Updating Evidence for Using Hypothermia

in Pediatric Severe Traumatic Brain Injury:

Conventional and Bayesian Meta-Analytic
Perspectives
Robert C. Tasker, MD, FRCP1,2; Frederick W. Vonberg, MBBS1; Elizabeth D. Ulano, MD1;
Alireza Akhondi-Asl, PhD1

Objective: To evaluate clinical trials of hypothermia management
on outcome in pediatric patients with severe traumatic brain injury
using conventional and Bayesian meta-analyses.
Data Sources: Screening of PubMed and other databases to
identify randomized controlled trials of hypothermia for pediatric
severe traumatic brain injury published before September 2016.
Study Selection: Four investigators assessed and reviewed ran-
domized controlled trial data.
Data Extraction: Details of trial design, patient number,
Glasgow Coma Scale score, hypothermia and control normother-
mia therapy, and outcome of mortality were collated.
Data Synthesis: In conventional meta-analysis, random-effects models
were expressed as odds ratio (odds ratio with 95% credible-interval).
Bayesian outcome probabilities were calculated as probability of odds
risk reduction of death > 20% hypothermia vs normothermia) gives a
chance of relative risk reduction of death by greater than 20% of 1-in-2.
Conclusions: Conventional meta-analysis shows the null hypothe-
sis—no difference between hypothermia versus normothermia on
mortality and poor outcome—cannot be rejected. However, Bayes-
ian meta-analysis shows chance of relative risk reduction of death
greater than 20% with hypothermia versus normothermia is 1-in-3,
which may be further altered by one’s optimistic or skeptical belief
about a patient. (Pediatr Crit Care Med 2017; XX:00–00)
Key Words: Bayesian analysis; meta-analysis; therapeutic
hypothermia; traumatic brain injury

ratio greater than or equal to 1. In seven randomized controlled trials
(n = 472, patients 0–17 yr old), there was no difference in mortality
(hypothermia vs normothermia) with pooled estimate 1.42 (credible-
interval, 0.77–2.61; p = 0.26). Duration of hypothermia (24, 48, or
72 hr) did not show difference in mortality. (Similar results were found
using poor outcome.) Bayesian analyses of randomized controlled
trials ordered by time of study completed recruitment showed, after
the seventh trial, chance of relative risk reduction of death by greater
than 20% is 1-in-3. An optimistic belief (0.90 probability that relative
1
Division of Critical Care Medicine, Department of Anesthesiology, Peri-
operative and Pain Medicine, Boston Children’s Hospital and Harvard
Medical School, Boston, MA.
2
Department of Neurology, Boston Children’s Hospital and Harvard Medi-
cal School, Boston, MA.
Supplemental digital content is available for this article. Direct URL cita-
tions appear in the printed text and are provided in the HTML and PDF
versions of this article on the journal’s website (http://journals.lww.com/
pccmjournal).
The authors have disclosed that they do not have any potential conflicts
of interest.
For information regarding this article, E-mail: Robert.tasker@childrens.
harvard.edu
Copyright © 2017 by the Society of Critical Care Medicine and the World
Federation of Pediatric Intensive and Critical Care Societies
DOI: 10.1097/PCC.0000000000001098
T
he PICU management of severe traumatic brain injury
(sTBI) is challenging. There are no “level I Brain Trauma
Foundation” (BTF) recommendations (1). For “tem-
perature control,” the 2012 BTF recommendations were “weak”
based on “moderate” quality evidence from contradictory class
II/III studies. The “level II” recommendations included “moder-
ate hypothermia (32–33°C) beginning early after sTBI for only
24-hour duration should be avoided; and, moderate hypother-
mia beginning within 8 hours after sTBI for up to 48-hour dura-
tion should be considered to reduce intracranial hypertension.”
Since 2012, there has been new randomized controlled trial
(RCT) evidence of hypothermia versus normothermia after
sTBI in children (2, 3). Recent meta-analyses (4–7) conclude
that in pediatric patients there is no benefit of hypothermia, and
it may even increase risk of mortality. In conventional “frequen-
tist” interpretation of RCTs, the null hypothesis assesses signifi-
cance using effect size, p value, and the CI. The p value is the
probability of observing the effect (or more extreme one) given
the null hypothesis (8); it is not the probability that the null is
true (9). (The 2016 American Statistical Association statement
is “P-values do not measure the probability that the studied
hypothesis is true, or that the data were produced by random
chance alone.” [10]). The problem with decision-making based
on p value is that it is misleading (11): statistically significant

