How to Write a Systematic Review: A Step-by-Step Guide for 2026

A systematic review is a structured synthesis of all available evidence on a defined research question, conducted using a pre-specified, reproducible methodology. It is not a literature review. It is not a narrative summary of papers you found interesting. It is a scientific investigation with its own study design, its own reporting standards, and its own peer review criteria. Most research teams underestimate the specific combination of skills, personnel, and time it requires.

This guide covers the complete 10-step workflow from question formulation through to a PRISMA 2020-compliant manuscript, updated for 2026 with the Cochrane Handbook version 6.5 (August 2024), the PROSPERO February 2025 platform update, ROBINS-I V2 (November 2025 revision), and the January 2024 RevMan update that changed the default random-effects estimator.

Ten steps, each with a specific methodology standard, at a mean of 1,139 hours of direct work. Most teams use ScribeLab Writer for the stages where specialist skills are hardest to source internally: search strategy, meta-analysis, GRADE.

Quick Answer:

A systematic review follows 10 stages: (1) formulate a PICO research question, (2) write and register a protocol on PROSPERO, (3) develop a PRESS-standard search strategy, (4) search the databases, (5) screen records in duplicate, (6) extract data in duplicate, (7) assess risk of bias using the correct tool for each study design (RoB 2 for RCTs, ROBINS-I V2 for non-randomized studies), (8) synthesize the evidence, (9) apply GRADE certainty ratings, and (10) write a PRISMA 2020-compliant manuscript. Each stage has a specific methodology standard. Failure to meet the standard at any stage compromises the entire review's eligibility for peer-reviewed publication at a Tier 1 journal.

What Makes a Systematic Review Different From Other Literature Reviews

The term "systematic review" is sometimes used loosely to describe any structured search and summary of published research. In the methodological sense, a systematic review has four characteristics that distinguish it from a narrative or scoping review.

First, the question and eligibility criteria are defined prospectively, before the search begins, and registered on a public register such as PROSPERO. This prevents the research team from adjusting the question or criteria based on what the search retrieves.

Second, the search is comprehensive and reproducible. The search must be documented in enough detail for another team to replicate it and retrieve the same record set. Every database, every search string, and every date limit must be recorded.

Third, screening, data extraction, and risk-of-bias assessment are conducted by at least two independent reviewers, with disagreements resolved through discussion or third-reviewer adjudication. This requirement is specified in Cochrane MECIR Standard M2 and is the reason a systematic review requires a minimum team of two people for its most labor-intensive stages.

Fourth, the findings are synthesized using a pre-specified method, whether narrative, meta-analytic, or qualitative, and the certainty of the evidence is graded using the GRADE framework.

The 10-Step Overview

Table 1: Systematic Review 10-Step Workflow With Tools and Standards (2026)

Step 1: Formulate the Research Question Using PICO

The PICO framework structures the research question into four components: Population (who are the participants?), Intervention (what is the exposure or treatment?), Comparator (what is it compared against?), and Outcome (what is being measured?). For systematic reviews of interventions, PICO is the standard framework. Reviews of diagnostic accuracy use PICO-D or PICO with an index test replacing the intervention. Reviews of prognosis use PICOTS, adding a time component.

A well-formed PICO question drives every subsequent decision in the review: the eligibility criteria, the database search terms, the data extraction form, and the synthesis approach. Vague or broad PICO components lead to oversized retrieval sets, poorly defined eligibility criteria, and a synthesis that cannot answer a specific question.

Common errors at this stage include population definitions that mix pediatric and adult patients without specifying which. Intervention definitions that include too many variants without pre-specifying subgroup analyses are also frequent. Outcome definitions that change between the protocol and the final manuscript will draw scrutiny at peer review.

Step 2: Write and Register a Protocol on PROSPERO

A systematic review protocol specifies the research question, eligibility criteria, search strategy, screening process, data extraction plan, risk-of-bias approach, and synthesis method before any of these steps are carried out. The protocol is registered on PROSPERO before the database search begins.

Under PROSPERO's February 2025 update, every named co-author must individually approve the registration by email before the record publishes. This approval step should be coordinated before the protocol is submitted, as the record cannot go live until all approvals are received. Researchers who begin their database search before PROSPERO registration is complete will not be able to describe their review as prospectively registered.

