Article type
Abstract
Background
Bayesian methods are becoming more popular in various areas of medical research, including standard pairwise meta-analysis.
Objectives
We aim to explore the methodological design and its reporting in health research, assess the overall quality, and provide some guidance on what items should be reported in systematic reviews with Bayesian meta-analysis.
Methods
We searched (as of 06-08-2023) PubMed and Cochrane Database of Systematic Reviews using the phrase ‘Bayesian meta-analysis’ in the title/abstract of published pairwise systematic reviews. Inclusion criteria were any pairwise systematic reviews with Bayesian meta-analysis, searched at least two databases, and published in English language. We excluded studies if these were not related to human health, protocols, conference abstracts, network meta-analysis, individual participant data meta-analysis, methodological reviews/articles, and if full texts were unavailable. A descriptive synthesis approach has been used to present findings and overall quality will be assessed by AMSTAR 2 tool.
Results
With initial screening, we identified 204 articles that met the eligibility criteria (final results will be presented during the meeting). Among 204 studies, over two-third studies published from Europe or North America (74.0%, n=151), primarily with observational designs (52.0%, n=106), published in specialised journals (78.9%, n=161). Median number of included studies were 19 (IQR 10-39) with 5900 participants (IQR 1609-26595) and published in a median of 4.84 (IQR 0.06-7.51) impact factor journals. Most frequent health condition under study was cardiovascular diseases (31.7%), followed by cancers (12.9%), infectious diseases (6.9%), and mental health problems (6.9%). Of 204 studies, 63.2% (n=129) were fully Bayesian meta-analysis, 65.2% (n=133) provided some form of justification for utilising Bayesian method, 54.9% (n=112) reported information on prior distribution, 31.4% (n=64) reported burn-in frequency, 34.3% (n=70) reported number of replication, 26.5% (n=54) reported sampling method, and 22.1% (n=45) reported diagnostic for posterior distribution. R statistical software (45.6%, n=93) was commonly used for conducting Bayesian meta-analysis. Only 7.8% (n=16) studies included statistical codes for Bayesian meta-analysis.
Conclusion
The preliminary results indicate suboptimal methodological design and its reporting. Therefore, standard reporting guidelines could facilitate better design and reporting of Bayesian meta-analysis.
Bayesian methods are becoming more popular in various areas of medical research, including standard pairwise meta-analysis.
Objectives
We aim to explore the methodological design and its reporting in health research, assess the overall quality, and provide some guidance on what items should be reported in systematic reviews with Bayesian meta-analysis.
Methods
We searched (as of 06-08-2023) PubMed and Cochrane Database of Systematic Reviews using the phrase ‘Bayesian meta-analysis’ in the title/abstract of published pairwise systematic reviews. Inclusion criteria were any pairwise systematic reviews with Bayesian meta-analysis, searched at least two databases, and published in English language. We excluded studies if these were not related to human health, protocols, conference abstracts, network meta-analysis, individual participant data meta-analysis, methodological reviews/articles, and if full texts were unavailable. A descriptive synthesis approach has been used to present findings and overall quality will be assessed by AMSTAR 2 tool.
Results
With initial screening, we identified 204 articles that met the eligibility criteria (final results will be presented during the meeting). Among 204 studies, over two-third studies published from Europe or North America (74.0%, n=151), primarily with observational designs (52.0%, n=106), published in specialised journals (78.9%, n=161). Median number of included studies were 19 (IQR 10-39) with 5900 participants (IQR 1609-26595) and published in a median of 4.84 (IQR 0.06-7.51) impact factor journals. Most frequent health condition under study was cardiovascular diseases (31.7%), followed by cancers (12.9%), infectious diseases (6.9%), and mental health problems (6.9%). Of 204 studies, 63.2% (n=129) were fully Bayesian meta-analysis, 65.2% (n=133) provided some form of justification for utilising Bayesian method, 54.9% (n=112) reported information on prior distribution, 31.4% (n=64) reported burn-in frequency, 34.3% (n=70) reported number of replication, 26.5% (n=54) reported sampling method, and 22.1% (n=45) reported diagnostic for posterior distribution. R statistical software (45.6%, n=93) was commonly used for conducting Bayesian meta-analysis. Only 7.8% (n=16) studies included statistical codes for Bayesian meta-analysis.
Conclusion
The preliminary results indicate suboptimal methodological design and its reporting. Therefore, standard reporting guidelines could facilitate better design and reporting of Bayesian meta-analysis.