The authors reply
Rhodes et al (1) present several queries relating to our article (1). It is incorrect to say that there was no a priori protocol for the planned analysis. Our group has extensive experience in conducting meta-analyses all of which have been published previously and subjected to international peer review. As is common with such studies, the protocol was detailed but it is not usual to publish such protocols within the main article due to space concerns.
Furthermore, it is incorrect to say that inclusion and exclusion criteria are not given as they are published in the Materials and Methods section of our article (2). Heterogeneity is potentially an indication of population admixture, and one well-established means of dealing with this is to use iterative analyses to identify the studies that account for most of the heterogeneity. This was undertaken and is presented in the Results section of our article (2). Furthermore, we are at pains in our article to discuss the limitations of our study and the well-known limitations of the use of meta-analyses. There are many ways to test the quality of publications, and we used heterogeneity and funnel plot analyses.
Finally, we fundamentally disagree with Rhodes et al (1) about the problem of pooling large volumes of data. It is the very fact that data are so large that errors from smaller studies can be minimized. Our data set is, to the best of our knowledge, the largest analyzed to date. The analysis presented by Rhodes et al (1) has been undertaken on a hypothetical and extrapolated data set. Although we respect their analysis, such an extrapolation cannot be a substitute for real data.
We do however fully agree with Rhodes et al (1) that the use of hypothermia in the management of traumatic head injury is a complex area, and we would endorse a comprehensive international large-scale prospective study.