There are insufficient data to support a Level I recommendation for this topic.

There are insufficient data to support a Level II recommendation for this topic.

Pooled data indicate that prophylactic hypothermia is not significantly associated with decreased mortality when compared with normothermic controls. However, preliminary findings suggest that a greater decrease in mortality risk is observed when target temperatures are maintained for more than 48 h.

Prophylactic hypothermia is associated with significantly higher Glasgow Outcome Scale (GOS) scores when compared to scores for normothermic controls.

As stated in the Method Section of this guideline, to determine the recommendation level derived from a metaanalysis, three criteria are considered: (1) are all included studies of the same quality class, (2) are the findings of the studies in the same or contradictory directions, and (3) what are the results of sub-analyses that examine concerns about potential confounding factors? In this meta-analysis, although all included studies were Class II, the sub-analyses findings introduced sufficient concern about unknown influences to render the recommendation a Level III.

Although hypothermia is often induced prophylactically on admission and used for ICP elevation in the ICU in many trauma centers, the scientific literature has failed to consistently support its positive influence on mortality and morbidity. Four meta-analyses of hypothermia in patients with TBI have been published. All analyses concluded that the evidence was insufficient to support routine use of hypothermia, and recommended further study to determine factors that might explain variation in results. Thus, for this topic a meta-analysis was conducted of induced prophylactic hypothermia that includes studies published subsequent to the last meta-analysis, using specific inclusion criteria designed to minimize heterogeneity. Only studies assessed to be Class II evidence or better were included. Also excluded was literature about induced hypothermia for ICP control because there were inconsistent inclusion criteria and outcome assessments across studies.

Selection criteria were as follows:

• Patients with TBI, age >= 14 years (studies that enrolled patients under age 14 were included if at least 85% of patients were >= 14 years)
• Hypothermia therapy used as prophylaxis, regardless of intracranial pressure (ICP) (studies in which hypothermia was used as treatment for uncontrollable ICP, and those that enrolled only patients with controlled ICP (e.g., <20 mm Hg), were excluded)
• Assessed all-cause mortality

All-cause mortality at the end of the follow-up period was the primary outcome evaluated. Secondary outcomes included favorable neurological status, defined as the proportion of patients that achieved a Glasgow Outcome Scale score (GOS) of 4 or 5 (good outcome) at the end of the follow-up period.

Only data from the moderate (Level II) to good (Level I) quality trials were used to calculate the pooled relative risk (RR) and 95% confidence intervals (CIs) for all cause mortality and good neurological outcome using a random-effects model. Analyses were conducted using RevMan version 4.2 (Update Software). Statistical heterogeneity was calculated using the chi-squared test.

A priori particular aspects of hypothermia treatment were identified, and a sensitivity analysis was conducted to examine their relationship to all-cause mortality. These aspects were as follows:

• Target cooling temperature (32-33°C or >33°C)
• Cooling duration (<48 h, 48 h, or >48 h)
• Rate of rewarming (1°C per hour, 1°C per day, or slower)

A post hoc analysis was conducted of the relationship between trial setting (single center vs. multicenter) and mortality.

Reference lists of the four previous good-quality systematic reviews provided the basis for identification of all eligible randomized controlled trials from 1966 through September, 2002. Electronic databases included MEDLINE (OVID), EMBASE, Cochrane Library, Current Contents, EMBASE, CENTRAL, Science Citation Dissertation Abstract, AANS and CNS abstract center, and Specialist Trials Register for the Injuries Group. Searches included various combinations of MeSH (Medical Subject Headings) terms and text words for hypothermia, brain injury, craniocerebral trauma, and neurosurgery. A supplemental literature search was conducted of MEDLINE (2002 through April 2006) using the search strategy for this question (see Appendix B). Of 29 potentially relevant trials, 13 met the inclusion criteria for this report. Of those, six trials were assessed as Level II (moderate quality) and seven as Level III (poor quality) Only the moderate quality trials are included in the meta-analysis (Evidence Table I).

