-hospital cardiac arrest, compared MIH with control, and were reported in abstract or paper form. 2.2. Data selection and extraction NN screened the titles and abstracts for relevant studies. NN and JW independently extracted data from the retrieved trials in a predened data extraction form. Disagreement in the initial interpretation was resolved by discussion and residual disagreement was resolved by discussion with CG. For extracted trial characteristics see Table 2. We evaluated the validity and design characteristics of each trial and potential sources of bias (adequate random sequence generation, adequate allocation concealment, adequate blinding, baseline differences, adequate use of interim analyses, adequate stopping rules and the possibility to perform intention to treat analysis) according to the Cochrane Handbook (Table 3) [38]. We were aware of the inherent problems with blinding of the intervention with MIH and considered blinding adequate if the outcome assessors had been blinded to allocation group. Trials were dened as having a low risk of bias if they fullled the above criteria. The relevant outcomes were mortality, neurological function, and adverse events. Neurological function was evaluated according to cerebral performance category (CPC) [43], where CPC 1 and 2 was dened as a good neurological outcome and 35 a poor neurological outcome. We used data from the longest follow-up for each trial. 2.3. Meta-analysis

1) Randomised clinical trials 2) Cardiac arrest OR out of hospital cardiac arrest OR OHCA OR in hospital cardiac arrest OR IHCA OR circulatory arrest OR heart stop OR resuscitation OR cardiopulmonary resuscitation OR CPR 3) Hypothermia OR therapeutic hypothermia OR induced hypothermia OR mild hypothermia OR temperature control OR temperature management OR thermoregulatory management OR thermoregulatory control OR chill therapy OR cooling OR body temperature 4) Neurological function OR neurological recovery OR neurological outcome OR cerebral performance category OR CPC OR survival OR outcome OR temperature OR ischemia OR brain ischemia OR cerebral ischemia OR global ischemia 5) Humans 1) AND 2) AND (3) OR 4)) AND 5) IHCA, in-hospital cardiac arrest; OHCA, out-of-hospital cardiac arrest.

meta-analyses [38] and trial sequential analyses (TSA) [39]. We also summarised the evidence using GRADE [40]. Finally we wanted to establish if clinical equipoise exist with regard to further clinical trials on MIH after cardiac arrest [41].
2. Methods We used the Cochrane Collaboration-methodology [38], the PRISMA-guidelines [42], and the GRADE-guidelines [40] when conducting this systematic review. A protocol can be found on www.ctu.rh.dk. 2.1. Selection of trials We searched the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE and EMBASE until May 2009. For search terms see Table 1. There were no language restrictions. We hand searched the reference lists of the retrieved trials and relevant reviews for additional trials. We contacted experts in the eld and searched ClinicalTrials.gov, Centre Watch Clinical Trials Listing Service and International Clinical

Data were summarised as relative risks (RR) with 95% condence intervals (CI). A priori, we assumed that substantial clinical heterogeneity would be present in the included trials. We therefore planned to apply both random-effects [44] and xedeffects models [45] for all meta-analyses. Heterogeneity was identied by visual inspection of the forest plots and by using a standard Chi2-test with a signicance level of P = 0.1. Heterogeneity was examined with I2, where I2 values of 50% and more indicate a substantial level of heterogeneity [38] and diversity (D2) [46]. All forest plots and meta-analytic estimates were calculated with RevMan5 [47]. 2.4. Trial sequential analysis Meta-analyses may result in type-I errors due to an increased risk of random error when few data are collected and due to repeated signicance testing when a cumulative

Table 2 Characteristics of the ve included trials with a total of 478 patients. Mori 2000 (abstract) Duration Participants HachimiIdrissi 2001 HACA 2002 65 months Unconscious CA patients, cardiac cause of arrest, initial rhythm VF or non-perfusing VT Air cooling induced hypothermia to 3234 C for 24 h, passive rewarming for 8 h Bernard 2002 33 months Unconscious OHCA patients, cardiac cause of arrest, initial rhythm VF or VT Laurent 2005

Not reported 6 months OHCA patients with GCS b 8 Unconscious OHCA patients, cardiac cause of arrest, initial rhythm asystole or PEA Helmet cooling to 34 C. When temperature of 34 C achieved or more than 4 h elapsed from start of cooling, passive rewarming for 8 h. Standard ICU care, acetaminophen if temperature over 38 C OHCA of cardiac origin Asystole or PEA as initial rhythm, N 18 years, temp N 30 C, GCS b 7

Experimental MIH to 3234 C for 72 h, intervention method of cooling not described, rewarming rate not reported Control 36 C for 72 h, method of intervention temperature control not described Inclusion OHCA and GCS b 8 criteria

23 months Unconscious OHCA patients, cardiac cause of arrest, initial rhythm VF or asystole Ice-pack induced hypothermia CVVH to 32-33 C (CVVH to 33 C for 12 h (started pre- for 8 h and surface cooling hospitally), active rewarming for 16 h), passive for 6 h rewarming Standard ICU care, no temperature control OHCA with VF as initial rhythm, persistent coma CVVH maintaining 37 C for 8 h, thereafter no temperature control OHCA of cardiac origin, VF or asystole, 1875 years, b 10 min to start of CPR, b 50 min to ROSC

Standard ICU care, no temperature control Witnessed CA of cardiac origin, VF or non-perfusing VT as initial rhythm, 1875 years, 515 min from arrest to CPR and b 60 min to ROSC b 30 C, coma because of drugs before CA, pregnancy, response to verbal command, MAP b 60 for N 30 min, hypoxemia N 15 min, terminal illness, factors making follow-up unlikely, coagulopathy, other study, CA after arrival of medical personnel 6 months 3551 275

Exclusion criteria

Not dened

Pregnancy, coagulopathy, CNS depressant medication before CA, cardiogenic shock (MAP b 60), GCS 7

b 18 years for men b 50 years Pregnancy, response to for women cardiogenic shock verbal command, terminal b 90 SBP despite epinephrine, illness before CA other causes of coma than CA, no available ICU bed

Follow-up time 1 month Patients Not reported screened (n) Patients 54 included (n)

14 days Not reported 30

Hospital discharge Not reported (84 eligible) 77

6 months 244 42

CA = cardiac arrest. CNS = central nervous system. CPR = cardiopulmonary resuscitation. CVVH = continuous veno venous ltration. GCS = Glasgow Coma Score. HACA = hypothermia after cardiac arrest. ICU = intensive care unit. MAP = mean arterial pressure. MIH = mild induced hypothermia. OHCA = out-of-hospital cardiac arrest. PEA = pulseless electrical activity. ROSC = return of spontaneous circulation. SBP = systolic blood pressure. VF = ventricular brillation. VT = ventricular tachycardia.

