Health literature review: set a focused question, search key databases, screen by criteria, appraise bias, synthesize results, and report with PRISMA.
What A Good Health Literature Review Does
A strong review maps what is known, what is uncertain, and where methods or data fall short. It pulls scattered studies into a tight story that a clinician, policymaker, or graduate student can act on. It also lays down a trail another team can follow and repeat.
The heart of the work is planning. Clear scope, explicit inclusion rules, and a repeatable search beat hunches. You are curating evidence, not collecting links. That means log every choice, from filters to excluded papers, with short reasons.
Planning Grid For Health Literature Reviews
| Element | What To Decide | Tips |
|---|---|---|
| Primary Question | Who, what, compared with what, and outcomes | Phrase with PICO or PECO so scope stays tight |
| Review Type | Systematic, scoping, rapid, umbrella, realist | Pick the design that matches your aim and timeline |
| Eligibility Criteria | Population, interventions, comparators, outcomes, study designs, time window | Write them before any screening starts |
| Databases | Where you will search | Plan at least two major indexes plus a trial register |
| Grey Literature | Theses, reports, preprints, guidelines | Note sources and how you will judge them |
| Language | Which languages you will include | State the reason for any limits |
| Search Strategy | Keywords, subject headings, Boolean, filters | Pilot strings in one database, then translate |
| Screening | Title/abstract then full text | Use two reviewers or a calibrated single screener |
| Data Extraction | What fields you will capture | Build a shared form and test it on five papers |
| Risk Of Bias | Which tool fits each design | Predefine domains and judgement rules |
| Synthesis Plan | Narrative, meta-analysis, or mixed | State how you will group studies and handle heterogeneity |
| Reporting | Checklist and flow diagram | Use the PRISMA 2020 checklist for transparent reporting |
Compiling A Literature Review In Health Research: Step-By-Step
1) Define The Question
Start by stating the problem in PICO or PECO terms. Spell out the population, the exposure or intervention, the comparator, and the outcomes that matter. Add setting and time frame if they change decisions. Tight wording helps later when you judge borderline papers.
For structure and examples, see the Cochrane Handbook chapter on review PICO. Link your question to a short list of outcomes that readers value, not just what databases return.
2) Write A Protocol
Put your plan on paper before you search. State aims, criteria, databases, search dates, screening steps, data items, bias tools, and the synthesis plan. Name roles and who breaks ties. A public protocol is even better; systematic teams often register with PROSPERO.
3) Design The Search
List the databases that match your topic. Health reviews usually draw from MEDLINE/PubMed and Embase, and add CINAHL, PsycINFO, Web of Science, or a trial register when relevant. Use both subject headings and free text. Mix synonyms, spelling variants, and acronyms.
Sample Boolean Pattern
(asthma OR wheez*) AND (inhaled corticosteroid* OR ICS) AND (step-down OR dose reduction) AND (random* OR trial OR cohort)
Build one strategy well, then translate to other platforms. Save the exact strings and export a copy to your appendix. Record the final search date and any filters, such as humans or age bands.
4) Run Searches And De-Duplicate
Export results with full fields. Import into your reference manager, then remove duplicates. Keep the original files untouched. If you hand-search key journals or scan reference lists, log the source and date.
5) Screen Systematically
Screen in two passes. First, scan titles and abstracts against your criteria. Second, read the full texts that pass. Use two reviewers when stakes are high or samples are large. If only one person screens, run a calibration set together first and log agreement.
Record why you exclude a full text. Short reasons like “wrong population” or “no comparator” make your PRISMA flow easy later and help future updates.
6) Extract Data You Can Use
Build a table with fields that answer the question. Typical items are setting, sample size, arms, doses, follow-up, outcomes, effect estimates, and notes on missing data. Pilot the form on a few papers, then refine labels so two people would fill it the same way.
7) Judge Risk Of Bias
Pick a bias tool that matches design. Randomized trials call for domain-based checks such as randomization process, deviations, missing data, measurement, and reporting. Non-randomized studies need a tool that handles confounding, selection, and measurement. Summarize judgments by domain, not just an overall word like low or high.
8) Choose Your Synthesis
When studies align on PICO and outcomes, a meta-analysis can serve readers well. Define the effect measure first, such as risk ratio, odds ratio, mean difference, or standardized mean difference. Plan how you will pool (fixed or random effects), check heterogeneity, and probe it with subgroup or sensitivity checks. If methods or outcomes differ widely, write a structured narrative that groups studies by design, dose, setting, or risk level.
