revisia · Blog (English)

What is a systematic review?

2026-10-12T00:00:00Z

A systematic review is a secondary study that gathers, appraises and synthesizes all the available evidence on a well-defined research question, following an explicit, transparent and reproducible method. In one sentence, it is the opposite of an opinion. Instead of citing the studies the author remembers or likes, it finds every relevant study, assesses its quality and combines the results using rules set in advance. That disciplined way of working is what makes it the foundation of evidence-based medicine and one of the most reliable sources for clinical, academic and health policy decisions.

This guide explains what a systematic review is, how it differs from a literature review or a narrative review, what its stages are and how to do one step by step. It is the central article of our blog on systematic review methods. From here we link to a dedicated guide for each stage, from the PICO question to the PRISMA flow diagram, so you can go deeper wherever you need to.

What is a systematic review?

A systematic review is a piece of research whose subjects are not patients or primary data. Its subjects are other studies that have already been published (and, when possible, unpublished literature too). Its defining feature is the systematic method. Every decision, from the question to the inclusion criteria, is set out in a protocol before you start and documented so that any other team could repeat the process and arrive at the same set of studies.

The Cochrane Handbook for Systematic Reviews of Interventions puts it this way: "A systematic review attempts to collate all the empirical evidence that fits pre-specified eligibility criteria in order to answer a specific research question." JBI (formerly the Joanna Briggs Institute) frames it in much the same way and provides methods for different types of question (effectiveness, prevalence, experiences and so on). Those two references, together with the PRISMA 2020 statement (Preferred Reporting Items for Systematic Reviews and Meta-Analyses), make up the methodological framework we follow throughout this guide.

What a systematic review is for

A well-conducted systematic review lets you:

Summarize what is known about a specific question without the bias of picking only the studies that confirm an existing idea.
Resolve contradictions between individual studies that point in different directions, by weighing their quality and, where appropriate, combining them statistically in a meta-analysis.
Spot evidence gaps, meaning questions that still lack enough studies, or enough good ones.
Ground clinical practice guidelines, public policy and academic work (undergraduate theses, master's dissertations, PhD theses) on a reproducible base. If yours is a thesis or dissertation, we have a dedicated guide on the literature review and the systematic review in a dissertation.

Six features separate it from any other review: an explicit research question, inclusion and exclusion criteria set in advance, a reproducible search across several databases, study selection by two independent reviewers, a formal risk of bias assessment, and a transparent synthesis (narrative or quantitative).

Systematic review vs literature review and narrative review

One of the most common mix-ups is treating "systematic review" and "literature review" as synonyms. They are not. A systematic review is a specific method. "Literature review" is a generic term that covers many ways of reviewing the literature, the narrative review included. The key difference is how systematic the process is and how much it controls bias. A systematic review defines everything in advance and documents every step. A narrative review usually depends on the author's judgment.

The table below sums up the main differences. If you are not sure which one to do, the extended comparison and decision guide has its own article. For a full classification of review formats (scoping, umbrella, integrative, rapid), see our guide on the types of literature reviews.

Aspect	Systematic review	Literature / narrative review
Research question	Specific and structured (e.g. PICO)	Broad, general, often implicit
Protocol in advance	Yes, registered (e.g. PROSPERO)	Usually not
Search strategy	Explicit, reproducible, across several databases	Not described or not reproducible
Inclusion/exclusion criteria	Defined before you start	Vague or missing
Study selection	By two independent reviewers	By the author, with no bias control
Quality assessment	Systematic (risk of bias)	Optional or missing
Synthesis	Structured; sometimes a meta-analysis	Interpretive discussion
Reproducibility	High (another team reaches the same result)	Low

A narrative review is not "bad" by definition. It is useful for giving a broad overview of a wide topic or for introducing a field. The problem is different, and it shows up when a narrative review is presented as if it were systematic, claiming a methodological strength that its way of working (no protocol in advance, no reproducible search, no independent double review) simply cannot guarantee.

Is yours a systematic review? In revisia you work through every stage with guidance, with each decision documented. Create your free account.

The stages of a systematic review (overview)

Doing a systematic review is a process in stages. Each stage feeds the next, and all of them are documented so the result can be reproduced. There are eight, and each one links to a detailed guide where we explain it step by step.

1. Research question (PICO)

Everything starts with a clear question. A vague question produces a vague review. The most widely used framework to structure it is PICO: Population, Intervention, Comparison and Outcome. Defining these four components well hands you, almost for free, your search terms and your inclusion criteria. PICO is built for interventions. For exposures, scoping, prevalence, diagnosis, prognosis or experiences, you can choose among seven question frameworks.

Learn how to frame your question, with examples by discipline, in our guide on how to build your research question with the PICO model. And if you want to see how that question ends up in the manuscript, we analyzed what 2,368 published reviews do with theirs. They write a declarative objective of about 26 words, almost never a question with a question mark.

2. Protocol and registration (PROSPERO)

Before you search for a single article, you write a protocol. It sets in advance the question, the eligibility criteria, the search strategy, the selection methods, data extraction and the synthesis plan. Registering it (in health, the usual place is PROSPERO, the international register of systematic reviews) reduces the risk of bias. It stops the criteria from shifting along the way to "fit" the results you are finding, and it makes your intentions public before you know the outcome.

The protocol is the document that makes your work verifiable, not a bureaucratic formality. PRISMA even has a dedicated extension for protocols, PRISMA-P.

3. Search strategy

The search must be comprehensive and reproducible. That means translating your PICO question into a search string with Boolean operators (AND, OR, NOT), controlled vocabularies (MeSH terms in PubMed, DeCS in Spanish- and Portuguese-language databases) and truncation, and running it in several databases, not just one.

We show you how to build that string step by step in the guide on the literature search strategy with Boolean operators and MeSH, and which sources to use in the one on databases for a systematic review. The general rule in Cochrane and JBI guidance is to search more than one database so you do not depend on the coverage of a single one.

4. Screening and selection

The search returns far more results than will end up being relevant (the full funnel, with how many studies usually remain at the end, has its own guide). Now you filter. Screening happens in two steps. First on title and abstract, then by reading the full text of the records that passed the first filter. To control bias, two reviewers assess each record independently, and disagreements are resolved by consensus or by a third reviewer.

Who gets in and who does not is decided by your eligibility criteria, which you should have defined precisely. We explain how to derive them from PICO in the guide on inclusion and exclusion criteria. To manage thousands of records and work blinded, dedicated tools are the practical choice; we compare them in the guide on Rayyan vs Covidence for study screening.

5. Data extraction

From the studies that pass screening, you extract the data of interest systematically, ideally with a data extraction form designed and piloted in advance. You record the population characteristics, the details of the intervention, the sample size, the outcome measures and the numerical data for a possible synthesis. The screening rule applies again here. Two people should extract independently and compare their results to reduce transcription and interpretation errors. We cover the table fields, how to pilot the form and the typical mistakes in the data extraction guide.

6. Risk of bias assessment

Not every included study deserves the same weight. Before drawing conclusions you need to assess their risk of bias and methodological quality, and each study design has its own tool: RoB 2 for randomized controlled trials, ROBINS-I for non-randomized studies, the Newcastle-Ottawa Scale for observational studies, AMSTAR 2 for appraising other reviews, and the GRADE approach for rating the overall certainty of the evidence.

This stage is what separates a rigorous review from a collection of studies. We cover it in the guide on risk of bias assessment with RoB 2, AMSTAR 2 and Newcastle-Ottawa.

7. Synthesis and meta-analysis

With the studies selected and appraised, it is time to synthesize. The synthesis can be narrative (structured and honest about its limitations) or quantitative. When the studies are similar enough in population, intervention and outcome, their results can be combined statistically in a meta-analysis, which produces a pooled estimate of the effect (with its forest plot, its confidence interval and its heterogeneity analysis).

A meta-analysis is not always possible or desirable. If the studies are too heterogeneous, combining them can produce a misleading number. How to read a forest plot, what a high I² means and when a meta-analysis makes sense (or not) are all explained in our guide on meta-analysis and how to interpret it.

8. Reporting according to PRISMA

The review is reported following the PRISMA 2020 statement, the international standard for reporting systematic reviews and meta-analyses completely and transparently. PRISMA includes a 27-item checklist and the well-known flow diagram showing how many records were identified, screened, excluded (and why) and finally included.

You will find the editable template and the guide to filling it in in the article on the PRISMA 2020 flow diagram. PRISMA does not change your results, but it makes your review assessable and publishable; many journals require it.

Recommended tools

You do not need paid software to do a rigorous systematic review. What does save weeks of work and cut errors along the way is the right tool at each stage. Here is a selection; each category links to its detailed guide.

Stage	Tools	What for
Search	Databases. PubMed/MEDLINE, Scopus, Web of Science, Cochrane Library, SciELO	Find the evidence across several sources
Reference management	Mendeley or Zotero (free), EndNote (paid)	Organize references, remove duplicates, cite in Word
Screening	Rayyan (free) or Covidence (paid)	Blinded selection of titles, abstracts and full texts
Synthesis / meta-analysis	Meta-analysis software. JASP, jamovi, RevMan, R (metafor package)	Calculate the pooled effect and draw the forest plot

One practical tip. Define your workflow before you start and keep each stage traceable to the next. For example, the exclusions you log in Rayyan or Covidence must match the numbers you later put in the PRISMA flow diagram. When the tools talk to each other (or at least their numbers add up), the final report writes itself.

Common mistakes

Knowing the most frequent errors saves you repeated work and criticism in peer review. These are the ones that come up most:

A question that is too broad or poorly defined. If the question is not structured (with PICO, for example), the search gets out of hand and the review loses focus.
Not registering the protocol. Changing the criteria along the way to fit the results is a classic source of bias. A protocol written in advance (ideally registered in PROSPERO) prevents it.
Searching a single database. Sticking to PubMed leaves out evidence indexed in Scopus, Web of Science or regional and non-English databases. Cochrane and JBI recommend several sources.
Not documenting the search strategy. If your search cannot be reproduced, it is not a systematic review.
Screening and extracting with a single reviewer. Without independent double review, the risk of error and selection bias shoots up.
Skipping the risk of bias assessment. Combining good and poor studies as if they were worth the same distorts the conclusions.
Forcing a meta-analysis on heterogeneous studies. A summary number from studies that are not comparable is worse than no number at all.
Reporting the review incompletely. Leaving out the flow diagram or PRISMA items makes the review impossible to assess and harder to publish.

Would you rather not get lost in the process? In revisia you do the review guided stage by stage, from the PICO question to the manuscript. Create your free account and get started.

Frequently asked questions

What is the difference between a systematic review and a literature review?

A systematic review follows an explicit, reproducible method (a protocol, a documented search and criteria set in advance). "Literature review" is a generic term that guarantees none of those controls. The dimension-by-dimension comparison with the narrative review, and which one suits each case, is in systematic review vs narrative review.

Do I need to register a protocol?

It is strongly recommended and more and more often required. Registering the protocol in PROSPERO before screening reduces bias, and many journals ask for the registration number.

Can I do a systematic review on my own?

You can, but screening and extraction by two independent people is the standard way to reduce errors. If you work alone, keep a record of the borderline decisions.

How long does a systematic review take?

The published average is over a year: 67.3 weeks from protocol registration to publication (Borah 2017), because screening, extraction and risk of bias assessment need two reviewers working independently. Shorter formats, such as the rapid review or the scoping review described in the guide on the types of reviews, cut the timeline by openly relaxing steps of the method. The stage-by-stage breakdown, with every figure and where you can save time, is in the guide on how long a systematic review takes.

What is the difference between a systematic review and a meta-analysis?

They are not the same, although they often go together. A systematic review is the whole method for identifying, selecting, appraising and synthesizing evidence. A meta-analysis is only one synthesis technique, the statistical procedure that combines the results of several studies into a pooled estimate of the effect. Every systematic review has a synthesis stage, but that synthesis can be narrative; you only run a meta-analysis when the studies are similar enough. Put another way, you can have a systematic review without a meta-analysis, but a rigorous meta-analysis should start from a systematic review. We cover it in the guide on what a meta-analysis is and how to interpret it.

How many reviewers do you need?

At least two independent reviewers for screening, data extraction and risk of bias assessment, with a third reviewer (or team consensus) to resolve disagreements. This double review is one of the features that controls bias and sets a systematic review apart from a narrative review written by one person. Both the Cochrane Handbook and JBI stress this point. You can do a review alone (for example, for an undergraduate thesis), but then you should state that limitation explicitly.

What is PRISMA?

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) is the set of international guidelines for reporting systematic reviews and meta-analyses transparently and completely. The current version, PRISMA 2020, includes a 27-item checklist and a flow diagram that tracks studies from identification to final inclusion. PRISMA does not tell you how to do the review (Cochrane and JBI cover that). It tells you how to report it so anyone can appraise and reproduce it. The template and the full guide are in the article on the PRISMA 2020 flow diagram.

What is a meta-analysis? How it is done and how to read one

2026-10-12T00:00:00Z

A meta-analysis is the statistical method that combines the results of several studies answering the same research question to get a single, more precise quantitative estimate of the effect. If you are asking what a meta-analysis is in plain terms, think of it as a "weighted average" of the evidence. Instead of reading ten trials one by one, you merge them into a single number (with its uncertainty) that sums up what the data say as a whole. Meta-analysis is the quantitative stage that sometimes caps a systematic review, although not every review ends in one.

Reading a meta-analysis critically takes more than checking whether the result is "significant". You need to know when it is legitimate to run one, which effect measures are used and, above all, how to interpret its two signature charts, the forest plot and the funnel plot. That is what this guide covers.

What a meta-analysis is (and when you can do one)

A meta-analysis takes the result of each study (its effect size) and weights it, usually by the inverse of its variance, so that larger and more precise studies count for more in the final estimate. The product is a pooled effect with its confidence interval.

The big advantage is statistical power. Small individual studies may be unable to detect a real effect, but once you add them up the meta-analysis can reveal it. The big trap is that combining things that should not be combined produces a precise but meaningless number (the classic "comparing apples and oranges").

That is why a meta-analysis is only defensible when:

The studies answer a comparable question (populations, intervention and outcome similar enough).
There is a common effect measure that can be calculated for all of them.
There is a reasonable number of studies and the heterogeneity between them is not so high that it invalidates the summary.
A systematic search and a risk of bias assessment of the included studies have been done first.

If the clinical or methodological heterogeneity is excessive, do not force the calculation. Present a narrative synthesis and explain why pooling is not appropriate.

Meta-analysis vs systematic review

This is the most common mix-up, so let's pin it down. A systematic review is the complete, reproducible process of finding, selecting, appraising and synthesizing all the available evidence on a question. A meta-analysis is, optionally, the statistical technique that synthesizes the results of that review quantitatively.

Aspect	Systematic review	Meta-analysis
What it is	A complete research method	A statistical technique
Scope	Search, screening, appraisal and synthesis	Only the quantitative synthesis
Always possible?	Almost always	Only if the studies can be combined
Result	A reasoned answer to the question	A pooled effect with its interval
Relationship	It is the "container"	It is one possible step inside it

The relationship runs one way. Every rigorous meta-analysis starts from a systematic review, but not every systematic review includes a meta-analysis. To understand the whole process around this quantitative synthesis, see the guide on how to do a systematic review step by step.

Effect measures (odds ratio, risk ratio, mean difference)

Before you combine anything, you have to choose how each study's effect is expressed (the numbers come from your data extraction table, so extract with the measure already in mind). The measure depends on the type of outcome.

For dichotomous outcomes (it happens / it does not, such as getting ill or not, surviving or not):

Risk ratio (RR), also called relative risk. The ratio between the probability of the event in the exposed group and in the control group. An RR of 0.70 means 30% less risk; it is intuitive and widely used in cohort studies and trials.
Odds ratio (OR). A ratio of odds (the probability that something happens divided by the probability that it does not). It is the typical measure in case-control studies and logistic regression. With rare events, OR and RR are similar; with common events, the OR exaggerates the size of the effect and should not be read as if it were an RR.

For continuous outcomes (blood pressure, a score on a scale, weight):

Mean difference (MD). When all the studies measure the outcome in the same unit.
Standardized mean difference (SMD). When they measure the same construct on different scales; it is expressed in standard deviations (Cohen's d is the best known).

For all of them, the "no effect" value is the reference point worth memorizing to read any meta-analysis.

Measure	Outcome type	"No effect" value
Risk ratio (RR)	Dichotomous	1
Odds ratio (OR)	Dichotomous	1
Mean difference (MD)	Continuous	0
Standardized mean difference (SMD)	Continuous	0
Hazard ratio (HR)	Time to event	1

The fact that the null value is 1 for ratios and 0 for differences is the key to reading the chart that comes next.

Synthesis starts long before the forest plot. In revisia you reach this stage with your studies screened, your data extracted and risk of bias assessed. Create your free account.

How to read a forest plot and its confidence interval

The forest plot is the standard visual display of a meta-analysis. Each row is a study; at the bottom, one row sums up the pooled effect. Learning to read it is the single most useful practical skill in this article.

Example forest plot. Each square is a study's effect estimate, its size reflects the study's weight, the horizontal lines are the confidence intervals and the diamond at the bottom is the pooled effect.

What to look at, in order:

The square marks each study's point estimate of the effect. Its area is proportional to the weight that study contributes to the result. Bigger squares count for more.
The horizontal line through each square is its confidence interval (usually 95%). The longer it is, the less precise the study.
The vertical "no effect" line is the reference. It sits at 1 for OR, RR or HR, and at 0 for mean differences. If a study's confidence interval crosses that line, that study on its own is not statistically significant.
The diamond at the bottom is the pooled effect of all the studies. Its center is the overall estimate and its width is the confidence interval for the whole set.

The reading rule is simple. If the diamond does not touch the no-effect line, the pooled effect is statistically significant; if it crosses it, you cannot rule out no effect. Also check which side of the chart the result points to, because "left" or "right" favors one group or the other depending on how the axis is labeled.

The confidence interval, in brief

The 95% confidence interval is the range of values compatible with the data. If we repeated the study many times, about 95% of the intervals built this way would contain the true effect. In practice, a narrow interval means a precise estimate and a wide one means a lot of uncertainty (typical of small studies). Do not confuse precision with size. An effect can be large and, at the same time, poorly estimated if its interval is huge.

How to measure heterogeneity with I² and Cochran's Q

Heterogeneity is the variability between study results beyond what you would expect from chance alone. It is the central question of any meta-analysis. Are the studies measuring "the same thing", or am I averaging different things?

