Define a sharp question, pre-register a protocol, search widely, screen in pairs, appraise bias, synthesize, and report with PRISMA and GRADE.
Medical readers want clear answers that pull together the best available studies. A well-run scientific review does exactly that, turning many papers into one coherent take you can trust.
This guide lays out each step from idea to submission. It suits systematic reviews, scoping reviews, and meta-analyses across clinical and public health topics. You’ll see repeatable methods, common pitfalls, and simple habits that save time.
Pick The Right Review Type
| Type | What You Get | When To Use |
|---|---|---|
| Systematic Review | Structured question, exhaustive search, dual screening, bias appraisal, synthesis. | When you need a complete, reproducible answer. |
| Meta-analysis | Quantitative pooling of effect sizes from comparable studies. | When studies measure the same outcome in compatible ways. |
| Scoping Review | Maps what exists, methods, and gaps without firm effect estimates. | When a topic is broad, emerging, or heterogeneous. |
| Rapid Review | Streamlined version of a systematic review with time-saving concessions. | When a decision deadline won’t allow a full process. |
| Narrative Review | Expert overview with selective searching and flexible structure. | When context and interpretation matter more than completeness. |
| Umbrella Review | Review of reviews across related questions or outcomes. | When many systematic reviews already exist. |
Doing A Scientific Review In Medicine: Start Strong
Start by shaping a focused question. PICO (Population, Intervention, Comparator, Outcome) fits trials; PEO (Population, Exposure, Outcome) often fits observational topics. Write the primary outcome, time frame, and setting in plain words.
Draft strict eligibility rules before any searching. Define designs you’ll include, participant traits, exposure or intervention details, comparators, outcomes, and minimum follow-up.
Register a protocol so others can see your plan and to lock in decisions. Use the international registry PROSPERO for most health topics (PROSPERO). Upload the full protocol or link to an open repository.
Assign roles. At minimum you need two independent reviewers for screening and bias appraisal, a content lead, and a method lead. Add an information specialist if you can.
Sketch milestones with dates: protocol, search runs, calibration, screening, extraction, bias, synthesis, manuscript. Timeboxing keeps momentum.
From day one, plan your report with PRISMA 2020. Keep the checklist open while you work; it doubles as a build sheet for your manuscript and flow diagram.
Search Strategy That Finds The Right Studies
List databases that match your field: MEDLINE via PubMed or Ovid, Embase, CENTRAL, Web of Science, CINAHL, PsycINFO, and regional indexes when relevant. Add trial registries and preprints where policy allows.
Turn your question into structured terms. Combine controlled vocabulary (e.g., MeSH) with text words. Use Boolean logic, truncation, and proximity operators. Pilot the query and check that known test articles are returned.
Capture grey literature: trial registries, theses, conference abstracts, regulatory reports, and preprint servers. Note any exclusions you must make for peer-review only analyses.
Export all results to a reference manager or screening platform. De-duplicate thoroughly and record the exact counts and dates per source.
Rerun searches before final analysis to catch recent studies. Report all run dates, interfaces, and full search strings in an appendix.
Invite an information specialist to peer-review the strategy when possible. Use a short pilot to calibrate sensitivity and precision.
Screening: Two Sets Of Eyes On Every Record
Set up dual, independent screening at title/abstract stage, then again at full text. Resolve conflicts by consensus or a third reviewer.
Train with a brief calibration round. Screen 50–100 records together, compare decisions, and refine rules. This prevents avoidable drift later.
Log exclusion reasons at full text using standard categories. Keep the wording short and consistent for the PRISMA flow diagram.
Keep a living log of contact with authors for missing data or clarifications. File responses in your project folder with dates.
Document the process well enough that another team could replay it. That level of clarity raises trust and speeds peer review.
Data Extraction That Reduces Errors
Build a pilot extraction form that mirrors your question and outcomes. Include setting, design, arms, sample size, follow-up, effect metrics, and funding.
Extract in duplicate for a sample first, then assign single extraction with verification on the remainder if the sample error rate is low.
Pre-define how you’ll handle multiple time points, multi-arm trials, cluster designs, cross-overs, and adjusted vs crude estimates.
Track unit conversions and imputed values in a separate sheet with formulas shown. Keep raw numbers alongside any calculated ones.
Spot duplicate publications by trial registry numbers, author groups, and identical baseline tables. Merge data carefully to avoid double counting.
Risk Of Bias And Study Quality
Pick tools that match each design. Common picks include RoB 2 for randomized trials, ROBINS-I for non-randomized studies of interventions, QUADAS-2 for diagnostic accuracy, and QUIPS for prognostic studies.
