Hypoxic ischemic encephalopathy (HIE) following perinatal asphyxia contributes significantly to neonatal mortality and morbidity including long-term neurodevelopmental sequelae in up to 25%-60% of survivors 123. Despite significant research there is still no proven intervention for neuroprotection in HIE 4. The literature on therapeutic hypothermia as a treatment for "white asphyxia" dates as far back as early 60s 56. Experimental studies have shown that mild to moderate hypothermia (33–34°C), applied within the first hours of an acute hypoxic event, is neuroprotective 789. A Cochrane review incorporating two small randomized controlled trials (RCT) has reported no evidence of benefit or harm related to therapeutic hypothermia in term neonates (N = 50) with HIE 4. Three large RCTs have been published since the last substantive amendment of this systematic review in July 2003 101112. Given the significance of the condition, and the encouraging results of the recent RCTs, an up to date systematic review was conducted to evaluate the efficacy and safety of therapeutic hypothermia in term neonates with HIE 1314.

RCTs comparing therapeutic hypothermia, by either selective head cooling or whole body cooling, with normothermia in term neonates with perinatal asphyxia and HIE were eligible. Asphyxia was considered to be present if at least one of the following criteria was met: Apgar score of ≤ 5 at 10 minutes, at least one cord pH or arterial pH ≤ 7.1 or base deficit ≥ 12 within the first hour of life, ongoing resuscitation or mechanical ventilation at 10 minutes of life. HIE had to be defined by standardized neurological examination 1516. Neonates with major congenital abnormalities were excluded.

The primary outcome was a composite of death or neurodevelopmental disability at ≥ 18 months of life. Disability included cerebral palsy (CP) according to the Gross Motor Function Classification System (GMF) 17 or another validated scale, developmental delay as measured by Griffiths or Bayley assessment 1819, intellectual impairment (IQ > 2 SD below the mean), blindness (vision < 6/60 in both eyes), and hearing loss requiring amplification. Secondary outcomes included individual components of the primary composite outcome and adverse events such as sinus bradycardia, arrhythmia, arterial hypotension (mean arterial pressure < 40 mm Hg), thrombocytopenia, coagulopathy, anemia, hypoglycemia, abnormal renal function (urine output < 0.5ml/kg/hour and/or serum creatinine > 0.09 mmol/l), hepatic dysfunction (aspartate aminotransferase > 200 IU/l, alanine aminotransferase > 100 IU/l), sepsis, seizures and hypokalemia.

A systematic literature search was conducted in July 2006 according to the methodology of the Cochrane Neonatal Review Group. The databases searched included the Cochrane Database of Systematic Reviews (Issue 3, 2006), the Cochrane Central Register of Controlled Trials, EMBASE, CINAHL, and MEDLINE databases using the following search strategy: "Infant, Newborn" [MeSH] AND ("Hypothermia" [MeSH] OR "Hypothermia, Induced" [MeSH]) AND ("Asphyxia" [MeSH] OR "Asphyxia Neonatorum" [MeSH] OR "Hypoxia-Ischemia, Brain" [MeSH] OR "hypoxic ischemic encephalopathy" OR "hypoxic ischemic encephalopathy"). Cross-references of publications were checked. A hand search of the proceedings of the Pediatric Academic Societies published in Pediatric Research from 1980 was conducted. No language restrictions were applied. SKP and SMS designed the review protocol. All authors searched the literature independently and assessed inclusion criteria and quality of the trials. SMS and SR independently extracted the data. Inconsistencies were sorted out by discussion. Trial quality was assessed by method of randomization, concealment of patient allocation, blinding of intervention, blinding of outcome assessors, and completeness of follow-up.

Meta-analysis was performed using Review Manager software (RevMan, version 4.2.7 for Windows, Oxford, England: The Cochrane Collaboration, 2003). Relative risk (RR) and risk difference (RD) were calculated with 95% confidence intervals (CI). Preplanned subgroup analysis for selective head cooling and whole body cooling was carried out. The number needed to treat (NNT) or harm (NNH) was calculated for significant comparisons. Heterogeneity was estimated by the I² statistic. A fixed effects model was used. Reporting follows the QUOROM guidelines 20.