Pediatric Critical Care Medicine www.pccmjournal.org 1

Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
 Tasker et al
difference is not necessarily clinically important; an insignifi-
cant difference does not indicate no difference; and studies with
the similar p values do not provide the same level of evidence
to reject the null hypothesis (9, 12). Therefore, decision-making
is based on effect size and the CI (9), but these measures do not
help at the bedside. For example, the 95% CI indicates that if
the RCT is repeated, then 95% of the CIs obtained will con-
tain the true effect size. We want to know the interval that the
effect size lies within, with the probability greater than or equal
to 95%. Another vulnerability of this approach is the poten-
tial that patients of differing risk stratification, such as Glasgow
Coma Scale (GCS) score, are lumped together.
Frequentist analysis starts with a hypothesis and uses data
to judge plausibility. In contrast, Bayesian analysis is a directed
decision-making approach starting with a prior belief and
using outcome of an analysis to generate a “posterior” or “cur-
rent” probability, which characterizes a renewed belief after
observing data. That is, using both our prior belief (which is
discounted in the frequentist framework) and observed data
to evaluate a hypothesis (13). These posterior and prior beliefs
are represented by probability distributions and the observed
data described by a likelihood distribution (i.e., probability of
observed data conditional on the phenomenon). In hypother-
mia for sTBI, for example, the posterior probability of an effect
of treatment could incorporate the prior probability generated
from previous RCTs. Furthermore, aggregate statistical assess-
ment can be translated into clinically relevant information using
what the clinician believes from experience or interpretation of a
patient’s condition. In the Bayesian framework, the “subjective”
prior may be enthusiastic, noninformative, or skeptical (14, 15).
In the past, it was impractical to implement Bayesian approaches
to meta-analyses because calculating the posterior distribution
requires major computation. However, using the Markov Chain
Monte Carlo algorithm and computer simulation means that
Bayesian analyses are now used frequently in medicine (16–20).
We have used frequentist and Bayesian approaches to assess
data from pediatric RCTs of hypothermia after sTBI in order
to better understand current clinical recommendations (1). We
show that Bayesian analysis can add clinically useful informa-
tion to the interpretation of clinical trials that cannot be pro-
vided by the frequentist decision framework.
METHODS
Search Strategy
A systematic search of English language literature using
PubMed (2000 to 31 August 2016) search terms “hypother-
mia,” “traumatic brain injury,” “head injury,” and “intracranial
pressure.” RCTs in pediatric patients were identified. We hand-
searched reference lists of four meta-analyses of therapeutic
hypothermia in pediatric sTBI (4–7). We looked at the catalog
of RCTs in the PICU population at www.picutrials.net. Last,
we checked a federated literature search engine (ACCESSSS,
see: http://plus.mcmaster.ca/ACCESSSS/). The abstracts were
analyzed for relevance and eligible articles were assessed. Refer-
ence lists from RCTs were also checked for further resources.
2 www.pccmjournal.org XXX 2017 • Volume XX • Number XXX
Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
Study Selection Criteria, Quality Assessment and
Data Extraction
The minimum inclusion for meta-analysis was use of induced
systemic hypothermia for greater than or equal to 12 hours and
survival at PICU discharge. Quality of evidence was assessed
using the Grading of Recommendations Assessment, Develop-
ment, and Evaluation system (21). Data extraction included
trial design, patient number, GCS score; hypothermia and
control normothermia therapy; and outcomes of mortality
and poor/unfavorable outcome. The division between poor/
unfavorable and favorable outcome was defined using the
Glasgow Outcome Scale (GOS) or the Pediatric Cerebral Per-
formance Category (PCPC) (22, 23). In reports that used GOS,
patients with scores 3 (severe outcome), 4 (vegetative state),
or 5 (death) were grouped as poor/unfavorable outcome. In
reports that used the PCPC, patients with scores 4 (severe dis-
ability), 5 (vegetative state), or 6 (death) were grouped as poor/
unfavorable outcome.
Statistical Analysis
In the conventional meta-analysis, we used risk ratio (RR) of
death (or poor outcomes) as the effect measure to compare
therapy (24). “R-Studio software” (R Studio, Inc., Boston, MA;
https://www.rstudio.com/) was used to model data and per-
form meta-analysis with “meta-package” (https://cran.r-proj-
ect.org/web/packages/meta/index.html). The random-effects
model was used to account for heterogeneity of populations,
where the number of deaths and sample size of each trial was
used to compute effect size and CI using the inverse variance
method and the DerSimonian-Laird heterogeneity estimator
(25). A Forest plot was performed using meta-package. Het-
erogeneity was investigated by comparing relative risks of sub-
groups defined by hypothermia duration.
In the cumulative Bayesian random-effects meta-analysis,
we used log (RR) in R-Studio software, where we assumed
Gaussian distribution (26) (rjags package, https://cran.r-proj-
ect.org/web/packages/rjags/index.html). JAGS software (Just
Another Gibbs Sample v4.2.0; http://mcmc-jags.sourceforge.
net/) was used to sample the posterior density. A noninfor-
mative prior was used for both mean and variance of the log
(RR) (27). At each time point, this prior and the studies up to
that time were used in the meta-analysis (14, 15, 28). Then,
the posterior probability of relative risk reduction (RRR) being
greater than 0% and 20% was calculated after each study.
Last, since Bayesian analysis can incorporate prior belief, we
have investigated the impact of prior skeptical and optimistic
beliefs with two informative priors for mean log (RR), that is,
probabilities of 0.01 or 0.90, respectively. These probabilities
are commonly used in Bayesian clinical decision-making stud-
ies (28–33). We have also calculated the posterior probability
of relative risk increase (RRI) in mortality greater than 20%.
RESULTS
The literature search is illustrated in Figure S1 (Supplemental
Digital Content 1, http://links.lww.com/PCC/A389). Thirty-five
 Neurocritical Care

