Reporting of statistically significant results at ClinicalTrials.gov for completed superiority randomized controlled trials.

Background: Publication bias and other reporting bias have been well documented for journal articles, but no study has evaluated the nature of results posted at ClinicalTrials.gov. We aimed to assess how many randomized controlled trials (RCTs) with results posted at ClinicalTrials.gov report statistically significant results and whether the proportion of trials with significant results differs when no treatment effect estimate or p-value is posted.

Methods: We searched ClinicalTrials.gov in June 2015 for all studies with results posted. We included completed RCTs with a superiority hypothesis and considered results for the first primary outcome with results posted. For each trial, we assessed whether a treatment effect estimate and/or p-value was reported at ClinicalTrials.gov and if yes, whether results were statistically significant. If no treatment effect estimate or p-value was reported, we calculated the treatment effect and corresponding p-value using results per arm posted at ClinicalTrials.gov when sufficient data were reported.

Results: From the 17,536 studies with results posted at ClinicalTrials.gov, we identified 2823 completed phase 3 or 4 randomized trials with a superiority hypothesis. Of these, 1400 (50%) reported a treatment effect estimate and/or p-value. Results were statistically significant for 844 trials (60%), with a median p-value of 0.01 (Q1-Q3: 0.001-0.26). For the 1423 trials with no treatment effect estimate or p-value posted, we could calculate the treatment effect and corresponding p-value using results reported per arm for 929 (65%). For 494 trials (35%), p-values could not be calculated mainly because of insufficient reporting, censored data, or repeated measurements over time. For the 929 trials we could calculate p-values, we found statistically significant results for 342 (37%), with a median p-value of 0.19 (Q1-Q3: 0.005-0.59).

Conclusions: Half of the trials with results posted at ClinicalTrials.gov reported a treatment effect estimate and/or p-value, with significant results for 60% of these. p-values could be calculated from results reported per arm at ClinicalTrials.gov for only 65% of the other trials. The proportion of significant results was much lower for these trials, which suggests a selective posting of treatment effect estimates and/or p-values when results are statistically significant.