You are on page 1of 10

JOURNAL OF NEUROTRAUMA 25:62–71 (January 2008)

© Mary Ann Liebert, Inc.


DOI: 10.1089/neu.2007.0424

Hypothermia Treatment for Traumatic Brain Injury:


A Systematic Review and Meta-Analysis

KIM PETERSON, SUSAN CARSON, and NANCY CARNEY

ABSTRACT

In this study, we conducted an updated meta-analysis of the effects of hypothermia therapy on mor-
tality, favorable neurologic outcome, and associated adverse effects in adults with traumatic brain
injury (TBI) for use by Brain Trauma Foundation (BTF)/American Association of Neurological Sur-
geons (AANS) task force to develop evidence-based treatment guidelines. Our data sources relied
on handsearches of four previous good-quality systematic reviews, which all conducted electronic
searches of primarily MEDLINE (OVID), EMBASE, and Cochrane Library. An independent, sup-
plemental electronic search of MEDLINE was undertaken as well (last searched June 2007). Only
English-language publications of randomized controlled trials of therapeutic hypothermia in adults
with TBI were selected for analysis. Two reviewers independently abstracted data on trial design,
patient population, hypothermia and cointervention protocols, patient outcomes, and aspects of
methodological quality. Pooled relative risks (RR) and associated 95% confidence intervals (CIs)
were calculated for each outcome using random-effects models. In the current study, only 13 trials
met eligibility criteria, with a total of 1339 randomized patients. Sensitivity analyses revealed that
outcomes were influenced by variations in methodological quality. Consequently, main analyses were
conducted based on eight trials that demonstrated the lowest potential for bias (n  781). Reduc-
tions in risk of mortality were greatest (RR 0.51; 95% CI 0.33, 0.79) and favorable neurologic out-
comes much more common (RR 1.91; 95% CI 1.28, 2.85) when hypothermia was maintained for
more than 48 h. However, this evidence comes with the suggestion that the potential benefits of hy-
pothermia may likely be offset by a significant increase in risk of pneumonia (RR 2.37; 95% CI
1.37, 4.10). In sum, the present study’s updated meta-analysis supports previous findings that hy-
pothermic therapy constitutes a beneficial treatment of TBI in specific circumstances. Accordingly,
the BTF/AANS guidelines task force has issued a Level III recommendation for optional and cau-
tious use of hypothermia for adults with TBI.

Key words: induced hypothermia; meta-analysis; traumatic brain injury

Oregon Evidence-Based Practice Center (EPC), Department of Medical Informatics and Clinical Epidemiology (DMICE),
Oregon Health and Science University (OHSU), Portland, Oregon.

