Publication bias in systematic reviews (SR) could result in serious over- or underestimation of effects. Statistical methods to detect this bias are methodologically limited and depend on the number of included trials. Therefore the search of clinical trial registries is a substantial part in the conduct of Cochrane Reviews to identify unpublished data. While almost all current Cochrane Reviews search registries, it remains unclear how authors incorporate their findings.
To assess how unpublished data contributes to the analyses and conclusions of Cochrane Reviews.
We conducted a SR of the latest published 50 Cochrane Reviews evaluating drug interventions. We extracted data and dates of the conducted registry searches and on the handling of unpublished studies. Two authors selected the reviews and extracted data independently. In a next step we will repeat the registry searches of selected Cochrane Reviews to assess whether study data reported in registries only, affect the overall results of the review.
Most of the reviews included randomized trials only (46; 92%). The most searched resource was clinicaltrials.gov (47; 94%). Detailed results from the registry search were reported in 27 (54%) reviews. Overall, four (8%) reviews identified eight completed studies with outcome data included in the registry. Two (4%) reviews included data from these studies in the meta-analysis. Forty completed studies without outcome data were identified in 15 (30%) reviews. Seven (14%) reviews reported these among the ongoing studies, and seven (14%) in the studies awaiting classification section.
While Cochrane authors generally search trial registers, their findings are variably reported and only used infrequently in formulating conclusions. In the next step we will clarify whether study data included in registries only influence the overall results of Cochrane Reviews. We will present these findings at the Colloquium.
Patient or healthcare consumer involvement:
This review is valuable for the patients who participated in 40 studies with no available outcome data.