clinical trials were hand-searched to assess suitability for inclu-
sion. At first, nine RCTs were reviewed (2, 3, 34–39), and these
studies were compared with RCTs in recent meta-analyses (4–
7). None of the studies in previous meta-analyses were missed.
However, we found that two reports used subsets of previously
published RCTs: Bourdages et al (34) reported cases from
Hutchison et al (38) and Salonia et al (35) reported cases from
Adelson et al (37). The authors reached consensus on the qual-
ity of the remaining RCTs (Table 1). The overall quality was
low: two trials were high (3, 38), and in the other trials one was
moderate (36), and four were low (2, 37, 39).
There were outlier studies (Table 2). Mean time to starting
cooling after injury was 4.2–6.8 hours in six studies; in contrast,
in the study by Adelson et al (37) mean time was 15 hours. Site
of temperature measurement differed across studies: three used
esophageal temperature (2, 36, 38); two, rectal temperature (37);
one, intracranial temperature (39); and one, either intracranial
or rectal temperature (3). Intended temperature in the normo-
thermia group was 36–37.5°C, but in one study that used exclu-
sively intracranial temperature, the range was 37.5–38.5°C (39),
which is equivalent to pyrexia. Hypothermia ranged 32–34°C
but the duration of temperature management varied: 24 hours
(38), 48 hours (3, 36, 37), or 72 hours (2, 39). Rewarming after
hypothermia was given in six studies, and varied: short, 12–18
hours (36–38); intermediate duration, 16–35 hours (2); or pro-
longed, 42–54 hours (3).
Classic Meta-Analysis
Statistical heterogeneity was low (p = 0.27; I2 = 20.4%). None
of the trials found benefit of hypothermia (Fig. 1). The fun-
nel plot of effect of hypothermia versus normothermia on
mortality did not indicate publication bias (Fig. 2). Using RR

Table 1. Summary of Randomized Controlled Trials Examining Temperature Management

in Pediatric Traumatic Brain Injury
Randomized
Controlled Trial
GCS Exclusions Exclusions Statistical Plan
Reference. Age Score (TBI-Specific) (General) Recruitment (a priori)

Biswas et 0–17 yr ≤8 Ventriculoperitoneal CA March 1998 to Unclear

al (36) shunt or Refractory shock April 1999
ventriculoatrial shunt Coagulopathy
Suspected brain death Abdominal injury
Adelson < 13 yr (a); ≤ 8 Penetrating brain Hypotension July 1999 to Unclear
et al (37) ≤ 17 yr (b) injury Coagulopathy December
Normal CT 2003 (a)
Suspected brain death August 2001 to
December
2003 (b)
Hutchison 1–17 yr ≤8 Penetrating brain Premorbid severe February 1999 to 222 patients
et al (38) injury NDD October 2004 ARR 20%
Suspected brain Prolonged CA (unfavorable)
death Refractory shock 80% power
Cervical SCI Suspected iTBI
Isolated EDH Suspected pregnancy
Li et al Child ≤8 Not given Premorbid severe January 2006 to Unclear
(39) (0.5–9 yr) NDD August 2007
Premorbid chronic
disability
Other injuries
Adelson 0–17 yr ≤ 8 and < motor FD pupils and GCS 3 Hypotension November 2007 340 patients
et al (3) score 6 in the Normal CT Coagulopathy to February 2011 ARR 10% (death)
motor subscale Hypoxia (trial suspended) 80% power
score of the Acute injury score ≥ 4
GCS Suspected pregnancy
Beca 1–15 yr ≤8 Penetrating brain Premorbid > mild November 2006 Unclear
et al (2) injury developmental delay to June 2008
FD pupils and GCS 3 Refractory shock (trial suspended)
Cervical SCI Suspected iTBI
Acute EDH
Posttraumatic seizure
and normal CT
ARR = absolute risk reduction, CA = cardiac arrest, EDH = extradural hemorrhage, FD = fixed and dilated, GCS = Glasgow Coma Scale, iTBI = inflicted
traumatic brain injury, NDD = neurodevelopmental delay, SCI = spinal cord injury.