187 abstracts were identified. Twelve potentially relevant reports were retrieved for detailed evaluation. A total of eight reports 1011122122232425 of five RCTs involving 552 neonates were eligible for inclusion in the analysis. Two RCTs involved selective head cooling by a cooling cap 1021, the other three involved whole body cooling 111224. Tables 1, 2, 3 summarize their characteristics and quality assessment. No significant heterogeneity was noted using the I² statistic. Gunn et al published their results as 3 different reports using 4 sequential temperature ranges for cooled infants 212223. From these 3 reports, we present the combined data on all neonates randomized to a rectal temperature of 34.0–35.5°C and refer to the trial as Gunn et al 1998. Eicher et al reported the safety 25 and efficacy 12 outcomes of therapeutic hypothermia in two separate publications. For the purpose of this meta-analysis their study is referred to as Eicher et al 2005. The studies Eicher et al 2005 and Shankaran et al 2002 24 were included in the review only for analysis of mortality and adverse events because the follow-up rate in Eicher et al 2005 was low (68%) and Shankaran et al 2002 did not report long-term follow up. The reasons for excluding four trials 26272829 from the meta-analysis and their characteristics are summarized in Table 4.

Our results suggest that in general, cooling of neonates with HIE has a beneficial effect on the primary composite outcome of death or disability at 18–22 months. The similarity of outcomes between trials despite the heterogeneity related to various factors including methodology of cooling (head vs. body, devices, target temperatures, site of monitoring, duration of intervention etc.), patient characteristics (place of birth, temperature and age at enrolment etc.), and the definition and degree (moderate and/or severe) of neuro-developmental disability (Table 1) is reassuring for the generalisability of the findings. Differences in the behavior of the composite outcome vs. its individual components are an important consideration 30. The selection of death and neurodevelopmental disability as components of the prespecified primary composite outcome is justified. The frequency of the components of the primary composite outcome is significant and comparable, assuring that no individual component is driving it in any specific direction. The data show reasonably convincingly that the benefit for the primary composite outcome is significant and is probably a fair reflection of benefit for its individual components. Individual components of the primary composite outcome show significant benefits of cooling but have wide CIs suggesting that more data is needed to minimize the uncertainty. However, it is important to note that interpretation of CI is a personal and subjective issue. Overall, there seems to be reasonably good evidence of a real benefit of cooling on the primary composite outcome and its components.

The definition of disability in the whole body cooling trial by Shankaran et al is quite different from that in the selective head cooling trials by Gluckman et al and Gunn et al (moderate/severe vs. only severe disability). However the effect of this difference in definitions on the results of the meta-analysis is limited because the proportion of infants with moderate disability is low in all trials. Subgroup analysis seems to suggest that the results of the selective head cooling trials are not as convincing as those of the whole body cooling trials. However, there might be no clinically important difference between these cooling techniques for several reasons: Firstly, this subgroup analysis is clearly dominated by the two major trials Gluckman et al and Shankaran et al, and therefore of limited value. The overall trend for both cooling techniques is towards a benefit (Figure 1). Secondly, differences in the severity of neuronal injury may explain these findings considering the differences in the inclusion criteria of those two trials (Table 2). The proportion of neonates with very low Apgar scores, severe aEEG background activity and severe clinical encephalopathy was higher in the intervention group in Gluckman et al, possibly reducing the chances to demonstrate selective head cooling benefits. A significant reduction in death or major disability was however noted in the prespecified subgroup analysis of neonates with only moderate injury defined by aEEG criteria 1031. Thirdly, palliation bias in the form of a higher rate of withdrawal of treatment (27 vs. 12) in the control group may also have played a role in the significant benefit reported in the whole body cooling trial by Shankaran et al 32. In addition, 41/106 neonates in the control group of Shankaran et al, at least once had a temperature > 38°C within the 72 hours of the intervention, which may have influenced for their outcome 32.