Two statistics are mainly used to quantify it:

Cochran's Q. A hypothesis test of whether the differences between studies are larger than chance would explain. Its problem is power. With few studies it barely detects real heterogeneity, and with many it flags heterogeneity even when it is trivial. So do not base the decision on its p-value alone.
I². It expresses the percentage of total variability that is due to real heterogeneity rather than sampling error. It is the most cited indicator because it depends less on the number of studies. The table below gives a rough guide, taken from the Cochrane Handbook.

I² value	Rough interpretation
0% to 40%	Might not be important
30% to 60%	May represent moderate heterogeneity
50% to 90%	May represent substantial heterogeneity
75% to 100%	Considerable heterogeneity

These ranges overlap on purpose. I² is read together with the clinical context and the direction of the effects, not as a rigid threshold. A high I² does not automatically invalidate the meta-analysis, but it forces you to act.

What should you do about substantial heterogeneity? You can look for its causes through subgroup analysis (by population type, dose, study quality) or meta-regression, use a random-effects model (next section) or, if the variability cannot be reconciled clinically, give up on the pooled effect and present a narrative synthesis of the studies. What you must never do is ignore a high I² and present the summary as if the studies were interchangeable.

Fixed-effect vs random-effects models

The choice of model determines how studies are weighted and what exactly the pooled effect means.

The fixed-effect model assumes there is a single true effect shared by all the studies and that the differences between them come only from sampling error. Under that assumption, studies are weighted by their precision alone (large studies dominate). It is defensible when the studies are very similar and heterogeneity is practically zero.

The random-effects model assumes the true effect can vary from one study to another (because of differences in populations, settings or protocols) and that those effects are distributed around a mean. It builds the between-study heterogeneity into the calculation, which widens the confidence interval of the result and gives relatively more weight to small studies. It is usually the default in biomedicine, because it is rarely realistic to assume an identical effect in every setting.

Criterion	Fixed effect	Random effects
Assumption	A single true effect	The effect varies between studies
What it estimates	That common effect	The mean of the distribution of effects
Study weights	Precision only	Precision + between-study variance
With heterogeneity	Not very appropriate	More appropriate
Resulting interval	Narrower	Wider (more conservative)

As a practical rule, decide a priori based on the assumptions, not after the fact by picking the model that gives the result you like best. If there is noticeable heterogeneity, the random-effects model is almost always the honest choice.

Publication bias and the funnel plot

Publication bias is the tendency for studies with positive or statistically significant results to be published more often, sooner and in more visible journals than "negative" studies. If your meta-analysis only finds the published studies, it may be systematically overestimating the effect. It is one of the most serious threats to the validity of any synthesis, and it is related to other biases covered in the guide to risk of bias assessment.

The graphical tool for exploring it is the funnel plot. It plots each study's effect size (horizontal axis) against its precision (vertical axis, usually the standard error). The logic:

Large, precise studies sit at the top and cluster near the pooled effect.
Small, less precise studies spread out at the base.
Without bias, the points form a symmetrical inverted funnel around the effect.

An asymmetrical funnel (a "gap" in the lower corner where the small studies with unfavorable results should be) is the classic warning sign of publication bias. To go beyond visual inspection, statistical tests of asymmetry such as Egger's test are used.

Two important caveats. A funnel plot is only informative with enough studies (roughly ten or more), and asymmetry is not always publication bias. It can also come from heterogeneity, from smaller studies being of lower quality or simply from chance. It is a lead to investigate, not a verdict.

Which software to use

You do not need to code from scratch to run a meta-analysis. Effect measures, weighting, I² and the plots are produced by dedicated tools, from free point-and-click interfaces to R packages. For a practical comparison of the options (JASP, jamovi, Cochrane's RevMan and the R package metafor), with their pros, cons and which to pick for your level, see the dedicated guide to meta-analysis software.

Once you have the pooled effect, remember that the meta-analysis is part of a review's synthesis. When you report it, the number of studies synthesized must match your PRISMA flow diagram, while the effect, its interval and I² go in the results section of the report.

Before the meta-analysis comes the whole review. In revisia you do it from start to finish. Create your free account.

Frequently asked questions

How many studies do you need for a meta-analysis?

There is no fixed minimum. Two studies are technically enough, but with very few studies the estimate is unstable and caution is in order.

What is an effect size?

It is the measure that quantifies the size of a difference or association (for example, an odds ratio, a risk ratio or a mean difference), not just whether it is significant.

When should you not do a meta-analysis?

When the studies are too heterogeneous in population, intervention or measurement. In that case the synthesis is done narratively.

What is the difference between a meta-analysis and a systematic review?

A systematic review is the complete, reproducible process of searching for, selecting, appraising and synthesizing all the evidence on a question. A meta-analysis is only the statistical technique that combines the results of those studies quantitatively into a single effect. Every rigorous meta-analysis starts from a systematic review, but not every systematic review includes a meta-analysis. If the studies cannot be combined, the synthesis is done narratively.

What does a high I² mean?

A high I² (roughly, above 50-75%) means that a substantial part of the variability between study results comes from real heterogeneity, not chance. In practice it suggests the studies may not be measuring "the same thing". It does not automatically invalidate the meta-analysis, but it forces you to investigate the causes (subgroup analysis, meta-regression), to use a random-effects model or, if the heterogeneity cannot be reconciled, to give up on the pooled effect.

How do you read a forest plot?

Each study is a row with a square (its effect, with an area proportional to its weight) and a horizontal line (its confidence interval). The diamond at the bottom is the pooled effect. The key is the vertical line of "no effect", at 1 for odds ratios and risk ratios and at 0 for mean differences. If a study's interval (or the final diamond) crosses that line, the result is not statistically significant; if it does not touch it, it is.

What is a funnel plot?

A funnel plot is used to explore publication bias. It plots each study's effect size against its precision. Without bias, the points form a symmetrical funnel around the pooled effect. Asymmetry (a gap where the small studies with unfavorable results should be) suggests that negative studies are missing because they were never published. Asymmetry is checked with tests such as Egger's test, and it can only be interpreted with enough studies.

Types of literature review: systematic, narrative, scoping and umbrella

2026-10-12T00:00:00Z

When you start doing research you quickly find out that there are many types of literature review, and they are not all after the same thing. Each question calls for a different approach. Needing a comprehensive, reproducible synthesis is not the same as wanting a first map of a field you barely know, or discovering that so many reviews are already published that the job now is to summarize them. Picking the wrong format is costly, because it usually forces you to redo the work halfway through. This guide covers the types of literature review that exist, from the narrative review to the scoping review and the umbrella review, how they differ and when each one fits.

Why there are so many types of literature review

What separates one type from another is how systematic it is: how explicit and reproducible the method is, and how far the search goes.

At one end sits the narrative review, flexible and dependent on the author's judgment. At the other, the systematic review, with a protocol registered in advance and methods anyone could replicate. Between and around the two, intermediate formats have appeared (scoping review, rapid review), along with "second-level" formats that synthesize reviews already published (umbrella review).

Methodology groups such as Cochrane and JBI (formerly the Joanna Briggs Institute) have defined and standardized these types, and guidelines such as PRISMA and its extension PRISMA-ScR (for scoping reviews) set out how to report them. Knowing this typology keeps you from mixing up terms that people use as synonyms but that are not.

Here they are, ordered from least to most structured, with the special formats at the end.

Narrative review

The narrative review (or traditional literature review) is the oldest and most flexible format. An expert on the topic gathers, reads and summarizes the relevant literature to give an overview or a state of the art, usually without a formal protocol or a documented comprehensive search.

Its main features:

Search. Not necessarily comprehensive or reproducible; it usually depends on what the author selects.
Study selection. Implicit criteria, often with no documented screening process.
Quality appraisal. As a rule, no formal risk of bias assessment.
Result. A qualitative, interpretive synthesis that reads easily.

In exchange for that speed and freedom it pays the price of selection bias. Because there is no explicit method, two authors can reach different conclusions from the same literature, and readers have no way to check the process.

When to use it. To introduce a topic, set out a theoretical framework, write the literature review chapter of an undergraduate thesis or master's dissertation, or when the field is too broad or too new for a structured synthesis. If your goal is to understand the big picture rather than answer a specific question with pooled evidence, a narrative review is enough.

Systematic review

The systematic review is the reference standard for synthesizing evidence and answering a well-defined research question. Unlike the narrative review, it follows an explicit, predefined method designed precisely to minimize bias.

Its defining features:

A structured question. It is framed with a framework suited to its purpose, such as PICO for interventions or PCC, CoCoPop, PIRD, PEO and SPIDER for other types of question.
A protocol in advance. The method is defined and, ideally, registered before you start, for example in PROSPERO, the international register of review protocols.
A comprehensive, reproducible search. In several databases, with a documented strategy someone else could repeat.
Paired screening. Two reviewers select studies independently against explicit inclusion and exclusion criteria.
Risk of bias assessment. Each included study is appraised with standardized tools.
Synthesis. Qualitative and, when the data allow, quantitative through meta-analysis.

It is the most rigorous and transparent type, but also the most expensive in time and resources (the published average is over a year of work, usually as a team). To understand the whole process step by step, see our guide to what a systematic review is and how to do one.

When to use it. When you have a specific, answerable question, there is enough primary literature, and you need a reliable, defensible answer (clinical decisions, recommendations, doctoral theses, scientific publication).

Going with a systematic review? In revisia you do the whole thing, from the question to the manuscript. Create your free account.

Scoping review

The scoping review shares the methodological rigor of the systematic review (structured search, paired screening, transparent method), but it is after something else. Instead of answering a closed question about whether something works, it sets out to map the scope and nature of a field.

A scoping review answers questions such as what has been researched on a topic, which types of studies exist, where the gaps in the evidence are, or how a concept has been defined in the literature.

Key differences from a systematic review:

The question is broader and exploratory; it does not focus on a specific intervention.
The risk of bias of the included studies is usually not formally assessed.
The synthesis is usually descriptive (maps, tables, counts), with no meta-analysis.
Its specific reporting guideline is PRISMA-ScR, the PRISMA extension for scoping reviews; JBI provides a widely adopted reference methodology.

When to use it. When the topic is broad or loosely defined, when you want to find gaps in the literature before planning a systematic review, or when you need to clarify concepts and map the types of evidence available. In fact, it is often done as a step before a systematic review.

Umbrella review

The umbrella review is a "second-level" review. Instead of synthesizing primary studies, it synthesizes systematic reviews and meta-analyses already published on the same topic.

When a topic has produced so many systematic reviews that the picture is once again too big to take in (sometimes with conflicting conclusions), an umbrella review brings all of them together, appraises their quality and gives an overall view.

Its features:

Unit of analysis. Systematic reviews (and meta-analyses), not individual studies.
Quality appraisal. The methodological quality of the included reviews is appraised with tools such as AMSTAR 2, which we explain in the guide to risk of bias assessment.
Result. A synthesis of the overall state of the evidence, showing where the reviews agree, where they contradict each other and how much weight each conclusion carries.

When to use it. When several systematic reviews already exist on a topic and you need a high-level, integrated overview, for example for clinical practice guidelines or to summarize a heavily researched field. An umbrella review makes no sense if there are hardly any earlier reviews.

Integrative review and rapid review

Two more formats complete the map.

Integrative review

The integrative review is very common in fields such as nursing and the social sciences. What sets it apart is that it lets you combine quantitative, qualitative and theoretical studies in the same synthesis. Its method is more structured than a narrative review (documented search, inclusion criteria, systematic analysis of the data), but more flexible than a classic systematic review about the kinds of sources it brings together.

When to use it. When your question benefits from combining empirical results with theoretical or qualitative literature, or when you want a broad view but with more rigor than a narrative review.

Rapid review

The rapid review applies the methods of a systematic review but simplifies or speeds up some steps to deliver results in weeks instead of months. For example, it may limit the number of databases, restrict dates or languages, or leave screening to a single reviewer.

It gains speed at the cost of comprehensiveness and a higher risk of bias, so it must state openly which shortcuts it took.

When to use it. When an urgent decision needs informing (health policy, management, emergencies) and there is no time for a full systematic review.

Which type of literature review to choose for your goal

The table sums up the goal of each format and when it fits.

Type of review	Goal	When to use it
Narrative review	Give a general, interpretive overview of a topic	Introducing a broad or emerging topic; literature review chapter of a thesis or dissertation
Systematic review	Answer a specific question by synthesizing the evidence reproducibly	Clear question, enough literature and a need for a reliable answer
Scoping review	Map the scope, types of evidence and gaps in a field	Broad or loosely defined topic; finding gaps before a systematic review
Umbrella review	Synthesize systematic reviews and meta-analyses already published	Many earlier reviews exist and an integrated overview is needed
Integrative review	Combine evidence from diverse designs (quantitative, qualitative, theoretical)	A question that benefits from empirical and theoretical sources together
Rapid review	Accelerated synthesis with simplified systematic methods	Urgent decisions with no time for a full systematic review

As a practical rule, the more specific your question and the more decisive the answer, the further you should move toward the systematic review (or the meta-analysis). The more exploratory your goal, the more sense a scoping review or a narrative review makes.

Doing a systematic review? In revisia it is guided from start to finish. Create your free account and get started.

Frequently asked questions

What is an integrative review?

A type of review that combines studies of different designs, quantitative and qualitative, to give a broader view of a topic.

What is a rapid review?

A review that simplifies some steps of the systematic method to deliver an answer sooner, while being open about its limitations.

Does a scoping review need a risk of bias assessment?

Usually not. Its aim is to map the existing literature, not to appraise quality or synthesize effects the way a systematic review does.

What is the difference between a systematic and a narrative review?

A systematic review follows an explicit, predefined method that requires a registered protocol, a documented comprehensive search, screening by two reviewers and a risk of bias assessment. A narrative review is flexible and depends on the author's judgment, with no comprehensive search or formal quality appraisal. The systematic review minimizes bias and lets others verify the process; the narrative review is faster but less transparent. Here is the full comparison between systematic and narrative reviews, dimension by dimension, with a guide to help you decide.

What is a scoping review?

A scoping review is a review with a structured method, similar to a systematic review, but its aim is to map the scope and nature of a topic rather than answer a closed question. It shows what has been researched, which types of studies exist and where the gaps are. Its reporting guideline is PRISMA-ScR, and it usually includes neither a formal risk of bias assessment nor a meta-analysis.

What is an umbrella review?

An umbrella review is a review of reviews. Instead of analyzing primary studies, it synthesizes systematic reviews and meta-analyses already published on the same topic. It appraises the quality of those reviews (for example with AMSTAR 2) and gives an integrated overview of the evidence. It is used when a topic already has many earlier reviews.

Which one should I do for my dissertation?

It depends on your aim and the time you have. To set a topic in context (the literature review chapter), a narrative review is usually enough. If you want to answer a specific question rigorously and your supervisor allows it, you can do a systematic review, although it takes more time and method. A middle option is a scoping review to map a field. To see how the review fits into the structure of the thesis, read our guide on the literature review for a dissertation.

Narrative review vs systematic review: which one to choose

2026-10-12T00:00:00Z

The difference between a narrative review and a systematic review fits in two sentences. A systematic review answers a specific question with an explicit method, set before the work starts and reproducible by someone else, with an exhaustive search, public selection criteria and a bias assessment of every study. A narrative review gives an overview of a topic based on the sources its author considers relevant, with methodological freedom and no duty to document the process. Neither one is "better" in the abstract. Each answers a different kind of question, and choosing wrong costs you months of work or an unconvinced examining committee. This narrative review vs systematic review comparison goes dimension by dimension, with a realistic way to decide and the middle ground that most guides skip.

The real narrative review (it is not "read and summarize")

The narrative review, the traditional literature review, has a reputation as the format with no rules, the one you write when there is no time for a systematic review. That is a caricature. A good narrative review, written by someone who knows the field, brings expert interpretation, historical context and a reading of the topic that connects pieces no structured search would put together. The major journals still publish them, because that expert reading, the one that organizes a field and makes sense of it, does not come out of any protocol.

It also has its own yardstick, even if it is rarely cited outside methods circles. The SANRA scale (Scale for the Assessment of Narrative Review Articles, Baethge and colleagues, 2019) rates a narrative review on 6 items, each scored 0 to 2, for a maximum of 12 points:

Justification of the article's importance for the readership
Statement of concrete aims or formulation of questions
Description of the literature search
Referencing (key statements backed by appropriate references)
Scientific reasoning, including the choice of evidence that is cited
Appropriate presentation of data

Look closely at item 3. Not even the quality standard for narrative reviews asks for an exhaustive, reproducible search. It asks you to say how you searched. That is the real border between the two formats, and it is why a narrative review, even an excellent one, cannot underpin clinical decisions or recommendations. Its built-in risk is selection bias. Two experts acting in good faith, reading the same literature, can reach opposite conclusions, and the reader has no way to audit the path either of them took.

What a systematic review requires

The systematic review exists precisely to close that gap. Everything that is the author's judgment in a narrative review becomes a method declared in advance. A structured question, a protocol registered before study selection starts, a search across several databases that someone else could repeat, explicit inclusion and exclusion criteria, screening by two independent reviewers, a formal risk of bias assessment and a synthesis reported according to the PRISMA guideline. The whole process, phase by phase, is in the guide on what a systematic review is.

The price of that transparency is well known. The published average is more than a calendar year (67.3 weeks from registration to publication, according to Borah and colleagues, 2017), and the method needs at least two people for independent screening. In return, the result can be checked, replicated and defended. That is why it is the standard for synthesizing evidence.

Narrative review vs systematic review: the differences, dimension by dimension

The table sets out the full comparison, based on what the methods guidance for each format says and on published data.

Dimension	Systematic review	Narrative review
Starting question	Specific and answerable, framed with a structure such as PICO	A broad topic or overview
Protocol in advance	Defined and registered before starting (PROSPERO)	Usually none
Search	Exhaustive, across several databases, documented and reproducible	Up to the author, no duty to be exhaustive
Study selection	Explicit criteria and screening by two independent reviewers	Implicit, rarely documented
Bias assessment	A standardized tool for every included study	Not done formally
Synthesis	Structured, with meta-analysis when the data allow	Interpretive and qualitative
Reproducibility	High. The method lets you repeat the whole process	Low. Another author would end up with a different selection
Typical time	More than a year on average (Borah 2017)	Weeks or a few months
Minimum team	Two reviewers for independent screening	One person can write it
Reporting and quality guide	PRISMA 2020 (27 items)	SANRA (6 items)

Read from top to bottom, the table tells one story. A systematic review is not a narrative review "done with more effort". Every decision that stays inside the author's head in a narrative review goes into a public document here, before the work begins. The extra effort buys one specific property, verifiability, and that is the property that clinical decisions, practice guidelines and journals with methodological peer review ask for.

If your question calls for a systematic method, you do not have to build that method yourself. In revisia the whole workflow is guided, from the PICO question to the PRISMA diagram. Create your free account.