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx Table 3 Description of how risk of bias was evaluated, based on the Cochrane Handbook. Risk of bias/ Type of bias Sequence generation Allocation concealment Blinding Incomplete data outcomes Low risk of bias If the allocation sequence is generated by a computer or random number table or similar. If the allocation of patients involves a central independent unit, on-site locked computer, or consecutively numbered, sealed envelopes. If the outcome assessors are blinded and the method of blinding is described. If there are no post-randomization drop-outs or withdrawals. Uncertain risk of bias If the trial is described as randomised, but the method used for the allocation sequence generation was not described. If the trial is described as randomised, but the method used to conceal the allocation is not described. If the outcome assessors are blinded and the method of blinding is not described. If it is not clear whether there are any drop-outs or withdrawals or if the reasons for these drop-outs are not clear. High risk of bias If a system involving dates, names, or admittance numbers are used for the allocation of patients (quasi-randomised). If the allocation sequence is known to the investigators who assigned participants or if the study is quasi-randomised. If the outcome assessors are not blinded. If the reasons for missing data are likely to be related to true outcomes, as-treated analysis is performed, potential for patients with missing outcomes to induce clinically relevant bias in effect estimate or effect size. If not all the pre-specied outcomes are reported, or if the primary outcomes are changed, or if some of the important outcomes are incompletely reported. If there was a baseline imbalance due to chance or due to imbalanced exclusion after randomization. If the trial is stopped early without formal stopping rules. If the trial is funded by an instrument, equipment, or drug manufacturer. If the author of the trial has conducted previous trials addressing the same interventions.

Selective outcome reporting

Baseline imbalance Early stopping

If all the important outcomes are reported or if the trial's protocol is available and all of the trial's pre-specied (primary and secondary) outcomes that are of interest in the review have been reported in the pre-specied way. If there was no baseline imbalance in important characteristics. If sample-size calculation is reported and the trial is not stopped or the trial is stopped early by an adequate stopping rule. If the trial is without specic funding, or is not funded by an instrument, equipment, or drug manufacturer. If the author of the trial has not conducted previous trials addressing the same interventions.

If there is insufcient information to assess whether the risk of selective outcome reporting is present.

If the baseline characteristics were not reported.

If sample size calculations are not reported and it is not clear whether the trial is not stopped early. If the source of funding is not clear.

Sponsor bias

Academic bias

If it is not clear if the author has conducted previous trials addressing the same interventions.

meta-analysis is updated with new trials [39,48]. To assess the risk of type-I errors, we used TSA. TSA combines information size estimation for meta-analysis (cumulated sample size of included trials) with an adjusted threshold for statistical signicance in the cumulative meta-analysis [39,48,49]. The latter, called trial sequential monitoring boundaries, reduce type-I errors. In TSA, the addition of each trial in a cumulative metaanalysis is regarded as an interim meta-analysis and helps to clarify whether additional trials are needed or not. The idea in TSA is that if the cumulative Z-curve crosses the trial sequential monitoring boundary, a sufcient level of evidence has been reached and no further trials are needed. If the Z-curve does not cross the boundary and the required information size has not been reached there is insufcient evidence to reach a conclusion [39,48,50,51]. We applied TSA since it reduces the risk of type-I error in a cumulative meta-analysis and may provide important information on how many more patients need to be included in further trials. Information size was calculated as diversity-adjusted information size (DIS) [46], suggested by the relative risk reduction (RRR) of the intervention in the included trials. 2.5. GRADE-criteria We summarised the evidence applying GRADE-levels [40] (high, moderate, low, and very low) by evaluating design, quality, consistency, precision, directness and possible publication bias of the included trials using GRADEpro-version 3.2.2-software [52]. 3. Results 3.1. Search results We identied 6165 references and excluded 6126 after screening titles and abstracts for hypothermia, cardiac arrest, and randomised trial. We retrieved 39 articles in full paper format of which four fullled our search criteria [20,21,53,54]. One additional trial [55], only presented as an abstract, was found in the reference list of one of the retrieved trials (Fig. 1). 3.2. Characteristics of included randomised trials The ve trials included 478 patients. The included patients were with few exceptions adult patients with out-of-hospital cardiac arrest randomised to MIH of 32 34 C versus control intervention. Three trials used external cooling [20,21,53], one did not report the method of cooling [55], and one used continuous veno-venous hemoltration to achieve target temperature [54]. Four trials had neurological function as primary outcome [20,21,54,55] and one was a feasibility study aimed at evaluating a cooling strategy [53]. Four trials reported on mortality [20,21,53,54]. Length of followup ranged from hospital discharge to six months (Table 2).

3.3. Systematic errors None of the trials had low risk of bias [38]. Two trials [21,54] reported adequate generation of allocation sequence, allocation concealment, blinding of outcome assessors (procedure not specied), intention-to-treat analyses and provided 180day follow-up. However, these trials did not report the level of coma prior to randomisation, and reported baseline imbalance. Also, both trials were terminated early without reporting robust power calculations, one stated lack of funding and low enrolment [21] and the other [54] because of results derived from other trials [20,21]. We characterised them as having the least risk of bias among the trials included.

