Impact of searching clinical trials registers in systematic reviews of pharmaceutical and non-pharmaceutical interventions: Reanalysis of meta-analyses

Abstract

Systematic reviewers are advised to search trials registers to minimise risk of reporting biases. However, there has been little research on the impact of searching trials registers on the results of meta-analyses. We aimed to evaluate the impact of searching clinical trials registers for systematic reviews of pharmaceutical or non-pharmaceutical interventions. We searched PubMed, Scopus, Science Citation Index and Social Sciences Citation Index, and Education Collection for systematic reviews with meta-analyses indexed from 2 November to 2 December 2020. A random sample of systematic reviews was initially drawn, and for reviews which considered randomised trials eligible for inclusion, which had not searched a trials register, we searched ClinicalTrials.gov, EudraCT, ANZCTR, and the WHO ICTRP search portal for eligible trials. We compared meta-analytic effect estimates before and after including results from additional trials identified. We found additional trials for 63% (63/101) of eligible reviews; however, trials with results that could contribute to a meta-analysis were identified for only 20% (20/101) of the reviews. On average, there was no difference in the meta-analytic effect estimates before versus after adding the new trials. In summary, searching clinical trial registers led to identification of additional trials for many reviews; however, very few trials had results available for inclusion in meta-analyses. Including results from the new trials led to no change in the meta-analytic estimates, on average. Trials registers would be even more valuable to systematic reviewers if more trialists made use of them (i.e., registered their trials and posted results in a timely manner).

Keywords: meta-analysis; publication bias; systematic review; trial registration.

FIGURE 1

PRISMA 2020 flow diagram of identification, screening and inclusion of systematic reviews. *There were 6292 unique records after duplicates were removed, but we only needed to screen 2000 randomly sorted records to reach our target sample size of 302 reviews. **We only needed to screen 438 of the 603 full text reports retrieved to reach our target sample size of 302 reviews [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 2

Random‐effects meta‐analysis (MA) of the difference in meta‐analytic risk ratios (ratio of risk ratios [RRR]) before versus after adding the new trials. RRR estimates below 1 indicate that the original meta‐analytic estimate was more favourable to the experimental intervention than the revised meta‐analytic estimate [Colour figure can be viewed at wileyonlinelibrary.com]

FIGURE 3

Random‐effects meta‐analysis (MA) of the difference in meta‐analytic standardised mean differences (dSMD) before versus after adding the new trials. dSMD estimates below 0 indicate that the original meta‐analytic estimate was more favourable to the experimental intervention than the revised meta‐analytic estimate [Colour figure can be viewed at wileyonlinelibrary.com]