62
META-ANALYSIS OF HYPOTHERMIA TREATMENT FOR TBI

INTRODUCTION develop evidence-based treatment guidelines for hy-


pothermia therapy are detailed in the third edition of the
Guidelines for the Management of Severe Traumatic
I N 1995, the Brain Trauma Foundation (BTF) and the
American Association of Neurological Surgeons
(AANS) collaborated in the development of the first edi-
Brain Injury (Brain Trauma Foundation, 2007). How-
ever, the work reported herein provides more detail of
tion of evidence-based Guidelines for the Management the standard systematic review and meta-analysis
of Severe Traumatic Brain Injury (Bullock et al., 1996). methodologies than reflected in the Guidelines publica-
Because the Guidelines could not address all important tion, and also represents an expansion of the original re-
aspects of TBI management, a task force of clinical ex- view objectives. Additionally, in a more recent update
perts was asked to select topics offering the highest po- literature search we found one additional RCT of hy-
tential for positively impacting patient outcomes. Hy- pothermia therapy that we added for this publication (Liu
pothermic therapy was not one of those topics, as at that et al., 2006).
time, clinical research on use of hypothermia to manage
TBI was in its infancy. Scientific evidence from the few,
early trials was unimpressive and indicated that hy- METHODS
pothermia was of no benefit in reducing risk of death in
adults with TBI (Clifton et al., 1992, 1993; Hirayama et Literature Search
al., 1994). Reference lists of four previous good-quality system-
Since 1995, however, scientific evidence regarding the atic reviews (Alderson et al., 2004; McIntyre et al., 2003;
use of hypothermia therapy has been mounting and the Harris et al., 2002; Henderson et al., 2003) provided the
controversy around its place in care has continued. Nu- basis for identification of all eligible RCTs available from
merous randomized controlled trials (RCTs) of hy- January 1, 1966 through week 1 of September 2002.
pothermic therapy have emerged. Although effects on Searches in these reviews involved numerous electronic
mortality have been variable across RCTs, many have databases including MEDLINE (OVID), EMBASE,
shown promise for hypothermia as an effective treatment Cochrane Library, Current Contents (week 27 of year
for TBI. Therefore, when the BTF/AANS-sponsored clin- 1993 to week 40 of year 2002), abstract center for the
ical expert task force reconvened in 2004 to update the American Association of Neurological Surgeons and the
TBI management guidelines, they elected to add thera- Congress of Neurological Surgeons, Science Citation
peutic hypothermia to the array of priority topics. Dissertation Abstracts, CENTRAL, and Specialist Trials
Preliminary literature searches identified four previous Registry for the Injuries Group. They used various com-
systematic reviews on hypothermia for TBI that were binations of MeSH (Medical Subject Headings) terms
published between 2002 and 2004 (Alderson et al., 2004; and text words for hypothermia, brain injury, cranio-
Harris et al., 2002; Henderson et al., 2003; McIntyre et cerebral trauma, and neurosurgery. Complete search
al., 2003), and these were considered for primary use in strategies and sources are described in detail in each sys-
guideline development. However, a detailed examination tematic review’s methods section.
of the previous reviews found them to have limited util- We also conducted a supplemental literature search of
ity for providing a reliable estimate of treatment effect MEDLINE (2002 through June Week 4 2007) by com-
for hypothermia. This was primarily due to inconsisten- bining the following terms: “Brain Injuries,” “hy-
cies in numbers of included RCTs, meta-analysis meth- potherm$,” “(brain or cerebr$) adj3 temperature$.” Fil-
ods and overall findings. Additionally, we found two new ters for English language, human, and controlled trial
RCTs that had been published subsequent to 2004, in- were applied.
volving a total of 482 TBI patients, and we noted that
their findings could potentially overturn the conclusions
from the previous systematic reviews (Qiu et al., 2005;
Study Selection Criteria
Zhi et al., 2003). Given these factors, we decided to build We included English-language publications of RCTs
upon the previous works by formally conducting a sys- that compared the benefits and harms of hypothermia to
tematic review and meta-analysis to further examine the standard care upon hospital admission in adults with TBI.
effects of various aspects of hypothermia therapy on mor- Adult populations were defined as being comprised of at
tality, overall neurological outcome, and its associated least 85% of patients aged 14 years or above. Two re-
adverse effects. viewers (K.P. and S.C.) independently assessed abstracts
Findings from our original systematic review and for inclusion using these criteria. Full-text articles of po-
meta-analysis and a description of the process carried out tentially relevant abstracts were retrieved and a second
by the BTF/AANS task force in using these findings to review for inclusion was conducted by reapplying the in-