PROSPERO now uses a dropdown menu for risk-of-bias tools drawn from the LATITUDES taxonomy. Select RoB 2 for RCTs, ROBINS-I V2 for non-randomized intervention studies, and QUADAS-2 for diagnostic accuracy studies. The protocol should also specify the planned synthesis method (narrative synthesis or meta-analysis) and the heterogeneity statistics that will be reported.

Step 3: Develop the Search Strategy

A PRESS-standard search strategy uses three components: controlled vocabulary terms, free-text synonyms for all major concepts, and Boolean operators connecting them. Controlled vocabulary includes MeSH for MEDLINE, Emtree for Embase, and CINAHL headings for CINAHL.

The strategy must be adapted for every database searched, because each database uses a different controlled vocabulary and different field coding. A MEDLINE search cannot be directly copied into CINAHL. The PRISMA 2020 reporting guidelines (Page et al., BMJ, 2021;372:n71) now require that full search strategies for all databases be reported in the manuscript, not just for the primary database.

Peer review of the search strategy by a second information specialist before the searches are run is recommended practice. The PRESS guideline specifies six review elements: translation of the question, Boolean and proximity operators, subject headings, text-word searching, spelling variants and phrase searching, and limits and filters.

Step 4: Search the Databases

The database scope for a clinical systematic review should include, at a minimum, MEDLINE (via PubMed or Ovid), Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL). Most clinical intervention reviews also search CINAHL for nursing and allied health literature and PsycINFO for psychological outcomes. One or more regional or gray literature databases are added depending on the topic.

All searches must be date-limited in a consistent way and documented with the date of the search. Gray literature sources, including clinical trial registries (ClinicalTrials.gov, WHO ICTRP), conference proceedings, and regulatory databases, should be searched to minimize publication bias.

After the searches are run across all databases, the records are deduplicated before screening begins. Deduplication removes records retrieved from multiple databases. Covidence and Rayyan both have built-in deduplication functions, though manual review of deduplicated records is recommended to ensure records are not erroneously merged.

Step 5: Screen Records in Duplicate

Cochrane MECIR Standard M2 requires at least two independent reviewers to determine eligibility at both the title-abstract and full-text stages. Screening in duplicate means that both reviewers assess the same records independently before comparing their decisions. Disagreements are resolved by discussion between the two reviewers, and by third-reviewer adjudication when discussion does not resolve the disagreement.

Inter-rater agreement is measured as Cohen's kappa and should be reported in the manuscript's methods section. A kappa above 0.6 indicates substantial agreement and supports the reliability of the screening process.

AI-assisted screening tools such as ASReview use active machine learning to prioritize likely-relevant records for human review. These tools reduce the manual screening workload by 64 to 92 percent while maintaining 95 percent recall of relevant studies (Ferdinands et al., Systematic Reviews, 2023). AI tools can be used to prioritize the order of screening, but do not replace the requirement for two human reviewers to independently assess each record.

The PRISMA 2020 flow diagram records the number of records identified, screened, and excluded at title-abstract with reasons, assessed at full text with reasons, and included in the final review.

Step 6: Extract Data in Duplicate

Data extraction uses a pre-piloted form that specifies which variables are extracted from each included study. The form is piloted on three to five included studies before full extraction begins. Piloting identifies ambiguous fields and any data not adequately captured by the current form.

Extraction is performed independently by two reviewers. The two extraction forms are then compared and reconciled. Disagreements are resolved by discussion or third-reviewer adjudication. This independent extraction process is specified in Cochrane MECIR Standard M5.

The extraction form typically includes study design and setting, population characteristics, intervention details, comparator details, outcome measures and time points, effect sizes and confidence intervals, and funding source. When data are not reported in a format suitable for synthesis, the study authors should be contacted to request the needed data.

Step 7: Assess Risk of Bias

Risk-of-bias assessment identifies the methodological limitations in the included studies that might systematically affect the estimated effect. The tool used must match the study design.

RoB 2 for randomized controlled trials. RoB 2 (Sterne et al., BMJ, 2019;366:l4898) is the Cochrane-recommended tool for RCTs. It assesses five domains: bias from the randomization process, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome, and bias in selection of the reported result. A key requirement is that RoB 2 is applied per outcome, not per study. A review with three outcomes and 20 included RCTs requires 60 separate RoB 2 assessments.

ROBINS-I V2 for non-randomized intervention studies. ROBINS-I V2, released in November 2024 and revised in November 2025, is the Cochrane-recommended tool for non-randomized studies of interventions in cohort or follow-up study designs. V2 introduced algorithm-based domain-level judgment mapping, explicit handling of immortal-time bias and prevalent-user bias, and expanded the missing-data domain. Use ROBINS-I V2, not the original 2016 version, for any review registered or submitted in 2026.

