How To Do A Literature Review In Medicine | Step-By-Step

What A Medical Literature Review Delivers

Medical decisions rest on published evidence. A well run review pulls together data from trials, observational reports, and guidelines to answer one sharp question. You set clear rules, search widely, filter out noise, and present what the studies show and what they do not. Readers get a single, dependable point of reference for a topic and a map of gaps worth tackling next.

Choosing The Right Review Type

Not every project needs a full systematic review. Pick the design that matches your aim, time, and team skills. The table below compares common formats so you can pick the right lane early and avoid midstream resets.

Doing A Literature Review In Medicine: Stepwise Plan

The steps below fit clinical, public health, and lab topics. Adjust the depth to match your deadline and the stakes of the decision you want to inform.

Frame The Question With PICO Or Variants

Turn a vague idea into one sentence you can search. Use PICO for interventions, PECO for exposure, or PICo for qualitative work. Write out the population, the intervention or exposure, the comparator, and the outcomes that matter. Add setting if that changes the search. Keep one primary outcome to steer scope and analysis.

PICO Variants

PICO, PECO, and PICo are the common setups also.

Pre-Register Your Plan (If Systematic)

For systematic work, write a protocol with your objectives, eligibility criteria, outcomes, and planned synthesis. Register it on PROSPERO or within an institutional registry. Registration prevents silent scope changes and signals that your review follows a public plan. It also helps teams avoid duplicate reviews on the same topic.

Build A Reproducible Search Strategy

Pick databases that fit the question. For medicine, PubMed or MEDLINE is a must. Add Embase for drug and device work, CENTRAL for trials, and CINAHL for nursing or allied health. Tailor spelling and field tags to each database. Combine controlled vocabulary such as MeSH with free-text terms and common synonyms. Pilot the search on a handful of known studies and tweak until you retrieve them reliably. Record full strings for each database, note any limits or filters, and save exported files with timestamps. Store strings and exports in a versioned folder and document librarian input if you have it.

Set Clear Eligibility Criteria

Write inclusion and exclusion rules before you search at scale. Define study designs you will accept, the minimum sample size if relevant, and any language or date limits you can justify. Set outcome windows for follow-up. Explain how you will treat preprints and conference abstracts. Create a short list of edge cases and decide now, not during screening, how to handle them.

Screen Titles And Abstracts Without Drift

Import all records into a manager such as EndNote, Zotero, or Covidence. Remove duplicates, then screen titles and abstracts in pairs. Train on a small batch, check agreement, and adjust the rules if your team reads the same phrases differently. Move to full-text screening with two reviewers for each paper. Track reasons for exclusion in a standard menu so your PRISMA flow later is painless.

Extract Data With A Standard Form

Design one form and pilot it on three to five studies. Capture study setting, design, sample size, eligibility, follow-up time, interventions or exposures, outcome definitions, and effect estimates with precision. Log funding and conflicts. Add fields for subgroup data that match your PICO. Keep free-text notes for quirks that may affect synthesis. When in doubt, extract both adjusted and unadjusted effects and record the model.

Team Roles And Tools

Two minds see more. Pair each step where feasible: two screeners, two extractors, two appraisers. Use tracking tools that match your budget: spreadsheets for small sets, or web platforms like Covidence or Rayyan for large sets. Name a guarantor who resolves conflicts and keeps the protocol on track. If you can, involve an information specialist for search tuning and a statistician for synthesis plans. Set a shared folder plan so everyone files work the same way.

Appraise Study Quality

Use the right tool for the design. RoB 2 fits randomized trials. ROBINS-I fits non-randomized intervention studies. QUADAS-2 fits diagnostic accuracy studies. AMSTAR 2 appraises reviews. CASP and JBI checklists work across designs. Two reviewers should rate each study independently and settle differences by consensus or a third reader. Keep judgments and quotes from the paper so readers can see the basis for each call.

Synthesize The Evidence

Pick a plan that matches heterogeneity. If studies are close in design and outcomes, run a random-effects meta-analysis. Convert effect sizes onto a common scale and check directions match. If designs or outcomes vary, group by design or outcome family and narrate patterns without forcing averages that mislead. Use tables and forest plots to keep readers oriented. Note where data are too thin for pooling and explain why.

Write And Share With PRISMA

Use PRISMA 2020 to structure reporting. Show a clear flow diagram of records from search to final set. Provide full search strings in an appendix, a list of included studies, and a table of excluded studies with reasons. Present risk-of-bias judgments beside each main outcome. Add a short, plain-language summary for busy clinicians. State limits and where new trials or real-world studies would shift the answer. Share data extraction sheets and any analysis code as supplements.