Fig. 1. PRISMA-ow chart of search for relevant references.

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx included trials using a random-effects model (RRR of 16% regarding mortality and 979 patients; RRR of 22% regarding poor neurological function and 729 patients). TSA indicated lack of rm evidence for a benecial effect of MIH for both mortality (Fig. 2) and a poor neurological outcome, since the monitoring boundaries were not nally surpassed and the required information size not reached. None of the estimated required information sizes were accrued, and an intervention effect of 16%, for a reduction of mortality, or of 22%, for a reduction of poor neurological function, could not be excluded.

None of the included trials reported the full details on how the decision to withdraw intensive care treatment was determined and if the assessor of the prognosis was blinded to allocation of intervention. The time for outcome evaluation was hospital discharge for two trials [20,53], and 30 days for one trial [55]. One trial [20] used quasi-randomisation with odd and even dates. This trial also had major baseline differences and conducted a non-scheduled interim analysis after inclusion of 80% of the patients with no adjustment of the P-value (Table 4). Because of differences in risk of bias, we present the effect estimates of MIH in the trials with least risk of bias [21,54], high risk of bias [53,55], and the quasi-randomised trial [20] separately, but also a meta-analysis of all trials using xed- and randomeffects models. 3.4. All-cause mortality In the trials with least risk of bias the RR of all-cause mortality for MIH versus control intervention was 0.92 (95% condence interval 0.56 to 1.51). In the trial with high risk of bias the RR was 0.88 (0.66 to 1.15). In the quasi-randomised trial the RR was 0.76 (0.52 to 1.10). A meta-analysis of all trials showed a RR of 0.84 (0.70 to 1.01). These analyses were performed using a random-effects model and none were statistically signicant (Table 5). In a xed-effects model of all trials, the RR did not change substantially but the results became statistically signicant (RR 0.80 (0.68 to 0.96)). 3.5. Neurological function In the trials with least risk of bias the RR of a poor neurological outcome was 0.92 (0.56 to 1.50). In the trials with high risk of bias the RR was 0.72 (0.43 to 1.20). In the quasirandomised trial the RR was 0.70 (0.49 to 0.99). All the trials showed a RR of 0.78 (0.64 to 0.95). These analyses were performed using a random-effects model (Table 6). In a xedeffect model, the RR of all trials did not change substantially (RR 0.76 (0.66 to 0.88)). 3.6. Adverse events There were only two trials reporting on adverse events, but since these two trials did not report the same set of adverse events, a meta-analytical estimate could not be calculated. However, there was a trend towards more complications in the MIH group in one of the trials [21]. 3.7. Random errors TSA were calculated with = 0.05 and = 0.20 (power 80%) and a required diversity-adjusted information size based on the intervention effect suggested by the

3.8. Summary of evidence according to GRADE Randomised trials per se are rated high on the GRADE scale. As indicated above there were variable but serious risks of bias of all the trials, leading us to downgrade the quality of the evidence. Apart from one trial [54] there was no serious inconsistency between trials. One of the trials, representing 59% of the randomised participants in this review, included only 8% of the screened cardiac arrest population presenting at the emergency department with ROSC [21]. Two trials included only patients with ventricular brillation or pulseless ventricular tachycardia without circulatory instability [20,21]. One trial included exclusively patients with asystole or pulseless electrical activity [53]. Hence there was a questionable directness. The accumulated sample size and event rates were low and the two trials with least risk of bias had, when data were pooled, a wide condence interval spanning both the potential for benet and harm. However according to our judgement and in context with the other measures, this may not implicate serious imprecision. Our application of GRADE-methodology led us to conclude that the accumulated evidence is of low quality. For a GRADE prole see Table 7.

4. Discussion In summary, we found ve randomised trials including 478 patients investigating MIH after cardiac arrest. These trials were all associated with substantial risk of systematic errors. The summarised point estimates indicate a potential benecial effect of MIH to reduce mortality and poor neurological function for patients resuscitated from cardiac arrest, albeit the reduction of mortality was not signicant in the random-effects model. The combined effect estimates for the trials with least risk of bias were neutral for both mortality and neurological outcome. TSA could not conrm or reject the intervention effect. Using

Table 4 Components of risk of bias for the individual randomised clinical trials. Trial/ Type of bias Sequence generation Allocation concealment Blinding Mori 2000 Not described U Not described U Not described U Not described U HachimiIdrissi 2001 Not described U Not described U Not described U No lost to follow-up L HACA 2002 Adequate L Sealed envelopes L Assessors blinded, procedure not described U One lost to follow-up in each allocation group. Intention to treat analysis performed. L All important outcomes reported L Imbalance in baseline characteristics H Trial stopped early without formal stopping rules H No funding with conicts of interest L No previous trials on hypothermia L Bernard 2002 Odd- and even days H Quasi-randomised H Assessors blinded, procedure described L No lost to follow-up H Laurent 2005 Adequate L Sealed envelopes L Not described U Not described U

Incomplete data outcomes

Selective outcome reporting Baseline imbalance

Sample size, power calculations, interim analysis and early stopping Sponsor bias

Only neurological outcome reported H Important baseline characteristics not described. U No sample size reported U

All important outcomes reported L Important baseline characteristics not described. U No sample size reported U No funding with conicts of interest L No previous trials on hypothermia L

Not described U No previous trials on hypothermia L

Academic bias

All important outcomes reported L Imbalance in baseline characteristics. Comorbidities not reported H Interim analysis without adequate correction of signicance H No funding with conicts of interest L No previous trials on hypothermia L

All important outcomes reported L Important baseline characteristics not described. U Trial stopped early without formal stopping rules H No funding with conicts of interest L No previous trials on hypothermia L

H = high risk of bias, L = low risk of bias, U = uncertain risk of bias.