63
PETERSON ET AL.

clusion criteria. All disagreements were resolved through For all variables, we calculated pooled relative risks
consensus. (RR) and associated 95% confidence intervals (CIs) us-
ing random-effects models (Deeks, 1998). Statistical het-
Data Abstraction erogeneity was calculated using the chi-squared test.
Two reviewers (K.P. and S.C.) independently ab- Most sensitivity analyses were specified a priori and con-
stracted data from the primary studies into an Excel ducted to examine potential effects of variations in in-
spreadsheet using a prespecified form. Reviewers were ternal validity, target cooling temperature, cooling dura-
masked to author and journal. Disagreements were resolved tion, and rate of rewarming. Post-hoc sensitivity analyses
through consensus. The following data were abstracted: were also conducted to examine the effects of hypother-
setting, brief inclusion/exclusion criteria; detailed descrip- mia when co-administered in settings involving use of
tions of resuscitation protocols; surgical/pharmacother- barbiturates for ICP-control and to examine differences
apy/nutritional interventions; goals for and strategies to between short-term and long-term trial durations. All in-
control intracranial pressure and cerebral perfusion pres- dependent variables were treated as dichotomous and cat-
sure; hypothermia intervention methods including timing egories were identified a priori. Target cooling tempera-
of induction, target and actual mean cooling temperature tures below 33°C were classified a priori as “moderate”
depth, temperature measurement method, duration of and temperatures of 33°C and above were classified as
cooling, shivering prevention protocol, and rate of re- “mild.” Cooling duration was analyzed using a prespec-
warming; normothermia group treatment protocol and ified cut-off of 48 h. Rewarming methods were classified
target temperature; population characteristics including as either “passive” or “active.” ICP management strate-
mean age, gender and race, cause and type of injury, pres- gies were classified based on use or nonuse of barbitu-
ence of pupillary abnormalities, rates of hypotension/ rates. For trial duration, groups were defined as “3–6
hypoxia at baseline, comorbid chest/abdominal injury months” and “1–2 years.” For study quality, we con-
and/or pelvic/leg/arm fractures, and baseline intracranial ducted sensitivity analyses based on the following indi-
pressure, cerebral perfusion pressure and Glasgow Coma vidual components: randomization (adequate or inade-
Scale scores when available. quate), allocation concealment (adequate or inadequate),
handling of attrition (intent-to-treat analysis or per-pro-
Quality Assessment tocol analysis) and baseline differences (present/unclear
or absent). All analyses were conducted using RevMan
Two reviewers (K.P. and S.C.) independently assessed version 4.2 (Update Software).
the internal validity of individual trials using predefined
criteria based on methods used for randomization and al-
location concealment, between-groups similarity in base-
RESULTS
line demographic and prognostic factors, blinding of out-
come assessors, adequacy of sample size, use of Overview
intention-to-treat analysis, follow-up rates, and associ-
ated maintenance of comparable groups. The quality as- Our supplemental search produced 72 potentially rel-
sessment tool was created based on criteria developed by evant citations. Ultimately, 13 trials met our inclusion
the U.S. Preventive Services Task Force (Harris et al., criteria and were included in this review (Aibiki et al.,
2001), the National Health Service Centre for Reviews 2000; Clifton et al., 1992, 1993, 2001; Hirayama et al.,
and Dissemination (U.K.) (Centre for Reviews and Dis- 1994; Jiang et al., 2000; Liu et al., 2006; Marion et al.,
semination, 2001), and the Cochrane Collaboration (Hig- 1997; Qiu et al., 2005; Yan and Tang, 2001; Zhi et al.,
gins and Green, 2006). Internal validity raters were 2003; Shiozaki et al., 2001; Shiozaki et al., 1999). Table
masked to author and journal. Disagreements were re- 1 describes characteristics of all included trials.
solved by consensus and, in some cases, involved con-
sultation with a third masked reviewer (N.C.). Patient Characteristics and Interventions
A total of 1339 patients were randomized across trials.
Statistical Analysis Not all trials reported patient demographics (Table 1).
The primary effectiveness outcome was all-cause mor- Among those that did (Aibiki et al., 2000; Clifton et al.,
tality. The secondary effectiveness outcome was favor- 2001; Jiang et al., 2000; Liu et al., 2006; Marion et al.,
able neurological response, defined as the proportion of 1997; Qiu et al., 2005; Yan and Tang, 2001; Zhi et al.,
patients that achieved a Glasgow Outcome Scale score 2003; Shiozaki et al., 1999, 2001), patients were 73% male,
of 4 or 5 at various time points. For safety, we examined with a mean age of 37 years. Sixty-one percent of all ran-
rates of arrhythmia and pneumonia. domized patients were Asian. Baseline Glasgow Coma

64
META-ANALYSIS OF HYPOTHERMIA TREATMENT FOR TBI

TABLE 1. CHARACTERISTICS OF INCLUDED TRIALS

Cooling
induction
Follow-up Mean target, Target Cooling
Author, duration age Male Mean no. of hours cooling duration Rate of
year (months) N (years) (%) GCS postinjury (ºC) (days) rewarming

Aibiki 6 26 36 77 5.7 3–4 32–33 3–4 1ºC/day


2000
Clifton 3 10 NR NR 5.7 6 30–32 1 Gradual/day
1992
Clifton 3 46 NR NR NR 6 32–33 2 1ºC/4 h
1993
Clifton 6 392 31 NR 5.7 6 33 2 0.5ºC/2 h
2001
Hirayama 3 22 NR NR NR 6 32–33 2 1ºC/4 h
1994
Jiang 12 87 41 83 5.1 Immediately 33 3–14 1ºC/h
2000
Marion 12 82 33 84 NR 10 32–33 1 1ºC/h
1997
Qiu 24 86 41 65 NR Immmediately 33–35 3–5 Naturally
2005 or 3–5 days
postcraniotomy
Yan 120 hours 44 42 68 Group A: NR 32–34 3–5 Slowly,
2001 4.1 naturally
Group B:
7.4
Zhi 6 396 42 73 5.8 20 32–33 1–7 1ºC/4 h
2003
Shiozaki 6 16 36 81 5.5 As soon as 33.5–34.5 2 1ºC/day
1999 possible
Shiozaki 3 91 38 72 5.3 As soon as 33.5–34.5 2 1ºC/day
2001 possible
Liu 24 66 41 64 NR NR 33–35 3 Spontaneously
2006

NR, not reported.