The systematized review, the middle ground almost nobody mentions

Between the two poles there is a format with its own name that most guides overlook. The classic typology by Grant and Booth (2009), which describes 14 review types, calls it a systematized review. It borrows elements of the systematic method (a structured question, a documented search, explicit selection criteria), but it deliberately stops short of the full standard. It is almost always written by one person, who cannot do dual screening or search every relevant database.

The authors themselves describe it as "typically conducted as postgraduate student assignment". That is where its value lies for an undergraduate thesis or a master's dissertation screened alone. It gives you a defensible method without promising what you cannot deliver. The key is that the limitations are declared, not hidden. "Search in two databases, screening by a single reviewer, no formal bias assessment" is an honest sentence that examiners understand and respect. In practice you declare it in the work itself. Call your review a systematized review in the title or in the methods section, and list there what it does not cover from the full standard. What examiners do not forgive is finding those same limitations when nobody declared them.

If you are weighing more formats (scoping review, rapid review, umbrella review), the full map is in the guide to types of literature reviews.

Which one to choose for your thesis or dissertation

Four questions are enough to place yourself. Answer them in order, and be honest with the second one.

Is your question specific or broad? "Does intervention X work for problem Y?" calls for a systematic method, in health and in any other field. "What is known about topic Z?" calls for a narrative review or, if you want to map the evidence with a method, a scoping review.
How much calendar time do you really have? A full systematic review rarely takes less than several months of steady work. If your deadline is ten weeks away, planning a full systematic review is planning to miss it.
Are you screening alone? Independent dual screening is part of the standard. Without a second person, the most defensible option is a systematized review that says so, not a "systematic" review that keeps quiet about it.
What do your supervisor and examiners expect? Some master's programs expect a systematic review as the final project, and some committees value a well-argued narrative review more. That conversation with your supervisor is worth more than any guide.

To see how each format fits into the structure of the whole project, read the guide to the dissertation literature review.

The mistake that fails you (in both directions)

Worse than picking the wrong format is dressing one format up with the other's name, and it is the most common error.

The classic version is calling a piece of work a "systematic review" when it has no protocol, a single-database search that nobody could repeat and no written selection criteria. A journal reviewer or a methodologist on the committee spots it within minutes, because the first things they look for are the protocol, the search string and the flow diagram. A badly done systematic review is not a narrative review. It is a systematic review you cannot defend.

The reverse also happens. You present as an "expert" narrative review something that actually needed a method, for example when the conclusion of the work is a practice recommendation. If your review ends with "therefore, X should be done", the question was specific and the evidence had to be gathered systematically. The examiner who reads that recommendation will ask how you know the study that contradicts it was not left out, and a hand-picked selection has no answer to that.

The rule is short. The name of the format is a methodological promise, and the reader will check whether you keep it.

From narrative to systematic, step by step

Another way to understand the border is to cross it. Turning a narrative review into a systematic one has little to do with "working harder" and a lot to do with adding four specific pieces, in this order:

Close the question. Go from "what is known about X?" to a question with a defined population, intervention, comparison and outcome. The PICO framework exists for exactly this.
Write the protocol before the formal search. Which databases, which criteria, how disagreements get resolved. Registering it in PROSPERO fixes the rules while they are still cheap to change.
Turn the search into a search string. Go from following references to a documented strategy that anyone can run in the same databases and get the same results.
Screen against written criteria. Each record passes or fails according to the inclusion and exclusion criteria, with the exclusion reason recorded for the flow diagram.

Seen this way, it is also clear what makes the jump expensive: volume. An exhaustive search multiplies the number of records, and screening is the phase that takes the most hours in the whole method. Keep in mind that a tool speeds up that first pass and the checking, but it does not replace the second reviewer of the full standard. If you screen alone with help from artificial intelligence, your format is still a systematized review, declared as such.

That mechanical part is the one you can compress. In revisia artificial intelligence does the first screening pass and you check each decision with the quote from the abstract in front of you, even if you screen alone. Create your free account and try it with your question.

Frequently asked questions

Can a narrative review count as an undergraduate thesis or master's dissertation?

It depends on what your program and your supervisor ask for. As the literature review inside an empirical project, a narrative review is the natural format. As the whole project, many examiners accept it if it is well argued, although more and more master's programs expect at least a systematized review, with a documented search and explicit criteria. Confirm it with your supervisor before you decide on the method, not after.

What is the SANRA scale?

SANRA (Scale for the Assessment of Narrative Review Articles) is a scale published by Baethge, Goldbeck-Wood and Mertens in 2019, in the journal Research Integrity and Peer Review, for rating the quality of narrative reviews. It has 6 items scored from 0 to 2 (12 points maximum), covering the justification of the article, its aims, the description of the literature search, referencing, scientific reasoning and the presentation of data.

What is a systematized review?

It is the named middle ground in the Grant and Booth (2009) typology. It borrows elements of the systematic method, such as a documented search and explicit selection criteria, but it does not meet the full standard, usually because only one person screens. It is a common and defensible format in academic work, as long as its limitations are declared.

How many references does a narrative review need?

There is no published standard number, and be wary of anyone who gives you one. The size depends on the topic and on the journal or degree program. The quality criterion is not quantity but whether the key claims are backed by relevant citations, which is exactly what the referencing item of SANRA scores.

Can I turn my narrative review into a systematic review halfway through?

You can, but it costs more than starting from scratch with a method, because almost everything you have done has to be redone to make it reproducible. The search has to be rewritten as a documented search string, the selection has to be repeated against written criteria, and the protocol has to be fixed before that new selection. If you are considering it, do it as early as possible and register the protocol at that point.

Risk of bias assessment: the Cochrane risk of bias tool (RoB 2), ROBINS-I, AMSTAR 2 and Newcastle-Ottawa

2026-10-12T00:00:00Z

Risk of bias is the likelihood that a study gets the wrong answer because of flaws in its design, conduct or analysis, not because of a real effect. A risk of bias assessment means judging how far you can trust the findings of each study included in a review, and the Cochrane risk of bias tool (RoB 2) is the standard way to do it for randomized trials. Why does it matter so much? A systematic review that pools badly biased studies does not produce reliable evidence. It only adds up the error. Two poorly conducted studies do not cancel each other out; combining them gives the same bias a false look of precision. So before you synthesize anything, you have to appraise the methodological quality of each study with a validated tool chosen for its design.

The standard tools for that job are RoB 2 for randomized trials, ROBINS-I for non-randomized studies of interventions, the Newcastle-Ottawa Scale for observational studies and AMSTAR 2 for published systematic reviews. At the end of the process, GRADE rates the overall certainty of the evidence.

What risk of bias is (and why it matters)

Bias is any systematic error that pushes a study's results away from the true value. Random error shrinks as samples get larger. Bias does not. Adding participants will not remove it, because it is built into the way the study was run. A trial in which patients know which treatment they are getting can overestimate the benefit, and no sample size will fix that.

In a systematic review, assessing risk of bias does three jobs:

It tells you how much weight to give each study. A study at low risk of bias counts for more in the interpretation than one at high risk.
It helps explain heterogeneity. When results disagree, part of the disagreement may come from differences in methodological quality, not from the true effect.
It justifies your confidence in the conclusions. It is the foundation for rating the overall certainty of the evidence with GRADE, the system behind evidence-based medicine.

Three terms often get mixed up. Risk of bias is about how well the study was done (its internal validity). Methodological quality is a broader term that usually covers risk of bias plus other aspects of reporting. And critical appraisal is the general process of reading a paper with a critical eye; risk of bias assessment is the formal, reproducible part of it.

Types of bias (selection, performance, detection, publication)

Modern tools are organized by domains, but it helps to know the main families of bias, because each tool covers them differently.

Selection bias. It appears when the groups being compared differ at baseline in something other than the intervention. In a trial it is controlled by proper, concealed random allocation (nobody can predict or manipulate which group each participant ends up in). Selection bias is the classic threat in observational studies, where groups are not formed at random.
Performance bias. It comes from differences in the care the groups receive beyond the intervention under study, usually because participants or staff know which treatment is being given. Blinding (masking) reduces it.
Detection bias. It happens when the person measuring the outcome knows the allocation and that shapes how they record it, especially for subjective outcomes. Blinding the outcome assessor is the main defense.
Attrition bias. It stems from loss to follow-up and incomplete data when those losses are not spread evenly across groups.
Reporting bias. It occurs when only some outcomes are published, usually the significant ones, and the rest are hidden.
Publication bias. It works at the level of the whole body of studies, not a single one. Studies with positive results get published more often and sooner than negative or inconclusive ones. That distorts any meta-analysis built on the published record. It is explored with methods such as the funnel plot and assessed at the synthesis stage, not with single-study tools.

Each study type has its own instrument, and using the wrong one invalidates the assessment. The tools below split the work by design.

RoB 2, the Cochrane risk of bias tool (randomized trials)

RoB 2 (Risk of Bias 2) is Cochrane's tool for assessing risk of bias in randomized controlled trials (RCTs). It was published in 2019 as an update of the original 2008 version. If your review includes randomized trials, this is the instrument to use.

RoB 2 does not produce a numeric score. Instead it structures the judgment into five domains, and for each one you answer a set of signalling questions that lead to a judgment guided by an algorithm. The five domains are:

Bias arising from the randomization process.
Bias due to deviations from intended interventions.
Bias due to missing outcome data.
Bias in measurement of the outcome.
Bias in selection of the reported result.

Each domain is judged as "Low risk of bias", "Some concerns" or "High risk of bias". The overall judgment follows conservative logic. A single domain at high risk is enough to put the whole trial at high risk. RoB 2 is also applied to a specific outcome rather than to the study as a whole, because the same trial can be at low risk for an objective outcome (mortality) and at high risk for a subjective one (patient-reported pain).

ROBINS-I (non-randomized studies)

When the included studies are not randomized but still estimate the effect of an intervention (for example, cohort studies comparing two treatments assigned in routine practice), the right tool is ROBINS-I (Risk Of Bias In Non-randomised Studies of Interventions), also from Cochrane.

ROBINS-I starts from the "target trial", the ideal randomized trial that would answer the same question. The assessment measures how far the real study strays from it. That is why it includes domains an RCT would not need, above all confounding, the main problem of non-randomized studies. Its seven domains are:

Bias due to confounding.
Bias in selection of participants into the study.
Bias in classification of interventions.
Bias due to deviations from intended interventions.
Bias due to missing data.
Bias in measurement of outcomes.
Bias in selection of the reported result.

The judgment categories go one step further than in RoB 2: low, moderate, serious, critical, or no information. A well-conducted observational study rarely reaches "low risk" in ROBINS-I, because without randomization the door to residual confounding is always open. That is not a flaw in the tool. It reflects a real limitation of the design.

Newcastle-Ottawa Scale (observational studies)

The Newcastle-Ottawa Scale (NOS) is one of the most widely used tools for rating the quality of observational studies, specifically cohort and case-control studies. Unlike RoB 2 and ROBINS-I, the Newcastle-Ottawa Scale does use a star system. It awards points in three blocks (selection of the study groups, comparability of the groups, and ascertainment of the outcome or the exposure), with a maximum of around nine stars depending on the version.

Its main strengths are simplicity and speed, which explain much of its popularity. Its limitations are just as well known. The scoring can hide serious problems (a study with a critical flaw in a single item can keep collecting stars on the others), and the cut-offs for calling a study "good" or "poor" are not well standardized across reviews. That is why, in reviews that aim for the highest methodological rigor, many teams prefer ROBINS-I for observational studies of interventions and keep the Newcastle-Ottawa Scale for exposure-outcome studies or for when the protocol specifies it. Either way, state in advance which tool you will use and why.

In revisia, the risk of bias assessment of your included studies happens in the same workflow and comes out ready for the manuscript. Create your free account.

AMSTAR 2 (quality of systematic reviews)

The tools above appraise primary studies. AMSTAR 2 (A MeaSurement Tool to Assess systematic Reviews, version 2, from 2017) moves up a level and appraises the methodological quality of systematic reviews themselves. You use it when your objects of study are not trials or cohorts but other reviews (for example in an umbrella review), or when you want to judge a published review critically before trusting its conclusions.

AMSTAR 2 has 16 items. It does not produce a total score. Instead it identifies seven critical domains (such as registering a protocol in advance, the adequacy of the literature search, justification for excluding studies, and the risk of bias assessment of the included studies). The number and weight of the weaknesses in those domains set four levels of confidence in the review: high, moderate, low or critically low. A single critical flaw is usually enough to drop confidence to low.

One of its items checks exactly whether the review assessed the risk of bias of its studies. Without that step, no review can hope for a high rating. If you want to go deeper into how the pooled result of a review is interpreted, read our guide on meta-analysis and how to read a forest plot, where publication bias comes back through the funnel plot.

ROBIS and other specific tools

For systematic reviews there is also ROBIS (Risk of Bias in Systematic Reviews), a tool focused specifically on risk of bias in the review process. It is organized into four domains (study eligibility criteria; identification and selection of studies; data collection and study appraisal; and synthesis and findings). ROBIS and AMSTAR 2 complement each other. AMSTAR 2 looks at overall methodological quality and ROBIS at specific risk of bias. Despite the name, ROBIS has nothing to do with similarly named commercial brands.

For particular study types or outcomes there are further tools (for example, the JBI critical appraisal checklists for different designs, or QUADAS-2 for diagnostic accuracy studies). The rule does not change. Pick the tool validated for your design instead of forcing a generic one.

Which tool to use for each study design

A quick reference.

Tool	Study type	What it assesses
RoB 2	Randomized controlled trials (RCTs)	Risk of bias in 5 domains (randomization, deviations, missing data, measurement, selective reporting); judgment per outcome
ROBINS-I	Non-randomized studies of interventions (cohorts, etc.)	Risk of bias in 7 domains, with emphasis on confounding; compares against the "target trial"
Newcastle-Ottawa Scale (NOS)	Observational studies (cohort, case-control)	Selection, comparability and outcome/exposure through a star system
AMSTAR 2	Systematic reviews (methodological quality)	16 items with 7 critical domains; overall confidence (high to critically low)
ROBIS	Systematic reviews (risk of bias in the review process)	4 domains of the review process plus an overall judgment
GRADE	Body of evidence per outcome (not a single study)	Certainty of the evidence on 4 levels (high, moderate, low, very low)

GRADE and levels of evidence

The tools above appraise studies or reviews one at a time. GRADE (Grading of Recommendations, Assessment, Development and Evaluations) works on a different plane. It rates the overall certainty of the evidence for each outcome, that is, how confident we can be that the estimated effect is close to the true one. It is not a single-study risk of bias tool but the final step that combines risk of bias with other factors.

GRADE rates the certainty of the evidence on four levels: high, moderate, low and very low. The starting point depends on the design (randomized trials start at high certainty and observational studies at low), and from there certainty can be rated down for five reasons:

Risk of bias (this is where the results of RoB 2, ROBINS-I and similar tools come in).
Inconsistency (unexplained heterogeneity between studies).
Indirectness (the population, intervention or outcome do not fully match the question).
Imprecision (wide confidence intervals, few events).
Publication bias (suspicion that unpublished studies are missing).

For observational studies, certainty can also be rated up for factors such as a large magnitude of effect or a dose-response gradient. The resulting level of evidence then underpins the strength of clinical recommendations in evidence-based medicine. That is why GRADE closes the process. If you do not appraise each study well, you cannot rate the body of evidence well.

How to report it under PRISMA

Assessing risk of bias is worthless if it is not documented in a transparent, reproducible way. The PRISMA 2020 statement devotes several items to it. You have to state which tool you used, how many reviewers assessed each study independently and how disagreements were resolved, and present the results of the assessment (usually with a traffic light plot by domain for RoB 2 and ROBINS-I, or a star table for the Newcastle-Ottawa Scale). Publication bias is reported separately, usually with a funnel plot and, where appropriate, an associated statistical test.

For the report to follow the guideline and get past peer reviewers and journals, stick to the official structure. You can see how the risk of bias assessment fits into the flow and the full checklist in our guide to the PRISMA 2020 flow diagram and its template.

Risk of bias assessment is one of the core steps of the method. To place it within the whole process, from the question to the synthesis, go back to the main guide on how to do a systematic review step by step.

In revisia you assess risk of bias inside the same workflow as the rest of your review. Create your free account.

Frequently asked questions

Are risk of bias and study quality the same thing?

Not exactly. Risk of bias focuses on flaws in design and conduct that distort the result; quality is a broader concept.

Which tool should I use for observational studies?

ROBINS-I is the usual choice for non-randomized studies, and the Newcastle-Ottawa Scale for cohort and case-control studies.

Who assesses risk of bias?

Ideally two reviewers working independently, who then resolve their disagreements, just as in screening.

What is risk of bias?

Risk of bias is the likelihood that a study's results are distorted by systematic errors in its design, conduct or analysis, rather than reflecting the true effect. Assessing it tells you how much confidence each study deserves within a systematic review. Unlike random error, bias does not go away with a larger sample.

Which tool should I use for clinical trials?

For randomized controlled trials, use RoB 2, the Cochrane tool that covers five domains (randomization, deviations from the intended interventions, missing outcome data, measurement of the outcome and selection of the reported result). For non-randomized studies of interventions (cohort studies, for example) the right tool is ROBINS-I, and for classic observational studies the Newcastle-Ottawa Scale is the usual choice.

What is AMSTAR 2?

AMSTAR 2 is a 16-item tool that appraises the methodological quality of systematic reviews (not primary studies). It identifies seven critical domains and, depending on the weaknesses found in them, rates overall confidence as high, moderate, low or critically low. It is used, for example, to judge published reviews or within an umbrella review.

What is GRADE?

GRADE is the system that rates the overall certainty of the evidence for each outcome on four levels: high, moderate, low and very low. It combines risk of bias with other factors (inconsistency, indirectness, imprecision and publication bias) and is the basis for the strength of clinical recommendations.

Research question frameworks: PICO, PECO, PCC, CoCoPop, PIRD, PEO or SPIDER?

2026-10-12T00:00:00Z

If you want to know whether a treatment works, PICO fits well. If you want to estimate how many people have a condition, it forces you to invent an intervention and a comparator that don't exist. That is where you need CoCoPop. A scoping review usually starts with PCC. For a qualitative synthesis, JBI recommends PICo. SPIDER can help organize the search and the criteria, but it doesn't replace that recommendation in every case. Those are the research question frameworks this guide compares.

Choosing a framework is not about memorizing acronyms. It is about recognizing what answer you are after. That decision shapes the question, the inclusion criteria and the way you search. A wrong box can narrow the review before you find the first article.

This comparison doesn't try to list every possible framework. It focuses on the seven that revisia can suggest today and shows their limits. The tool doesn't yet structure PICo, PICOTS, ECLIPSE or other specialized mnemonics.