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx Table 5 Forest plot of the effect of mild induced hypothermia on mortality suggested by the randomised and quasi-randomised clinical trials (424 patients). Study or subgroup MIH Events Total Control Events Total 138 20 158 Weight 36.0% 12.6% 48.6% Risk ratio M-H, Random, 95% CI 0.74 [0.58, 0.95] 1.24 [0.76, 2.02] 0.92 [0.56, 1.51] Risk ratio M-H, Random, 95% CI

1.1.1 Trials with least risk of bias HACA 2002 56 137 76 Laurent 2005 15 22 11 Subtotal (95% CI) 159 Total events 71 87 Heterogeneity: Tau2 = 0.09; Chi2 = 3.39, df = 1 (P = 0.07); I2 = 70% Test for overall effect: Z = 0.34 (P = 0.74) 1.1.2 Trials with high risk of bias HachimiIdrissi 2001 13 Subtotal (95% CI) Total events 13 Heterogeneity: Not applicable Test for overall effect: Z = 0.95 (P = 0.34)

16 16

13 13

14 14

31.6% 31.6%

0.88 [0.66, 1.15] 0.88 [0.66, 1.15]

1.1.3 Quasi-randomised trials Bernard 2002 22 43 23 Subtotal (95% CI) 43 Total events 22 23 Heterogeneity: Not applicable Test for overall effect: Z = 1.47 (P = 0.14) Total (95% CI) 218 Total events 106 123 2 2 Heterogeneity: Tau = 0.01; Chi = 3.86, df = 3 (P = 0.28); I2 = 22% Test for overall effect: Z = 1.90 (P = 0.06) Test for subgroup differences: Not applicable

34 34

19.9% 19.9%

0.76 [0.52, 1.10] 0.76 [0.52, 1.10]

206

100.0%

0.84 [0.70, 1.01]

Risk ratio with 95% condence interval. df = degrees of freedom. I2 = heterogeneity. HACA = hypothermia after cardiac arrest. MIH = mild induced hypothermia.

GRADE-methodology we are led to conclude that the quality of the evidence is low. Observational data indicate a similar or better survival today for cardiac arrest patients compared to historical controls and case series [5661]. Observational studies and registries may answer questions of a descriptive nature, elucidate rare adverse events, generate hypotheses, and can thus be used in the design of clinical trials [31,33].

However, they can rarely prove efcacy of an intervention because of inherently high risks of bias caused by confounding by indication, possible selective reporting, and the inclusion of non-comparable groups. Also, improved intensive care with careful attention to vital parameters, metabolic control, circulatory support and coronary interventions may have had an effect on the reported outcome [61]. Many factors are tentative contributors to the potential improvement

Table 6 Forest plot of the effect of mild induced hypothermia on neurological function suggested by the randomised and quasi-randomised clinical trials (478 patients). Study or subgroup MIH Events Total Control Events Total 138 20 158 Weight 26.2% 11.9% 38.1% Risk ratio M-H, Random, 95% CI 0.74 [0.59, 0.93] 1.24 [0.76, 2.02] 0.92 [0.56, 1.50] Risk ratio M-H, Random, 95% CI

1.2.1 Trials with least risk of bias HACA 2002 62 37 84 Laurent 2005 15 22 11 Subtotal (95% CI) 159 Total events 77 95 Heterogeneity: Tau2 = 0.09; Chi2 = 3.47, df = 1 (P = 0.06); I2 = 71% Test for overall effect: Z = 0.35 (P = 0.73) 1.2.2 Trials with high risk of bias HachimiIdrissi 2001 14 16 14 Mori 2001 18 36 16 Subtotal (95% CI) 52 Total events 32 30 Heterogeneity: Tau2 = 0.11; Chi2 = 5.81, df = 1 (P = 0.02); I2 = 83% Test for overall effect: Z = 1.27 (P = 0.20) 1.2.3 Quasi-randomised trials Bernard 2002 22 43 25 Subtotal (95% CI) 43 Total events 22 25 Heterogeneity: Not applicable Test for overall effect: Z = 2.00 (P = 0.05) Total (95% CI) 254 Total events 131 150 2 2 Heterogeneity: Tau = 0.03; Chi = 8.52, df = 4 (P = 0.07); I2 = 53% Test for overall effect: Z = 2.42 (P = 0.02) Test for subgroup differences: Not applicable

14 18 32

26.9% 17.2% 44.1%

0.88 [0.71, 1.10] 0.56 [0.39, 0.81] 0.72 [0.43, 1.20]

34 34

17.7% 17.7%

0.70 [0.49, 0.99] 0.70 [0.49, 0.99]

224

100.0%

0.78 [0.64, 0.95]

Risk ratio with 95% condence interval. df = degrees of freedom. I2 = heterogeneity. HACA = hypothermia after cardiac arrest. MIH = mild induced hypothermia.

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx

Fig. 2. Trial sequential analysis for a relative risk reduction of all-cause mortality of 16% of hypothermia after cardiac arrest in 4 trials with 424 patients reporting mortality. A required diversity-adjusted information size (DIS) of 979 patients was calculated based on a control event proportion of 59%, a hypothermia induced relative risk reduction of mortality of 16% suggested by all trials, = 0.05 two-sided, = 0.20 (power = 80%), and diversity D2 = 23%. The cumulated Z-curve (blue) crosses the traditional boundary (P = 0.05) but not the trial sequential monitoring boundary indicating lack of rm evidence for a benecial effect of 16% relative risk reduction of the intervention when the analysis is adjusted for repetitive testing on accumulating data. There is insufcient information to reject or detect an intervention effect of 16% RRR of all-cause mortality as the required information size is not yet reached.

for cardiac arrest patients and the role for MIH must in our opinion be further investigated. 4.1. Systematic errors Two studies were adequately randomised, two were inadequately reported, and one was quasi-randomised. Inadequate allocation sequence generation and non-concealed allocation is associated with an overestimation of intervention effects, even more than what may be caused by inadequate blinding of intervention and outcome assessment [62]. Such inadequacies create concern about the reliability of the trials. The quasi-randomised trial [20] continued inclusion of patients for one year after rst stopping and analysing the trial at 80% of the nal sample, without adjusting the signicance level, why the results may be substantially biased. This interim-like-analysis should be regarded as an unplanned adaptive design necessitating an adjustment of the level of signicance [63]. This trial used hospital outcome as the endpoint, which may be troublesome since the neurological status for survivors evolve over the rst six months after the arrest [64].