Score (GCS) information was only reported in 54% of tri- ods of randomization and allocation concealment (Table
als (Aibiki et al., 2000; Clifton et al., 2001; Jiang et al., 2). With regard to other components of methodologic
2000; Yan and Tang, 2001; Zhi et al., 2003; Shiozaki et quality, the only source of variation found was based on
al., 1999, 2001) and averaged 5.6 points. One trial divided whether or not randomization resulted in between-
patients by baseline GCS, which was 4.1 points for Group groups comparability in important baseline patient char-
A and 7.4 points for Group B (Yan and Tang, 2001). acteristics (e.g., age, gender, GCS, type of injury). Sen-
Hypothermic treatment protocols varied across trials sitivity analyses were performed to assess the influence
with regard to type of device used, time to induction, of variability in between-groups comparability and
cooling depth and duration, temperature monitoring site, found evidence of bias in treatment effect estimates for
and rate of rewarming. Trials provided inadequate detail mortality and good neurologic outcome. When we ex-
to determine adherence to these protocols. cluded trials suspected of failing to study comparable
groups (Yan and Tang, 2001; Clifton et al., 1992; Hi-
rayama et al., 1994; Qiu et al., 2005; Zhi et al., 2003),
Methodological Quality estimates of treatment benefits were more conservative
Assessment of the overall methodological quality of and differences in outcome between hypothermia and
these trials was limited by incomplete reporting of meth- control groups were no longer statistically significant

65
PETERSON ET AL.

(Figs. 1 and 2). Therefore, we proceeded with a cautious have also been an important source of variability among
approach and conducted final meta-analyses based only trials, in that hypothermia was associated with a signifi-
on the 8 trials that clearly studied comparable groups cant reduction in risk of mortality only when data were
(Clifton et al., 1993, 2001; Aibiki et al., 2000; Jiang et pooled from longer-term trials (1–2 years).
al., 2000; Marion et al., 1997; Shiozaki et al., 1999, 2001;
Liu et al., 2006). Favorable Neurological Outcome
In trials that studied comparable groups, hypothermia
Mortality was associated with a nonsignificant increase of 25% in
In trials that studied comparable groups, hypothermia improved neurological outcomes, as measured by the
reduced mortality by 20% relative to conventional ther- GOS (RR 1.25; 95% CI 0.96, 1.62; Fig. 2 and Table 4).
apies in patients with traumatic brain injury (RR 0.80; Results of subgroup analyses of the effects of hy-
95% confidence interval (CI) 0.59, 1.09), but this effect pothermia on improving neurological outcomes (Table 4)
was not found to be statistically significant (Clifton et al., were consistent with findings from subgroup analyses of
1993, 2001; Aibiki et al., 2000; Jiang et al., 2000; Mar- mortality outcomes. Again, there were differences in like-
ion et al., 1997; Shiozaki et al., 1999, 2001; Liu et al., lihood of favorable neurological outcome based on vari-
2006) (Fig. 1). ations in cooling duration, ICP management strategy, and
In subgroup analyses (Table 3), variations in cooling trial duration (Aibiki et al., 2000; Clifton et al., 1992,
duration, ICP management strategy, and trial duration 1993, 2001; Hirayama et al., 1994; Jiang et al., 2000;
were found to have significant influences on risk of mor- Marion et al., 1997; Zhi et al., 2003). Increased likeli-
tality. Consistent with previous meta-analyses, our find- hood of a favorable neurological outcome was greatest
ings confirmed that benefits of hypothermia were great- in trials where hypothermia was maintained for over 48
est when cooling was maintained for more than 48 h (RR h (RR 1.91; 95% CI 1.28, 2.85), did not prespecify bar-
0.51; 95% CI 0.33, 0.79). We also found that use of ag- biturate therapy as a standard component of ICP man-
gressive ICP management strategies that involved ad- agement (RR 1.79; 95% CI 1.27, 2.52), and in trials that
ministration of barbiturates was a significant moderator followed patients for a year or longer (RR 1.72; 95% CI
of hypothermia therapy. Hypothermia therapy was of sig- 1.24, 2.38).
nificant benefit only to patients who were not enrolled in
trials in which barbiturate administration was part of the
Adverse Events
standard ICP management protocol (RR 0.58; 95% CI Rates of serious adverse events were reported in very
0.40, 0.85) (Aibiki et al., 2000; Clifton et al., 1993; Jiang few trials and this should be taken into consideration
et al., 2000; Liu et al., 2006). Duration of follow-up may when interpreting results of these analyses (Clifton et al.,