9) Handle Heterogeneity Sensibly
Variation is normal in health studies. Before pooling, scan forest plots and the I² statistic, but also think about clinical and methodological spread. Predefine thresholds for pooling and when to switch to narrative only. Explain large effects that hinge on a single small trial.
10) Rate Confidence In The Body Of Evidence
State how certain readers should be. Many teams grade certainty by outcome across risk of bias, inconsistency, indirectness, imprecision, and publication bias. Link these ratings to how you phrase practice or policy takeaways.
11) Report With Clarity
Write the flow from identification to included studies. Provide a table of characteristics, a bias summary, and the synthesis with figures. Align section headers with your protocol. Use PRISMA 2020 resources to check that every method choice and result is visible.
Efficient Search Habits That Save Time
Sketch a small concept map before you touch a keyboard. That quick step cuts missed terms later. Keep a bank of tested strings for common methods, measures, and study designs. Translate one polished strategy across platforms rather than writing five mediocre ones.
Use subject headings where they exist, and add free text for new terms. Truncate carefully so you add variants without noise. Logins expire and platforms change; save screenshots of filters and limits on the day you search.
Writing That Satisfies Reviewers
Front-load the bottom-line message in the abstract. State the question, what you did, what you found, and how sure you are. Use plain words for methods: who, what, where, and when. Keep paragraphs short. Put numbers in tables and figures, then use the text to tell readers what the numbers say.
Be consistent with terms. If you call an outcome “severe flare” in one place and “exacerbation” in another, readers will miss links across sections. Define abbreviations once and stick to them. Label supplementary files clearly so they are easy to spot.
Common Pitfalls And How To Avoid Them
Vague eligibility rules blow up screening time. Write bright lines before you start, then pilot them on a small stack and tweak wording where people disagree. Keep those examples with the rules so future updates read them the same way.
Over-filtering at the search stage hides useful work. Start broad, then tighten during screening. Field tags and human limits can drop key studies in older databases. When in doubt, run a sensitivity run with fewer limits and note any extra inclusions.
Mixing outcomes in a single forest plot confuses readers. Pool like with like. If two trials use different pain scales, convert first or keep them separate. Spell out any conversions so a reader can re-run your math with a calculator.
Letting one dramatic study steer the story is a trap. Run leave-one-out checks and show both pooled and unpooled views when impact is large. Report small-study effects and match your claims to the certainty level, not to the most striking number.
Data Management And Reproducibility
Give files stable names that sort well: 01_searches, 02_screening, 03_extraction, 04_analysis. Freeze a read-only copy of each stage. If you use code to clean data or run models, keep scripts with comments and version numbers. Small habits like these shorten peer review and make updates easy.
Store the data-extraction sheet with headers that match the write-up. Save one row per study per outcome so totals are simple to trace. If teams work in parallel, assign unique IDs early and carry them through figures and supplements.
Synthesis Choices And When They Fit
| Scenario | Better Fit | Notes |
|---|---|---|
| Homogeneous PICO, same outcome scale | Meta-analysis | State effect measure and model; test influence of key studies |
| Similar questions, mixed measures | Standardize or convert | Use standardized mean difference or convert odds to risk ratios with care |
| Different designs or wide clinical spread | Narrative synthesis | Group by design, dose, or setting; explain patterns, not just counts |
| Adverse events across many small trials | Pooled rates or rare-event models | Watch zero cells; use exact or continuity-corrected methods |
| Time-to-event outcomes | Hazard ratios | Prefer log HR; align follow-up windows |
| Complex interventions | Logic model plus mixed synthesis | Map components, context, and dose; narrate pathways |
Visuals That Clarify
A PRISMA flow chart shows where records came from and why things were excluded. Forest plots should include study IDs, weights, and confidence limits. Funnel plots can live in the supplement with brief notes on what they show and where they are limited.
Tables carry most of the weight. Put study characteristics, bias judgments, and outcome data where a reader expects them. Good labels beat long prose. If a plot or table answers a question at a glance, the text can stay crisp.
Quality Checks Before You Submit
Rerun the search right before submission and add new hits if the window is long. Rebuild the PRISMA flow to reflect any late changes. Recheck your bias tables against source texts, and confirm that effect directions align across figures and captions.
Scan captions and footnotes for units and scales. Make sure tables include denominators. Confirm that exclusions list matches your criteria. Archive your data extraction sheet and code so an update can start quickly. A tidy package makes peer review smoother and gives future teams confidence in your process.