Judge bias per outcome where the tool requires it. Work in pairs, blind to each other at first, then reach agreement with notes that cite pages or figures.
Present bias visually with domain-level plots and traffic light summaries. Keep the narrative short and tied to your synthesis plan.
Bias Appraisal Tools That Fit The Design
| Study Type | Tool | Notes |
|---|---|---|
| Randomized Trials | RoB 2 | Outcome-level judgements across randomization, deviations, missing data, measurement, and reporting. |
| Non-randomized Interventions | ROBINS-I | Pre-intervention confounding, selection, classification; at- and post-intervention domains. |
| Diagnostic Accuracy | QUADAS-2 | Patient selection, index test, reference standard, flow and timing. |
| Prognostic Studies | QUIPS | Study participation, attrition, prognostic factor measurement, outcome measurement, confounding. |
| Systematic Reviews | AMSTAR 2 | Protocol, search, selection, extraction, bias appraisal, meta-analysis methods, publication bias. |
Synthesis And Meta-analysis
Decide upfront when you will pool and when you will narrate. Pool only like with like: comparable designs, measures, and time points.
Pick effect metrics that fit your outcomes. Risk ratio or odds ratio for binary data; mean difference or standardized mean difference for continuous data; hazard ratio for time-to-event.
Choose a model that matches expected variability. A random-effects model handles between-study spread; a fixed-effect model assumes one common effect.
Quantify inconsistency with I² and tau². Probe obvious outliers with planned subgroup or sensitivity checks rather than ad-hoc fishing.
Guard against small-study effects with funnel plots and tests where counts allow. Weigh real-world causes before blaming bias alone.
When pooling is off the table, write a tight narrative that pairs effect directions with study quality and sample size.
Assess Certainty With GRADE
Rate certainty by outcome, not by study. Randomized trials start higher; non-randomized designs start lower. Then rate down for risk of bias, inconsistency, indirectness, imprecision, or publication bias. Rate up when large effects, dose-response, or bias that would shrink the effect are present.
Create a Summary-of-Findings table with absolute and relative effects, baseline risk, and certainty labels (High, Moderate, Low, Very low). State plain-language take-aways beside the numbers.
Keep value judgements transparent. Write one-line justifications for each rating move so readers can follow your path from evidence to certainty.
Report And Share Your Work
Write the manuscript in the same order readers search: title and abstract that match the question and design; methods that mirror the protocol; results that flow from the PRISMA diagram; then a balanced take on strengths and limits.
Cite the Cochrane Handbook when you use standard methods. That quick reference helps reviewers see the logic behind each step.
Include the full search strategies, extraction forms, bias decisions, and data files in a public repository. That transparency speeds reuse and updates.
Share plain-language summaries and visual abstracts for clinicians and patients. Clear, jargon-light wording widens the reach of your findings.
How To Do A Medical Scientific Review: Final Checks
Reproduce your own flow. Start with the raw exports and see if a teammate can reach the same included set using your notes alone.
Spot-check calculations against the original papers. Confirm effect directions, units, and any conversions.
Run a last search update shortly before submission. Record the date and whether it changed the pooled estimate or the narrative.
Scan conflicts of interest, funding, and author affiliations for both included studies and your own team. State any ties plainly.
Set an update signal now. If new trials or approvals appear, be ready to refresh the review on a schedule or as a living version.
Ethics, Data Care, And Fair Citation
Even without patient contact, ethics still matter. Avoid selective reporting, cite preprints with care, and flag any retractions you encounter.
Store files in a structured workspace with write-protected raw data. Keep a change log for every sheet and script.
Credit prior reviews that shaped your plan. Where you reuse text such as standard methods, paraphrase and cite to avoid overlap concerns.
Ready-To-Use Checklist
- Question framed with PICO/PEO; primary outcome set.
- Protocol registered and publicly accessible.
- Databases listed; full search strings saved.
- Grey literature sources planned and justified.
- Dual screening with calibration completed.
- Exclusion reasons standardized and logged.
- Extraction form piloted; duplicate checks done.
- Bias tools matched to design and applied in pairs.
- Pooling rules written; model choice justified.
- Heterogeneity, small-study effects, and outliers handled by plan.
- GRADE ratings by outcome with one-line reasons.
- PRISMA checklist completed; flow diagram built.
- All materials shared in a public repository.
- Update plan set; triggers defined.
Helpful resources: Register your protocol at PROSPERO; build and report with PRISMA 2020; and lean on methods from the Cochrane Handbook.