Pediatric Critical Care Medicine www.pccmjournal.org 3

Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
 Tasker et al

Table 2. Quality Assessment and the Temperature Intervention in Pediatric Traumatic

Brain Injury Randomized Controlled Trials
Grading of Evidence of Ad-
Recommenda- Setting equate Allocation Temperatures and Duration
tions Assessment, (Single or Concealment/ (Normothermia vs Site of Temperature
Development, and Multiple Outcome Assessor No. of Hypothermia Measurement; Time to
Reference Evaluation Quality Centers) Blinding Cases [Rewarming Time]) Start Cooling

Biswas Moderate Single Yes/yes 21 36.5–37.5°C vs 32–34°C Esophageal;

et al (36) 4.6 ± 1.4 hr (mean
48 hr vs 48 (12) hr ± sd)
Adelson Low (a) Multiple (a) Yes/yes 48 (a) 36.5–37.5°C vs 32–33°C Rectal; 4.6 ± 1.1 hr (a)
et al (37)
Low (b) Single (b) 26 (b) 48 hr vs 48 (12–18) hr 15.0 ± 7.1 hr (b)
Hutchison High Multiple Yes/yes 225 36.5–37.5°C vs 32–33°C Esophageal;
et al (38) 6.3 ± 2.3 hr
24 hr vs 24 (14–18) hr
Li et al (39) Low Single Unclear/unclear 22 37.5–38.5°C vs 34.1– Intracranial;
34.4°C 6.8 ± 1.2 hr
72 hr vs 72 (unclear) hr
Adelson High Multiple Yes/yes 77 36.5–37.5°C vs 32–33°C Rectal or
et al (3) intracranial;
48 hr vs 48 (42–54) hr 5.1 ± 0.6 hr
Beca Low Multiple Yes/unclear 50 36–37°C vs 32–33°C Esophageal;
et al (2) 4.2–6.0 hr (range)
72 hr vs 72 (16–35) hr

of death in hypothermia versus normothermia, the pooled
estimate with a random-effects model is 1.42 (95% CI, 0.77–
2.61; p = 0.26). Table 3 shows the changes in the overall esti-
mates when conducting the sensitivity analysis. That is, after
removing one study at a time and rerunning the model, the
estimates changed little—which implies the results are statis-
tically reliable.
Figure 3 shows the Forest plot of hypothermia ver-
sus normothermia with studies grouped by duration of
hypothermia (24, 48, and 72 hr). The statistical heteroge-
neity of these subgroups was low (24 hr, no analysis since
only one study; 48 hr, p = 0.16, I2 = 42.2%; 72 hr, p = 0.20,
I2 = 39.1%). The results showed no difference in hypother-
mia versus normothermia. Effect sizes and 95% CIs were
24 hours, RR = 1.78 (95% CI, 0.97–3.28; p = 0.06); 48 hours,
RR = 1.29 (95% CI, 0.42–3.95; p = 0.65); and 72 hours,
RR = 1.18 (95% CI, 0.16–8.84; p = 0.87). Duration of hypo-
thermia did not show a difference in the magnitude of the
association with mortality (p = 0.84). These same analyses
were repeated using poor outcome and were no different
to the results of mortality (Fig. S2, Supplemental Digital
Content 1, http://links.lww.com/PCC/A389).

Figure 1. Forest plots for risk ratio (RR) of mortality in severe traumatic brain injury associated with hypothermia (HT) versus normothermia (NT)
treatment. Size of the squares is proportional to the size of the cohorts. Error bars represent 95% CI. The diamond shape represents the pooled estimates
within each analysis.

4 www.pccmjournal.org XXX 2017 • Volume XX • Number XXX

Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
 Neurocritical Care

normothermia. An optimistic belief (0.90 probability that RRR

of death > 20%) gives a current probability of 0.50 (1-in-2) of
RRR of death greater than 20% on theoretically using hypother-
mia rather than normothermia. The probability of RRI of death
greater than 20% is 0.28 (1-in-3) if choosing hypothermia rather
than normothermia. The Bayesian analysis for poor outcome
provides similar findings and is given in Figure S4 (Supplemental
Digital Content 1, http://links.lww.com/PCC/A389).