TABLE 2. METHODOLOGICAL QUALITY COMPONENTS OF INCLUDED TRIALS

Significant Percentage patients


Adequate Adequate between-groups excluded
randommization allocation baseline from
Author, year method concealment differences analyses

Aibiki 2000 NR NR No None


Clifton 1992 NR NR NR None
Clifton 1993 NR NR No 2%
Clifton 2001 NR NR No 6%
Hirayama 1994 NR NR Yes Unclear
Jiang 2000 NR NR No None
Marion 1997 NR No No 1%
Qiu 2005 NR NR Unclear None
Yan 2001 NR NR NR None
Zhi 2003 NR NR No None
Shiozaki 1999 NR NR No None
Shiozaki 2001 NR NR No None
Liu 2006 Yes NR No None

NR, not reported.

66
META-ANALYSIS OF HYPOTHERMIA TREATMENT FOR TBI

FIG. 1. Relative risk of mortality for use of hypothermia to manage traumatic brain injury in randomized controlled trials (RCTs).

FIG. 2. Relative benefit (95% confidence interval [CI]) on neurological outcome (Glasgow Outcome Scale [GOS] 4 or 5) for
hypothermia therapy versus normothermia in trials of adults with traumatic brain injury (TBI).

67
PETERSON ET AL.

TABLE 3. RELATIVE RISK OF MORTALITY FOR HYPOTHERMIA VERSUS CONTROL TREATMENTS IN TBI PATIENT SUBGROUPS

Subanalyses RR (95% CI) No. of trials N  2 (p-value) I2

Cooling duration
Cooled up to 48 h 1.03 (0.79, 1.36) 5 602 0.42 (0.94) 0%
Cooled over 48 h 0.51 (0.33, 0.79) 3 179 0.58 (0.75) 0%
Target cooling depth
Moderate depth 0.87 (0.65, 1.16) 5 608 4.47 (0.35) 10.5%
Mild depth 0.77 (0.28, 2.10) 3 173 3.05 (0.08) 67.2%
Rewarming strategy
Passive rewarming 0.49 (0.26, 0.93) 1 66 N/A N/A
Active rewarming 0.91 (0.70, 1.18) 7 715 5.15 (0.40) 3%
ICP management strategy
Barbiturate use 1.05 (0.78, 1.41) 4 556 0.32 (0.85) 0%
No barbiturate use 0.58 (0.4, 0.85) 4 225 2.19 (0.53) 0%
Trial duration
3–6 months 1.02 (0.76, 1.36) 5 547 2.15 (0.54) 0%
1–2 years 0.61 (0.42, 0.90) 3 234 1.85 (0.40) 0%

RR, relative risk; CI, confidence interval.

1992, 1993; Jiang et al., 2000; Qiu et al., 2005) (Table DISCUSSION
5). Significantly more cases of pneumonia were reported
within 12 months following hypothermia versus nor- Our meta-analysis did not resolve existing uncertain-
mothermia (RR 2.37; 95% CI 1.37, 4.10). Risk of pneu- ties concerning the effectiveness of hypothermia for treat-
monia was much greater, however, in the subset of trials ment of TBI in adults; however, it does provide impor-
that involved barbiturate administration (RR 6.45; 95% tant information about sources of variation in trials that
CI 2.47, 16.85) (Shiozaki et al., 1999, 2001). Similar rates can be used to guide future research. Additionally, we
of arrhythmia were reported following hypothermia as have identified a clinical circumstance in which hy-
following normothermia (RR 1.07; 95% CI 0.53, 2.18), pothermia appears to be harmful.
and barbiturate use did not have significant differential Findings from subgroup analyses suggested that some
effects for this outcome. variations in hypothermic treatment and co-intervention

TABLE 4. RELATIVE BENEFIT OF IMPROVEMENTS IN NEUROLOGICAL OUTCOME FOR HYPOTHERMIA VERSUS CONTROL
TREATMENTS IN TBI PATIENT SUBGROUPS IN TRIALS THAT STUDIED COMPARABLE GROUPS