QUADAS-2 for diagnostic accuracy studies. QUADAS-2 assesses patient selection, index test, reference standard, and flow and timing domains.

Risk-of-bias results are presented in a summary table and, for RoB 2, as a traffic-light plot generated using the robvis visualization tool.

Step 8: Synthesize the Evidence

The synthesis approach should be pre-specified in the protocol. If a meta-analysis is planned, it is conducted when at least two studies with comparable populations, interventions, comparators, and outcome measures are available, and their data can be combined.

Meta-analysis. The January 2024 update to RevMan replaced the DerSimonian-Laird random-effects estimator with REML (restricted maximum likelihood) as the default tau-squared estimator and added the HKSJ confidence interval method and prediction intervals as standard outputs. These updates align RevMan with the Cochrane Handbook v6.5 recommendations. A meta-analysis in 2026 that still uses DerSimonian-Laird without a documented justification may be flagged by peer reviewers at methods-focused journals.

Heterogeneity is assessed using I-squared (the proportion of variability due to between-study variation), tau-squared (the absolute between-study variance), and the 95 percent prediction interval. The prediction interval estimates the range within which the true effect will fall in a new study. It is now a standard output for all random-effects meta-analyses under the Cochrane Handbook v6.5.

Narrative synthesis. When quantitative pooling is not appropriate because of clinical or methodological heterogeneity, a narrative synthesis using a pre-specified framework (such as the Evidence Synthesis Conceptual Framework, Popay et al., 2006) describes the direction and magnitude of effects across included studies without numerical pooling.

Step 9: Apply GRADE

GRADE (Grading of Recommendations Assessment, Development and Evaluation) rates the certainty of evidence for each outcome as High, Moderate, Low, or Very Low. Certainty starts at High for randomized controlled trials and Very Low for observational studies and is downgraded based on five criteria: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Observational studies can be upgraded based on large effect sizes, dose-response relationships, or the absence of plausible confounding.

GRADE results go into a Summary of Findings table, one per outcome. Each table shows the number of studies and participants, the effect estimate with confidence interval, the GRADE certainty rating, and a brief explanation. Most clinical journals and all Cochrane reviews require a Summary of Findings table.

GRADEpro GDT software (gradepro.org) is used to construct Summary of Findings tables and is available free for non-commercial use.

Step 10: Write the Manuscript to PRISMA 2020

The PRISMA 2020 statement (Page et al., BMJ, 2021;372:n71) provides a 27-item checklist for reporting systematic reviews and an updated flow diagram. Every PRISMA 2020 item must be addressed in the manuscript. The completed checklist is submitted alongside the manuscript as required documentation.

The PRISMA 2020 flow diagram tracks records from initial retrieval through to final inclusion, with reasons for exclusion at each stage. PRISMA 2020 requires reasons for exclusion at full-text screening to be listed by category, not just as a total number. The number excluded at the title-abstract stage must also be reported separately from full-text exclusions.

PRISMA 2020 also requires that the full search strategy for every database searched be reported, either in the main text or as a supplementary appendix. Reporting the search strategy for only one database (commonly seen in older reviews) does not meet PRISMA 2020 reporting requirements.

Common Errors That Lead to Rejection

Not registering before searching. A systematic review that was searched before the protocol was registered cannot be described as prospective. Most Tier 1 journals check the PROSPERO registration date against the recorded search date. A post-hoc registration is not the same as a prospective one, and peer reviewers will note the discrepancy.

Single-reviewer screening. Single-reviewer screening is not acceptable under Cochrane MECIR standards. PRISMA 2020 item 8 requires disclosure of the number of reviewers and their independence. A manuscript that discloses single-reviewer screening signals a methodological limitation immediately.

Using an outdated risk-of-bias tool. Applying NOS to intervention studies, applying ROBINS-I V1 to non-randomized studies in 2026, or applying RoB 2 at the study level rather than per outcome are all errors that peer reviewers at methods-focused journals will flag.

Reporting I-squared without prediction intervals. Under Cochrane Handbook v6.5, prediction intervals are a standard output of random-effects meta-analysis. A manuscript that reports only I-squared, or that reports I-squared without tau-squared, will receive reviewer comments requesting these statistics.