Figure 2. Funnel plot of all included randomized controlled trials
examining the effect of publication bias in trials reporting mortality of
hypothermia versus normothermia in pediatric traumatic brain injury. The
symmetry of this plot suggests that publication bias is unlikely.
Bayesian Meta-Analysis
After ordering the studies by month/yr of recruitment closure, we
carried out a Bayesian cumulative meta-analysis using log (RR)
with a noninformative prior (Fig. 4). The figure shows that after
the seventh RCT (2), the current probability of reducing mor-
tality with hypothermia compared with normothermia is 0.40
(with RR < 1 or RRR > 0%). The current probability of RRR of
death greater than 20%, with hypothermia rather than normo-
thermia, is 0.28 (close to 1-in-3, see RRR > 20% curve in Fig. 4).
Figures 5 and S3 (Supplemental Digital Content 1, http://
links.lww.com/PCC/A389) show the effect of an a priori belief
on current probability of RRR of death greater than 20%. Based
on the seven RCTs, a skeptical belief (0.01 probability that RRR of
death > 20% when using hypothermia rather than normother-
mia) gives a current probability of 0.16 (1-in-6) of RRR of death
greater than 20% on theoretically using hypothermia rather than
normothermia. However, the probability of RRI of death greater
than 20% is 0.61 (3-in-5) if selecting hypothermia rather than
DISCUSSION
We have evaluated data from seven pediatric RCTs of hypother-
mia versus normothermia after sTBI using both conventional
and Bayesian meta-analyses in order to better understand the
basis for clinical recommendations.
The updated conventional meta-analysis of hypothermia in
sTBI in pediatric patients confirms that the null hypothesis—
no difference between hypothermia versus normothermia on
mortality and poor/unfavorable outcome—cannot be rejected.
However, we should be circumspect about the use of hypother-
mia for two main reasons. First, the overall quality of evidence
is low. Second, lumping together these RCTs is questionable.
There is heterogeneity with patients of differing risk stratifica-
tion (range in GCS score) across the studies. Studies also used
different body sites for temperature measurement, as well as
different durations of hypothermia and rewarming. And, there
were two obvious outliers: one study had considerably later
timing (mean 15 hr vs mean 4.2–6.8 hr in the other studies)
for starting cooling (37) and one study used a pyrexial target
temperature in the normothermia group (39).
In regard to clinical decision-making, a previous conven-
tional meta-analysis of five RCTs went so far as to say “because
of safety concerns with hypothermia, we do not recommend
further RCTs of the intervention in children with sTBI” (5). This
conclusion was based on an overall estimate showing hypother-
mia could increase mortality and poor/unfavorable outcome
with sTBI but without statistical significance. We too have a

Table 3. The Changes in the Overall Estimates When Conducting the Sensitivity Analysis
Mortality

p for Between-Study
Groups Risk Ratio (95% CI) p for Test of Association I2 (%) Heterogeneity

Overall 1.42 (0.77–2.61) 0.26 20.4 0.27

Excluded
  Biswas et al (36) 1.32 (0.72–2.43) 0.37 19.9 0.28
  Adelson et al (2005a) (37) 1.74 (1.06–2.84) 0.03 0 0.53
  Adelson et al (2005b) (37) 1.51 (0.72–3.16) 0.27 28.9 0.22
  Hutchison et al (38) 1.24 (0.52–2.96) 0.63 26.8 0.23
  Li et al (39) 1.55 (0.83–2.88) 0.17 19.7 0.28
  Adelson et al (3) 1.25 (0.62–2.52) 0.53 25.9 0.24
  Beca et al (2) 1.31 (0.66–2.59) 0.44 29.0 0.22
I = heterogeneity statistic, where I = {100% × (Cochran Q-statistic – degrees of freedom)/Q}.
2 2

Pediatric Critical Care Medicine www.pccmjournal.org 5

Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
 Tasker et al

Figure 3. Forest plots for risk ratio (RR) of mortality in severe traumatic brain injury associated with hypothermia (HT) versus normothermia (NT)
treatment in studies ordered by duration (24, 48, and 72 hr) of treatment. Size of the squares is proportional to the size of the cohorts. Error bars
represent 95% CI. The diamond shape represents the pooled estimates within each analysis.