Subanalyses RR (95% CI) No. of trials N  2 (p-value) I2

Cooling duration
Cooled up to 48 h 1.05 (0.82, 1.36) 5 601 7.51 (0.11) 46.8%
Cooled over 48 h 1.91 (1.28, 2.85) 3 179 0.30 (0.86) 0%
Target cooling depth
Moderate depth 1.41 (1.04, 1.91) 5 608 7.49 (0.11) 46.6%
Mild depth 1.02 (0.61, 1.71) 3 172 5.82 (0.05) 65.7%
Rewarming strategy
Passive rewarming 2.05 (0.98, 4.31) 1 66 N/A N/A
Active rewarming 1.19 (0.91, 1.54) 7 7 13.16 (0.04) 54.4%
ICP management strategy
Barbiturate use 1.00 (0.77, 1.31) 4 556 5.75 (0.12) 47.8%
No barbiturate use 1.79 (1.27, 2.52) 4 225 0.69 (0.87) 0%
Trial duration
3–6 months 1.02 (0.79, 1.35) 5 546 7.00 (0.14) 42.9%
1–2 years 1.72 (1.24, 2.38) 3 234 0.29 (0.86) 0%

RR, relative risk; CI, confidence interval.

68
META-ANALYSIS OF HYPOTHERMIA TREATMENT FOR TBI

TABLE 5. RATES OF SERIOUS ADVERSE EVENTS come were consistently positive in trials that followed
IN CONTROLLED TRAILS OF HYPOTHERMIA patients for 1–2 years. Although this may offer some op-
timism that the potential benefits of hypothermia are sim-
Adverse Hypothermia, Normothermia,
ply more evident later in the course of recovery, when a
event n/N (%) n/N (%)
patient’s cognitive status has become more stable, these
Pneumonia 51/120 (51%) 28/120 (23%) post-hoc findings must be interpreted with caution in light
Arrhythmia 14/120 (12%) 13/120 (11%) of the potential treatment-related confounders discussed
above.
Regarding variations in components of methodologi-
cal quality, our sensitivity analysis detected evidence of
protocols had differential effects on patient outcome and bias in treatment effects in trials that did not study com-
these findings may have practical implications for use in parable groups. Compared to trials studying comparable
the acute care setting. Consistent with findings from pre- groups, mortality risk reductions and improvements in
vious systematic reviews, we observed more pronounced neurological outcomes appeared inflated in trials with
improvements in patient outcomes when hypothermia disproportionately more high-risk patients (e.g., lower
was maintained for over 48 hours. Influence on outcome heart rate or GCS scores) in the hypothermia group rel-
was less evident for variations in other measured aspects ative to the control groups. Therefore, to minimize bias,
of hypothermia treatment protocols. our meta-analyses were restricted to only trials that
Another important source of heterogeneity among tri- clearly studied comparable groups. Notably, this led to
als mentioned in previous systematic reviews was use of less optimistic conclusions about the treatment effects of
cointerventions. Until now, no attempts have been made hypothermia.
to examine in any depth whether variation in any aspects It is unclear why analyses from previous systematic re-
of TBI management cointerventions potentially affected views did not find any differences in treatment effects
patient outcomes. Although there were limitations in the based on variations in methodologic quality. Possible rea-
level of detail provided about how cointerventions were sons include differences in study quality scoring systems
actually administered, it was noted that there was a di- and/or differences in statistical methods used to account
vision among trials regarding whether the ICP-manage- for variances in study quality. Also, our meta-analysis in-
ment protocol clearly specified use of barbiturate ther- cluded 45% more patients than in the previous reviews
apy. Due to the controversial nature of using barbiturates and had improved statistical power to detect potential bi-
to lower ICP (Brain Trauma Foundation, 2007), we ases in treatment effects.
opted to conduct post-hoc analyses of the sensitivity of Although this review offers new insights about tri-
hypothermia effects to cointervention with barbiturates. als of hypothermic therapy, it suffers from the same
Although exploratory in nature, results of our sensitiv- limitations as did previous reviews. The chief limita-
ity analyses suggested that the almost threefold increased tion of this review is the relatively small number of tri-
risk of pneumonia associated with hypothermia may als with low potential for bias that were available for
largely outweigh its negligible potential benefits when analysis. When used under such circumstances, meta-
coadministered in the presence of barbiturate therapy. analysis methods are at higher risk of instability and
While these post-hoc findings should be interpreted with require a more critical approach to interpretation. This
caution, their potential implications for future research is especially true with regard to results of our pre-
and possibly clinical practice could include limiting us- planned and post-hoc sensitivity analyses, which eval-
age of hypothermia to the subgroup of patients that are uated more finely divided subsets of trials. We were
initially responsive to standard ICP-lowering strategies also limited by the inadequate detail provided by the
and avoiding its use in the most extreme of clinical sit- trials regarding important aspects of trial design and
uations. patient characteristics and, consequently, may have
This systematic review and meta-analysis also repre- overlooked other sources of heterogeneity that may
sents the first attempt to examine the time course of re- have affected patient outcomes. Also, due to the under-
covery from TBI at fixed intervals following hypother- reporting of adverse event outcomes, the full extent of
mia therapy. Results of our sensitivity analyses based on potential harms associated with hypothermia remains
trial duration suggest that whereas there was much het- largely unknown. Finally, whereas our focus was pri-
erogeneity in GOS scores across trials that looked at only marily on global GOS scores, judgments about the
the first 3–6 months post-TBI, the effects of hypother- overall benefit of hypothermia therapy should also take
mia in significantly reducing risk of death and in in- into account its effects on more specific aspects of re-
creasing the likelihood of a favorable neurological out- covery such as return to work, driving, and indepen-

