Article type
Year
Abstract
Background: Prospective meta-analyses (PMA) can reduce many of the problems associated with traditional meta-analyses by specifying study selection criteria, hypotheses and analyses before the results of included studies are known. This can reduce risk of publication bias and selective outcome reporting, and enables researchers to harmonise their ongoing research efforts to answer important questions with greater certainty. However, despite rising numbers of PMA, the terminology and definitions used to date have lacked clarity and consistency, and there is little guidance on how to conduct PMA. This threatens successful implementation of PMA.
Objective: To develop step-by-step guidance on how to conduct PMA.
Methods: We, the Cochrane PMA Methods Group, developed step-by-step guidance based on 1) a scoping review of methodology papers, 2) a scoping review of existing PMA 3) expert opinions 4) experiences with previous PMA. We illustrate each step with a recent case study.
Results: We describe seven steps for PMA (Figure). First, a protocol needs to be developed, including details on collaboration policies (Step 1). Next, a systematic search for planned/ongoing studies should be conducted, including a search of trial registries, medical databases and contacting stakeholders (Step 2) and eligible studies need to be identified for inclusion (Step 3). Importantly, only studies for which the results are not known can be included in a PMA. These studies are then invited to form a collaboration (Step 4), ideally including a steering and data analysis committee, with representatives from each study. Next, core outcomes, common intervention features and a statistical analysis plan are agreed upon within the collaboration (Step 5). This can be particularly useful for rare but important outcomes such as adverse side effects, that individual studies would not have enough power to test statistically. There is usually a waiting period while all studies are being completed, before the evidence is synthesised. Certainty of evidence is assessed by adapting tools such as GRADE (Step 6). Results should be reported using adapted versions of reporting tools such as PRISMA (Step 7).
Conclusion: PMA allow for greatly improved use of data, while reducing bias and research waste. PMA could be integral to rapid learning health systems since evidence gaps are identified, and ongoing studies are initiated, tracked and meta-analysed as soon as their results become available, which can then inform policy and practice. Adaptive trial methodology can be utilised to adapt ongoing PMA to emerging evidence. With rising trial registration compliance and new technical advances in machine learning and data processing, we see new horizons for PMA. This step-by-step guidance will improve the understanding of PMA in the research community and enable more researchers to conduct successful PMA.
Patient or healthcare consumer involvement: We will invite healthcare consumers to comment on this research project, to increase its accessibility from their perspective.
Objective: To develop step-by-step guidance on how to conduct PMA.
Methods: We, the Cochrane PMA Methods Group, developed step-by-step guidance based on 1) a scoping review of methodology papers, 2) a scoping review of existing PMA 3) expert opinions 4) experiences with previous PMA. We illustrate each step with a recent case study.
Results: We describe seven steps for PMA (Figure). First, a protocol needs to be developed, including details on collaboration policies (Step 1). Next, a systematic search for planned/ongoing studies should be conducted, including a search of trial registries, medical databases and contacting stakeholders (Step 2) and eligible studies need to be identified for inclusion (Step 3). Importantly, only studies for which the results are not known can be included in a PMA. These studies are then invited to form a collaboration (Step 4), ideally including a steering and data analysis committee, with representatives from each study. Next, core outcomes, common intervention features and a statistical analysis plan are agreed upon within the collaboration (Step 5). This can be particularly useful for rare but important outcomes such as adverse side effects, that individual studies would not have enough power to test statistically. There is usually a waiting period while all studies are being completed, before the evidence is synthesised. Certainty of evidence is assessed by adapting tools such as GRADE (Step 6). Results should be reported using adapted versions of reporting tools such as PRISMA (Step 7).
Conclusion: PMA allow for greatly improved use of data, while reducing bias and research waste. PMA could be integral to rapid learning health systems since evidence gaps are identified, and ongoing studies are initiated, tracked and meta-analysed as soon as their results become available, which can then inform policy and practice. Adaptive trial methodology can be utilised to adapt ongoing PMA to emerging evidence. With rising trial registration compliance and new technical advances in machine learning and data processing, we see new horizons for PMA. This step-by-step guidance will improve the understanding of PMA in the research community and enable more researchers to conduct successful PMA.
Patient or healthcare consumer involvement: We will invite healthcare consumers to comment on this research project, to increase its accessibility from their perspective.