The Meta-Analysis Workflow: 10 Stages from Question to Submission, Who Does What and How Long Each Takes

• A meta-analysis runs through ten stages: question, protocol, search, screening, extraction, risk of bias, synthesis, heterogeneity and sensitivity analyses, GRADE and reporting.
• Screening, extraction and risk of bias are done by two independent reviewers; the statistician owns synthesis.
• Plan for about 22–45 weeks of work, and more than a year from registration to publication.

A meta-analysis is the statistical end point of a systematic review, and the meta-analysis process runs from a precise question to a reporting-guideline-compliant manuscript. It typically involves a review team of several people and often takes more than a year: an analysis of PROSPERO-registered reviews found a mean of 67.3 weeks from registration to publication3.

This flowchart breaks the process into ten stages, names who is responsible for each one and gives indicative durations, so teams can plan resources and deadlines realistically12.

The flowchart groups the ten stages into four phases: planning (stages 1–2), evidence (3–6), analysis (7–9) and reporting (10). Each box names the accountable owner and the indicative time for the stage.

Durations are Pubrica planning estimates for a review of moderate size, not published benchmarks; they depend on the number of records, team capacity and topic complexity. Run in sequence, the stages total about 22–45 weeks (5–10 months) of work. Journal peer review and revisions add further time, which is why registration to publication averaged 67.3 weeks in PROSPERO-registered reviews3.

Define the question with the PICO question framework (population, intervention, comparator, outcomes) and turn it into explicit eligibility criteria1. Write a protocol that pre-specifies outcomes, effect measures, the synthesis model and planned subgroup and sensitivity analyses, and register it prospectively, for example through PROSPERO registration, before screening begins to reduce the risk of selective reporting4.

These stages carry most of the workload. Screening, data extraction and risk-of-bias assessment should be done independently by two reviewers, and the search strategy should be peer reviewed1:

• Search strategy: an information specialist searches several bibliographic databases plus trial registers and grey literature, and records full search strings for reporting.
• Study screening: screen titles and abstracts, then full texts, independently and in duplicate, resolving disagreements by discussion or a third reviewer.
• Data extraction: pilot the extraction form and extract outcome data, study characteristics and funding in duplicate.
• Risk of bias: use RoB 2 for randomised trials and ROBINS-I for non-randomised studies of interventions; our risk of bias tool decision tree shows which applies.

The statistician pools results with the pre-specified effect measure and model, produces forest plots, quantifies heterogeneity and runs planned subgroup and sensitivity analyses, plus small-study-effect analyses where at least 10 studies are available1. The methods to follow are set out in our Cochrane Handbook Chapter 10 and MECIR guide. The review team then rates the certainty of evidence for each key outcome with GRADE, considering risk of bias, inconsistency, indirectness, imprecision and publication bias, and summarises it in a summary of findings table1.

Write the manuscript to PRISMA 2020, including the flow diagram and the completed checklist, and state any deviations from the registered protocol2. Rerun the searches within 12 months before publication (MECIR C37, mandatory)5, and confirm that data, code and extraction forms are available as the target journal requires.

A typical review team combines several roles, and one person may hold more than one; reviews in the PROSPERO analysis had a mean of five authors13.

A well-run meta-analysis process is a managed project: a registered protocol, duplicate screening and extraction, pre-specified data synthesis, GRADE assessment and PRISMA reporting, each with a named owner and a realistic deadline. Planning for months rather than weeks protects quality and makes the final submission defensible. For common technical questions, see our meta-analysis Q&A; for support at any stage, see meta-analysis services.