Background: Information on the effectiveness and safety of health care interventions should be valid, precise, up-to-date, comprehensible and freely available. Currently none of these criteria is adequately satisfied. Many systematic reviews contain a biased sample of under-powered trials (some include fabricated trials) and reviews are often long and unreadable.
Methods:
In response to these concerns, the Cochrane Injuries Group developed a new policy (Roberts 2015). In addition to following standard Cochrane methods, authors of new and updated reviews should:
1. Only include a trial if it was prospectively registered (unless the trial report was published before 2010).
2. Provide a sample size estimate showing how many participants need to be included in a meta-analysis for reliable results. Trial sequential methods should be used to explore all 'significant' treatment effects obtained before reaching the required sample size.
3. The editorial team will conduct statistical checks on data included in a review. Authors of selected trials will be asked to provide the original trial data for checking. If authors decline to provide it, the trial will be removed from the review.
Conclusions: We invite critical discussion of our policy and urge other review groups to consider adopting it.

Roberts I, Ker K, Edwards P, Beecher D, Manno D, Sydenham E. The knowledge system underpinning health care is not fit for purpose and must change. BMJ 2015;350:h2463