69
PETERSON ET AL.

dent living. However, we are aware of no trials that Clifton, G.L., Allen, S., and Barrodale, P., Plenger, P., Berry,
have evaluated these types of real-life outcomes. J., Koch, S., Fletcher, J., Hayes, R.L., and Choi, S.C. (1993).
In summary, although limited, the best available evi- A phase II study of moderate hypothermia in severe brain in-
dence suggests that hypothermic therapy may reduce the jury. J. Neurotrauma 10, 263–273.
risk of mortality and improve prospects for a favorable Clifton, G.L., Allen, S., Berry, J., and Koch, S.M. (1992). Sys-
neurological outcome, particularly when maintained for temic hypothermia in treatment of brain injury. J. Neuro-
a duration of greater than 48 hours and when used in pa- trauma 9, S487–S495.
tients that respond well to standard measures of ICP-con- Clifton, G.L., Miller, E.R., Choi, S.C., Levin, H.S., McCauley,
trol that exclude high-dose barbiturates. Based on simi- S., Smith, K.R., Jr., Muizelaar, J.P., Wagner, F.C., Jr., Mar-
lar findings from our original systematic review and ion, D.W., Luerssen, T.G., Chesnut, R.M., and Schwartz, M.
meta-analysis, in their third edition of the Guidelines for (2001). Lack of effect of induction of hypothermia after acute
the Management of Severe Traumatic Brain Injury (Brain brain injury. N. Engl. J. Med. 344, 556–563.
Trauma Foundation, 2007), the BTF/AANS task force is- Deeks, J. (1998). When can odds ratios mislead? Odds ratios
sued a Level III recommendation for only optional use should be used only in case-control studies and logistic re-
of hypothermia in clinical practice at this time. Until more gression analyses. Br. Med. J. 317, 1155–1156.
evidence from well-conducted trials becomes available,
Harris, R.P., Helfand, M., Woolf, S.H., Lohr, K.N., Mulrow,
clinicians should continue to exercise caution when con- C.D., Teutsch, S.M., Atkins, D., Methods Work Group, Third
sidering administering hypothermia for treatment of TBI. US Preventive Services Task Force. (2001). Current meth-
ods of the third U.S. Preventive Services Task Force. Am. J.
Prevent. Med. 20, 21–35.
ACKNOWLEDGMENTS Harris, O.A., Colford, J.M., Jr., Good, M.C., and Matz, P.G.
(2002). The role of hypothermia in the management of
We thank Cindy Davis-O’Reilly, B.Sc., for her valu- severe brain injury: a meta-analysis. Arch. Neurol. 59,
able contributions to all phases of this project. We also 1077–1083.
thank Ben K.S. Chan, M.S., and Rongwei Fu, Ph.D., for
Henderson, W.R., Dhingra, V.K., Chittock, D.R., Fenwick, J.C.,
providing statistical advice. and Ronco, J.J. (2003). Hypothermia in the management of
traumatic brain injury. A systematic review and meta-analy-
sis. Intensive Care Med. 29, 1637–1644.
REFERENCES Higgins, J.P.T, and Green, S., editors. Cochrane Handbook for
Systematic Reviews of Interventions 4.2.6 [updated Septem-
Aibiki, M., Maekawa, S., and Yokono, S. (2000). Moderate hy-
ber 2006]. John Wiley & Sons: Chichester, UK.
pothermia improves imbalances of thromboxane A2 and
prostaglandin I2 production after traumatic brain injury in Hirayama, T., Katayama, Y., Kano, T., et al. (1994). Impact of
humans. Crit. Care Med. 28, 3902–3906. moderate hypothermia on therapies for intracranial pressure
control in severe traumatic brain injury. In: Intracranial Pres-
Alderson, P., Gadkary, C., and Signorini, D.F. (2004). Thera-
sure IX: 9th International Symposium Held in Nagaya,
peutic hypothermia for head injury [update of Cochrane Data-
Japan. H. Nagai, S. Ishii, and M. Maeda (eds), Springer-Ver-
base Syst Rev. 2002;(1):CD001048; PMID: 11869586].
lag: New York, pps. 233–236.
Cochrane Database of Systematic Reviews 4, CD001048.
Jiang, J., Yu, M., and Zhu, C. (2000). Effect of long-term mild
Brain Trauma Foundation, American Association of Neurolog-
hypothermia therapy in patients with severe traumatic brain
ical Surgeons (AANS), Congress of Neurological Surgeons
injury: 1-year follow-up review of 87 cases. J. Neurosurg.
(CNS), AANS/CNS Joint Section on Neurotrauma and Crit-
93, 546–549.
ical Care. (2007). Guidelines for the Management of Severe
Traumatic Brain Injury, 3rd edition. J. Neurotrauma 24, Liu, W.G., Qiu, W.S., and Zhang, Y., Wang, W.M., Lu, F., and
S1–S106. Yang, X.F. (2006). Effects of selective brain cooling in pa-
tients with severe traumatic brain injury: a preliminary study.
Bullock, R., Chesnut, R.M., and Clifton, G. (1996). Guidelines
J. Int. Med. Res. 34, 58–64.
for the management of severe head injury. J. Neurotrauma
13, 641–734. Marion, D.W., Penrod, L.E., Kelsey, S.F., Obrist, W.D.,
Kochanek, P.M., Palmer, A.M., Wisniewski, S.R., and
Centre for Reviews and Dissemination, University of York.
Dekosky, S.T. (1997). Treatment of traumatic brain injury
(2001). Undertaking Systematic Reviews of Research on Ef-
with moderate hypothermia. N. Engl. J. Med. 336, 540–546.
fectiveness: CRD’s Guidance for Those Carrying Out or
Commissioning Reviews. CRD Report Number 4, 2nd ed. McIntyre, L.A., Fergusson, D.A., Hebert, P.C., Mohrer, D., and
NHS Centre for Reviews and Dissemination: York, UK. Hutchison, J.S. (2003). Prolonged therapeutic hypothermia