Research question frameworks: choose by what you want to find out

I want to find out	Framework	Elements
Whether an intervention produces an effect	PICO	Population, Intervention, Comparison, Outcome
Whether an exposure is associated with an outcome, with a comparison	PECO	Population, Exposure, Comparator, Outcome
What has been studied on a topic	PCC	Population, Concept, Context
How many people have a condition	CoCoPop	Condition, Context, Population
How accurately a test detects a condition	PIRD	Population, Index test, Reference test, Diagnosis of interest
How an exposure influences an outcome without a common reference	PEO	Population, Exposure, Outcome
How people experience or perceive a phenomenon	SPIDER	Sample, Phenomenon of Interest, Design, Evaluation, Research type
How a phenomenon is experienced within a context	PICo	Population, phenomena of Interest, Context

This classification follows the question types covered in the JBI Manual and the methods literature on review types. It is not a hierarchy. A framework with five boxes is not more rigorous than one with three. It fits when its boxes match real decisions in the protocol.

The same decision can show up in different places. CoCoPop can treat the time period as part of the context, PICo includes context in the acronym, and PEO leaves time and setting outside its three letters. That is not a contradiction. Follow the methods guidance you chose and document where you set each limit.

Write the question however it comes out. revisia suggests one of the seven frameworks it offers and flags possible gaps. It can get it wrong, and it doesn't offer PICo yet. You review every field before saving it. See which framework revisia suggests.

PICO to test the effect of an intervention

PICO answers questions about treatment, prevention and other compared interventions. Its elements are population, intervention, comparison and outcome. The key word is intervention. Someone deliberately applies a treatment, program or practice, and its effect is compared with an alternative.

One example would be evaluating the effects of metformin versus sulfonylureas on cardiovascular complications in adults with type 2 diabetes.

Population: adults with type 2 diabetes.
Intervention: metformin.
Comparison: sulfonylureas.
Outcome: cardiovascular complications.

Our guide to the PICO question explains how to tune each element and turn it into a search. PICO stops fitting when the question contains no intervention, or when the goal is to map a field, estimate frequency or understand experiences.

PECO to study an exposure with a comparison

PECO swaps the intervention for an exposure. It is used to study associations in etiology or risk when there is a reference group, level or condition that matters from the protocol stage.

Imagine you want to determine the association between long-term exposure to traffic noise and high blood pressure in the adult urban population, compared with living in low-noise areas.

Population: adult urban population.
Exposure: long-term traffic noise.
Comparator: living in low-noise areas.
Outcome: high blood pressure.

The exposure is not assigned like a treatment. It is observed. That difference separates PECO from PICO, even though both have four boxes and share population, comparison and outcome.

PCC to map a field

PCC is usually used in scoping reviews. The goal is to describe what evidence exists, how a topic has been studied and where the gaps are. These questions tend to be broader than effectiveness reviews, but they can also have precise limits.

For example, you could map the evidence on the use of telemedicine for people receiving palliative care in rural areas.

Population: people receiving palliative care.
Concept: use of telemedicine.
Context: rural areas.

JBI recommends PCC for scoping review questions and asks you to align those elements with the objective, eligibility, screening and extraction. It also acknowledges that some questions can drop an element or need another framework. The current JBI guidance on PCC lets you check those nuances.

PCC doesn't turn every loosely defined topic into a scoping review. First you have to decide whether you want to map a field or answer a narrow question about effects, associations or frequency.

CoCoPop for prevalence and incidence questions

CoCoPop structures questions that measure how often a condition occurs in a population and a context. The JBI Manual proposes it for systematic reviews of prevalence and incidence.

One example would be estimating the prevalence of iron-deficiency anemia in women of reproductive age seen in primary care.

Condition: iron-deficiency anemia.
Context: primary care.
Population: women of reproductive age.

There is no intervention or comparator here. Nor does the condition work like a PICO outcome. It is the thing whose frequency you want to estimate. If you are studying incidence, you also need to define the period in which new cases are counted, even though time is not one of the letters in CoCoPop.

PIRD for diagnostic accuracy

PIRD is for when you want to know how a test performs against a method that establishes the reference condition. A clean example is determining the accuracy of a rapid antigen test versus laboratory RT-PCR for detecting influenza in adults with acute respiratory symptoms.

Population: adults with acute respiratory symptoms.
Index test: rapid antigen test.
Reference test: laboratory RT-PCR applied to the whole sample.
Target condition: influenza virus infection.

A different question would be whether using ultrasound reduces unnecessary procedures or speeds up care. There you are measuring the clinical effect of a practice, and PICO may be the right fit. PIRD measures accuracy. It is not enough for a diagnostic test to appear in the sentence.

PEO for an exposure without a comparison box

PEO uses population, exposure and outcome. It fits when you don't want to fix a comparison box in the question.

For example, determining the association between different levels of exposure to traffic noise and high blood pressure in the adult urban population, without imposing a common reference category on every study.

Population: adult urban population.
Exposure: traffic noise.
Outcome: high blood pressure.

The line between PEO and PECO is drawn in the protocol. If it requires a specific reference group or level, use PECO. If it doesn't reserve a box for that reference, PEO saves you from inventing one. The studies still need variation in the exposure and some analytic contrast, such as categories, dose-response or continuous change.

PEO is not the full standard for a formal prognosis review. The 2026 Cochrane handbook for prognosis reviews recommends PICOTS, which adds the index prognostic factor or model, comparator, timing and setting.

SPIDER for experiences and perceptions

SPIDER was created as an alternative tool for turning qualitative and mixed-methods questions into a search strategy. It has since also been used to organize questions and criteria through sample, phenomenon of interest, design, evaluation and research type.

One example would be exploring perceptions of family support among people with diabetes, drawing on qualitative interview studies.

Sample: people with diabetes.
Phenomenon of Interest: family support.
Design: interviews.
Evaluation: perceptions.
Research type: qualitative studies.

JBI recommends PICo for framing qualitative synthesis questions. SPIDER can add detail on design and study type, but it doesn't require you to turn its five boxes into five blocks joined with AND. In a published comparison of SPIDER and PICO, SPIDER produced a more specific search but missed five relevant studies in the case analyzed.

The three decisions people mix up most

PICO or PECO

Ask who determines the central element. If the research team applies a treatment, program or practice, start with PICO. If it observes an exposure such as pollution, smoking or noise, start with PECO.

PECO or PEO

Look for a comparison that is part of the question. It could be an unexposed population, another level of exposure or a reference condition. If there is one and it determines eligibility, use PECO. If you don't want to fix it in advance, use PEO. Dropping the C doesn't remove the analytic contrast that lets you estimate an association.

PICO or PCC

Look at the kind of final answer. PICO tries to estimate an effect between alternatives. PCC aims to describe the extent, characteristics and gaps of a field. A broad PICO question doesn't become PCC just because it is poorly focused.

Paste your question in plain language. revisia suggests PICO, PECO, PCC, CoCoPop, PIRD, PEO or SPIDER. It can get it wrong, and it doesn't replace the methods guidance behind your protocol. You can correct or reject every field. Try my question for free.

The framework is not the search strategy

The framework does three related jobs. It helps you write the question, anticipates the eligibility criteria and supplies concepts for the search. It doesn't require you to search for every element.

An intervention search usually works best with population and intervention. Adding outcome and comparator can remove relevant articles, because they don't always appear in the title or abstract. In SPIDER, requiring design, evaluation and research type can cause the same problem. In PIRD, the test and the condition usually supply the core concepts.

The sensible order is this:

Write the full question.
Turn each element into clear criteria.
Test several search combinations.
Check whether you retrieve studies you already know about.
Document the final version so it can be reproduced.

The search strategy guide explains how to combine controlled vocabulary, synonyms and operators without confusing the question with the search string.

Templates for writing your question

PICO Evaluate the effect of [intervention] versus [comparison] on [outcome] in [population].
PECO Determine the association between [exposure] and [outcome] in [population] compared with [comparator].
PCC Map the evidence on [concept] in [population] within [context].
CoCoPop Estimate the prevalence or incidence of [condition] in [population] within [context].
PIRD Determine the accuracy of [index test] versus [reference test] for detecting [target condition] in [population], with [clinical role and threshold] where relevant.
PEO Determine the association between [exposure] and [outcome] in [population] without fixing a common reference category.
SPIDER Explore [evaluation] of [phenomenon] in [sample] through [design] in [type] studies.
PICo Explore [phenomena of interest] in [population] within [context]. revisia doesn't offer this template yet.

These templates help you get started. Next, turn the question into a declarative objective and check that every term decides something real in the protocol.

Frequently asked questions

Which framework is used for a systematic review?

It depends on the question. PICO fits interventions, PECO and PEO fit exposures, CoCoPop fits frequency, PIRD fits diagnostic accuracy and PCC fits scoping reviews. JBI recommends PICo for framing qualitative syntheses. SPIDER can help organize their search and eligibility.

Can I use PICO for any question?

PICO works well for the effects of interventions. Forcing it onto prevalence, experiences, scope or diagnostic accuracy creates artificial boxes. In those cases, choose a framework built for the answer you are looking for.

What is the difference between PICO and PCC?

PICO compares the effect of an intervention on an outcome. PCC organizes population, concept and context to map a field in a scoping review. The difference lies in the purpose of the review.

What is the difference between PECO and PEO?

PECO keeps a box for the comparison. PEO doesn't. Use PECO when a specific reference group or level is part of the protocol. PEO lets you accept different contrasts across studies, but it doesn't remove the need to compare levels, categories or changes in the exposure.

Do I have to use every part of the framework in the search?

No. The framework should appear in full in the question and the criteria, but a search can leave out the comparator, outcome, design or other parts that cut sensitivity too much. You need to test the strategy and check that it retrieves relevant studies.

How do I know if I chose the right framework?

Read the question without the acronym and say what answer you expect. An effect, an association, a map of the field, a frequency, the accuracy of a test, a prognosis or an experience. If the boxes describe that answer without inventing elements, the framework probably fits.

Methods sources consulted in September 2026. Living manuals, such as JBI and Cochrane, may update their recommendations.

Start with your own sentence. revisia suggests a framework and flags possible gaps. It can get it wrong, and you review every field before saving it. The suggestion doesn't replace the methodological decisions in your protocol. Check my question for free.

Rayyan vs Covidence in 2026: which screening tool to choose

2026-10-12T00:00:00Z

Screening is the phase of a systematic review in which you decide which studies make it into your analysis and which ones are dropped. After the literature search you end up with hundreds or thousands of references, and you can rarely read every full text. Rayyan and Covidence exist to organize that work, split it between reviewers and keep a record of every decision.

The comparison no longer comes down to "Rayyan is free and Covidence is complete". Rayyan offers paid plans, data extraction, a risk of bias tool and a PRISMA diagram. Covidence is still more expensive, but it brings paired screening, dual extraction, quality assessment and PRISMA reporting together in one workflow. The choice depends on the size of the team, the budget and how much of the work you want to keep inside the platform.

If you want to see screening within the whole process, go back to the guide to the systematic review step by step, where it fits as one of the core phases of the method.

What screening is in a systematic review

Article screening is the process of study selection: starting from the full set of references retrieved by the search, you filter until you keep only the studies that answer your research question and meet your eligibility criteria.

Selection is always made against the inclusion and exclusion criteria you defined in advance (population, intervention, study design, language, dates and so on). It is not about discarding on instinct but about applying explicit, reproducible rules, so that any other researcher would reach the same result with your criteria.

Screening matters for two reasons:

Rigor: what sets a systematic review apart from a narrative review is precisely the transparency of the selection process. Anyone who reads your work must be able to reconstruct why each study ended up in or out.
Traceability: every exclusion counts. The number of references imported, the duplicates removed and what is discarded at each phase is what feeds the PRISMA flow diagram, the chart that documents all the filtering visually.

That is why "cleaning up the list" falls short as a description: screening is a methodological phase with its own rules, two separate steps and, almost always, two reviewers working in parallel.

Title and abstract screening vs full-text screening

Screening happens in two successive phases. Keep them separate, because they have different goals and different levels of strictness.

Phase 1: title and abstract screening

Title and abstract screening is the first filter. Here you only read the surface information of each reference (the title and, if there is one, the abstract) and decide quickly: could this study be relevant, or is it clearly not? It is also the part that takes the most hours in the whole review. The guide on how long a systematic review takes has the measured times, phase by phase.

What this phase looks like:

It is fast and high volume: you can go through hundreds of records in a session.
The criterion is lenient: when in doubt, include. If an abstract does not give enough information to rule the study out, it moves on to the next phase. A false positive (a study that will be excluded later) is better than losing a relevant one.
You do not need the full document, which makes it ideal for tools like Rayyan that import references from a RIS file or straight from the database.

Phase 2: full-text screening

In the second phase you retrieve the full text of the studies that survived the first filter and read them closely to confirm they meet every eligibility criterion.

What this phase looks like:

It is slower and more demanding: now you do read the methods, population and results.
The criterion is strict: every study you exclude at this phase must carry a recorded reason for exclusion (for example, "non-randomized design", "population outside the age range", "outcome not measured"). Those reasons are required for the PRISMA diagram.
This is where access becomes a problem: you will not always have the PDF at hand. A reference manager such as Mendeley or Zotero helps you find and store full texts before this phase.

The difference between the two phases: at title and abstract you filter for quantity; at full text you confirm eligibility and document why you drop each study.

Blinded dual screening

In both phases, the methodological standard is duplicate screening with decisions hidden between reviewers (usually called "blinded" or "dual" screening). Two reviewers screen independently, without seeing each other's decisions, and then compare results. Disagreements are resolved by consensus or with a third reviewer.

Why do it this way? Because study selection introduces bias when a single person does it. Blinding reduces reviewer bias and lets you calculate the agreement between the two (for example, with Cohen's kappa). Both Rayyan and Covidence are designed to hide the other reviewer's decisions during screening and reveal them only at the conflict resolution stage.

Rayyan: what the free plan includes today

Rayyan started out as a screening tool at the Qatar Computing Research Institute (QCRI). That is still what it is best known for, but the current platform also includes full-text screening, data extraction, risk of bias assessment and a PRISMA diagram.

The workflow starts outside Rayyan. First you search PubMed, Scopus or other bibliographic databases and export the references. Then:

You import RIS, EndNote, CSV or BibTeX files.
Rayyan detects duplicates and organizes the references for screening.
Each reviewer marks "Include", "Exclude" or "Maybe". Blind mode hides decisions until the team resolves the conflicts.
The studies that pass move on to full-text screening. You can attach the PDF and record the reason for each exclusion.
The platform lets you build extraction forms and assess the risk of bias of randomized trials with RoB 2.

According to Rayyan's pricing page, the Free plan allows three active reviews, two invited reviewers, one sample and relevance predictions. That AI is a classifier that ranks references based on your earlier decisions. It helps you prioritize, but it does not replace the reviewer's vote.

There are two limits worth knowing before you start:

The PRISMA diagram is not included in Free. It is available from Essential, which costs US$4.99 per seat per month billed annually and also adds automatic duplicate resolution and AI-assisted PICO extraction. Advanced costs US$8.33 on the same terms.
ResearchPilot, the suite that includes AI Reviewer and automated extraction, is reserved for institutional contracts. The official getting started guide separates this suite from the relevance predictions in the free plan.

Rayyan also has structured data extraction and a risk of bias tool. For the latter, Rayyan announces support for RoB 2, ROBINS-I, QUADAS-2, Newcastle-Ottawa, JBI and AMSTAR 2, plus its own templates. The interface and the help center are in English.

Covidence: an integrated workflow priced per review

Covidence covers import, title and abstract screening, full-text screening, extraction, quality assessment and PRISMA export. It is part of Cochrane's review production environment, although it is also used outside Cochrane reviews.

Its usual workflow is this:

You import the references. Covidence detects duplicates and keeps a history of the sources.
Two reviewers vote independently at title and abstract. Conflicts go to consensus.
At full text you attach the PDFs, record the reasons for exclusion and resolve disagreements again. Covidence tries to retrieve some open-access PDFs automatically.
Extraction can be done in duplicate with customizable templates and a consensus step.
Quality assessment can be built into the template. Covidence's current documentation says its standard template is based on the original Cochrane risk of bias tool (RoB 1) and that Extraction 2 lets you customize it. It should not be described as a native RoB 2 form.
The PRISMA diagram updates as the project moves forward and can be downloaded as a DOCX file so you can correct any fields that need manual adjustment.

The public price is US$339 for one review for twelve months, with unlimited collaborators. The three-review package costs US$907 a year. The free trial allows up to 500 records; it is not a free full review.

Before you pay, ask your university library. Covidence sells institutional licenses, and many organizations let their members create reviews under that subscription.

If you would rather screen with automatic PRISMA traceability, in revisia artificial intelligence speeds up screening and every uncertain call stays in your hands. Create your free account.

Rayyan vs Covidence (table)

Rayyan is the budget option for getting started with screening. Covidence costs quite a bit more and pays off when the team needs a stable dual-reviewer workflow, from screening through extraction and quality assessment, or already has an institutional license.

Feature	Rayyan	Covidence
Public price	Free; Essential from US$4.99/month per seat, billed annually	US$339/year for one review; 500-record trial
Entry limit	Free: 3 active reviews and 2 invited reviewers	One review and unlimited collaborators on the Single plan
Database searching	No; imports files from other sources	No; imports files from other sources
Title and abstract screening	Yes	Yes
Full-text screening	Yes, with PDFs and exclusion reasons	Yes, with PDFs, reasons and retrieval of some open-access papers
Blinded dual screening	Yes	Yes
Duplicates	Detection on Free; automatic resolution from Essential	Automatic detection and quarantine
AI assistance	Relevance prediction on Free; PICO from Essential; ResearchPilot for institutions	RCT classifier and occasional extraction suggestions
Data extraction	Structured forms; automation with ResearchPilot	Customizable templates, dual extraction and consensus
Risk of bias assessment	RoB 2, ROBINS-I, QUADAS-2 and other tools, according to Rayyan	RoB 1 template and customizable quality assessment
PRISMA diagram	From Essential; mixes automatic and manual fields	Included; updates with the workflow and exports to DOCX
Backing / endorsement	QCRI (origin)	Cochrane
Interface language	English	English
Best fit if...	You want to screen on a small budget and can handle the missing steps elsewhere	You need paired reviewer coordination in a single environment or have an institutional license

If you choose Rayyan Free, decide from the start where you will build the PRISMA diagram and how you will handle extraction and the quality tool that matches your study designs. If you choose Covidence, check your institution's license first, and remember that it does not run the literature search either: the references still come from external databases.

How to record exclusions for the PRISMA diagram

Having the list of included studies is not enough to close screening: you also need to be able to explain every exclusion. The PRISMA flow diagram requires you to report how many records came in, how many were duplicates, how many were excluded at title and abstract screening and how many at full text, the latter with their reasons.

