Possible explanations for the results of CRASH/Author's reply

• Published in: The Lancet (British edition) Vol. 365; no. 9455; p. 212
• Main Authors: Douglas, Ivor S, Kasperlik-Zaluska, Anna A, Slowinska-Srzednicka, Jadwiga, Bonicki, Wieslaw
• Format: Journal Article
• Language: English
• Published: London Elsevier Limited 15.01.2005
• Subjects: Head injuries; Mortality
• ISSN: 0140-6736; 1474-547X

The 3% excess of deaths among the head injury patients in the CRASH trial1 who were allocated active methylprednisolone is significant (P=0.0001, if no allowance is made for the data-dependent stop halfway through the study). But, if the results of this trial are considered together with those from the other randomised trials of such treatment, which collectively indicate no hazard,1 then the excess becomes 2%, and the p value becomes 0.001, which is ten times less extreme. Hence, the apparent excess of deaths in these studies could be largely or wholly due to an extreme play of chance that was made to seem still more extreme by the datadependent halt to recruitment to CRASH. If so, the accompanying Comment2 should not have described the apparent increase in mortality as indisputable. In my view, the cautious conclusion of the CRASH investigators is more appropriate: they review all the randomised evidence, and then claim only that the evidence shows that such treatment does not materially reduce mortality.