How To Conduct A Medical Literature Review: Search That Finds What Matters

Good searches start with a seed set. List three to five articles that must show up. Pull their MeSH terms and keywords, then expand. Split concepts into blocks and link with AND. Within blocks, add synonyms with OR. Use truncation and adjacency where available. Apply study filters only if you know their recall and precision. Capture grey literature by checking trial registries, preprint servers, and conference proceedings. Search reference lists and use forward citation alerts to catch fresh papers that appear after your initial run.

Practical Tips For Databases And Sources

PubMed or MEDLINE gives broad coverage and indexing. Embase adds European journals. CENTRAL concentrates trials. Web of Science helps with forward citation chasing. ClinicalTrials.gov and WHO ICTRP reveal unpublished or ongoing studies. For guidelines, scan specialty society sites. For qualitative work, add PsycINFO and SocINDEX. Record contact with study authors when you need missing numbers or clarifications and save email threads for your audit trail.

From Search To Clean Library

Export results from each source in a consistent format like RIS. Deduplicate with a structured rule set: same DOI, same title and first author, or same trial registry ID. If the same trial produces several papers, group them under one record and note which paper holds the primary outcome. Tag each record with the source, the date searched, and the search string version. These tags make updates and audits simple.

Managing Bias And Quality

Bias can tilt results. Selection, performance, detection, attrition, and reporting bias all show up in clinical trials. Confounding and selection issues dominate non-randomized work. Use structured tools, keep the process blind to study results where possible, and present judgments alongside outcomes so readers can weigh them.

Common Appraisal And Synthesis Tools

Data Synthesis Without Headaches

Before any meta-analysis, align effect measures. Risk ratio, odds ratio, and risk difference each tell a slightly different story. Pick one and convert as needed. Check for unit issues such as per-protocol rates versus intention-to-treat. For continuous outcomes, convert to mean difference or standardized mean difference. Handle clustering, crossover, and multi-arm trials with care so you do not double-count participants. Run leave-one-out checks and small-study bias tests. When heterogeneity is large, avoid a single pooled estimate; compare strata, try planned sensitivity analyses, and use meta-regression with restraint.

Model Choice And Diagnostics

Use random-effects models when clinical or methodological diversity is present. Fixed-effect models suit tight sets where one underlying effect is plausible. Inspect heterogeneity with tau-squared and I-squared, but do not treat thresholds as laws. Read forest plots and prediction intervals. Compare DerSimonian-Laird, REML, or Paule-Mandel estimates to see if results swing on method choices. For rare events, use methods that handle zeros, such as continuity corrections or Peto odds ratios, without dropping studies that carry information. Document each choice and show when conclusions change.

From Numbers To Meaning

Translate pooled effects into numbers that clinicians and patients can use. Provide absolute risks and numbers needed to treat or harm using baseline risks from typical settings. Mark any outcome where certainty is low using GRADE language. If results are fragile, say so plainly. Keep subgroup claims modest unless they rest on prespecified rules and strong interaction tests. When data resist pooling, write a tight narrative that groups like with like and explains patterns without spin.

Writing For Trust And Clarity

Clinicians skim. Put the answer up front, then show how you got there. Use short paragraphs and direct verbs. Label tables and figures so they stand alone. Report absolute effects along with relative ones so readers can estimate real-world impact. State the size and direction of any change instead of leaning on p values. Explain what the evidence means for typical patients and settings.

Reporting Extras That Help Readers

Add a one-page summary box with the question, main findings, and what they mean. Place a table that lists each outcome, the effect size, confidence interval, and the certainty rating. Provide downloadable supplements: extraction sheet, risk-of-bias tables, and code. If space allows, add a lay summary so patients can follow along.

Common Pitfalls And Fixes

Vague questions spread searches thin. Start with a tight PICO. Missing methods make readers doubt findings. Post your protocol and stick to it. Single-screening lets errors slip. Use pairs for each stage. Unclear outcomes muddle pooling. Predefine outcomes and time points. Poor search logs block updates. Save strings and dates in one place. Selective reporting skews impressions. Present both positive and null studies without cherry-picking quotes. Rushed synthesis invites overreach. If data are sparse or inconsistent, keep to a narrative approach and mark the gaps for later studies.

Ethics, Transparency, And Registration

Reviews influence care and policy. Describe funding and any ties that could sway choices. Share your full dataset and code where feasible. Register systematic protocols and link the record in your paper. Use open-access supplements for appendices so busy readers can find the technical pieces fast. If you update a prior review, state what changed in the question, time window, and methods, and supply a new PRISMA flow.

Resources You Can Trust

Use the PRISMA 2020 statement for reporting checklists. Open the Cochrane Handbook for methods across planning, search, bias, and synthesis. Learn MeSH from NLM to build strong searches that mix controlled vocabulary and text. These links serve as core references for planning and reporting.