Planning to do a meta-analysis? Decide these ten things first

What decides whether a meta-analysis succeeds is settled before the first search: whether the question is worth answering, whether enough comparable studies exist, who will do the work, how long it will take and what you will do when the data surprise you. Most of the effort, and most of the regret, comes from skipping that stage.

This page is the plan, in order. Each step links to a guide that goes deeper. If you have an area but no topic yet, start at step 1. If you already have a topic, begin at step 2 and check that nobody has done it.

Ten decisions before you start

Topic: choose one that is novel, valid and impactful. How to find a novel, valid and impactful meta-analysis topic gives the three tests, and project ideas for students and residents suits a first review.
Existing reviews: check that nobody has done it, or that what exists is out of date. See how to check whether a systematic review already exists.
Type of review: decide whether you need a meta-analysis at all. See systematic review, meta-analysis, scoping review or umbrella review.
Question: write it as population, intervention, comparator and outcome. See how to write a PICO question.
Feasibility: count the studies that measure the same outcome in a comparable way. See how many studies you need.
Team: decide who screens, who extracts and who does the statistics. See who you need for a meta-analysis.
Time: budget by the number of records you expect, not by weeks. See how long a meta-analysis takes.
Protocol: write down your methods before you search, and register them. See the protocol checklist and how to register on PROSPERO.
Tools: choose the software and decide whether AI will help, and where. See meta-analysis software and AI tools for meta-analysis.
Reporting: plan to follow PRISMA 2020 from the start, including any use of AI. See how to report AI use.

How long to plan for

Plan for months, not weeks. In an analysis of 195 completed systematic reviews registered on PROSPERO, the average time from registration to publication was 67 weeks, and the teams averaged about five authors. A review with a narrow question, a small search and a clear protocol can be much quicker, and one with a broad question can take far longer.

Let SynthGap run the first pass

Enter an area. SynthGap proposes candidate topics, checks each against existing systematic reviews, and lists the primary studies behind it. Start on the Free plan: no card needed, and every run shows every topic it finds.

Warning signs that you should stop and rethink

Fewer than about three comparable studies: there may be nothing to pool.
A recent, good review already answers the question, and you have no clear reason why yours adds anything. An old review may be worth updating instead: see when an old systematic review is worth updating.
The studies measure different things: outcomes, populations or time points that cannot sensibly be combined.
Nobody can be your second reviewer, and you have no plan for the extra checks that requires.
You cannot get the full texts of many of the key papers.
You cannot say who would use the answer or what they would do differently.

Where SynthGap fits in the plan

SynthGap helps with steps 1, 2 and 5. You describe a clinical area; it proposes candidate topics, checks them against existing systematic reviews and lists the primary studies behind each one, so that you can see whether a topic is new and whether enough studies exist. It does not do steps 6 to 10, and its suggestions are AI-generated, so verify them before you register a protocol.

Sources

The figures and rules on this page come from these sources.

Borah R, Brown AW, Capers PL, Kaiser KA. Analysis of the time and workers needed to conduct systematic reviews of medical interventions using data from the PROSPERO registry. BMJ Open 2017;7:e012545.

Questions

How early should I register my protocol?

As early as you can. Before screening starts is best practice, and PROSPERO does not accept a review that has already started data extraction. Registering late weakens the point of registering, which is to fix your methods before you see the results.

Can a student do a meta-analysis alone?

A student can do most of the work, but standards ask for at least two people to make the key decisions independently, so you need a second reviewer for at least part of the screening and the data extraction. The guide on who you need for a meta-analysis explains the options.

What if I only have a few weeks?

Narrow the question, limit the outcomes, or choose a different design, such as a rapid review or a scoping review. A full systematic review with a meta-analysis rarely fits into a few weeks.

Do I need to know the statistics before I start?

You need enough to write the analysis plan, which means choosing an effect measure, a model and the way you will handle heterogeneity. You can learn that from the Cochrane Handbook and get advice from a statistician at the protocol stage, so it should not stop you from starting.

Keep reading

General methodological guidance, not medical advice. SynthGap's suggestions are AI-generated from the published literature, so verify them before registering a protocol.