Steps in a Systematic Review: The Full Sequence

The core steps in a systematic review are: frame a focused question and register a protocol, run a reproducible search, screen studies in duplicate against eligibility criteria, extract the data, assess each study’s risk of bias, synthesise the results, and report everything against PRISMA. Each step exists to remove a specific source of bias, so the order is fixed and no stage can be skipped without weakening the whole review.

Why the sequence is locked before you start

What makes these steps systematic is that they are decided in advance, not improvised as results arrive. Fixing the method up front is the safeguard against cherry-picking: you commit to the question, the databases, the eligibility rules, and the planned analysis before seeing a single result. That is why the very first step produces a written, often registered, protocol, and why every later step simply executes what the protocol promised. Treat the list below as a contract you sign before the evidence can talk back.

Step 1: Frame the question and write a protocol

Begin with a precise, answerable question, usually structured with a framework such as PICO. Then fix the methods in a written protocol that names the databases, the inclusion and exclusion criteria, the outcomes, and the planned synthesis. For health questions the protocol is registered on PROSPERO so your methods are timestamped publicly.

Step 2: Build and run a reproducible search

Design a comprehensive search across multiple databases, combining subject headings and free-text terms with Boolean operators. The goal is to find every eligible study, not a convenient sample, and the search must be documented in enough detail that someone else could rerun it and get the same records. Many teams have the strategy peer-reviewed using the PRESS checklist before running it.

Step 3: Screen studies in duplicate

After removing duplicate records, two reviewers independently screen titles and abstracts, then full texts, against the eligibility criteria. Working in duplicate catches studies a single reviewer would miss, and every full-text exclusion is logged with a reason for the PRISMA flow diagram.

Step 4: Extract the data

Pull the same fields from every included study using a piloted data extraction form, ideally in duplicate. Consistent extraction is what allows the results to be compared and, where appropriate, pooled. Piloting the form on a few studies first prevents costly rework later.

Step 5: Assess risk of bias

Trained assessors rate each study’s risk of bias with the tool that fits its design, such as RoB 2 for randomised trials or ROBINS-I for non-randomised studies. These judgements feed directly into how much weight the synthesis can carry.

Step 6: Synthesise the evidence

If the studies are similar enough, pool their results in a meta-analysis to produce a single weighted estimate, with heterogeneity quantified. When pooling is not appropriate, combine the evidence in a structured narrative synthesis. Either way, the overall certainty of evidence is rated with GRADE.

Step 7: Report against PRISMA

Write the review up against the PRISMA 2020 checklist, presenting the flow diagram, the full search, the risk-of-bias judgements, and the synthesis so a reader can audit every decision. Complete reporting is what turns a careful process into a trustworthy, citable result. For more on how these steps fit into a full timeline, see how long a systematic review takes.

What each step hands to the next

The steps are not independent tasks ticked off a list; each produces a documented output that becomes the input to the one after it. Seeing the chain of hand-offs is what makes the fixed order feel inevitable rather than arbitrary:

1. The protocol hands the search a set of concepts and the eligibility rules that the query must operationalise.
2. The search hands screening a de-duplicated set of records, with the line counts that will later populate the flow diagram.
3. Screening hands extraction a defined included set, plus a logged reason for every full-text exclusion.
4. Extraction hands the synthesis a clean dataset of effect estimates and their variances, in a consistent shape across studies.
5. Risk of bias hands the synthesis the weight each study can credibly carry, feeding any sensitivity analysis that removes high-risk studies.
6. The synthesis hands the report a pooled estimate or structured narrative, and a GRADE certainty rating for each outcome.

Because each hand-off is a documented artefact, a reviewer can audit the join between any two steps, which is precisely what a traditional literature review cannot offer.

Common mistakes at each step

Most reviews that struggle in peer review fail at a predictable point. At the question stage, the error is a scope so broad the search returns tens of thousands of records, which a sharper research question would have avoided. At the search stage it is relying on a single database, which cannot credibly claim to have found all the evidence. At screening it is a single tired reviewer working alone, reintroducing the selection bias the method exists to remove, which is why teams agree how many reviewers screen up front.

The later steps fail more quietly. At extraction, pulling figures inconsistently across studies makes pooling impossible without rework. At appraisal, treating a high risk-of-bias study as if it were sound lets a flawed trial dominate the estimate. At synthesis, the classic error is forcing a heterogeneous set of studies into one number that looks precise but means nothing. Pre-specifying the method for each step in the protocol is what closes the door on all of them at once.

A final point ties the sequence together: the steps are easiest when the earliest decisions are the most carefully made. A sharp question shrinks the search; a clean search shrinks screening; tight criteria shrink conflicts; a piloted extraction process shrinks rework at synthesis. Effort invested at the front of the sequence pays back several times over at the back, which is the opposite of how most teams instinctively allocate their time. If you want the same lifecycle expanded into a full methods walkthrough, our guide to writing a systematic review end to end covers each step in more depth.