similar finding with two more pediatric RCTs: the RR for death
is 1.42 (95% CI, 0.77–2.61; p = 0.26). So, would, should, or could
we use hypothermia in our sTBI patients? Our response to this
quandary was to reframe the meta-analysis using a Bayesian
approach (11, 13–20).
The Bayesian approach considers all sources of preexist-
ing knowledge admissible for analysis. Both previous RCT
results and different opinions are used to facilitate informed
decisions likely to be meaningful to clinicians and guideline
developers. First, in regard to information from RCTs, we
have calculated the probability of a hypothesis (i.e., RRR of
death > 20% with hypothermia rather than normothermia)
given information from each trial in historical sequence. After
the seventh pediatric RCT, the current probability of RRR of
death greater than 20% is close to 1-in-3. The 20% threshold
in RRR for mortality is used since a new therapy for critical
illness is likely to be accepted by clinicians and regulators at
this level (15). Second, in regard to the more subjective aspect
of clinician’s prior skepticism or optimism (i.e., beliefs), we
have used the Bayesian approach to incorporate a range of
views in assessing the effect of hypothermia in a particular
clinical setting. For example, given a current probability for
RRR of death greater than 20% of 1-in-3, a skeptical (0.01
probability that RRR of death > 20%) belief in the effect of
hypothermia treatment on outcome changes the probabil-
ity from 1-in-3 to 1-in-6, that is, less chance of RRR death
greater than 20%. If we also consider this change in probabil-
ity along with a current probability of RRI in death greater
than 20% of 3-in-5 by being skeptical about effectiveness of
hypothermia, you would have to conclude that hypothermia
should not be used. That is, in comparison with normother-
mia, hypothermia has less chance of RRR death greater than
20% and more than even chance of RRI in death greater than
20%. In contrast, an optimistic (0.90 probability that RRR of
death > 20%) belief in the effect of hypothermia treatment
on outcome changes the probability of 1-in-3 to 1-in-2, that
is, even chance in RRR death greater than 20%. In addition,
the current probability of RRI in death greater than 20% of
1-in-3 means a less than even chance of RRI in death greater
than 20%. Under these circumstances, the clinician with an
optimistic belief about the effect of hypothermia in a par-
ticular setting may consider that the choice of hypothermia
is a valid therapeutic option. In fact, this observation adds
support to the recent proposal by Lazaridis and Robertson
(40) that “the benefit-risk ratio of HT has not been shown to
be unfavorable.”
The different degrees of clinical belief that are formally trans-
lated into mathematical calculations in Bayesian meta-analysis
warrant further discussion. We do not fully understand what
clinical features lead treating physicians to have skeptical or
optimistic beliefs about effectiveness of hypothermia in a par-
ticular pediatric patient with sTBI. Our specialty has not fully
codified “optimism” and “doubt” about particular outcomes in
sTBI in relation to use of hypothermia management. In contrast,
investigators of hypothermia for out-of-hospital cardiac arrest
(41) and neonatal hypoxic-ischemic encephalopathy (33) have
better characterized subgroups that are less likely to respond to
the intervention. Now, too, there is a systematic review of meth-
ods to elicit clinical beliefs from expert and nonexpert clinicians
for Bayesian priors (18) that could also be applied to the design

6 www.pccmjournal.org XXX 2017 • Volume XX • Number XXX

Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
 Neurocritical Care

of future pediatric critical care

studies (14, 42, 43).

CONCLUSIONS
In our updated conven-
tional meta-analysis of seven
RCTs, we show that the null
hypothesis—no difference
between hypothermia ver-
sus normothermia on mor-
tality and poor/unfavorable
outcome in pediatric severe
TBI—cannot be rejected.
We also found no influence
on outcome when compar-
ing duration of hypothermia
(24, 48, and 72 hr). These
findings have an impact on
the BTF 2012 recommenda-
tions: we cannot conclude
that hypothermia for only 24
hours “should be avoided,”
and we cannot conclude that
hypothermia for 48 hours
“should be considered.” This
Figure 4. Cumulative probability of survival with current probability of reducing mortality with hypothermia lack of guidance means that
compared with normothermia, using either relative risk reduction (RRR) > 0% or RRR > 20%.
clinicians are left making the
decision, and we know that
some are choosing to use
hypothermia (40, 44).
The Bayesian meta-analysis
provides a framework for deci-
sion-making in circumstances
of uncertainty (45). By taking
the cumulative probability of
RRR of death greater than 20%
and chance of RRI of death
greater than 20% from seven
RCTs and including one’s prior
degree of skepticism or opti-
mism about treatment effec-
tiveness, one may be dissuaded
or persuaded about treatment
in a particular clinical set-
ting. Therefore, more work is
needed to understand clinical
skepticism and optimism about
treatment effect. Only in this
way we will know how to deal
with the results of seven RCTs
on hypothermia in sTBI and be
able to decide when hypother-
mia should be “avoided” and
Figure 5. Cumulative probability of survival with current probability of reducing (relative risk reduction [RRR]) or
increasing (relative risk increase [RRI]) mortality with hypothermia compared with normothermia, using the setting when it should be “considered”
of an a priori skeptical or optimistic belief (0.01 or 0.90 probability of RRR/RRI of death > 20%, respectively). in pediatric management.

