How to Conduct a Systematic Review in Biomedicine

Systematic reviews start with a structured question. The most common framework is PICO: Population, Intervention, Comparison, Outcome. For diagnostic accuracy reviews, use PIRD (Population, Index test, Reference standard, Diagnosis). For qualitative questions, PEO (Population, Exposure, Outcome) works better. A well-defined PICO question drives every subsequent decision: which databases to search, what inclusion criteria to apply, and what outcomes to extract.

PROSPERO (International Prospective Register of Systematic Reviews, run by NIHR) is the standard registration platform for health-related systematic reviews. Registration establishes priority, reduces duplicate reviews, and is now required or strongly recommended by most clinical journals (Lancet, BMJ, JAMA, Cochrane). Register before you start searching. After you've screened studies, most journals won't accept late registration. The registration record is public and citable immediately. PROSPERO assigns a registration number (CRD number) you'll include in your manuscript.

A full protocol describes your search strategy, inclusion/exclusion criteria, data extraction plan, and analysis approach before you've seen the results. Protocols reduce bias by preventing outcome switching. You can publish the protocol as a standalone article (JMIR Research Protocols, Systematic Reviews journal, BMJ Open) and cite it in your final review. If you publish the protocol, it becomes part of the peer-reviewed record and strengthens the review's credibility. Use the SPIRIT 2013 checklist when writing a protocol for a clinical trial; for SR protocols, Systematic Reviews journal has its own requirements.

A systematic review requires a reproducible, multi-database search. Work with a medical librarian or information specialist to build the search. The strategy should include controlled vocabulary terms (MeSH for PubMed, Emtree for Embase), free-text synonyms, and appropriate Boolean operators. The full search strategy must be published in the paper (usually as a supplementary table). Peer review of search strategies by a second information specialist using PRESS (Peer Review of Electronic Search Strategies) is increasingly standard at top journals.

Screening requires two independent reviewers for every title/abstract and every full text, with disagreements resolved by a third reviewer or consensus discussion. This is the most labor-intensive part of a systematic review. Deduplicate results first (across databases, the same paper often appears 3–5 times). Software like Covidence or Rayyan makes this manageable by presenting records one at a time and tracking reviewer decisions. Document inter-rater reliability (Cohen's kappa or percentage agreement) and report it. Track exclusion reasons at the full-text stage per PRISMA flow diagram requirements.

Data extraction should use a standardized form developed and piloted before extraction begins. Extract twice independently for included studies, then reconcile. At minimum, extract: study design, population characteristics, intervention/exposure, outcomes, follow-up duration, and funding source. For clinical trials, extract allocation concealment, blinding, and intention-to-treat analysis details. Pilot the extraction form on 3–5 studies before using it on all included papers. Changes to the extraction form after full extraction begins introduce inconsistency.

Risk of bias (RoB) assessment evaluates the methodological quality of each included study, not the quality of reporting. The tool depends on study design: RoB-2 for randomized trials, ROBINS-I for non-randomized studies of interventions, QUADAS-2 for diagnostic accuracy studies, and Newcastle-Ottawa Scale (NOS) for observational studies. Two reviewers assess each study independently. RoB assessment informs both the synthesis and the strength-of-evidence summary (GRADE). Don't confuse risk of bias with quality: a well-reported study can still have high risk of bias.

Not every systematic review should include a meta-analysis. Pool results statistically only when studies are sufficiently similar in population, intervention, and outcome measurement. When pooling, choose the appropriate model: fixed-effects when heterogeneity is low and you assume one true effect size; random-effects when heterogeneity exists or populations vary. Report I² as a measure of heterogeneity. Explore heterogeneity through subgroup analyses and meta-regression. If studies are too heterogeneous to pool, a narrative synthesis is the honest choice. Use the SWiM (Synthesis Without Meta-analysis) reporting guideline.

GRADE (Grading of Recommendations Assessment, Development and Evaluation) rates the overall certainty of evidence for each outcome as High, Moderate, Low, or Very Low. It starts from the study design (RCTs start at High, observational studies at Low) and adjusts up or down based on risk of bias, inconsistency, indirectness, imprecision, and publication bias. GRADE is required for Cochrane reviews and increasingly expected at Lancet, JAMA, BMJ, and NEJM systematic reviews. A summary of findings (SoF) table is the standard way to present GRADE ratings alongside the quantitative findings.

PRISMA 2020 (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) is the required reporting checklist for systematic reviews at virtually every major biomedical journal. The 27-item checklist covers the full review process from title to discussion. Two parts are mandatory at most journals: the PRISMA flow diagram (showing records identified, deduplicated, screened, assessed for eligibility, and included) and the completed PRISMA checklist with page numbers. Use the PRISMA-P extension if you're reporting a protocol. PRISMA-NMA for network meta-analysis, PRISMA-IPD for individual participant data meta-analysis.

Most systematic reviews go to field-specific journals, not general journals. Top-tier SRs with major clinical implications can go to NEJM, Lancet, JAMA, or BMJ. Field-specific reviews typically target journals in their specialty. The Cochrane Database of Systematic Reviews publishes only systematic reviews and is indexed in MEDLINE, making Cochrane reviews highly visible. For rapid/living reviews, check journals that support these formats (BMJ, Cochrane). Most journals require the PROSPERO registration number, completed PRISMA checklist, and data availability statement.