Systematic Review and Meta-Analysis with PRISMA

How to conduct and report a systematic review, protocol, search strategy, screening, PRISMA flow diagram, and when meta-analysis is appropriate.

A systematic review answers a focused question by finding, appraising and synthesising all relevant studies using an explicit, reproducible method. Meta-analysis goes further, statistically combining results. Both sit near the top of the evidence hierarchy, and both are judged on the transparency of their method, for which PRISMA is the reporting standard.

Start with a Protocol and a Focused Question

Systematic reviews are pre-planned. Register a protocol (e.g. PROSPERO for health reviews) before you begin, to guard against selective reporting. Frame the question with a structured format such as PICO (Population, Intervention, Comparison, Outcome), which also drives your search terms and inclusion criteria.

Search Strategy

Search comprehensively across multiple databases (e.g. MEDLINE, Embase, CINAHL, Cochrane), using a documented combination of keywords and controlled vocabulary (MeSH). The search must be reproducible: report the exact strings and dates so another team could repeat it. Supplement with reference-list and citation searching.

Screening and Selection

Apply pre-set inclusion and exclusion criteria in two stages, title/abstract, then full text, ideally with two independent reviewers to reduce bias. Record numbers at each stage for the PRISMA flow diagram, which shows records identified, screened, excluded (with reasons) and finally included.

Quality Appraisal

Assess the risk of bias in each included study using an appropriate tool (e.g. Cochrane RoB 2 for trials, Newcastle-Ottawa for observational studies). The strength of your conclusions depends on the quality of the underlying evidence, so this step is not optional.

Synthesis: Narrative or Meta-Analysis

If studies are too heterogeneous to combine statistically, synthesise narratively around themes. If they are sufficiently similar in design, population and outcome, meta-analysis pools effect sizes, typically presented in a forest plot, and quantifies heterogeneity (e.g. the I² statistic). Do not force a meta-analysis onto incompatible studies. A misleading pooled estimate is worse than an honest narrative synthesis.

Review Types, and Which You Are Actually Doing

Students frequently label a piece a systematic review when the brief asked for something lighter. The distinction matters, because a systematic review makes a claim about completeness that you must then evidence.

TypeSearchAppraisalTypical use
Narrative reviewSelective, not exhaustiveInformalUndergraduate essays, background chapters
Scoping reviewBroad, mapping the fieldUsually noneEstablishing what evidence exists
Systematic reviewExhaustive, documented, reproducibleFormal risk-of-bias toolPostgraduate dissertations, publication
Meta-analysisAs systematic reviewAs systematic review, plus statistical poolingWhere studies are similar enough to combine
Rapid reviewSystematic method, deliberately limitedOften simplifiedTime-constrained work, stated as a limitation

If you cannot document the search well enough that another person could repeat it and get your result, you have written a narrative review. Say so. Calling it systematic when it is not is the fastest way to lose credibility with an examiner.

PRISMA and Where It Fits

PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses. It is a reporting guideline, not a method: it tells you what to disclose, not how to conduct the review. The current version is PRISMA 2020, published in the BMJ, and it is indexed alongside other reporting standards by the EQUATOR Network.

Two components matter for student work. The checklist lists the items your write-up must contain. The flow diagram shows how many records you identified, screened, excluded and included, with reasons for exclusion at full-text stage. Markers look for the diagram first, because it reveals instantly whether the process was systematic.

Framing the Question

FrameworkElementsBest for
PICOPopulation, Intervention, Comparison, OutcomeClinical effectiveness questions
PICOSPICO plus Study designWhere you restrict to trials or cohorts
SPIDERSample, Phenomenon of Interest, Design, Evaluation, Research typeQualitative evidence synthesis
SPICESetting, Perspective, Intervention, Comparison, EvaluationSocial science and service evaluation

The framework does more than shape the question. Each element becomes a concept block in your search string, so choosing it early saves rework.

Building a Search You Can Defend

The search is where most student reviews come apart, and it is also the part a marker can check most easily. Build it in layers.

  1. Split the question into concept blocks. Each element of your framework becomes one block.
  2. Gather synonyms within each block and join them with OR, including British and American spellings.
  3. Join the blocks with AND. Fewer blocks means more results. Three is usually the practical maximum.
  4. Add controlled vocabulary such as MeSH terms in MEDLINE, alongside your free-text terms rather than instead of them.
  5. Record the string, the database and the date every time you run it. Databases change, so an undated search is unreproducible.

Then supplement the database search. Check the reference lists of your included studies, look at what has cited them since, and search relevant grey literature such as theses and government reports. Reporting these steps signals a serious method.

Risk of Bias Tools by Study Design

Study designCommon tool
Randomised trialsCochrane RoB 2
Non-randomised interventionsROBINS-I
Observational, cohort or case-controlNewcastle-Ottawa Scale
Qualitative studiesCASP qualitative checklist
Mixed study typesMixed Methods Appraisal Tool

Name the tool you used and report the result per study, usually as a table. "Studies were of good quality" without a tool or a table is an assertion, not an appraisal.

Narrative Synthesis Versus Meta-Analysis

Meta-analysis is not the more impressive option. It is the appropriate option only when studies are similar enough in population, intervention and outcome measure to combine meaningfully.

A misleading pooled estimate is worse than an honest narrative synthesis, because it puts a number on a comparison that was never valid.

Common Mistakes and Fixes

MistakeFix
Search string not reportedReproduce the exact string per database, with the date run, in an appendix
One database onlySearch at least three relevant databases plus reference lists
No exclusion reasons at full-text stagePRISMA requires them. Record as you screen, not afterwards
Screening alone with no checkHave a second screener on a sample and report the agreement
Calling a narrative review systematicDescribe what you actually did and name the limitation
Pooling incompatible studiesSynthesise narratively and explain the heterogeneity

Discipline Notes

Where Ethical Support Fits

A librarian will help build and document your search strategy, and that is exactly the sort of support universities provide for this. Statistical advice on whether pooling is appropriate is also legitimate. What stays yours is screening the studies, appraising them and deciding what the evidence shows.

Frequently Asked Questions

Do I have to register a protocol?

For a student dissertation, usually not, though it is good practice and some supervisors require it. For anything intended for publication in health research, expect to.

How many studies should a systematic review include?

There is no target. The number is whatever meets your criteria. A review finding four eligible studies is a valid result, and the scarcity is itself a finding.

Can I do a systematic review alone?

Student reviews are often solo. Acknowledge single-screener bias as a limitation, and where possible have someone check a sample of your screening decisions.

What is I² telling me?

It estimates the proportion of variation across studies due to heterogeneity rather than chance. High values suggest the studies may not belong in one pooled estimate.

Do I need meta-analysis for a good mark?

No. Marks come from method transparency and appraisal quality. A well-justified narrative synthesis beats an inappropriate pooled estimate.

Your Next Step Today

Open the PRISMA 2020 checklist and mark which items your draft already reports. The gaps are your task list, and they are usually search documentation and exclusion reasons.

Checklist

Trusted Sources

Review requirements differ by discipline and by supervisor. Where this guide and your department's guidance differ, follow your department.

Related services

Related guides

All academic writing guides or browse our academic writing services.