Pediatric Critical Care Medicine www.pccmjournal.org 7

Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
 Tasker et al
REFERENCES
1. Kochanek PM, Carney N, Adelson PD, et al: Guidelines for the acute
medical management of severe traumatic brain injury in infants, chil-
dren, and adolescents—second edition. Pediatr Crit Care Med 2012;
13 (Suppl 1):S1–S82
2. Beca J, McSharry B, Erickson S, et al; Pediatric Study Group of the
Australia and New Zealand Intensive Care Society Clinical Trials
Group: Hypothermia for traumatic brain injury in children-A phase ii
randomized controlled trial. Crit Care Med 2015; 43:1458–1466
3. Adelson PD, Wisniewski SR, Beca J, et al; Paediatric Traumatic Brain
Injury Consortium: Comparison of hypothermia and normothermia
after severe traumatic brain injury in children (Cool Kids): A phase 3,
randomised controlled trial. Lancet Neurol 2013; 12:546–553
4. Georgiou AP, Manara AR: Role of therapeutic hypothermia in improv-
ing outcome after traumatic brain injury: A systematic review. Br J
Anaesth 2013; 110:357–367
5. Hutchison JS, Guerguerian AM: Cooling of children with severe trau-
matic brain injury. Lancet Neurol 2013; 12:527–529
6. Ma C, He X, Wang L, et al: Is therapeutic hypothermia beneficial for
pediatric patients with traumatic brain injury? A meta-analysis. Childs
Nerv Syst 2013; 29:979–984
7. Zhang BF, Wang J, Liu ZW, et al: Meta-analysis of the efficacy and
safety of therapeutic hypothermia in children with acute traumatic
brain injury. World Neurosurg 2015; 83:567–573
8. Korner-Nievergelt F, Roth T, von Felten S, et al: Bayesian Data
Analysis in Ecology Using Linear Models With R, BUGS, and Stan.
First Edition. Waltham, MA, Academic Press, 2015
9. Cumming G: Understanding the New Statistics: Effect Sizes,
Confidence Intervals, and Meta-Analysis. New York, NY, Routledge,
Taylor & Francis Group, 2013
10. Wasserstein RL, Lazar NA: The ASA’s statement on p-values: Context,
process, and purpose. Am Stat 2016; 70:129–133. Available at:
http://dx.doi.org/10.1080/00031305.2016.1154108
11. Boylan JF, Kavanagh BP, Armitage J: Randomised controlled trials:
Important but overrated? J R Coll Physicians Edinb 2011; 41:126–131
12. Goodman S: A dirty dozen: Twelve p-value misconceptions. Semin
Hematol 2008; 45:135–140
13. Spiegelhalter DJ, Abrams KR, Myles JP: Bayesian Approaches to
Clinical Trials and Health-Care Evaluation. Chichester, West Sussex,
United Kingdom, John Wiley & Sons, 2004
14. Kalil AC, Sun J: Bayesian methodology for the design and interpreta-
tion of clinical trials in critical care medicine: A primer for clinicians.
Crit Care Med 2014; 42:2267–2277
15. Kalil AC, Sun J: Why are clinicians not embracing the results from
pivotal clinical trials in severe sepsis? A Bayesian analysis. PLoS One
2008; 3:e2291
16. Chaloner K, Church T, Louis TA, et al: Graphical elicitation of a prior
distribution for a clinical trial. The Statistician 1993; 42:341–353
17. Hiance A, Chevret S, Lévy V: A practical approach for eliciting expert
prior beliefs about cancer survival in phase III randomized trial. J Clin
Epidemiol 2009; 62:431–437.e2
18. Johnson SR, Tomlinson GA, Hawker GA, et al: Methods to elicit
beliefs for Bayesian priors: A systematic review. J Clin Epidemiol
2010; 63:355–369
19. See CW, Srinivasan M, Saravanan S, et al: Prior elicitation and
Bayesian analysis of the Steroids for Corneal Ulcers Trial. Ophthalmic
Epidemiol 2012; 19:407–413
42. Schoenfeld DA, Hui Zheng, Finkelstein DM: Bayesian design
20. Sun CQ, Prajna NV, Krishnan T, et al: Expert prior elicitation and
Bayesian analysis of the Mycotic Ulcer Treatment Trial I. Invest
Ophthalmol Vis Sci 2013; 54:4167–4173
21. Morrison LJ, Gent LM, Lang E, et al: Part 2: Evidence evaluation
and management of conflicts of interest: 2015 American Heart
Association Guidelines Update for Cardiopulmonary Resuscitation
and Emergency Cardiovascular Care. Circulation 2015; 132 (Suppl
2):S368–S382
22. Jennett B, Bond M: Assessment of outcome after severe brain dam-
age. Lancet 1975; 1:480–484
23. Fiser DH: Assessing the outcome of pediatric intensive care. J Pediatr
