BJOG authors do not usually need to be told that ‘random’ treatment allocation reduces the risk that treatment and control groups will differ on unknown factors, but they may sometimes forget that randomisation is not the only important feature of good treatment evaluation.
Trials must also be of adequate size, the interventions in the treatment and control groups must be clearly described, and the primary endpoint and analysis plan must be defined before starting. There must be a reasonable level of compliance with the treatment allocated, follow-up rates must be both high and equal between the groups, and assessment of any subjective endpoints should be made without knowledge of the treatment allocated. Finally, the trial should be equally likely to get published whatever the result. If any of these other things go wrong, the trial may mislead readers.
This is why BJOG encourages trials where all these other potential sources of error have been eliminated as far as possible. The best way to achieve this is for researchers to write detailed trial protocols and register them in a public forum. This allows others to pick up errors before it is too late and makes it impossible to hide any changes between design and publication.
As from now BJOG is accepting trial protocols for registration. The Lancet has been doing this for some time, but we believe we are the first obstetric and gynaecological journal to do so. Protocols should be submitted in the usual way via http://bjog.allentrack.net. The instructions are available on the website, and reprinted on the Instructions to authors pages. Some model protocol headings are also printed on p. 762. These are not obligatory, but many researchers have indicated that they find them useful.
Our plan is to send protocols to referees and select the best for registration with BJOG. These will be published electronically on BJOG online and a summary will be printed in the journal. We hope that the main trial publication would be submitted first to BJOG and we promise to send it to referees. If their reports are satisfactory, we would normally agree to publish it, although we cannot guarantee this at the registration stage.
Our aim is to encourage good behaviour by both authors and editors. We want to stop authors giving up trials early and then analysing the incomplete sample as if that had always been their intended sample size; if early stopping is unavoidable, readers must know. We do not like authors to analyse many endpoints and present the one that is apparently statistically significant, as if it had been pre-specified. We want to stop editors behaving badly, too. Protocol registration will discourage them from letting the result of a trial colour their decisions and should reduce publication bias.
BJOG may never publish a paper entitled An independently randomised, double-blind trial of treatment X with pre-specified sample size, primary endpoint, and analysis plan, and high compliance and follow-up, but it may not be too long before we publish one entitled A pre-registered protocol-driven evaluation of treatment X. I look forward to the day.