The frequency of sinus bradycardia following therapeutic hypothermia was significant. A borderline effect on thrombocytopenia was noted and there was a trend towards a higher risk of anemia, coagulopathy, and hypotension (Table 5). However, sinus bradycardia is a physiological response rather than a true adverse event and did not compromise perfusion, and thrombocytopenia was not reported as of clinical importance. The adverse events therefore may be outweighed by the potential benefits.

Given the overall encouraging results without significant adverse effects it is not surprising that some centers may now consider therapeutic hypothermia as a standard treatment for HIE 33. However, many experts including a commission of the American Academy of Pediatrics 34 have suggested that further research should continue and therapeutic hypothermia should not be offered outside RCTs. Their suggestions are based on heterogeneity as discussed above and the possibility that neurological outcomes at 18 to 22 months may not reflect the true long-term benefits 3536. The rate of severe disability is very unlikely to change, however, more subtle neurodevelopmental problems that cannot be assessed at the age of 18 months may become apparent by school age 1. The unaddressed issues include the specific target population that is most likely to benefit, the most effective and safe method for cooling, the optimal age at onset and duration of cooling, and the field difficulties in applying any specific method for cooling, particularly for outborn neonates 34. In practice, hypothermia is quite frequent in asphyxiated neonates, whereas guidelines for rewarming are not standardised/uniform. The targets, methods as well as the speed of rewarming may influence the neuronal recovery/damage following HIE. This issue is especially important during transport of hypothermic neonates with HIE. The field difficulties have been addressed to some extent by Eicher et al who showed that it is feasible to cool outborn neonates with ice bags followed by cooling with a blanket on reaching the receiving hospital. This approach can help to reduce the time between birth asphyxia and initiation of cooling. Animal studies have clearly shown that there is a correlation between early onset of cooling and treatment effect 37.

Experts have advised that centers wishing to offer therapeutic hypothermia outside RCTs should adhere strictly to a trial protocol and have established the substantial resources required to cool neonates with HIE. The minimum resources include a transport team to retrieve neonates and start cooling before four to six hours of life and a multidisciplinary team for long-term neurodevelopmental follow-up 33.

At least three more RCTs of therapeutic hypothermia aiming at a combined total of over 650 neonates are currently in progress 383940. The long-term outcomes ≥ 18 months of age of the single trial with complete recruitment (according to the TOBY trial website) will not be ready for publication until end of 2008 at the earliest 38. Ideally it is preferable to include the long-term results of those ongoing studies in this systematic review to have definitive answers. However, those studies are not designed to answer all the unaddressed issues listed earlier, therefore waiting for their results is ethically complex. They may show beneficial effects of therapeutic hypothermia while narrowing the CI. In that case it is disturbing to think that while waiting for the CI to narrow the purists may have denied a beneficial intervention to neonates with HIE. Obviously ethics 41, resources, parents' wishes, and last but not the least, the anxiety related to future medicolegal challenges, will have to be balanced before deciding whether therapeutic hypothermia can be offered as a standard treatment for HIE. Continuing to participate in a trial of therapeutic hypothermia while offering it to neonates whose parents insist on it probably violates the principle of equipoise, the very justification for conducting a RCT.

Evidence from high quality RCTs indicates that overall, cooling of neonates with moderate to severe HIE reduces the risk of death or disability at 18 to 22 months without significant adverse effects. Despite the methodological differences, wide CIs, and lack of long-term follow-up data, the consistency of benefits and its sound scientific basis indicate that cooling may be an attractive option for neuroprotection in HIE – a condition that lacks any effective treatment at present.

The author(s) declare that they have no competing interests.

SKP and SMS designed the review protocol. All authors searched the literature independently and assessed inclusion criteria as well as quality of the trials. SMS and SR independently extracted the data. SMS and SKP wrote the manuscript. All authors read and approved the final manuscript.