1992; 121:68–74
8 www.pccmjournal.org XXX 2017 • Volume XX • Number XXX
Copyright © 2017 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies.
Unauthorized reproduction of this article is prohibited
24. Stangl D, Berry DA: Meta-Analysis in Medicine and Health Policy.
New York, NY, CRC Press, 2000
25. Hedges LV, Vevea JL: Fixed- and random-effects models in meta-anal-
ysis. Psychol Methods 1998; 3:486–504
26. Sutton AAJ, Abrams KR, Jones DR, et al: Methods for Meta-Analysis
in Medical Research. Chichester, West Sussex, UK, John Wiley &
Sons, 2000
27. Smith TC, Spiegelhalter DJ, Thomas A: Bayesian approaches to
random-effects meta-analysis: A comparative study. Stat Med 1995;
14:2685–2699
28. Kalil AC, Sun J: Low-dose steroids for septic shock and severe sep-
sis: The use of Bayesian statistics to resolve clinical trial controver-
sies. Intensive Care Med 2011; 37:420–429
29. Harrell FE Jr, Shih YC: Using full probability models to compute prob-
abilities of actual interest to decision makers. Int J Technol Assess
Health Care 2001; 17:17–26
30. Higgins JP, Spiegelhalter DJ: Being sceptical about meta-analyses: A
Bayesian perspective on magnesium trials in myocardial infarction. Int
J Epidemiol 2002; 31:96–104
31. Diamond GA, Kaul S: Prior convictions: Bayesian approaches to the
analysis and interpretation of clinical megatrials. J Am Coll Cardiol
2004; 43:1929–1939
32. Salpeter SR, Cheng J, Thabane L, et al: Bayesian meta-analysis of
hormone therapy and mortality in younger postmenopausal women.
Am J Med 2009; 122:1016–1022.e1
33. Pedroza C, Tyson JE, Das A, et al; Eunice Kennedy Shriver National
Institute of Child Health and Human Development Neonatal Research
Network: Advantages of Bayesian monitoring methods in deciding
whether and when to stop a clinical trial: An example of a neonatal
cooling trial. Trials 2016; 17:335
34. Bourdages M, Bigras JL, Farrell CA, et al; Canadian Critical Care Trials
Group: Cardiac arrhythmias associated with severe traumatic brain injury
and hypothermia therapy. Pediatr Crit Care Med 2010; 11:408–414
35. Salonia R, Empey PE, Poloyac SM, et al: Endothelin-1 is increased
in cerebrospinal fluid and associated with unfavorable outcomes in
children after severe traumatic brain injury. J Neurotrauma 2010;
27:1819–1825
36. Biswas AK, Bruce DA, Sklar FH, et al: Treatment of acute traumatic
brain injury in children with moderate hypothermia improves intracra-
nial hypertension. Crit Care Med 2002; 30:2742–2751
37. Adelson PD, Ragheb J, Kanev P, et al: Phase II clinical trial of mod-
erate hypothermia after severe traumatic brain injury in children.
Neurosurgery 2005; 56:740–754; discussion 740
38. Hutchison JS, Ward RE, Lacroix J, et al; Hypothermia Pediatric Head
Injury Trial Investigators and the Canadian Critical Care Trials Group:
Hypothermia therapy after traumatic brain injury in children. N Engl J
Med 2008; 358:2447–2456
39. Li H, Lu G, Shi W, et al: Protective effect of moderate hypothermia
on severe traumatic brain injury in children. J Neurotrauma 2009;
26:1905–1909
40. Lazaridis C, Robertson CS: Hypothermia for increased intracranial
pressure: Is it dead? Curr Neurol Neurosci Rep 2016; 16:78
41. Meert KL, Telford R, Holubkov R, et al; Therapeutic Hypothermia after
Pediatric Cardiac Arrest (THAPCA) Trial Investigators: Pediatric out-
of-hospital cardiac arrest characteristics and their association with
survival and neurobehavioral outcome. Pediatr Crit Care Med 2016;
17:e543–e550
using adult data to augment pediatric trials. Clin Trials 2009;
6:297–304
43. Moatti M, Zohar S, Facon T, et al: Modeling of experts’ divergent
prior beliefs for a sequential phase III clinical trial. Clin Trials 2013;
10:505–514
44. Fulkerson DH, White IK, Rees JM, et al: Analysis of long-term (median
10.5 years) outcomes in children presenting with traumatic brain
injury and an initial Glasgow Coma Scale score of 3 or 4. J Neurosurg
Pediatr 2015; 16:410–419
45. Tasker RC, Akhondi-Asl A: Targeted temperature management after
cardiac arrest due to drowning: “Frequentist” and “Bayesian” deci-
sion making. Pediatr Crit Care Med 2016; 17:789–791