None of the trials specied how a decision on withdrawal of intensive care was made and whether the assessor of the prognostication was blinded. The duration of active intensive care treatment will change prognostication of outcome and also the chance of surviving [65,66]. In future trials randomising cardiac arrest patients, it is paramount that prognostication and withdrawal from intensive care is predened, standardised and reported. We observed baseline differences in those trials that did report baseline characteristics [20,21]. However, none of the trials reported the level of coma before randomisation, hence we cannot be sure that this factor was equally distributed between the groups. Level of coma at admission is shown to be an important predictor of outcome [5,33]. The two trials with least risk of bias were both terminated early. A priori power calculations were not reported. Sample-size calculations for well-designed trials should be based on realistic event rates in the control group, expectations of plausible and moderate intervention effects, and appropriate type-I and II error rates [50]. A relative risk reduction (RRR) of 20% may be regarded as an appropriate a priori treatment effect for MIH when related to studies in cardiology on intervention effect for the outcomes myocardial infarction, stroke, and mortality [50]. With out-of-hospital cardiac arrest and mortality higher than 50% for patients admitted to intensive care, this would suggest a randomised trial with approximately 700 patients to detect or reject a RRR of 20% with a power of 80% and a type-1 error risk of 5%. The largest trial [21] was terminated after 285 included patients and 132 events for the primary outcome, because of low enrolment and lack of funding. However, we do not know if the stopping was informative of or dependent on the results [25]. A data monitoring committee would probably be reluctant to stop a trial for benet at a halfway interim analysis with a P-value as high as 0.02 for mortality and 0.009 for neurological outcome, as P-values of 0.001 (or lower) are recommended [67]. The trial may therefore have been underpowered and should possibly be considered as having been stopped too early for benet, which is associated with overestimated intervention effects [68]. 4.2. Random errors Since none of the trials on MIH had low risk of bias, one could argue that they should not be incorporated into meta-analyses or into TSA. However, to illustrate that the body of evidence is not conclusive, we have chosen to give the trials the full benet of doubt and include them in TSA for mortality and poor neurological function, as if they

Table 7 GRADE prole for assessing quality of evidence for mild induced hypothermia after out-of-hospital cardiac arrest. Quality assessment Summary of ndings No. of patients No. of Design studies Limitations Inconsistency Indirectness Imprecision Other Mild induced Control considerations hypothermia 106/218 (48.6%) Effect Relative (95% CI) Absolute Quality Importance

Mortality (follow-up 180 daysa) 4 Randomised Seriousb trials

No serious inconsistency

Seriousc

No serious None imprecisiond

123/206 RR 0.84 (59.7%) (0.7 to 1.01)

96 fewer per 1000 (from 179 fewer to 6 more)

CRITICAL LOW

Neurological function (follow-up 180 daysa; Cerebral performance categorye) No serious Seriousc No serious 5 Randomised Seriousb inconsistency imprecision trials

None

131/254 (51.6%)

150/224 RR 0.78 147 fewer per 1000 CRITICAL (67%) (0.64 to 0.95) (from 33 fewer LOW to 241 fewer)

ROSC, return of spontaneous circulation; RR, relative risk. a The two trials with least risk of bias had 180 days follow-up. The two trial with high risk of bias had follow-up until 14 days. The quasi-randomised trial had follow up until hospital discharge. b All trials had substantial risk of bias. c The largest trial included only 8% of the screened patients with ROSC in the emergency department. The second largest trial included only a subset of the cardiac arrest population. d The total sample size is limited and event rates are low. The two trials with least risk of bias have a wide condence interval spanning both the potential for benet and harm. We address this problem with Trial Sequential Analysis. In context with the other measures of evidence quality we choose not to downgrade on imprecision. e The second largest trial judged neurological outcome based on where to patients were discharged.

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx

were of low risk of bias and with impeccable design. TSA show that there is lack of rm evidence for a benecial effect and an insufcient information size to reject the anticipated intervention effect, even if the trials had been of high quality. Thus, the question whether MIH is benecial, neutral or harmful, for adult out-of-hospital cardiac arrest patients still needs an answer. 4.3. GRADE-directness The exclusion of 92% of the screened cardiac arrest patients with ROSC in the largest trial [21] was motivated to increase power. The investigators chose a population with an ischaemic insult that theoretically might have a high chance of getting a benecial effect of MIH (survival approximately 50%). In this context, the study was designed as a possible proof-of-concept trial, but it limits the directness of the results. Since the other trials also were selective in their inclusion this fact was, together with the risk of bias, the reason to downgrade the quality to low. 4.4. Limitations and strengths It may be that we did not nd all the relevant trials of MIH after out-of-hospital cardiac arrest, but our data from included trials are consistent with the recently published Cochrane review [27] and our search strategy was comprehensive using Cochrane-methodology. We have discussed the intervention with experts in the eld and we nd it highly unlikely that there are trials refuting our results. Contrary to other reviews of MIH, we have evaluated systematic and random errors and incorporated the GRADE classication, thus broadening the base for a well-founded judgement of the evidence. The risk of type-I errors has not previously been evaluated in this eld, but growing evidence highlights this as one of the major problems of spurious ndings due to meta-analyses [48,69] and may therefore provide a valuable addition. 4.5. Further aspects One major concern when discussing the evidence for MIH is that the majority of patients in the control groups were not treated for fever. The control groups in the trials investigating MIH were possibly allowed to follow their natural temperature course and had median temperatures between 37 C and 38 C with ranges of temperatures being denitely febrile. We do not know if the suggested intervention effect in favour of MIH was due to an increased temperature in the control group, to a benecial effect of MIH, or both. The temperature rise reported may look clinically insignicant, but observational data on survival from cardiac arrest populations in the pre-hypothermia era show odds ratios for a poor outcome of 2.7 (1.26.2) for temperatures N 37.8 C [4] and 2.26 (1.244.12) for every degree higher than 37 C [70]. These data are observational and it is impossible to distinguish between causation and mere association. The majority of the trials in this review compared MIH with no temperature control. It has not been tested whether MIH is superior to normothermia. In clinical practice, controlled normothermia would diminish the risk for possible adverse events related to MIH. The gradual implementation of MIH and the guidelines have shed light upon the cardiac arrest population, where therapeutic nihilism traditionally has prevailed. Recent observational studies indicate that intensive care survival after cardiac arrest is comparable to other emergency diagnoses. In that sense current guidelines and practice might reect the best available evidence, since previous therapies did not consider temperature management and active standardised intensive care. However, after performing this systematic review, we conclude that MIH must be evaluated further to merit its position. Strictly, there are two questions to address: is MIH superior to standard intensive care (often associated with fever) and is MIH