To make those numbers add up at the end, record them as you screen:

Total records identified from each database and other sources.
Duplicate records removed (Rayyan and Covidence detect and count them).
Records excluded at title and abstract (you do not need an individual reason here, but you do need the total).
Full-text reports assessed for eligibility and, of those, the ones excluded with reasons grouped by category (design, population, language, outcome, etc.).
Studies included in the final synthesis.

Covidence updates the diagram as you go: it records imported references, duplicates, decisions and exclusion reasons. Its PRISMA export guide notes that the DOCX may still need adjustments, for example if there were studies whose full text could not be retrieved.

Rayyan generates a PRISMA 2020 diagram on its paid plans. According to its PRISMA documentation, it fills in duplicates, records screened and excluded, reports sought for retrieval and exclusion reasons automatically. You have to enter by hand the databases searched, reports not retrieved, reports assessed for eligibility and studies included. Rayyan also pools the exclusion reasons from title/abstract and full-text screening, so you may need to separate those counts before publishing.

Once you have the numbers, carry them over to the final chart by following the guide to the PRISMA flow diagram and its editable template, where they slot into the identification, screening and included boxes.

In revisia the AI proposes the screening decision and you settle every uncertain one, with everything audited. Create your free account and start screening.

Frequently asked questions

How many phases does screening have?

Two: first by title and abstract, then by full text for the records that passed the first phase.

What happens if the two reviewers disagree?

Disagreements are resolved by consensus or by a third reviewer. It is part of the method and should be recorded.

Does screening AI decide for me?

It should not. A good tool prioritizes or suggests, but decisions and uncertain cases are settled with human judgment and recorded.

What is Rayyan?

Rayyan is a platform for managing systematic reviews. It imports references retrieved from other databases, detects duplicates and lets you screen by title, abstract and full text. It also includes data extraction, RoB 2, a PRISMA diagram on paid plans and various AI features depending on the plan. It was originally developed at the Qatar Computing Research Institute (QCRI).

Is Rayyan free?

Yes, with limits. Rayyan Free allows three active reviews, two invited reviewers, one sample and relevance predictions. The PRISMA diagram, automatic duplicate resolution and other features start on the paid plans. ResearchPilot, which includes AI Reviewer and automated extraction, is reserved for institutional contracts.

Rayyan or Covidence?

Choose Rayyan if you want to start on a small budget and its limits on reviews and collaborators fit your project. Choose Covidence if you need dual extraction, consensus and an integrated workflow, or if your institution already pays for the license. Both import references from external databases; neither replaces the literature search.

How does blinded dual screening work?

In blinded dual screening, two reviewers screen the same references independently, without seeing each other's decisions. Both Rayyan and Covidence let you turn on a "blind" mode that hides the rest of the team's votes during screening. Once both have finished, the tool shows the conflicts (the studies where they disagree), which are resolved by consensus or by a third reviewer. This procedure reduces reviewer bias and lets you measure inter-rater agreement.

PROSPERO registration: what it is and how to register your protocol

2026-10-12T00:00:00Z

Before you start screening studies, you should write down what you are going to do and register it. PROSPERO is the international register where systematic review protocols are published, and PROSPERO registration is one of the signs of rigor your examiners and peer reviewers will look for. The process is simpler than it looks. You open an account, fill in a form of about 40 fields in English and pass an administrative check. And if your review falls outside its remit, there are other places to register.

What PROSPERO is

PROSPERO (International Prospective Register of Systematic Reviews) is a free public register, run by the Centre for Reviews and Dissemination at the University of York and funded by the NIHR, the UK's National Institute for Health and Care Research. Teams publish the protocol of their systematic review there before carrying it out. You declare your question, criteria and methods in advance, so anyone can check later that you did not change the rules halfway through.

It accepts systematic reviews relevant to health and wellbeing (including nursing, physiotherapy, public health and clinical psychology). Scoping reviews and literature reviews are not accepted in PROSPERO, and purely methodological reviews do not always fit. For those cases there are alternatives (see below).

Why register the protocol (and when)

Registering the protocol gives you three things:

Transparency and bias control. By fixing the question and the inclusion and exclusion criteria before you see the results, you avoid the bias of adjusting the methods until you get the answer you expected.
Avoiding duplication. Before you start, you can check whether someone is already doing the same review.
Proof of rigor. Many journals, and the PRISMA 2020 statement itself, ask for the protocol's registration number.

The time to register is before screening starts (ideally after your test searches, but before you select studies). Registering once you have already extracted data defeats much of the purpose.

How to complete your PROSPERO registration step by step

Create an account on the PROSPERO website (crd.york.ac.uk/prospero) with your email address, preferably your institutional one.
Start a new registration and complete the form (about 40 fields, many of them optional).
Define the question using the PICO model: population, intervention/exposure, comparator and outcomes.
Describe the search. The databases you plan to use and, if you have it, the search strategy with its terms.
Set out the methods. Eligibility criteria, screening by two reviewers, data extraction and risk of bias assessment.
Submit the registration for the administrative check. PROSPERO verifies that it meets the basic requirements (it does not review the scientific quality of the protocol).

What the form asks for

The required fields cover the title, details of the team and the organization, the planned start and end dates, the research question, the eligibility criteria, the information sources, the screening and selection method, and the plan for risk of bias assessment and synthesis. You can save a draft and keep filling it in; there is no need to finish it in one go.

In total the form has about 40 fields, and roughly half are required; the rest can be left blank if they do not apply. Grouped, this is what it will ask you for:

Identification and team. Review title, contact details, team members with their affiliations and the organization responsible.
Funding and conflicts. Sources of funding or sponsorship and a declaration of conflicts of interest.
Question and scope. The condition or health problem studied, population, intervention or exposure, comparator, and primary and secondary outcomes (your PICO, moved over field by field).
Methods. Planned information sources, eligibility criteria, selection process, data extraction plan, risk of bias assessment and synthesis strategy, stating whether you plan a meta-analysis and which subgroup analyses.
Project status. The stage the review is at when you submit and the planned start and end dates.

A practical note if English is not your first language. PROSPERO works in English. Both the form and the published record are written in English, so set aside time to translate the protocol accurately (you will already have the terms of your PICO in English from preparing the search).

How long it takes and what happens next

The waiting time varies with the register's workload. At some times the administrative check takes days, and at others it takes several weeks or more. The PROSPERO website itself posts the current estimated time, so submit the protocol with time to spare and do not leave registration until the end. While you wait you can prepare the final searches; what you should not do is start screening before you have submitted the registration. Once accepted, your record gets a public PROSPERO number (of the form CRD42026XXXXXX) that you must cite in the manuscript. The record is not set in stone. You can and should update it if the methods change, and each change is dated, which keeps the audit trail. Keep in mind that this wait is only the first one in the project. Our guide on how long a full systematic review takes gives you the timeline phase by phase.

Your PICO, your criteria and your plan. What you declare in PROSPERO is the first thing you fix in revisia when you create your project, so the methods are locked from day one. Create your free account.

Search for reviews already registered (avoid duplication)

Before registering, search PROSPERO with keywords for your topic. If you find an ongoing review that is very similar, you save months of duplicated work, or you can reshape your question to add something different. It is also a good source for spotting evidence gaps.

Use the search tool on the PROSPERO website (it has a simple search and an advanced search by field). Search in English, the language of the records, combining synonyms for your population and your intervention. Look at the status of each record (ongoing, completed or discontinued) and at its dates. Finding a similar review does not mean you have to give up. Compare its scope with yours (population, comparator, outcomes) and judge whether your proposal offers an update or a different angle.

Alternatives to PROSPERO

If your review does not fit PROSPERO (for example, a scoping review or a methodological review), you have other ways to register the protocol publicly:

OSF (Open Science Framework). Free, with immediate time-stamped registration; it accepts any type of review (scoping reviews included) and lets you attach materials such as the search strategy.
INPLASY. A registry specifically for review protocols, with fast publication (paid).
Publishing the protocol in a journal. Publishing it as an article (for example in BMJ Open or Systematic Reviews) gets it peer reviewed.

Does PROSPERO accept student reviews?

PROSPERO does not set a minimum academic level in its criteria. What it assesses is whether the project is a systematic review within its remit (health and wellbeing) and at an early stage, not who signs it. In fact, records linked to PhD theses and master's dissertations are common. That said, because of the volume of submissions it receives, the register has adjusted its acceptance conditions in recent years. Check the current criteria on its website before submitting and, if you are a student, ask your supervisor to review the protocol before you send it.

For an undergraduate thesis or a master's dissertation, plan for two things. First, timelines. If the administrative check will not come back in time for your academic calendar, you can register the protocol on OSF (it is immediate) and cite that registration in your dissertation. Second, language. The form and the published record are in English. And if your review is the core of the project, our guide on the literature review in a dissertation explains how to fit it into the write-up.

Once the protocol is registered, you run the full review in revisia. Create your free account and get started.

Frequently asked questions

How long does PROSPERO take to accept my registration?

It depends on the register's workload. It can take days or several weeks. Check the estimated time posted on the PROSPERO website and submit with time to spare. The check is administrative (basic requirements), not a review of scientific quality.

Can I register a scoping review in PROSPERO?

No. PROSPERO does not accept scoping reviews. To register one, use OSF or another protocol registry.

What is the CRD number?

It is the public identifier PROSPERO assigns to your protocol (CRD followed by digits), and you must cite it in the manuscript.

Is PROSPERO registration mandatory?

It is not a legal requirement, but it is strongly recommended and more and more journals and examiners expect it. PRISMA 2020 asks you to state where the protocol is registered and its registration number.

Is PROSPERO free?

Yes. Creating an account and registering a protocol in PROSPERO is free.

Can I register if I have already started?

PROSPERO accepts registrations up to the point before data extraction starts. If you have already extracted data, registration loses its preventive purpose; be transparent about it in the manuscript.

What happens if I change the methods after registering?

You update the record. Changes are dated, so the audit trail is kept, and it is not considered bad practice as long as it is documented.

Does PROSPERO accept student reviews (undergraduate or master's)?

It does not require a minimum academic level. It accepts the review if it falls within its health remit and is at an early stage. Check the current criteria on its website and, if the timelines do not fit your academic calendar, register the protocol on OSF.

Protocol registration is one more stage of the method. To place it within the whole process, go back to the guide on what a systematic review is and how to do one step by step.

PRISMA 2020 flow diagram: free generator, template and guide

2026-10-12T00:00:00Z

The PRISMA flow diagram is the chart that shows how many records you identified, how many you screened, how many you excluded and why, and how many studies ended up in the synthesis. It is the most recognizable part of the PRISMA statement and the part reviewers look at first, because those numbers show whether the selection can be reproduced.

How you get yours depends on where you are:

Your review is done and you only need the drawing. Use our PRISMA 2020 flow diagram generator. You type your numbers, watch the diagram update as you type and download it as PDF, SVG, PNG or Word. It is free and needs no account. Downloads carry a small revisia mark in the corner, which goes away if you sign in. Ten minutes and you are done.
You are just starting the review. Then it pays to run the review in revisia, because the diagram builds itself from the real counts of your screening. You never copy a number by hand, so the counts always add up.

The rest of this article covers what PRISMA is, what changed in 2020, what goes in each box of the diagram and how the 27-item checklist fits in. If you want the full process, this article is part of our guide on how to do a systematic review step by step.

What the PRISMA statement is

PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses. It is not a method for doing a review. It is a standard for reporting what you did. The difference matters. PRISMA does not tell you how to design the search or how to analyze the data. It tells you what information to include so any reader can understand, assess and reproduce your work.

The PRISMA statement appeared in 2009 as the successor to the QUOROM guideline. Today most scientific journals that publish systematic reviews require it. When a journal asks for your manuscript to "follow PRISMA", it means two things:

The PRISMA checklist, with the items the text of the article must contain (title, structured abstract, methods, results and so on).
The PRISMA flow diagram, the visual outline of the study selection process, which sums up in numbers the whole path from identification to inclusion.

What readers get from PRISMA is the ability to see how many studies came in and went out at each step, and to repeat the process if they want to check it. For you, following it is in practice a condition for publishing in indexed journals and a quality criterion when undergraduate theses, master's dissertations and PhD theses are graded.

A common mix-up is treating PRISMA as an assessment of study quality. Reporting in line with PRISMA is separate from judging the risk of bias of the included studies, which you do with dedicated tools such as RoB 2 or AMSTAR 2. A review can be perfectly reported under PRISMA and still include low-quality studies. PRISMA only ensures you report what you did, not that what you did was good.

PRISMA 2020 vs PRISMA 2009 (what changed)

PRISMA 2020, the update that replaces the 2009 version, was published in 2021. It came because review methods had changed over those ten years. Teams search in more places, use automation tools and are expected to document better why each study was excluded.

These are the changes to know before you pick your template.

Aspect	PRISMA 2009	PRISMA 2020
Checklist items	27 items	27 items (rewritten, with more detailed sub-items)
Diagram phases	Identification, screening, eligibility, included (4 boxes)	Identification, screening, included (3 phases)
Sources of records	A single entry path	Separate paths for databases, registers and other sources (websites, citation searching, contacting authors)
Documenting exclusions	Generic	Detailed reasons for exclusion at full-text screening
Abstract	Recommended	A dedicated abstract checklist (PRISMA for Abstracts)
Automation	Not addressed	Distinguishes between screening by humans and by automation tools

The most visible change is in the structure of the diagram. PRISMA 2009 used four labels (identification, screening, eligibility, included), and PRISMA 2020 regroups them into three phases (identification, screening, included). Above all, it splits the sources of records into two columns. The diagram comes in four variants, depending on whether it is a new review or an update and whether the search covered only databases and registers or other methods too.

Unless a journal asks for the older version, use PRISMA 2020. It is the current one and the one reviewers expect in 2026.

The PRISMA flow diagram step by step

The PRISMA 2020 flow diagram is organized in three phases, each with its own boxes and counts. Keep your tally sheet at hand. Every number in the diagram must be traceable to a real record in your reference manager or your screening tool.

Phase one, identification

Here you record everything you retrieved, before you look at anything in detail.

Records identified from each database and register searched (PubMed, Scopus, Web of Science, Cochrane and so on), added together. Report the total and, if your template allows it, the breakdown by source.
Records identified from other methods. Organization websites, reference lists of included studies, contacting authors or hand searching. In PRISMA 2020 these go in a separate path.
Duplicate records removed before screening. This is the number reviewers audit most, so document how you found them (by hand or automatically with your reference manager).

Phase two, screening

This phase has two sub-steps, and it is where most exclusions happen.

Records screened on title and abstract. This is the total after removing duplicates.
Records excluded at title and abstract screening. A number is enough here (detailed reasons are not required, because these records are usually plainly irrelevant).
Reports sought for retrieval. The records that passed the first filter and for which you tried to get the full text.
Reports not retrieved. Those whose full text you could not get (no access, no reply from the author and so on).
Reports assessed for eligibility in full text.
Reports excluded at full text, with reasons. This is the requirement PRISMA 2020 tightened. Every exclusion at this stage carries its reason, for example "wrong study design, n=12", "different population, n=8" or "language, n=3". This is where your inclusion and exclusion criteria turn into justified numbers.

Phase three, studies included

Studies included in the review and the number of reports linked to those studies (one study can have several publications). This is the number that feeds the qualitative synthesis and, where appropriate, the meta-analysis. If yours looks small, check the data on how many studies a review usually includes. The typical Cochrane review includes six.

For the diagram to add up, at every step the records that come in minus those excluded must equal those that move on. If the counts do not match, fix them before you submit. Reviewers check them with a calculator.

With the numbers in front of you, the drawing takes ten minutes. Our PRISMA 2020 flow diagram generator is free, needs no account and warns you when two counts contradict each other. Make my PRISMA diagram.

How to make a PRISMA diagram for free (without drawing anything)

Drawing the diagram by hand in Word is the slow way to do it, and the one that leaves the most counting errors. If you already have your numbers, these are the three routes, from most to least convenient.

Our free generator works in the browser with no account

Make a PRISMA diagram is the tool we built for this. You type the counts and the diagram appears as you type, so you see the result before you commit to anything. It needs no sign-up and costs nothing. Downloads carry a small revisia mark in the corner, which goes away if you sign in.

What it does for you:

It checks the arithmetic. If the records screened do not match the records identified minus duplicates, or if the reasons for exclusion do not add up to the reports excluded, it tells you which number contradicts which. It warns, it does not block. Some gaps have a legitimate explanation, and sometimes you want to save a half-finished diagram.
It includes the four official PRISMA 2020 templates. New review or update, with or without the column for other methods. You switch between them with a drop-down and a checkbox, without redoing anything.
It exports to PDF, SVG, PNG and Word. PDF or SVG when the journal asks for vector artwork, PNG at 4x when the submission system wants a high-resolution image, and Word to embed the figure in the manuscript.
It speaks two languages. The labels come out in English or Spanish, depending on what your journal or university asks for. Both use the wording of the published PRISMA 2020 standard.
It does not upload your data anywhere. Everything happens inside your browser. Your counts never travel to our servers or anyone else's.
It reads and writes the CSV format of the PRISMA2020 R package, in case you work with that tool or want to keep the numbers separately.

The official PRISMA2020 app (Shiny and R package)

The team behind the statement built it. You enter the numbers into a template and it generates an interactive diagram you can export to PDF, PNG or an editable format. It is free and the closest to the official version. If you already work in R, you will have no trouble with it. If you do not, the interface is designed for people who do, and it shows.

The Word or PowerPoint template

It is still useful when your journal or university gives you its own template and you need to match its design. It is the manual route, so you check the arithmetic yourself. It is covered in the next section.

None of the three makes up numbers. The diagram reflects your data, so what decides its quality is screening that is well documented from day one. If you have not reached that point yet, jump to the last section.

PRISMA flow diagram template (Word / editable)

If you prefer the manual route, the most practical start is a PRISMA flow diagram template in Word (or PowerPoint) with the boxes already drawn, where you only replace the numbers and the reasons for exclusion with your own.

The official templates are on the statement's own website: PRISMA 2020 flow diagram templates in Word, free, editable and licensed under CC BY 4.0 (which lets you adapt them as long as you credit the source). They include:

The full PRISMA 2020 flow diagram with the three phases (identification, screening, included) and the two source paths.
Boxes with the "(n = )" placeholder in each count, so you know which number goes where.
One version for new reviews and another for updated reviews.
Variants with and without the "other methods" path, depending on how you ran the search.

To fill it in, download the Word version, replace each "(n = )" with your final counts and delete the boxes that do not apply. The official templates are in English, so the box labels already use the wording English-language journals expect. The PRISMA 2020 27-item checklist is a separate download on the same website.

