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

**Background:** Publication bias and selective non-reporting bias threaten the validity of systematic reviews. Searching clinical trial registries is recommended to identify unpublished trials and assess reporting bias, but prior studies show underutilization of registry searches. Only limited research has quantified how often registry searches yield additional trials and whether including them changes meta-analytic results. This study aimed to evaluate the impact of searching four clinical trial registries on meta-analyses of pharmaceutical and non-pharmaceutical interventions.

**Methods:** Systematic reviews with meta-analyses indexed between 2 November and 2 December 2020 were identified from PubMed, Scopus, Science Citation Index, Social Sciences Citation Index, and Education Collection. Reviews were eligible if they considered randomized trials, had not searched a trials register, and reported a search strategy with dates. For 101 eligible reviews, one author searched ClinicalTrials.gov, EudraCT, ANZCTR, and the WHO ICTRP search portal using search terms derived from each review's population and intervention criteria. Records were screened against each review's eligibility criteria. For trials with results available (posted in the register or published before the review's search date) that could contribute to a meta-analysis, outcome data were extracted. Meta-analyses were re-run including the new trials using the same effect measures and statistical models as the original reviews. The ratio of risk ratios (RRR) for binary outcomes and difference in standardized mean differences (dSMD) for continuous outcomes were synthesized using random-effects meta-analysis with Hartung-Knapp-Sidik-Jonkman confidence intervals.

**Key Results:** Among 101 reviews, additional eligible trials were identified for 63% (63/101). However, only 20% (20/101) had at least one new trial with results that could be included in a meta-analysis. A total of 29 new trials were added across 20 reviews (median 2 trials per review, IQR 1–2). The weight of new trials ranged from 0% to 36% (median 10%). For binary outcomes (5 meta-analyses), the combined RRR was 1.00 (95% CI 0.99–1.01; I²=0%; 95% predictive interval 0.99–1.01). For continuous outcomes (15 meta-analyses), the combined dSMD was 0.01 (95% CI −0.04 to 0.06; I²=0%; 95% predictive interval −0.04 to 0.06). Thus, on average, adding registry-identified trials did not change meta-analytic estimates. The proportion of reviews with new trials varied by intervention type: pharmaceutical (64%), surgical (60%), dental (80%), nutritional (79%), physiotherapy (86%), psychology (67%), and traditional Chinese medicine (11%).

**Clinical Implications:** Searching clinical trial registries can identify additional eligible trials for a majority of systematic reviews, but most identified trials lack available results. When results are available, including them is unlikely to change meta-analytic conclusions on average, though individual reviews may see shifts. The findings highlight the need for better compliance with trial registration and results posting requirements. Systematic reviewers should still search registries to assess reporting bias and identify ongoing trials, but should be aware that registry searches require time and resource investment, particularly for broad reviews of behavioral or physiotherapy interventions. The study's limitations include single-author screening for most records, exclusion of reviews yielding >500 registry records, and a small sample of 20 re-analyzable meta-analyses.