How to do a systematic review and meta-analysis: a step by step guide
Everything a medical student or doctor needs to go from a research question to a published systematic review and meta-analysis. Start at step 1 if you have never done one.
Written by Muhammad Farhan, Founder and Research Head of InsightMed Research Hub, with 27 peer reviewed publications. Updated November 2026.
In this guide
- Choose a question that has not been answered
- Write and register a protocol
- Build a reproducible search
- Screen studies with two reviewers
- Extract data in duplicate
- Assess risk of bias with the right tool
- Pool the results (meta-analysis)
- Rate the certainty of evidence (GRADE)
- Write it to PRISMA 2020
- Choose a journal, submit and respond to reviewers
Choose a question that has not been answered
Start with a clinical question you care about, then check it has not already been answered. Search PubMed for your topic AND systematic review, and search PROSPERO for registered reviews in progress. If a good review exists from the last two or three years, either change the angle (a new population, outcome or comparator) or pick another topic.
Frame the question with PICO: Population, Intervention, Comparator, Outcome. For prevalence or prognosis questions, use frameworks such as CoCoPop or PICOTS.
Read next: How to choose a meta-analysis topic.
Write and register a protocol
A protocol fixes your methods before you see the data, which protects the review from bias. Follow PRISMA-P 2015: objectives, eligibility criteria, information sources, the search, study selection, data items, risk of bias, synthesis plan and certainty of evidence (GRADE).
Register it on PROSPERO for health reviews, or on OSF if PROSPERO does not accept your topic. Many journals ask for the registration number.
Read next: How to write a systematic review protocol and How to register on PROSPERO.
Build a reproducible search
Build the search in PubMed first. Combine MeSH terms with free text synonyms for each PICO element, use OR within a concept and AND between concepts, then translate the strategy to Embase (Emtree), Cochrane CENTRAL and Scopus or Web of Science. Record the date, database and number of hits for every search, because you will report them in the PRISMA flow diagram.
Use our free search strategy builder and read How to build a PubMed search strategy.
Screen studies with two reviewers
Remove duplicates, then screen titles and abstracts with two independent reviewers who cannot see each other's decisions. Run a calibration round on 50 to 100 records first and agree on how to apply each criterion. Retrieve the full texts of everything that passes, screen again in pairs, and record a reason for every full text exclusion. Resolve conflicts by discussion or a third reviewer.
Read next: How to screen studies in Rayyan and Records, reports and studies in PRISMA 2020.
Extract data in duplicate
Design an extraction form before you start: study characteristics, participants, interventions, comparators, outcomes, follow up and the numbers you need for each effect measure. Two people extract independently and reconcile differences. Contact authors for missing data.
Studies often report medians and interquartile ranges instead of means and standard deviations. Convert them with recognised methods using our free effect size converter and the guide Median and IQR to mean and SD.
Assess risk of bias with the right tool
The tool depends on study design: RoB 2 for randomised trials, ROBINS-I for non randomised studies of interventions, the Newcastle-Ottawa Scale for cohort and case control studies, QUADAS-2 for diagnostic accuracy and PROBAST for prediction models. Two reviewers rate each domain independently and give an overall judgement for each study.
Pool the results (meta-analysis)
Choose the effect measure: risk ratio or odds ratio for binary outcomes, mean difference or standardised mean difference for continuous outcomes, hazard ratio for time to event data. Most clinical meta-analyses use a random effects model. Report heterogeneity with I squared and tau squared, run sensitivity analyses (for example leaving one study out or excluding high risk of bias studies), and assess small study effects with a funnel plot and Egger's test when you have at least 10 studies. RevMan, R (meta and metafor packages) and Stata all work.
If studies are too different to pool, do a structured narrative synthesis instead, following the SWiM guideline.
Rate the certainty of evidence (GRADE)
GRADE rates each outcome as high, moderate, low or very low certainty, based on risk of bias, inconsistency, indirectness, imprecision and publication bias. Present it in a Summary of Findings table. Reviewers increasingly expect it.
Write it to PRISMA 2020
Write the manuscript in PRISMA 2020 order and complete the 27 item checklist. Include the flow diagram (build it with our free PRISMA flow diagram tool), the full search strategy for every database as a supplement, risk of bias figures and forest plots. Your abstract should follow PRISMA for Abstracts.
Choose a journal, submit and respond to reviewers
Pick a journal that publishes systematic reviews in your specialty, is indexed in PubMed or Scopus, and fits your budget for article processing charges. Avoid predatory journals. Write a short cover letter, run a pre-submission check, and expect revisions. Answer reviewers point by point, politely and with exact changes.
Read next: How to choose a journal, How to respond to reviewer comments and get the free pre-submission checklist.
Three ways to get it done
Do it yourself
The DIY Review Kit: nine templates for every step above. AED 99.
Learn it live
The online bootcamp: the full method in 3 to 5 days on Zoom.
Do it with us
Join a mentored cohort project or book 1:1 mentorship.
Frequently asked questions
How long does a systematic review take?
For a well scoped question with a small team working steadily, three to six months from protocol to submission is realistic. Large reviews can take a year or more. Our mentored cohorts aim to reach submission in 4 to 8 weeks with experienced supervision.
Can a medical student do a systematic review?
Yes. Systematic reviews need no patients, ethics approval or lab, so they are one of the most common first research projects for medical students. What students usually lack is method training and supervision, which is why a mentor or a structured template set helps.
What is the difference between a systematic review and a meta-analysis?
A systematic review is the whole process of finding, selecting and appraising every relevant study on a question. A meta-analysis is the statistical step that pools the numerical results of those studies into one estimate. Every meta-analysis should sit inside a systematic review, but not every systematic review includes a meta-analysis.
Which software do I need?
Rayyan or Covidence for screening, Excel or Google Sheets for extraction, and RevMan, R or Stata for the meta-analysis. Rayyan has a free plan and R is free.
Do I need to register my review on PROSPERO?
It is strongly recommended and many journals require it. Registration shows your methods were set in advance and reduces duplication with other teams.