Rather not edit boxes by hand? Our PRISMA flow diagram generator offers the same four templates, checks that the numbers add up and exports to Word too, free and with no account.

Fill in the template at the end, once you have the final counts from your reference manager and your screening tool. Doing it before screening is closed almost always means doing it again.

PRISMA 2020 checklist (27 items)

The PRISMA 2020 checklist makes sure the text of your article reports everything it needs to. It is organized in seven sections that follow the usual structure of a scientific article. These are the 27 items, grouped.

Title (item 1)

Title. Identify the report as a systematic review.

Abstract (item 2) 2. Abstract. A structured abstract following the PRISMA 2020 for Abstracts checklist.

Introduction (items 3-4) 3. Rationale (what is already known and the gap you fill). 4. Objectives, ideally with a question in PICO format.

Methods (items 5-15) 5. Eligibility criteria. 6. Information sources (databases, registers, dates). 7. A complete, reproducible search strategy. 8. Selection process (who screened, how many reviewers, which tools). 9. Data collection process. 10. Data items. 11. Study risk of bias assessment. 12. Effect measures. 13. Synthesis methods. 14. Reporting bias assessment (for example, publication bias). 15. Certainty assessment (for example, GRADE).

Results (items 16-22) 16. Study selection (this is where the flow diagram goes). 17. Study characteristics. 18. Risk of bias in studies. 19. Results of individual studies. 20. Results of syntheses. 21. Reporting biases. 22. Certainty of evidence.

Discussion (item 23) 23. Discussion. Interpretation, limitations and implications.

Other information (items 24-27) 24. Registration and protocol (for example, in PROSPERO). 25. Support (funding). 26. Competing interests. 27. Availability of data, code and other materials.

Notice that item 16 is where the PRISMA flow diagram fits. The diagram does not replace the checklist, it complements it. The checklist covers the text; the diagram covers the numbers of the selection process.

If you have not started your review yet

Everything above assumes you already have the numbers and only need the drawing. If you are still at the start, there is a route that saves you that whole job.

The counts in the diagram are not new data. They are the output of your screening. The trouble is that almost nobody records them while screening, so at the end they have to be rebuilt by hand, counting in the reference manager and in the spreadsheet. That is where counts that do not add up come from.

In revisia, screening and the diagram are the same thing. Every decision you make is recorded, and the PRISMA flow diagram is built from those real counts. You do not copy a single number, you do not rebuild anything at the end, and the counts add up because they come from the process itself. You also get the 27-item checklist.

Start the review and get the diagram done for you. In revisia the PRISMA diagram comes from your own screening, with nothing to fill in by hand. Create your free account.

Frequently asked questions

How many phases does the PRISMA 2020 flow diagram have?

The three phases are identification, screening and included, with the record counts at each step and the exclusions with their reasons.

What is the difference between PRISMA 2009 and 2020?

PRISMA 2020 updated the checklist and the diagram to reflect searches across several sources and to report the reasons for exclusion more clearly.

Can the PRISMA diagram be generated automatically?

Yes. With the numbers in hand, our generator draws it for free with no account and checks that the counts do not contradict each other. If you are still screening, in revisia the diagram builds itself from the counts of your own screening, without you typing a single number.

What is PRISMA in a systematic review?

PRISMA is the international guideline for reporting a systematic review and meta-analysis transparently. It has a 27-item checklist (what the text must contain) and a flow diagram (the outline of the study selection process). It is not a method for doing the review but a standard for presenting it transparently, and most scientific journals require it.

How do you make a PRISMA flow diagram?

You fill it in once selection is finished, going through the three phases of PRISMA 2020. Under identification you record the records retrieved from each database and from other sources, and the duplicates removed. Under screening you record the records screened, those excluded on title and abstract, and the reports assessed in full text (with the reason for each exclusion). Under included you give the final number of studies. The fastest route is to type those numbers into a PRISMA flow diagram generator, which draws the boxes and checks that the counts add up at every step.

Where can I download the PRISMA template in Word?

The official template is a free Word download from prisma-statement.org, with versions for new and updated reviews (CC BY 4.0 license). If you want the diagram already built with your numbers, our generator also exports to Word, and it warns you if any count does not add up.

Is PRISMA mandatory?

PRISMA is not a legal requirement, but in practice it works like one. Most scientific journals that publish systematic reviews require authors to follow PRISMA and ask for the completed checklist as supplementary material, and examiners of undergraduate theses, master's dissertations and PhD theses use it as a quality criterion. Reporting in line with PRISMA 2020 is today the sensible default for any systematic review.

How to write a PICO question, with examples

2026-10-12T00:00:00Z

The PICO model turns a question about the effects of an intervention into a structured research question. Its name covers Population, Intervention, Comparison and Outcome. A good PICO question is useful in nursing, medicine, psychology and other fields whenever you compare an intervention with an alternative.

PICO is not the first step of every systematic review, and it does not work equally well for every design. A scoping review, a qualitative synthesis, a prevalence question or a diagnostic test accuracy review need other decisions. Even in an intervention review, the question, the inclusion criteria and the search strategy are not exact copies of one another.

What the PICO model is

The PICO model is an evidence-based practice tool that breaks a question about interventions into manageable parts. Richardson and colleagues described the approach in 1995. The Cochrane Handbook uses it to frame reviews of effects and then to separate the general review question from the specific comparisons that will be synthesized.

A question like "does exercise help depression?" is too broad to answer rigorously. The PICO format forces you to say who you mean, which specific intervention you are evaluating, what you compare it with and what outcome you expect. That gets you to a question you can answer by finding and synthesizing specific studies.

A well-built PICO question helps with these decisions:

It sets the scope of the review and keeps the project from becoming unmanageable.
It guides the search strategy, although not every element has to become a required search block.
It helps define the inclusion and exclusion criteria. Population, intervention and comparison usually drive eligibility. Outcomes are set in advance to prevent self-serving decisions later, but a study normally does not have to report them to be included.

That is why the PICO strategy should not be treated as a box-ticking exercise. It is the methodological decision that shapes everything that follows.

The four components of PICO

The PICO format has four elements. Define each one precisely, because each plays a different role in the review.

P, population or problem

The P describes who the question is about: the group of patients or people, or the clinical problem of interest. This is where you state characteristics such as age, sex, diagnosis, severity, comorbidities or care setting. Be specific without being so narrow that no literature is left to review. Two examples are "adults over 65 with type 2 diabetes" and "hospitalized patients at risk of pressure ulcers".

I, intervention

The I is the deliberate action you want to evaluate. It can be a treatment, a preventive technique, an educational program or a way of delivering care. Describe it in enough detail to recognize it in the studies, for example by dose, duration or format.

An observed exposure, such as smoking or air pollution, usually fits better in PECO or PEO. A formal diagnostic test accuracy review needs to separate the index test from the reference standard, as PIRD does. You can adapt PICO to those cases, but that does not turn an exposure into an intervention or a reference standard into a therapeutic comparator.

C, comparison

The C is the alternative the intervention is compared with: placebo, usual care, another intervention or no intervention. Not every question has an explicit comparison (prevalence or experience questions, for example, usually do without one), but in treatment questions it is key. Two possible comparators are "standard care" and "usual drug treatment".

O, outcome

The O is the outcome you want to measure. It can be clinical, functional or about safety. Prioritize what matters to patients and clinicians, not only what is easiest to measure. Two examples are a reduction in depressive symptoms on a validated scale and new pressure ulcers at 30 days.

Setting an outcome in advance does not mean excluding a study just because the article does not report it. Cochrane recommends including studies based on characteristics that can be known without looking at their results. Requiring them to report a specific outcome can introduce selective reporting bias.

Component	Guiding question	Example
P - Population / problem	Who is affected? What is the problem?	Adults with mild to moderate depression
I - Intervention	Which intervention am I evaluating?	Supervised aerobic exercise
C - Comparison	Compared with what?	Usual drug treatment
O - Outcome	Which outcome do I want to measure?	Reduction in depressive symptoms

How to write your PICO question step by step

Going from a vague idea to a structured question is an orderly process. These are the steps to apply the PICO strategy without getting lost.

Confirm that you are looking for the effect of an intervention. Start from a real question and decide what answer you need. If you are after frequency, experiences, scope, exposure or diagnostic accuracy, look at another framework before filling in PICO.
Define the population (P). Describe precisely who is affected: age, sex, diagnosis, setting and any relevant characteristic. If you narrow it too much there will be no evidence left to review; if you leave it too open, the question loses its focus.
Specify the intervention (I). Pin down the treatment, program or practice you want to evaluate. The clearer it is, the easier it will be to recognize it in the studies and to build the search.
Set the comparison (C). State what it is compared with: placebo, usual care, another intervention or nothing. If your question does not need a comparison, say so explicitly to justify its absence.
Decide on the outcome (O). Define a clinically relevant measure and state how it will be assessed, by scale, indicator or time point, where possible.
Write the question and check its limits. Join the four components in one clear sentence. Check whether the comparison answers the clinical decision and whether each detail can be applied in the protocol. Do not demand that the whole question fit literally into a single search string.

You can write it with this template: In [Population], what is the effect of [Intervention] compared with [Comparison] on [Outcome]?

That template helps you think the question through. For the manuscript, switch to the declarative form published reviews use ("To assess the effect of..."), because that is what nearly all of them do. In an analysis of how 2,368 published reviews state their question, only 0.7% of Cochrane objectives contained a question mark.

Write your intervention question in revisia. The tool suggests a PICO and flags possible gaps. It can be wrong. You review every field before saving it, and the suggestion does not replace the methodological decisions in your protocol. Try my question for free.

PICO question examples (nursing, medicine, psychology)

Seeing the model applied in different fields helps you absorb it. These are plausible PICO question examples (they do not refer to specific studies; they only illustrate the format).

PICO in nursing

PICO is used a lot in nursing, both in undergraduate theses and in evidence-based practice. An example:

P. Hospitalized adult patients at risk of pressure ulcers
I. Repositioning every two hours
C. Repositioning according to the facility's usual protocol
O. Incidence of pressure ulcers at 14 days

Question. In hospitalized adult patients at risk of pressure ulcers, does repositioning every two hours, compared with the facility's usual protocol, reduce the incidence of pressure ulcers at 14 days?

Medicine

An example from clinical medicine:

P. Adults with newly diagnosed mild hypertension
I. Structured dietary intervention (low-sodium diet)
C. General dietary advice
O. Reduction in systolic blood pressure at 3 months

Question. In adults with newly diagnosed mild hypertension, does a structured dietary intervention, compared with general dietary advice, lower systolic blood pressure more at 3 months?

Psychology

In psychological research, the comparison is sometimes another active therapy:

P. Adolescents with generalized anxiety
I. Cognitive behavioral therapy
C. Waiting list
O. Reduction in anxiety symptoms measured with a validated scale

Question. In adolescents with generalized anxiety, does cognitive behavioral therapy, compared with a waiting list, reduce anxiety symptoms at six months?

If you want more PICO question examples, look at the common pattern. The intervention and the comparison isolate one decision. The outcome is defined before seeing the results, but it is not used by default to exclude studies.

Other frameworks besides PICO

PICO is not the universal framework. Depending on the type of question, you may need to add components or choose a different structure. If you do not know yet which one fits, start by comparing the frameworks available in revisia and their limits.

PICOS. Adds an S for study design. It is a common variant in systematic reviews because it makes visible which designs can be included. The design must be defined in the criteria even if it does not appear in the acronym of the objective.
PICOT. Adds a T for time, the follow-up period. It is useful when the time point at which the outcome is measured is decisive, as in long-term effectiveness studies.
PECO. Replaces the I (intervention) with an E for exposure. It is used in epidemiology and etiology studies, where there is no deliberate intervention but an exposure, such as smoking or air pollution. PEO is the variant without an explicit comparison, and choosing between the two depends on whether your question needs a defined contrast.
PICo. JBI uses Population, phenomenon of Interest and Context to frame qualitative synthesis questions. revisia does not yet offer a PICo template.
SPIDER. It was created as an alternative tool for turning qualitative and mixed-methods questions into search concepts. It can also organize criteria through Sample, Phenomenon of Interest, Design, Evaluation and Research type. Its five blocks can reduce search sensitivity, so check it against known studies before relying on it.
PCC. Organizes Population, Concept and Context for a scoping review question.
CoCoPop. Separates Condition, Context and Population in prevalence and incidence questions.
PIRD. Distinguishes Population, Index test, Reference test and Diagnosis of interest in a diagnostic test accuracy question.

There are more frameworks, such as PICOTS for prognosis and ECLIPSE for health services management. Choose by the type of answer you need and the methods guidance for your review, not by the acronym you know best.

A PICO can exist at three levels

The Cochrane Handbook separates decisions that often get mixed up.

The review PICO sets the overall scope.
Each synthesis can use a more specific PICO, for example for one outcome or one follow-up time point.
Each included study has its own PICO and may contribute data to some syntheses but not to others.

This separation keeps you from forcing the whole review into a single comparison. A study can meet the general criteria and still report no usable data for a particular meta-analysis. That does not make it an excluded study.

From the PICO question to the search strategy

Once the question is settled, the search begins. The PICO elements supply candidate concepts, but not all of them should appear as required blocks.

The procedure is straightforward:

Start with the main concepts, usually population and intervention.
For each one, list free-text terms, synonyms and controlled vocabulary such as MeSH or DeCS.
Join the synonyms of one concept with OR and the different concepts with AND.
Add comparison or outcome only after checking that known relevant studies do not drop out.
Ask an information specialist for a PRESS peer review if the search will support a review you plan to publish.

The Cochrane Handbook advises against routinely searching for all outcomes. They may be missing from the title and abstract or be described in unpredictable words. The PRESS guideline lets you have the strategy reviewed before you run it for good.

Turning the question into a reproducible search is the next step in the process. To go deeper, read our guide on how to build a literature search strategy with Boolean operators, MeSH and DeCS, which walks from the PICO to the final search string.

The PICO question is one of the core stages of a systematic review. To see how it fits into the whole process, from the protocol to the PRISMA-compliant report, go back to the guide on what a systematic review is and how to do one step by step.

revisia suggests a framework and flags possible gaps. It can be wrong. You review every field before saving it and decide which elements to use in the search. Check my question for free.

Frequently asked questions

What is PECO?

It is a framework for exposure questions rather than intervention questions. It covers Population, Exposure, Comparator and Outcome and is used in epidemiology when there is a defined contrast.

Does PICO work for qualitative questions?

PICO is not usually the first choice for a qualitative synthesis. JBI recommends PICo: population, phenomenon of interest and context. SPIDER can help organize the search and eligibility, but its five blocks can reduce sensitivity.

Is the comparison always required?

Not always. Some intervention questions have no active comparator, and other review types use different frameworks. Do not delete the comparison from PICO to make a prevalence, experience or scoping question fit.

What does PICO stand for?

PICO stands for Population, Intervention, Comparison and Outcome. It structures questions about the effects of an intervention compared with an alternative.

What is a PICO question used for?

It turns a question about interventions into a clear research question. It helps define the scope, the comparisons and the outcomes. It does not force you to search for all four elements, and it does not turn every outcome into an inclusion criterion.

What is the difference between PICO and PICOS?

PICOS adds S for study design. While PICO describes population, intervention, comparison and outcome, PICOS makes visible which designs will be accepted. It is a common variant, but the design can also be set in the eligibility criteria without adding the letter to the objective.

How do I go from PICO to the search?

Usually start with population and intervention. Combine the synonyms of each concept with OR and the different concepts with AND. Add comparison or outcome only if tests against known studies show you do not lose sensitivity.

Methods sources checked in September 2026.

Meta-analysis software: JASP, jamovi, RevMan and R

2026-10-12T00:00:00Z

Your choice of meta-analysis software matters more than it seems. It decides how much time and stress the quantitative phase of your review will cost you. What you need is a program that combines the results of several studies into a single estimate, calculates heterogeneity, draws a readable forest plot and lets you check for publication bias with a funnel plot. Asking for all that for free sounds optimistic. It is not. Today there are free tools with a visual interface and more than enough statistical rigor that cover it without you writing a single line of code. The four candidates with the longest track record are JASP, jamovi, RevMan and R with the metafor package, each with its pros, its cons and a very different learning curve.

Still going over the concepts? Before you touch any software, make sure you are clear on what a meta-analysis is and how to read a forest plot. And if you are coming from the start of the process, this article is one piece of our complete guide to the systematic review.

What meta-analysis software needs to do

Not every statistics program can synthesize studies. Decent meta-analysis software has to give you, at a minimum:

The pooled effect size, calculated from each study's measures (odds ratio, relative risk, standardized mean difference, correlations and so on), with its confidence interval.
Fixed-effect and random-effects models, because you will almost never assume that all the studies measure exactly the same true effect.
An estimate of heterogeneity. The I² statistic, Cochran's Q and tau² to quantify how much the studies differ from each other.
A forest plot, the chart that sums up all the studies and the overall estimate at a glance.
A funnel plot and asymmetry tests (Egger, Begg) to explore publication bias.
Subgroup analysis and meta-regression, if you want to explain where the heterogeneity comes from.

With this list in front of you, the useful question changes. "Which one is best" matters less than which one fits your statistics level and the time you have.

JASP (free, visual interface)

JASP (jasp-stats.org) is a free, open-source statistics program developed at the University of Amsterdam. It runs on Windows, macOS and Linux. Its biggest strength is its point-and-click visual interface. You load your data, pick the analysis from a menu and the results appear instantly in a clean panel, ready to copy into your report.

For meta-analysis, JASP includes a dedicated module (which relies on the R package metafor under the hood, so the statistical engine is first-rate) that lets you:

Enter effect sizes and their standard errors, or let JASP calculate them.
Fit fixed-effect and random-effects models.
Get I², tau² and Cochran's Q automatically.
Produce a forest plot and a funnel plot in a couple of clicks.
Run Egger's test for publication bias.

As a bonus, JASP is very strong in Bayesian statistics, so if you want a Bayesian meta-analysis it is right there without installing anything else. Its limit is reproducibility. Because it is so visual, the workflow is not recorded in a script as easily as in R (although it saves a .jasp file with the whole analysis). For the vast majority of undergraduate theses, master's dissertations and PhD theses, JASP is the starting point we would recommend.

jamovi (free, MAJOR module)

jamovi is the other major free program in this comparison, also open source and with a philosophy very close to JASP's. It offers a visual interface, real-time results and the same R foundation underneath. That is no coincidence. Both came from teams close to the JASP project, so if you can find your way around one, the other will feel familiar.

In jamovi, meta-analysis runs through the MAJOR module (jamovi's R-based meta-analysis module), which you install for free from the program's own module library. MAJOR also relies on metafor and gives you:

Fixed-effect and random-effects meta-analysis.
Forest plot and funnel plot.
Heterogeneity (I², tau², Q).
Meta-regression and subgroup analysis.

