Reason for withdrawal from publication
There were 2 reasons for withdrawal:
1) Comments received have suggested the presence of a substantial number of additional trials claiming positive results that need to be reviewed, and that, if eligible, could affect the conclusions.
2) Some errors made in the extraction of data from two trials were pointed out by Peter Gøtzsche and colleagues. These errors involved only 3 in over 100 figures and we note that replacing our calculations with those suggested by Gøtzsche et al. would not affect the conclusions of the review.