So how do you find out what is already known about a question, fairly and without picking only the studies you like? That is the job of a literature review.
Companion to Research, simply explained.
A weak review lists studies one by one. A good one pulls them together into themes: what keeps coming up, where studies disagree, and what nobody has looked at yet.
They differ in how strict the method is and how focused the question is. Tap each one to see what it asks about our example.
Every review moves from a big messy pile to a small set of studies you can trust, and then to one clear story.
Narrative reviews follow these steps loosely. Scoping and systematic reviews follow them strictly and write every step down, so someone else could repeat them.
"Antibiotics and patients" could mean anything. Break your question into parts with PICO. Tap each letter.
Mapping a topic rather than testing something? Use PCC instead: Population, Concept, Context.
Each letter becomes a search concept in Step 2.
Before you start, search PROSPERO, the Cochrane Library and OSF to see whether someone is already doing your review.
AND narrows, OR widens, NOT removes. Tap each to see what a database would bring back.
Use OR between synonyms. A star finds every ending: antibiotic* catches antibiotic and antibiotics.
Each database covers different journals. Then follow the trail: check what your best papers cite, and who cited them later. This is called snowballing.
The funnel you just saw is a simplified PRISMA 2020 flow diagram. Systematic and scoping reviews are expected to include one, with numbers at every stage.
This is critical appraisal: checking how a study was done, not whether you like its answer. Tap each feature. Would it make you trust the result more, or less?
0 of 6 judged
Two people appraise each study independently. Report the judgement for every study, usually in a table or "traffic-light" chart.
Build a table before you start reading. Fill in the same columns for every included study, so nothing important slips through and studies can be compared side by side.
| Study | Who | Design | Main finding |
|---|---|---|---|
| A | 200 adults, city clinic | Survey | Most who stopped said they felt better |
| B | 24 adults | Interviews | Side effects made people stop |
| C | 310 adults, rural | Survey | Cost and forgetting both mattered |
| D | 450 adults | Trial | Reminder call: more finished the course |
Synthesis means organising by idea, not by study. Say where studies agree, where they clash, how strong the evidence is, and what's missing.
Then say how sure you are. GRADE rates certainty for effects; GRADE-CERQual does the same for qualitative findings.
Arksey and O'Malley (2005) set out five stages, plus an optional sixth:
JBI's updated guidance (Peters et al., 2020) refines these steps and lines up with PRISMA-ScR.
GRADE rates how sure we can be about each main result: high, moderate, low or very low.
Certainty goes down for risk of bias, inconsistency between studies, indirectness (studies don't quite match your question), imprecision (wide uncertainty) and publication bias.
The protocol follows PRISMA-P. Cochrane reviews register automatically; others register on PROSPERO, which only accepts reviews with a health-related outcome.
Used inside a systematic review when studies are similar enough to combine. In this forest plot, each line is a study and bigger boxes carry more weight. The diamond is the combined answer. Tap a study to leave it out.
Answer up to three questions. The right type depends on your question, not on which one sounds most impressive.
Predatory journals charge fees but skip real peer review. Warning signs:
Ask your librarian or supervisor, and use the Think. Check. Submit. checklist.
Written for laypeople and diploma students. The running example, its studies and all its numbers are made up.