One useful advantage of jamovi over JASP is that each analysis can generate the equivalent R code if you switch it on, which helps reproducibility and makes it easier to move to R later. And between the two? Mostly a matter of personal preference. Both are free, visual and rigorous.

How to download jamovi (and JASP)

Downloading jamovi takes a couple of minutes. Go to jamovi.org, click "Download", choose your operating system (Windows, macOS or Linux) and install the base program. Then open jamovi, go to the modules menu (the "+" icon at the top right) and add MAJOR from the library. That is all you need to run a meta-analysis. To download JASP the process is the same from jasp-stats.org ("Download JASP"), and there meta-analysis is included with no extra downloads (you only have to switch the module on if it does not appear). Both are small downloads. Neither needs a license, a sign-up or a payment.

Data extraction comes before the statistics software. In revisia the data from each study come out structured and linked to their source, ready for your meta-analysis. Create your free account.

RevMan (Cochrane)

RevMan (Review Manager) is Cochrane's official software for writing systematic reviews and building their meta-analyses into them. This is not a general statistics program. The whole tool is designed, from start to finish, around the Cochrane workflow: from the protocol to the characteristics of studies, risk of bias, summary of findings tables and, of course, forest plots.

Its strengths:

It is built exactly for the format of a Cochrane review, with templates that guide each section.
Its forest plots are the standard you see in thousands of published reviews.
It integrates risk of bias assessment and GRADE tables.

It has limitations. Its statistical range is narrower than R's (fewer advanced meta-regression methods, for example), and on price a common misunderstanding needs clearing up. The current web version, RevMan Web, is paid/subscription-based for general use, although Cochrane offers free access to authors working on a registered Cochrane review and to people in certain low-income countries. In other words, it is not freely available to everyone the way JASP or jamovi are. If your goal is to publish within Cochrane, RevMan is close to mandatory. Outside that circuit, it is rarely the most practical option.

R + metafor (advanced)

R is a free, open-source statistical programming language, and metafor is, without question, the reference package for meta-analysis in R. It is the very engine JASP and jamovi's MAJOR module use under the hood, so working directly with metafor gives you access to everything it offers without the limits of a graphical interface.

With R and metafor there are hardly any limits. Fixed-effect and random-effects models, multilevel and multivariate models, complex meta-regression, sensitivity analyses, fully customizable forest plots and funnel plots, and methods to correct for publication bias (trim-and-fill, selection models). Everything stays in a reproducible script. For a serious publication that is worth a great deal. Anyone can run your code and get the same results.

The price of that power is the learning curve. You need to get used to R syntax, prepare the data in the right format and understand which function to call (rma() for the model, forest() for the plot and so on). You do not need to be a professional programmer, but you do need to put in the time. R for meta-analysis is the option we recommend when the project is ambitious, when you will publish in a demanding journal or when you need a method the visual interfaces do not offer.

Comparison table (price, learning curve, forest plot)

Software	Price	Learning curve	Forest plot
JASP	Free (open source)	Low. Visual interface, everything through menus	Yes, in one click
jamovi (MAJOR module)	Free (open source)	Low. Visual, easy module install	Yes, in one click
RevMan	Paid/subscription (free only for Cochrane authors)	Medium. Built around the Cochrane workflow	Yes, standard Cochrane format
R + metafor	Free (open source)	High. Requires coding in R	Yes, fully customizable

The table leaves a clear recommendation. If you have never run a meta-analysis, start with JASP or jamovi, which will give you correct results in an afternoon without fighting the syntax. Keep R + metafor for when you need advanced methods or full reproducibility. RevMan only pays off inside the Cochrane ecosystem.

A forest plot in JASP, step by step

The typical workflow in JASP, from start to finish:

Prepare the data in a spreadsheet. One row per study, with columns for the effect size (for example, the log odds ratio) and its standard error (or the raw event counts and totals per group, letting JASP calculate the effect).
Load the file into JASP (CSV or Excel) by dragging it into the window.
Open the meta-analysis module (if it does not appear, switch it on from the modules menu, the "+" icon).
Assign the variables to their boxes and choose the random-effects model (the usual choice when there is heterogeneity).
Tick the Forest plot box and, while you are at it, Funnel plot and Egger's test.

In seconds you will have the plot with each study, its confidence interval and the diamond for the overall estimate at the bottom. Here is how to read it.

Each horizontal line is a study. The square marks its effect (sized in proportion to the study's weight) and the line, its confidence interval. The diamond at the bottom sums up the pooled effect. To go deeper into how to interpret I², the funnel plot and effect measures, see our guide on how to interpret a meta-analysis and its forest plot. Before combining studies, you should have assessed their risk of bias, because the quality of the studies limits the quality of the meta-analysis.

Before the meta-analysis, revisia gives you the full review and the PRISMA diagram. Create your free account.

Frequently asked questions

Do JASP and jamovi work for reviews outside health?

Yes. Both are general statistics programs and work for meta-analysis in any discipline, not only in the health sciences.

Can I move from a visual interface to R later?

Yes. jamovi can show the R code for each analysis, which makes it easier to learn and move to R with the metafor package.

Do I need the raw data from each study or just the effect?

For a standard meta-analysis, the effect size and its variability for each study are enough; you do not need individual participant data.

What is the best free meta-analysis software?

For most users, JASP and jamovi are the best free options. Both have a visual interface, are rigorous (they use the R package metafor under the hood) and let you run a full meta-analysis, with forest plot and funnel plot, without coding. If you need advanced methods or full reproducibility through code, R with metafor is the most powerful, also free, but with a steeper learning curve.

JASP or jamovi?

Both are free, open source and visual, and they share a statistical engine, so the difference is a matter of nuance. JASP stands out for its clean interface and strong Bayesian section. jamovi, with the MAJOR module, can show the equivalent R code for each analysis, which helps if you later want to move to R. Try both. The choice usually comes down to which interface you prefer.

Is RevMan free?

Not entirely. RevMan Web, the current version of Cochrane's software, runs on a paid subscription for general use. Cochrane offers free access to authors preparing a registered Cochrane review and to users in certain low-income countries. If you do not work within Cochrane, JASP, jamovi or R, which are free, are usually the better choice.

Do I need to know R?

Not to get started. With JASP or jamovi you can run a sound meta-analysis without writing a single line of code. Knowing R (and the metafor package) only becomes necessary when you need advanced models, highly customized analyses or a fully reproducible script for a demanding journal. Many researchers start with a visual interface and move to R when their projects call for it.

Zotero vs Mendeley: which reference manager to choose

2026-10-12T00:00:00Z

A reference manager is a program that collects, organizes and automatically cites the references for your sources (articles, books, reports) so you never have to type a bibliography by hand. If you are preparing a systematic review, a thesis or a paper, it is a tool you cannot do without. You will handle hundreds of references, and you need to insert them into the text, switch citation style in seconds and avoid duplicates. Zotero vs Mendeley is the comparison people ask about most, because they are the two most widely used free reference managers in academia. This guide looks at both in depth, adds EndNote as the paid alternative, and ends with a comparison table and a recommendation for each use case.

What a reference manager is (and why you need one)

A reference manager (also called a citation manager or bibliography manager) does four basic jobs:

Capture references from your browser, a database or a PDF, picking up the metadata automatically (authors, title, year, journal, DOI).
Organize your library into folders, tags and collections, with search and, in many cases, management of the full-text PDFs.
Cite inside the document with an add-in (plugin) that inserts the citation at the cursor while you write in Word, Google Docs or LibreOffice.
Generate the final bibliography in the style you need (APA, Vancouver, IEEE, Chicago) and switch from one style to another without redoing anything.

In a systematic review a reference manager matters even more. When you export results from several databases for systematic reviews (PubMed, Scopus, Web of Science), the manager lets you bring all those records together in one place and, above all, detect and remove duplicates before screening starts. That count of imported references and removed duplicates is exactly the number you will later have to enter in the PRISMA flow diagram, so working with a manager from day one saves you trouble.

The questions "what is Mendeley" and "what is Zotero" have the same basic answer. Both are reference managers. The difference is in how they do it, who develops them and what their limits are.

Mendeley features, Mendeley Cite and the Word plugin

Mendeley is a reference manager developed by Elsevier, the scientific publisher behind Scopus and ScienceDirect. It is free on its base plan, runs on Windows, macOS and Linux through its desktop app (Mendeley Reference Manager) and syncs your library to the cloud so it is available on several computers.

Its strengths:

Built-in PDF reader. You can open articles inside the app itself, highlight them and add notes that stay saved with each reference. For many people this is the main reason to use Mendeley.
Mendeley Cite. This is the current citation add-in for Microsoft Word. You install it from the Office add-in store, and it works with both the desktop and web versions of Word. With Mendeley Cite you search for the reference, insert it in the text, and the bibliography builds and updates itself. It replaces the old "Mendeley Cite-O-Matic" plugin, so when someone searches for how to use Mendeley in Word, what they need today is Mendeley Cite.
Simple importing. It has a browser importer (Mendeley Web Importer) for saving references from the web, and it imports RIS, BibTeX and similar files exported from databases.
Elsevier integration. It connects with Scopus and ScienceDirect, the publisher's catalog of scientific literature.

Limits to keep in mind:

It is proprietary software. It is not open source, so you depend on Elsevier's decisions about the product. Mendeley has in fact reorganized its apps over time (retiring the classic desktop version in favor of Reference Manager), which has caused friction with long-time users.
Limited cloud storage. The free plan includes a storage quota. If you manage many full-text PDFs you can run short and need a paid plan with more storage.
Account required. You need an Elsevier account to sync.

To download Mendeley, always go to the official website and install both the Reference Manager app and, inside Word, the Mendeley Cite add-in from the Office store.

Zotero features and browser connectors

Zotero is an open-source, nonprofit reference manager, originally developed in a university setting (the Center for History and New Media) and maintained by the Corporation for Digital Scholarship. It is free, cross-platform (Windows, macOS, Linux) and, because it is open, it does not depend on any commercial publisher.

Its strengths:

Browser connectors (Zotero Connector). This is probably its best feature. You install the extension in Chrome, Firefox, Edge or Safari and, with a single click in the browser toolbar, Zotero captures the reference you are looking at (an article in PubMed, an entry in Google Scholar, a book in an online bookstore) with its metadata and, when possible, the attached PDF. The connectors recognize the type of resource and adjust the fields to match.
Smart capture of multiple results. From a search results page you can select several records and import them at once, which is very handy when you pull results from a database.
Citation styles and word processor plugins. Zotero installs plugins for Microsoft Word, LibreOffice and Google Docs, and it includes thousands of citation styles (APA, Vancouver, IEEE and many more) through the Zotero Style Repository. Switching style is instant.
Extensible. Because it is open source, it has an ecosystem of third-party plugins (for example, to improve PDF management, generate reports or connect it to other tools).
Flexible storage. Zotero gives you a free cloud storage quota to sync your PDFs and attachments, and you can pay to expand it if you need more. It also lets you use your own storage (file syncing via WebDAV or external services) if you would rather not pay, something Mendeley does not offer.

Its limits:

More basic PDF reading. Zotero includes a PDF viewer and annotation, but its reading experience has traditionally been less polished than Mendeley's, although it has improved a lot.
Initial learning curve. The power of plugins and custom syncing can feel overwhelming at first if you only want "the basics".

To download Zotero, go to the official website, install the desktop app and, separately, the connector for your browser. That connector is what makes the difference day to day.

EndNote (the paid alternative)

EndNote is the leading paid reference manager in academia, developed by Clarivate, the company behind Web of Science. It is not free (it is sold under license, although many universities have institutional licenses for their community), so its typical users are institutions, research groups with a budget and authors who need advanced features.

What it adds compared with the free options:

Advanced features for managing very large libraries, strong deduplication and tools built for large-scale review workflows.
Integration with Web of Science and with its full-text finder (Find Full Text), which tries to locate and download the PDFs for your references automatically.
The "Cite While You Write" plugin for Word, the equivalent of Mendeley Cite, very mature and widely used in scientific publishing.
Support and training backed by a commercial vendor.

Its big drawback is cost. Without an institutional license, the outlay is considerable compared with Mendeley or Zotero, which are enough for most students and researchers without paying anything. So before you weigh "Zotero or EndNote", check whether your university gives you access to EndNote and whether you really need its advanced features.

Whichever manager you pick, the next step happens in revisia. Import your references and go straight to screening. Create your free account.

Comparison table (price, storage, collaboration, citing in Word)

The table below sums up the criteria that matter most when choosing. Specific storage quotas change over time, so confirm them on each tool's official website before you decide.

Criterion	Mendeley	Zotero	EndNote
Developer	Elsevier (commercial)	Corporation for Digital Scholarship (nonprofit)	Clarivate (commercial)
Model	Proprietary, free plan	Open source and free	Paid (institutional licenses are common)
Base price	Free plan; paid extra storage	Free; paid extra storage	Paid license
Web capture	Web Importer (browser extension)	Zotero Connector (one of its strengths)	Browser and Web of Science capture
Citing in Word	Mendeley Cite (Office add-in)	Plugin for Word, LibreOffice and Google Docs	Cite While You Write
PDF reading and annotation	Very good, built in	Good, much improved	Good
Cloud storage	Limited free quota, paid upgrades	Limited free quota, paid upgrades or your own storage (WebDAV)	Its own sync, aimed at institutional use
Collaboration / groups	Private groups (limited on the free plan)	Unlimited groups, public or private	Library sharing (depends on version/license)
Operating systems	Windows, macOS, Linux	Windows, macOS, Linux	Windows, macOS
Customization	Limited	High (third-party plugins)	Medium

Zotero wins on web capture and open philosophy. Mendeley stands out for its built-in PDF reader. EndNote only pays off if you get it free through your university or need its advanced features.

Zotero vs Mendeley: which one to choose for your case

There is no single winner in the Zotero vs Mendeley comparison. It depends on how you work:

Choose Zotero if you value free software, care about independence from a publisher, capture a lot of references from the browser and from databases, or want to collaborate in groups without limits. For systematic reviews and for undergraduate theses and master's dissertations, it is what many university libraries recommend today.
Choose Mendeley if you read and annotate a lot inside the manager itself and want a comfortable all-in-one PDF reader, or if you already work in the Elsevier ecosystem (Scopus, ScienceDirect).
Choose EndNote if your institution gives you the license for free and you need to manage huge libraries with advanced deduplication and deep Web of Science integration.

In reviews with several authors, all three let you cite in Word. The question "how do I cite in Word with Mendeley?" gets an equally good answer from Mendeley Cite, from the Zotero plugin or from EndNote's Cite While You Write. The difference lies in reference capture, storage and collaboration, not in inserting citations.

If you want to go deeper into where to search before you import anything into your manager, read our guide to databases for a systematic review, because choosing your sources well shapes everything else.

How to fit it into your review workflow

The reference manager connects the search to screening. An orderly workflow looks like this:

Search and export. You run your search strategy in each database and export the results as RIS or BibTeX.
Import into your manager. You load all those files into Mendeley or Zotero, in a collection set aside for the review.
Remove duplicates. You use the manager's duplicate detection to clean out records repeated across databases. Write down how many you remove. That number goes into the PRISMA diagram.
Move on to screening. You export the cleaned library (again as RIS) to your study selection tool. This is where screening tools such as Rayyan and Covidence come in: they take that reference file and let you select studies by title, abstract and full text, ideally with blinded dual screening.
Cite as you write. When you write the manuscript, you go back to the manager to insert citations and generate the bibliography in the style your journal or university requires.

The reference manager stays with the review from the first import to the final bibliography. If you want to see how each phase fits into the whole, read our complete guide on how to do a systematic review step by step, which walks through the process from start to finish.

Once you have your references, in revisia you screen them with artificial intelligence, and every uncertain call goes to a human. Create your free account and try it.

Frequently asked questions

What is Mendeley?

Mendeley is a free reference manager developed by Elsevier. It lets you save and organize references, read and annotate PDFs inside the app itself (Mendeley Reference Manager), sync your library to the cloud and cite in Word with the Mendeley Cite add-in.

What is Zotero?

Zotero is a free, open-source reference manager maintained by the Corporation for Digital Scholarship. It captures references from the browser with the Zotero Connector, organizes your library into collections and tags, and lets you cite in Word, LibreOffice and Google Docs.

What are Mendeley and Zotero used for?

Mendeley and Zotero are reference managers. You use them to collect the references for your sources, organize them in a library, insert citations while you write in Word or Google Docs and generate the final bibliography in the style you need (APA, Vancouver, IEEE). In a systematic review they also let you gather the results from several databases in one place and remove duplicates before screening.

What is the main function of a reference manager like Mendeley or Zotero?

The main function of a reference manager is to automate reference management. It captures the metadata of each source (authors, title, year, DOI), organizes it into folders or collections, inserts citations through an add-in and generates the bibliography without you typing it by hand. It also lets you switch citation style in seconds.

Can I move my library from Mendeley to Zotero?

Yes. You can export your references (for example as RIS or BibTeX) and import them into the other manager, then check that your PDFs and notes came across.

Do reference managers detect duplicates?

Yes. Both Mendeley and Zotero detect and merge duplicate references, a necessary step when you combine results from several databases.

Do they work with Google Docs or only with Word?

Zotero has a Google Docs integration as well as Word and LibreOffice. Mendeley focuses on Word through its citation add-in.

Which is better, Mendeley or Zotero?

It depends on your priority. Zotero wins on browser capture, open-source philosophy and group collaboration, and it is the default recommendation at many university libraries. Mendeley wins on reading and annotating PDFs inside the program. Both are free on their base plan and both let you cite in Word, so the best way to decide is to try each one for a week with your own references.

Is Mendeley free?

Yes. Mendeley has a free plan that covers what most students and researchers need. It includes the desktop app, the PDF reader, syncing and citing in Word with Mendeley Cite. What the free plan limits is cloud storage and group size. If you need more, there are paid plans with extra storage.

How do I cite in Word with Mendeley?

Install the Mendeley Cite add-in from the Microsoft Word add-in store (Insert > Add-ins) and sign in with your account. Place the cursor where you want the citation, search for the reference in the Mendeley Cite panel, insert it and, when you finish, generate the bibliography. Mendeley Cite builds it and keeps it up to date automatically. It replaces the old Cite-O-Matic plugin.

Zotero or EndNote?

Zotero is free, open source and enough for almost any student or researcher. EndNote is paid and only pays off if your institution provides the license or if you need to manage very large libraries with advanced features and tight integration with Web of Science. For an undergraduate thesis, a master's dissertation or a standard systematic review, Zotero is usually the sensible choice.