superior to strict normothermia? The available trials have addressed the rst question, but not conclusively, and our ndings suggest that clinical equipoise exists. Meanwhile, MIH has been introduced into clinical practice with a novel focus on temperature management, and there are experimental and observational ndings suggesting that hyperthermia may be detrimental after brain damage. Accordingly, a reasonable way to proceed would be a trial of MIH with the comparator temperature close to normothermia. In case of a benecial effect of the intervention in such a trial, the implementation of MIH will be founded on more solid evidence. If the ndings are negative the question whether normothermia is superior to standard care should be considered. In case of a neutral nding we believe that this probably would stimulate further trials trying to elucidate the optimal target temperature after cardiac arrest, rather than suggest succumbing to previous treatment traditions. Of note is that comparisons with more narrow temperature intervals may need larger sample sizes as there is reason to believe that the intervention effect would be smaller. 5. Conclusion The available randomised trials suggest a possible benecial effect of MIH after cardiac arrest, but the results are still inconclusive. The accumulated evidence is associated with high risks of systematic errors as well as random errors. Thus, the risk of spurious ndings for a benecial effect derived from the cumulative data on MIH for out-ofhospital cardiac arrest is substantial. Using GRADE-methodology we conclude that the quality of the evidence is low. We further conclude that clinical equipoise exists whether to perform new trials on MIH and that future research is very likely to have an important impact on the estimated effect of this intervention. Guidelines and Cochrane reviews should consider these limitations. Contributions of authors NIklas Nielsen (NN): Developed of the initial idea for the review, developed the protocol, undertook searches, selected studies, data extraction, quality assessment of studies, data analysis, contact person, developed nal review, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Hans Friberg (HF): Assessed clinical validity, developed nal review. Christian Gluud (CG): Developed protocol, advised on statistical methods to be used, developed nal review. Johan Herlitz (JH): Assessed clinical validity, developed nal review. Jrn Wetterslev (JW): Proposed the review, developed the protocol, advised on statistical methods to be used, data extraction, quality assessment of studies, data analysis, developed nal review. Conicts of interest None of the authors declare any conicts of interest and have no nancial disclosures. Acknowledgements This study was supported by grants from the Stig and Ragna Gorthon Foundation, Helsingborg, Sweden, Gyllenstierna Krapperup Foundation, Sweden and research grants from Region Skne, Sweden, and the Copenhagen Trial Unit, Denmark. The funding sources did not have any involvement in the study design; the collection, analysis and interpretation of data; in the writing of the review; or the decision to submit the review for publication. The authors of this manuscript have certied that they comply with the Principles of Ethical Publishing in the International Journal of Cardiology [71].

