2.3. Are there important inconsistencies between the publication and the registration documents?

• Where a trial has been registered (prospectively or otherwise), the reviewer may check for major inconsistencies between the publication(s) and registration entry.
• The reviewer should consider the first version of the registration page and any other versions posted before recruitment started, as well as the history of changes made to the registration page, rather than considering only the latest version.
• Unexplained discrepancies in planned sample size, interventions, study dates, or eligibility criteria could be grounds for concern. However, failure to achieve the planned sample size should not be considered a marker of untrustworthiness; many trials struggle with recruitment and considering adequacy of study sample size is not within the scope of INSPECT-SR.
• Disagreements between study publications and study registration documents posted after the trial start may be particularly concerning.
• The purpose of this check is not to investigate outcome reporting bias, which is covered by existing Risk of Bias frameworks. The focus should therefore be on other key aspects of the trial.
• Major unexplained discrepancies with a trial protocol, where available, could also lead to valid concerns about trustworthiness.
• If there is no study registration, reviewers should answer “Not applicable” for this check.
• The answer to this check should contribute to a domain-level judgement.

Example of check 2.3

A clinical trial has been retrospectively registered on the Iranian Registry of Clinical Trials, with the registration number IRCTXXXXXXXXXX. Despite being retrospectively registered, examination of the revisions made to the record indicates that the description of the control group differed on the registration record compared to the published manuscript, and was amended to match the manuscript over two years after publication. In addition, the recruitment date reported in the published manuscript (April 2016 to September 2016) differs compared to the recruitment dates described on the retrospective registration record (April 2016 to May 2016). The reviewer answers “yes” for this check, and this response contributes to the domain-level judgement.