70
META-ANALYSIS OF HYPOTHERMIA TREATMENT FOR TBI

after traumatic brain injury in adults: a systematic review. Yan, Y., and Tang, W. (2001). Changes of evoked potentials
JAMA 289, 2992–2999. and evaluation of mild hypothermia for treatment of severe
Qiu, W-S., Liu, W-G., Shen H., Wang, W.M., Hong, Z.L., brain injury. Chin. J. Traumatol. 4, 8–13.
Zhang, Y., Jiang, S.J., and Yang, X.F. (2005). Therapeutic Zhi, D., Zhang, S., and Lin, X. (2003). Study on therapeutic
effect of mild hypothermia on severe traumatic head injury. mechanism and clinical effect of mild hypothermia in pa-
Chin. J. Traumatol. 8, 27–32. tients with severe head injury. Surg. Neurol. 59, 381–385.
Shiozaki, T., Hayakata, T., Taneda, M., Nakajima, Y.,
Hashiguchi, N., Fujimi, S., Nakamori, Y., Tanaka, H., Shi- Address reprint requests to:
mazu, T., and Sugimoto, H. (2001). A multicenter prospec- Kim Peterson, M.S.
tive randomized controlled trial of the efficacy of mild hy- Department of Medical Informatics
pothermia for severely head injured patients with low
and Clinical Epidemiology
intracranial pressure. Mild Hypothermia Study Group in
Japan. J. Neurosurg. 94, 50–54.
Oregon Health & Science University
3181 SW Sam Jackson Park Road
Shiozaki, T., Kato, A., Taneda, M., Hayakata, T., Hashiguchi, Mail Code BICC
N., Tanaka, H., Shimazu, T., and Sugimoto, H. (1999). Lit-
Portland, OR 97239
tle benefit from mild hypothermia therapy for severely head
injured patients with low intracranial pressure. J. Neurosurg.
91, 185–191. E-mail: peterski@ohsu.edu

71

You might also like