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx [34] Oksanen T, Pettila V, Hynynen M, Varpula T. Therapeutic hypothermia after cardiac arrest: implementation and outcome in Finnish intensive care units. Acta Anaesthesiol Scand 2007;51:86671. [35] Laver SR, Padkin A, Atalla A, Nolan JP. Therapeutic hypothermia after cardiac arrest: a survey of practice in intensive care units in the United Kingdom. Anaesthesia 2006;61:8737. [36] Merchant RM, Soar J, Skrifvars MB, et al. Therapeutic hypothermia utilization among physicians after resuscitation from cardiac arrest. Crit Care Med 2006;34: 193540. [37] Wolfrum S, Radke PW, Pischon T, Willich SN, Schunkert H, Kurowski V. Mild therapeutic hypothermia after cardiac arresta nationwide survey on the implementation of the ILCOR guidelines in German intensive care units. Resuscitation 2007;72:20713. [38] Higgins JPT, Green Se. Cochrane handbook for systematic reviews of interventions, Version 5.0.2. In: Higgins JPT, Green S, editors. The Cochrane Collaboration; 2009. [39] Wetterslev J, Thorlund K, Brok J, Gluud C. Trial sequential analysis may establish when rm evidence is reached in cumulative meta-analysis. J Clin Epidemiol 2008;61:6475. [40] Atkins D, Best D, Briss PA, et al. Grading quality of evidence and strength of recommendations. BMJ 2004;328:1490. [41] Freedman B. Equipoise and the ethics of clinical research. N Engl J Med 1987;317: 1415. [42] Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. J Clin Epidemiol 2009;62: 100612. [43] Jennett B, Bond M. Assessment of outcome after severe brain damage. Lancet 1975;1:4804. [44] DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986;7: 17788. [45] Demets DL. Methods for combining randomized clinical trials: strengths and limitations. Stat Med 1987;6:34150. [46] Wetterslev J, Thorlund K, Brok J, Gluud C. Estimating required information size by quantifying diversity in random-effects model meta-analyses. BMC Med Res Methodol 2009;9:86. [47] The Nordic Cochrane Centre. Review Manager (RevMan). Version 5.0 ed. Copenhagen: The Cochrane Collaboration; 2008. [48] Brok J, Thorlund K, Wetterslev J, Gluud C. Apparently conclusive meta-analyses may be inconclusivetrial sequential analysis adjustment of random error risk due to repetitive testing of accumulating data in apparently conclusive neonatal meta-analyses. Int J Epidemiol 2009;38:28798. [49] Thorlund K, Devereaux PJ, Wetterslev J, et al. Can trial sequential monitoring boundaries reduce spurious inferences from meta-analyses? Int J Epidemiol 2009;38:27686. [50] Pogue J, Yusuf S. Overcoming the limitations of current meta-analysis of randomised controlled trials. Lancet 1998;351:4752. [51] Pogue JM, Yusuf S. Cumulating evidence from randomized trials: utilizing sequential monitoring boundaries for cumulative meta-analysis. Control Clin Trials 1997;18: 58093 discussion 661-6. [52] Brozek J, Oxman A, Schunemann H. GRADEpro for Windows. 3.2 ed2008. [53] Hachimi-Idrissi S, Corne L, Ebinger G, Michotte Y, Huyghens L. Mild hypothermia induced by a helmet device: a clinical feasibility study. Resuscitation 2001;51:27581. [54] Laurent I, Adrie C, Vinsonneau C, et al. High-volume hemoltration after out-ofhospital cardiac arrest: a randomized study. J Am Coll Cardiol 2005;46:4327. [55] Mori K, Takeyama Y, Itoh Y. Multivariate analysis of prognostic factors in survivors of out-of-hospital cardiac arrest with brain hypothermia therapy. Crit Care Med 2000;28:A168. [56] Becker L, Gold LS, Eisenberg M, White L, Hearne T, Rea T. Ventricular brillation in King County, Washington: a 30-year perspective. Resuscitation 2008;79:227. [57] Busch M, Soreide E, Lossius HM, Lexow K, Dickstein K. Rapid implementation of therapeutic hypothermia in comatose out-of-hospital cardiac arrest survivors. Acta Anaesthesiol Scand 2006;50:127783. [58] Hollenberg J, Herlitz J, Lindqvist J, et al. Improved survival after out-of-hospital cardiac arrest is associated with an increase in proportion of emergency crewwitnessed cases and bystander cardiopulmonary resuscitation. Circulation 2008;118:38996. [59] Knafelj R, Radsel P, Ploj T, Noc M. Primary percutaneous coronary intervention and mild induced hypothermia in comatose survivors of ventricular brillation with ST-elevation acute myocardial infarction. Resuscitation 2007;74:22734. [60] Oddo M, Schaller MD, Feihl F, Ribordy V, Liaudet L. From evidence to clinical practice: effective implementation of therapeutic hypothermia to improve patient outcome after cardiac arrest. Crit Care Med 2006;34:186573. [61] Sunde K, Pytte M, Jacobsen D, et al. Implementation of a standardised treatment protocol for post resuscitation care after out-of-hospital cardiac arrest. Resuscitation 2007;73:2939. [62] Wood L, Egger M, Gluud LL, et al. Empirical evidence of bias in treatment effect estimates in controlled trials with different interventions and outcomes: metaepidemiological study. BMJ 2008;336:6015. [63] Brannath W, Konig F, Bauer P. Estimation in exible two stage designs. Stat Med 2006;25:336681. [64] Arrich J, Zeiner A, Sterz F, et al. Factors associated with a change in functional outcome between one month and six months after cardiac arrest: a retrospective cohort study. Resuscitation 2009;80:87680. [65] Nolan JP, Neumar RW, Adrie C. Post-cardiac arrest syndrome: epidemiology, pathophysiology, treatment, and prognostication. A Scientic Statement from the International Liaison Committee on Resuscitation; the American Heart Association Emergency Cardiovascular Care Committee; the Council on Cardiovascular Surgery and Anesthesia; the Council on Cardiopulmonary, Perioperative, and Critical Care; the Council on Clinical Cardiology; the Council on Stroke. Resuscitation 2008;79: 35079.

