Reproducible research in the way that it is usually done is not a good guarantee against fraud. It is not difficult to create fairly convincing data sets when they are restricted only to the variables of interest. As an example if someone has an observational data to determine the relationship between prescription of a drug and an adverse event, all they need to do is with around the patients in each group. What does work are the requirements of the FDA and NIH where everything must be able to be tracked back to source data.

Richard Hockey wrote:

https://www.the-scientist.com/news-opinion/the-lancet-alters-editorial-practices-after-surgisphere-scandal-67953

The journal will also introduce “additional peer-review requirements,” including ensuring that “at least one peer reviewer is knowledgeable about the details of the dataset being reported”

Note in particular comments from UCSF infectious diseases researcher Matthew Spinelli and UCSF biostatistician David Glidden:

<<Spinelli and Glidden tell The Scientist that journals should put more emphasis on the conversation that takes place after a paper is published. “A process which reflects the capabilities of modern, on-line journals would provide a record of post-publication review and would recognize the work of external reviewers,” Spinelli writes. “This process would be more accountable, fairer, and more transparent.” >>

Post-publication review (and there should not be a time limit) is a crucial defense against superficially plausible claims.  Issues with the analysis, and/or its interpretation, appear to me more common than issues with the data.  Post-publication review provides important feedback for reviewers, apprising them of issues to which they have not paid attention or have missed -- a form of on -the-job training!  How much passes under the radar because claims made do not attract the same widespread attention as the Surgisphere papers?