Comparison of protocols and registry entries to published reports for randomised controlled trials

Background Publication of complete trial results is essential if people are to be able to make well‐informed decisions about health care. Selective reporting of randomised controlled trials (RCTs) is a common problem. Objectives To systematically review studies of cohorts of RCTs to compare the content of trial reports with the information contained in their protocols, or entries in a trial registry. Search methods We conducted electronic searches in Ovid MEDLINE (1950 to August 2010); Ovid EMBASE (1980 to August 2010); ISI Web of Science (1900 to August 2010) and the Cochrane Methodology Register (Issue 3, 2010), checked reference lists, and asked authors of eligible studies to identify further studies. Studies were not excluded based on language of publication or our assessment of their quality. Selection criteria Published or unpublished cohort studies comparing the content of protocols or trial registry entries with published trial reports. Data collection and analysis Data were extracted by two authors independently. Risk of bias in the cohort studies was assessed in relation to follow up and selective reporting of outcomes. Results are presented separately for the comparison of published reports to protocols and trial registry entries. Main results We included 16 studies assessing a median of 54 RCTs (range: 2 to 362). Twelve studies compared protocols to published reports and four compared trial registry entries to published reports. In two studies, eligibility criteria differed between the protocol and publication in 19% and 100% RCTs. In one study, 16% (9/58) of the reports included the same sample size calculation as the protocol. In one study, 6% (4/63) of protocol‐report pairs gave conflicting information regarding the method of allocation concealment, and 67% (49/73) of blinded studies reported discrepant information on who was blinded. In one study unacknowledged discrepancies were found for methods of handling protocol deviations (44%; 19/43), missing data (80%; 39/49), primary outcome analyses (60%; 25/42) and adjusted analyses (82%; 23/28). One study found that of 13 protocols specifying subgroup analyses, 12 of these 13 trials reported only some, or none, of these. Two studies found that statistically significant outcomes had a higher odds of being fully reported compared to nonsignificant outcomes (range of odds ratios: 2.4 to 4.7). Across the studies, at least one primary outcome was changed, introduced, or omitted in 4‐50% of trial rep