References
[1] Atwood C, Eisenberg MS, Herlitz J, Rea TD. Incidence of EMS-treated out-ofhospital cardiac arrest in Europe. Resuscitation 2005;67:7580. [2] Rea TD, Eisenberg MS, Sinibaldi G, White RD. Incidence of EMS-treated out-ofhospital cardiac arrest in the United States. Resuscitation 2004;63:1724. [3] Herlitz J, Engdahl J, Svensson L, Angquist KA, Silfverstolpe J, Holmberg S. Major differences in 1-month survival between hospitals in Sweden among initial survivors of out-of-hospital cardiac arrest. Resuscitation 2006;70:4049. [4] Langhelle A, Tyvold SS, Lexow K, Hapnes SA, Sunde K, Steen PA. In-hospital factors associated with improved outcome after out-of-hospital cardiac arrest. A comparison between four regions in Norway. Resuscitation 2003;56:24763. [5] Carr BG, Kahn JM, Merchant RM, Kramer AA, Neumar RW. Inter-hospital variability in post-cardiac arrest mortality. Resuscitation 2009;80:304. [6] Herlitz J, Bang A, Gunnarsson J, et al. Factors associated with survival to hospital discharge among patients hospitalised alive after out of hospital cardiac arrest: change in outcome over 20 years in the community of Goteborg, Sweden. Heart 2003;89:2530. [7] Bottiger BW, Grabner C, Bauer H, et al. Long term outcome after out-of-hospital cardiac arrest with physician staffed emergency medical services: the Utstein style applied to a midsized urban/suburban area. Heart 1999;82:6749. [8] Bunch TJ, White RD, Gersh BJ, et al. Long-term outcomes of out-of-hospital cardiac arrest after successful early debrillation. N Engl J Med 2003;348:262633. [9] Jorgensen EO, Holm S. The natural course of neurological recovery following cardiopulmonary resuscitation. Resuscitation 1998;36:11122. [10] Kuisma M, Alaspaa A. Out-of-hospital cardiac arrests of non-cardiac origin. Epidemiology and outcome. Eur Heart J 1997;18:11228. [11] Sauve MJ, Doolittle N, Walker JA, Paul SM, Scheinman MM. Factors associated with cognitive recovery after cardiopulmonary resuscitation. Am J Crit Care 1996;5:12739. [12] Schneider A, Bottiger BW, Popp E. Cerebral resuscitation after cardiocirculatory arrest. Anesth Analg 2009;108:9719. [13] Busto R, Dietrich WD, Globus MY, Ginsberg MD. Postischemic moderate hypothermia inhibits CA1 hippocampal ischemic neuronal injury. Neurosci Lett 1989;101:299304. [14] Busto R, Dietrich WD, Globus MY, Valdes I, Scheinberg P, Ginsberg MD. Small differences in intraischemic brain temperature critically determine the extent of ischemic neuronal injury. J Cereb Blood Flow Metab 1987;7:72938. [15] Coimbra C, Wieloch T. Moderate hypothermia mitigates neuronal damage in the rat brain when initiated several hours following transient cerebral ischemia. Acta Neuropathol (Berl) 1994;87:32531. [16] Kuboyama K, Safar P, Radovsky A, Tisherman SA, Stezoski SW, Alexander H. Delay in cooling negates the benecial effect of mild resuscitative cerebral hypothermia after cardiac arrest in dogs: a prospective, randomized study. Crit Care Med 1993;21: 134858. [17] Sterz F, Safar P, Tisherman S, Radovsky A, Kuboyama K, Oku K. Mild hypothermic cardiopulmonary resuscitation improves outcome after prolonged cardiac arrest in dogs. Crit Care Med 1991;19:37989. [18] 2005 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Medicine Part 7.5: Postresuscitation Support. Circulation 2005;112:IV848. [19] Nolan JP, Deakin CD, Soar J, Bottiger BW, Smith G. European Resuscitation Council guidelines for resuscitation 2005. Section 4. Adult advanced life support. Resuscitation 2005;67(Suppl 1):S3986. [20] Bernard SA, Gray TW, Buist MD, et al. Treatment of comatose survivors of out-ofhospital cardiac arrest with induced hypothermia. N Engl J Med 2002;346: 55763. [21] HACA-study-group. Mild therapeutic hypothermia to improve the neurologic outcome after cardiac arrest. N Engl J Med 2002;346:54956. [22] Biber B, Waldenstrom A. New data on hypothermia. Now we have to keep cool! Lakartidningen 2004;101:4389. [23] Fisher GC. Hypothermia after cardiac arrest: feasible but is it therapeutic? Anaesthesia 2008;63:8856 author reply 6. [24] Moran JL, Solomon PJ. Application of therapeutic hypothermia in the intensive care unit. Intensive Care Med 2004;30:2288 author reply 7. [25] Moran JL, Solomon PJ. Therapeutic hypothermia after cardiac arrestonce again. Crit Care Resusc 2006;8:1514. [26] Pechlaner C. Joannidis M (Therapeutic hypothermia after cardiopulmonary resuscitationpro and con.). Wien Med Wochenschr 2008;158:62733. [27] Arrich J, Holzer M, Herkner H, Mullner M. Hypothermia for neuroprotection in adults after cardiopulmonary resuscitation. Cochrane Database Syst Rev 2009:CD004128. [28] Bangalore S, Wetterslev J, Pranesh S, Sawhney S, Gluud C, Messerli FH. Perioperative beta blockers in patients having non-cardiac surgery: a meta-analysis. Lancet 2008;372: 196276. [29] Moja LP, Telaro E, D'Amico R, Moschetti I, Coe L, Liberati A. Assessment of methodological quality of primary studies by systematic reviews: results of the metaquality cross sectional study. BMJ 2005;330:1053. [30] Olsen O, Middleton P, Ezzo J, et al. Quality of Cochrane reviews: assessment of sample from 1998. BMJ 2001;323:82932. [31] Arrich J. Clinical application of mild therapeutic hypothermia after cardiac arrest. Crit Care Med 2007;35:10417. [32] Bouwes A, Kuiper MA, Hijdra A, Horn J. Induced hypothermia and determination of neurological outcome after CPR in ICUs in the Netherlands: Results of a survey. Resuscitation 2010;2010:30. [33] Nielsen N, Hovdenes J, Nilsson F, et al. Outcome, timing and adverse events in therapeutic hypothermia after out-of-hospital cardiac arrest. Acta Anaesthesiol Scand 2009;53:92634.

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008

N. Nielsen et al. / International Journal of Cardiology xxx (2010) xxxxxx [66] Young GB. Clinical practice. Neurologic prognosis after cardiac arrest. N Engl J Med 2009;361:60511. [67] Haybittle JL. Repeated assessment of results in clinical trials of cancer treatment. Br J Radiol 1971;44:7937. [68] Montori VM, Devereaux PJ, Adhikari NK, et al. Randomized trials stopped early for benet: a systematic review. JAMA 2005;294:22039.

[69] Borm GF, Lemmers O, Fransen J, Donders R. The evidence provided by a single trial is less reliable than its statistical analysis suggests. J Clin Epidemiol 2009;62:7115 e1. [70] Zeiner A, Holzer M, Sterz F, et al. Hyperthermia after cardiac arrest is associated with an unfavorable neurologic outcome. Arch Intern Med 2001;161:200712. [71] Coats AJ. Ethical authorship and publishing. Int J Cardiol 2009;131:14950.

Please cite this article as: Nielsen N, et al, Hypothermia after cardiac arrest should be further evaluatedA systematic review of randomised trials with meta-analysis and trial sequential analysis, Int J Cardiol (2010), doi:10.1016/j.ijcard.2010.06.008