Overall, the risk of all-cause mortality for patients treated with hypothermia was not significantly different from that observed in the control groups (RR 0.76; 95% CI 0.50, 1.05; p = 0.18) (Fig. 1). However, hypothermia was associated with a 46% increased chance of good outcome, defined as a GOS score of 4 or 5 (RR 1.46; 95% CI 1.12, 1.92; p = 0.006) (Fig. 2).

Interpretation of results from subgroup analyses based on aspects of hypothermia treatment protocols is limited due to small sample sizes.

Mortality. Cooling duration was the only aspect of hypothermia treatment, specified a priori, that was possibly associated with decreased rates of death. Preliminary results suggest that there was a significantly lower risk of death when hypothermia was maintained for more than 48 h (RR 0.51; 95% CI 0.34, 0.78). Target cooling temperature and rate of rewarming did not influence mortality.

The post hoc analysis indicated an influence of study setting on mortality. One of the six trials, which was the largest trial (n = 392) was conducted at multiple centers. When removed from the analysis, hypothermia was associated with a significant decrease in mortality (RR 0.64; 95% CI 0.46, 0.89).

GOS. Target temperature was the only aspect of hypothermia treatment protocols that was possibly associated with improved outcomes. There was significantly greater chance of better outcomes with target temperature ranges of 32-33°C (RR 1.67; CI 1.18, 2.35) and 33-35°C (RR 1.75; CI 1.12, 2.73). Findings from subgroup analyses did not suggest any clear relationship between cooling duration or rate of rewarming and improved outcomes.

As with mortality, the post hoc analysis of study setting showed a higher chance of good outcomes from studies conducted in single centers (RR 1.70; CI 1.33, 2.17)

A concern regarding interpretation of outcome, introduced in one RCT3 and a recent systematic review, is the interaction of the patient's baseline temperature at hospital admission with treatment group allocation. As illustrated in Table 1, at randomization, there are four potential patient categories: (a) hypothermic patient randomized to hypothermia; (b) hypothermic patient randomized to normothermia; (c) normothermic patient randomized to hypothermia; and (d) normothermic patient randomized to normothermia.

There is potential for either a confounding influence or an effect modification (interaction) of warming hypothermic patients who are randomized to the normothermic group, or of having patients in the normothermic group become hypothermic during the observation period. Clifton et al addressed this question in part by conducting a subanalysis of 102 patients who were hypothermic at hospital admission, and finding a non-significant trend toward poor outcomes in the control group (Table 1, category b) compared to the treatment group (category a). Data in the studies included in this meta-analysis were insufficient to address this question. Thus, all results reported must be considered in light of the possibility that baseline temperature either confounds or interacts with outcome. Furthermore, there is the possibility that patients who are hypothermic on admission have a decreased brain temperature and may have a pseudo-lowering of the GCS independent of the level of TBI.

Evidence from six moderate quality RCTs did not clearly demonstrate that hypothermia was associated with consistent and statistically significant reductions in all-cause mortality. However, patients treated with hypothermia were more likely to have favorable neurological outcomes, defined as GOS scores of 4 or 5. Preliminary findings suggest that hypothermia may have higher chances of reducing mortality when cooling is maintained for more than 48 hours. Interpretation of results from this and other subgroup analyses based on different aspects of the hypothermia treatment protocols were limited due to small sample sizes. Potential confounding and effect modifying factors that are not accounted for in the trials included in this analysis, such as patients' temperature at admission, limit these recommendations to Level III.

Although 13 RCTs of hypothermia meeting the inclusion criteria have been conducted, only six were included in the meta-analysis due to serious quality flaws in the remaining seven. Flaws, which are markers for improvement in future research, included the following:

• Inadequate or poorly described randomization or allocation concealment
• Inability to rule out confounding of treatment effects, due to differences in (or inadequately described) baseline prognostic factors
• No blinding of outcome assessors
• Inadequate management of missing outcome data

Improvements should also include use of independent event monitoring committees, larger sample sizes across multiple trauma centers, and increased standardization and reporting of control group temperature management protocols.