Incomplete PRISMA 2020 reporting. The most common PRISMA 2020 reporting errors are: no full search strategy for all databases, no reasons listed for full-text exclusions, no GRADE Summary of Findings table, and a flow diagram with irreconcilable numbers.

How Long Does Each Stage Take?

Table 2: Systematic Review Stage Duration Estimates (2026)

Duration estimates for academic teams are based on Borah et al. (BMJ Open, 2017) and Allen and Olkin (JAMA, 1999) data and represent typical ranges, not minimum times. Professional support timelines are indicative. Actual timelines depend on project scope, retrieval volume, and the number of included studies.

Table 2: Systematic Review Stage Duration Estimates (Based on Borah et al., 2017, and Allen and Olkin, 1999)

The mean time from PROSPERO registration to publication is 67.3 weeks, with a mean team of five researchers (Borah et al., BMJ Open, 2017). Allen and Olkin (JAMA, 1999) estimated 1,139 total hours. Searching and retrieval consumed 588 hours (52 percent). Statistical analysis consumed 144 hours (13 percent). Report writing took 206 hours (18 percent). Administration accounted for 201 hours (17 percent). These figures represent professional systematic review firm data and may underestimate the time for academic teams conducting their first review.

Frequently Asked Questions

How many databases should a systematic review search?

At a minimum, a clinical systematic review should search MEDLINE, Embase, and CENTRAL. Most reviews of clinical interventions also search CINAHL and at least one gray literature source. Scoping reviews and reviews in non-clinical disciplines may require different database combinations. The number of databases is determined by the research question and the target journal's requirements. Cochrane reviews are expected to search extensively; pragmatic reviews for clinical practice guidelines may use a more limited but targeted scope.

Can a systematic review be conducted by one person?

A solo systematic review cannot meet Cochrane MECIR Standards M2 and M5, which require at least two independent reviewers for screening and data extraction. A solo researcher can formulate the question, register the protocol, develop the search strategy, and coordinate the review. For the screening and extraction stages, a trained second reviewer must be involved. This can be a co-author, a supervisor, or a professional second-reviewer service.

Do I need to be a statistician to conduct a systematic review?

No, but the statistical stages of a systematic review require specialist knowledge. A researcher without formal training in random-effects meta-analysis, model selection, REML estimation, and GRADE can manage most stages of the review independently. The meta-analysis and GRADE stages benefit significantly from biostatistical expertise, either through a co-investigator with relevant training or through professional statistical support.

What is the difference between a systematic review protocol and the final review?

The protocol is the pre-registered plan specifying the research question, eligibility criteria, search strategy, screening process, data extraction form, risk-of-bias approach, and synthesis method. It is registered on PROSPERO before searching begins. The final review is the completed study, following the plan set in the protocol. Deviations from the protocol (changes to eligibility criteria, additional outcomes, different synthesis approaches) must be disclosed in the final manuscript's methods section.

What is the PRESS standard for search strategies?

PRESS (Peer Review of Electronic Search Strategies) is a guideline for the peer review of systematic review search strategies by a second information specialist before searches are run. Developed by McGowan et al. (Journal of Clinical Epidemiology, 2016), PRESS covers six review elements: PICO translation, Boolean and proximity operators, subject headings, text-word searching, spelling variants and phrase searching, and limits and filters. A PRESS-reviewed search strategy is more likely to retrieve a comprehensive, reproducible record set.

What happens if I find a similar review has already been published?

A recently published systematic review on the same PICO does not necessarily mean your review cannot proceed, but it requires justification. Your review must add something that the existing review does not. Valid additions include a different date range, additional databases, a different comparator, a different population subgroup, or an updated evidence base covering studies published since the prior review's search date. The justification for proceeding despite a similar existing review should be stated in the protocol and in the manuscript introduction.

Starting Your Systematic Review on the Right Methodological Foundation

The ten steps in this guide represent the current standard for a systematic review that will survive peer review at a Tier 1 journal. Each step has a specific methodology requirement, and missing or shortcutting any one of them creates a gap that peer reviewers will identify. The most preventable reasons for rejection are single-reviewer screening, outdated risk-of-bias tools, missing GRADE tables, and incomplete PRISMA 2020 reporting.

ScribeLab Writer's systematic review team, led by credentialed researchers with published systematic reviews in the biomedical literature, works with research teams at any stage of this workflow. From a full protocol-to-manuscript engagement to targeted support at the specific step where your team needs it, submit your project details, and a PhD methodologist will respond within 2-4 hours.