Literature search strategy: Boolean operators, MeSH terms and truncation

2026-10-12T00:00:00Z

A literature search strategy is the set of terms, Boolean operators and filters you use to retrieve every relevant study from a database in a systematic, reproducible way. Boolean operators (AND, OR, NOT) are the words that combine your terms to broaden or narrow what you retrieve. Typing a few words into PubMed is not enough. A strategy is a documented recipe that anyone else could run again and get the same results. In a systematic review, the literature search is the stage with the biggest impact on the validity of the work, because if you don't find a study, you simply can't include it. That is why it has to be built with method, combining concepts, synonyms, controlled vocabularies and a clear search string.

That method means combining Boolean operators (AND, OR, NOT), using truncation and wildcards, relying on MeSH and DeCS terms, and documenting every step so you meet PRISMA.

What a search strategy is (and why it must be reproducible)

The search strategy is the plan that turns your research question into concrete search strings for each database. It has two competing goals that you have to balance:

Sensitivity (recall). Retrieve as many relevant studies as possible, even if that brings noise. In a systematic review, sensitivity comes first. Screening too many articles is better than missing a key one.
Precision. Most of what you retrieve should be relevant, so you don't drown in thousands of irrelevant references.

What separates a methodical literature search from a casual one is reproducibility. A reproducible strategy meets three conditions:

Explicit. The search string is written out in full, not summarized.
Documented. You record the database, the date the search was run, the filters, the syntax and the number of results.
Transferable. It can be adapted to other databases while keeping the same conceptual logic.

The Cochrane Handbook and the PRISMA 2020 statement require you to report the full search strategy for every database searched (usually in the supplementary material). If you can't copy and paste your search string so someone else can rerun it, what you have for now is a hunch, not a strategy.

Before you write anything, settle your research question. If you haven't structured it yet, start by defining the PICO question for your review, because each component (population, intervention, comparison and outcome) will later become a block in your search string.

Boolean operators AND, OR and NOT

Boolean operators are the three words (AND, OR, NOT) that connect your terms and tell the database how to combine them. Most academic search engines (PubMed, Scopus, Web of Science) expect them in capitals, so the system reads them as operators and not as search words.

AND narrows the search

AND retrieves only records that contain every term. It reduces the number of results and raises precision. You use it to cross different concepts.

This string crosses three concepts.

diabetes AND exercise AND glycated hemoglobin

It only returns articles where all three concepts appear. The more ANDs you add, the narrower the result.

OR broadens the search

OR retrieves records that contain at least one of the terms. It increases the number of results and the sensitivity. You use it to group synonyms and variants of the same concept.

This string pulls together several synonyms.

exercise OR "physical activity" OR training OR "physical exercise"

All of those expressions describe the same concept, so you join them with OR to avoid losing any study that uses one or the other.

NOT excludes terms

NOT removes records that contain a term. It raises precision, but use it with care. It is the easiest way to lose relevant studies without noticing, because a relevant article may mention the excluded term in passing.

This string tries to exclude studies in mice.

breast neoplasms NOT mouse

Here you are trying to drop animal studies, but a human trial that cites a mouse model in its introduction would be left out. In systematic reviews it is usually better to filter species with the database's limits than with NOT.

Nesting with parentheses

A real search string is built by nesting blocks with parentheses, just as in math. Each concept is built with OR between synonyms and wrapped in its own parentheses. Then the concepts are crossed with AND. The full string looks like this.

(diabetes OR "type 2 diabetes mellitus" OR T2DM)
AND
(exercise OR "physical activity" OR training)
AND
("glycated hemoglobin" OR HbA1c OR "glycemic control")

This is the canonical structure of any search strategy: blocks of synonyms joined with OR and concepts crossed with AND.

Operator	What it does	Example
AND	Narrows. Retrieves records with every term (intersection)	`obesity AND adolescents`
OR	Broadens. Retrieves records with at least one of the terms (union, for synonyms)	`obesity OR overweight OR "body mass index"`
NOT	Excludes. Drops records that contain the term (subtraction)	`obesity NOT surgery`
( )	Nests. Groups blocks to control the order of combination	`(obesity OR overweight) AND (diet OR exercise)`

Truncation, wildcards and phrase searching

Boolean operators handle the logic between concepts. The following tools help you capture every morphological variant of each word without typing them one by one.

Truncation

Truncation replaces the ending of a word stem with a symbol (usually the asterisk *) to retrieve every word that starts with that stem. It is one of the most useful ways to gain sensitivity.

nurs*

This retrieves nurse, nurses, nursing, nursed, and so on. Be careful not to truncate too early. car* would bring cardiac, but also career, carbon and caries, and a lot of noise with them.

Wildcards

A wildcard replaces a single character (often ? or #, depending on the database) inside a word. It handles spelling variants, especially between British and American English.

randomi?ed

This retrieves both randomised and randomized. Depending on the system, it also handles other spelling variants, such as words with and without accents.

Phrase searching

Quotation marks force the database to search for the words together and in that exact order. You need them for compound terms.

"physical activity"
"health-related quality of life"

Without quotation marks, many search engines apply an implicit AND between the words and retrieve records where they appear separately, and precision suffers.

Watch out. Each database uses its own syntax. The asterisk is almost universal for truncation, but wildcards, quotation marks and the minimum stem length change from one system to another. Always check the database's help pages before you trust your search string.

Struggling to get from the question to the search string? In revisia the search strategy is built step by step from your PICO and run against several sources at once. Create your free account.

Controlled vocabularies: MeSH and DeCS

So far we have searched with natural language (the words as they appear in the text). The problem is that the same concept goes by many names. "Myocardial infarction", "heart attack", "MI", "AMI". If you rely only on keywords, you will miss the articles that use a synonym you didn't think of.

Controlled vocabularies solve this. They are standardized lists of terms (descriptors) that indexers assign to each article based on its content, not its wording. Searching with a descriptor retrieves every study on that concept, whatever they call it.

MeSH in PubMed

MeSH (Medical Subject Headings) is the National Library of Medicine thesaurus that indexes MEDLINE/PubMed. Every PubMed article is assigned several MeSH terms. The advantages:

You retrieve all the synonyms at once by using the descriptor.
You can use hierarchical explosion (explode). When you search a MeSH term, PubMed by default also includes all the narrower terms below it in the tree. For example, the descriptor Cardiovascular Diseases expands to Heart Diseases, Myocardial Infarction, and so on.
You can narrow by subheadings such as /drug therapy, /diagnosis or /epidemiology.

In PubMed syntax, a MeSH term is tagged with [Mesh].

"Diabetes Mellitus, Type 2"[Mesh] AND "Exercise"[Mesh]

A robust strategy combines controlled vocabulary and natural language, because the most recent articles are not yet indexed with MeSH.

("Diabetes Mellitus, Type 2"[Mesh] OR "type 2 diabetes"[Title/Abstract] OR "DM2"[Title/Abstract])
AND
("Exercise"[Mesh] OR "physical activity"[Title/Abstract] OR "exercise"[Title/Abstract])

DeCS for Spanish- and Portuguese-language sources

DeCS (Health Sciences Descriptors) is the trilingual controlled vocabulary (Spanish, Portuguese and English) that BIREME develops from MeSH, expanded with concepts specific to regional health. It is the MeSH equivalent for Ibero-American databases such as SciELO and LILACS.

DeCS lets you search with Spanish or Portuguese descriptors and retrieve literature that is not in PubMed. Here is an example string with DeCS descriptors in Spanish.

("Diabetes Mellitus Tipo 2" OR "diabetes tipo 2")
AND
("Ejercicio" OR "actividad física")

The MeSH/DeCS relationship is direct. Most DeCS descriptors have an equivalent MeSH descriptor, which makes it easy to translate the same strategy between international databases and Spanish-language ones.

Thesauri and descriptors

A thesaurus is the structured dictionary the descriptors come from. Unlike a simple alphabetical list, it organizes terms into hierarchies and relationships. Understanding its structure greatly improves your literature search.

Every thesaurus defines three types of relationships between terms:

Hierarchical. Broader term (BT) and narrower term (NT). This is where the explode function comes from.
Equivalence. Synonyms and variants that point to the preferred descriptor (the entry "heart attack" sends you to "myocardial infarction").
Associative. Related terms (RT) that suggest neighboring concepts you may not have considered.

How to use the thesaurus in practice:

Look up each concept from your question in the database's thesaurus (the MeSH Browser for PubMed, the DeCS browser for Spanish- and Portuguese-language sources).
Note the preferred descriptor, its synonyms (which you will use as natural-language keywords) and its related terms.
Decide whether to explode the descriptor or restrict it to the exact concept.

The most widely used thesauri in the health sciences are MeSH (PubMed/MEDLINE), DeCS (SciELO, LILACS), Emtree (Embase) and CINAHL Subject Headings (CINAHL). Each database has its own, and that is exactly why the same search has to be rewritten for each one.

How to build the search string from your PICO

The search string comes straight from your research question. If you have structured your PICO question well, your conceptual blocks are already done. The process is systematic:

Identify the key concepts. Take the PICO components. For most reviews two or three concepts are enough, typically Population and Intervention (Comparison and Outcome tend to cut sensitivity if you add them to the search, so they are often kept for screening the articles).
Gather the synonyms for each concept. For each block, list natural-language keywords and the thesaurus descriptors (MeSH/DeCS). This is where controlled vocabulary pays off.
Join the synonyms with OR. Each concept becomes a set of parentheses with all its terms linked by OR.
Cross the concepts with AND. Combine the parentheses with AND so the result contains every concept.
Add truncation, wildcards and filters. Refine with *, quotation marks and the database's limits (dates, language, study type).

A worked example from a PICO question. "In adults with type 2 diabetes (P), does physical exercise (I) improve glycemic control (O)?"

(diabet* OR "type 2 diabetes mellitus" OR T2DM OR "diabetes tipo 2")
AND
(exercis* OR "physical activity" OR training OR ejercicio OR "actividad física")
AND
("glycemic control" OR "glycated hemoglobin" OR HbA1c OR "control glucémico")

This is your master search string. From it you generate the specific version for each database. (The Spanish synonyms are there on purpose: they pick up records indexed in Spanish-language sources.)

Adapting the search to each database

The same conceptual strategy has to be rewritten for each source, because each one has its own thesaurus, field tags and syntax. Copying and pasting the same string into all of them is not enough. You have to translate it. These are the things that change from one database to another:

The controlled vocabulary. [Mesh] in PubMed, Emtree in Embase, DeCS in SciELO/LILACS, CINAHL Subject Headings.
The field tags. [Title/Abstract] in PubMed is equivalent to TITLE-ABS-KEY( ) in Scopus or TS=( ) in Web of Science.
The truncation symbol and the wildcards.
How you nest and combine lines (some databases number each search and then combine them by line number).

To decide which sources to search and how the syntax translates in each one, see the guide to databases for systematic reviews, which covers how many and which ones to search according to PRISMA and Cochrane recommendations.

How to document the search for PRISMA

Documenting the search is a methodological requirement, not an optional step. PRISMA 2020 item 7 asks you to "Present the full search strategies for all databases, registers and websites, including any filters and limits used" (a stricter requirement than PRISMA 2009, which asked for at least one). For each source, record:

The name of the database and the platform or interface (for example, MEDLINE via PubMed).
The full search string, copied verbatim, not shortened.
The date the search was run.
The filters and limits applied (publication dates, language, study type).
The number of results from each database.

Those numbers feed directly into the "records identified" count of the PRISMA flow diagram. Keep the strings in a table in the supplementary material. That way your literature search is documented, can be audited, and any reader can reproduce it.

To see how the search strategy fits into the whole process, from the question to the synthesis, go back to the guide on how to do a systematic review step by step.

In revisia your search strategy is versioned and reproducible, ready for the PRISMA appendix. Create your free account.

Frequently asked questions

What is truncation in a search?

Using a symbol (often an asterisk) at the end of a word stem to retrieve all its variants. For example, therap* finds therapy, therapies and therapeutic.

What are quotation marks for in a search?

To search for an exact phrase. Putting an expression in quotation marks retrieves those words together and in that order, not separately.

Do I have to adapt the search string to each database?

Yes. Each database has its own syntax, fields and controlled vocabulary, so the same strategy is translated for each one and documented separately.

What are Boolean operators?

Boolean operators are three words (AND, OR, NOT) that combine your search terms in a bibliographic database. AND narrows the search because every term must appear, OR broadens it because one term is enough (ideal for synonyms), and NOT excludes records that contain a term. Combined with parentheses, they let you build precise, reproducible search strings.

What is a MeSH term?

A MeSH (Medical Subject Headings) term is a descriptor from the National Library of Medicine thesaurus that PubMed/MEDLINE uses to index each article by its content, regardless of the exact words in the title or abstract. Searching with a MeSH term retrieves every study on that concept even when they use different synonyms, and lets you use the hierarchical explosion to include narrower terms from the tree.

What is the difference between MeSH and DeCS?

MeSH is the English-language thesaurus of the National Library of Medicine that indexes PubMed/MEDLINE. DeCS (Health Sciences Descriptors) is its trilingual equivalent (Spanish, Portuguese and English), developed by BIREME from MeSH and expanded with regional health concepts. DeCS is used in Ibero-American databases such as SciELO and LILACS, and most of its descriptors have a MeSH equivalent, which makes it easy to carry the same strategy across international and Spanish-language sources.

How do I document the search strategy for PRISMA?

PRISMA 2020 requires you to report the full search strategy for every database you searched (PRISMA 2009 asked for at least one). Record five things for each source: the database name and platform, the search string copied verbatim, the date you ran it, the filters and limits applied, and the number of results. Put these strings in the supplementary material and use the counts to build the PRISMA flow diagram. That way your literature search is documented, auditable and reproducible.

Dissertation literature review: how to write it (and when to make it systematic)

2026-10-12T00:00:00Z

A dissertation literature review (in science and engineering often called the state of the art) is the critical synthesis of what has already been researched on your topic before you add anything new. Put another way, it answers a simple but demanding question. What do we know today about this problem, and what is still unresolved? It is not a loose summary of articles but a reasoned map of the literature that justifies why your work deserves to exist. In a well-planned undergraduate thesis or master's dissertation, that map is the foundation the whole project stands on, which is why it pays to write it with a method and not at the last minute.

To write it well, the first step is to tell it apart from the theoretical framework and from related labels. The second is to follow a reproducible process. This guide covers both, along with the structure of a thesis or dissertation built on a review, the most common mistakes and a frequent question about when it makes sense to ask for methodological help without taking anything away from your own work.

What a dissertation literature review is (vs theoretical framework and state of the art)

The literature review is the analytical, up-to-date review of the most relevant research output on a specific topic. Rather than describing everything ever published, it aims to identify trends, consensus, contradictions and, above all, the gaps your research can fill. That is why it usually ends with a sentence along the lines of "despite the extensive literature on X, no studies have addressed Y", which is exactly the gap where your research question fits.

Three related concepts get mixed up all the time, but they do not mean the same thing:

State of the art. Focuses on the most recent and advanced work. It looks at what is being researched now, with which methods and with which results.
Theoretical framework. Brings together the theories, models and concepts that support your work. It does not describe specific studies but the conceptual scaffolding you will use to interpret your data.
Literature review (or "state of the question"). In many academic traditions, especially in the humanities and social sciences, this is used as a synonym for the state of the art. The difference is one of nuance and discipline more than substance.

The "state of the art or literature review" question has a practical answer. Check your university's guide or ask your supervisor, because the right label depends on your program's convention. In most cases you can use them interchangeably, as long as you stick to the same term throughout the whole document.

Aspect	State of the art	Theoretical framework	Literature review (state of the question)
What it covers	Recent empirical studies and current advances	Theories, models and concepts	Debate and accumulated output on the topic
Time focus	The present and the leading edge of the field	Timeless (foundations)	Historical and current
Main function	Find research gaps	Give the analysis conceptual support	Place the work in the academic conversation
Usual disciplines	Science, engineering, health	All	Humanities and social sciences
Expected result	Justify the research question	Define the variables and how to interpret them	Show command of the topic

How to write a dissertation literature review step by step

Knowing how to write the literature review for a thesis or dissertation comes down to following an orderly, reproducible process. These are the steps:

Narrow down the topic and frame your question. The more specific it is, the easier the search. A well-structured question, for example one built with the PICO model for defining your research question, gives you the keywords to search with straight away.
Define the search strategy. Decide which databases you will search (Scopus, PubMed, Web of Science), with which terms and with which date and language limits. Document that strategy. In serious work, someone else should be able to repeat your search and get the same results.
Collect and filter the sources. Apply clear inclusion and exclusion criteria (years, study type, relevance) so you keep only what is relevant. This is where you discard duplicates and work that does not answer your question.
Read critically and take structured notes. Summarizing each article is not enough. Record the aim, methods, results and limitations of each study in a table or literature matrix. That matrix is what you will later turn into prose.
Organize the information by theme, not by author. The most typical mistake is going article by article ("Smith says... Jones says..."). A literature review is written in thematic blocks or lines of research, grouping the authors who are in conversation about the same idea.
Write in your own voice and synthesize. Compare, contrast and point out agreements and disagreements. Your value lies in the critical synthesis, not in piling up citations. Link each block to the next so the text has a clear thread.
Close by identifying the gap. End the literature review by pointing out what remains to be researched. That gap is the justification for your thesis or dissertation, and it leads naturally into your aims.

A good literature review example stands apart from a list of summaries because every paragraph argues one idea supported by several sources. If, after reading it, it is clear where there is agreement, where there is debate and which gap is still open, you have done it right.

Steps 2 to 4 (search, criteria and screening) are exactly what revisia handles for you if your literature review turns into a systematic review. Create your free account.

Systematic review vs traditional literature review in a student thesis

Not every review is a systematic review. Be clear on the difference before you choose the method for your project, because it determines the effort and the rigor you will have to apply.

A traditional (or narrative) literature review gives an overview of the topic based on a selection of sources chosen at the author's discretion. It is more flexible and faster, but also more subjective. It does not guarantee that all the relevant evidence has been found.
A systematic review follows a predefined, reproducible protocol, with a structured question, a documented search strategy, explicit selection criteria and an assessment of study quality. It reduces bias and is the most demanding method.

The full comparison between the two, with a decision guide written for people who screen alone, is in narrative review vs systematic review: which one to choose. To go deeper into other formats (scoping review, umbrella review, integrative review), this guide to the types of literature reviews and which one to choose for your aim will help. In academic settings, the undergraduate thesis or master's dissertation built as a systematic review is more and more common, because it lets you do solid research without collecting primary data, which is especially useful when you have no access to a sample of your own.

Can you do your thesis as a systematic review? Yes (the guide on what a systematic review is and how to do one step by step has the full method), and in fact it is one of the best options when you have limited time and a small budget. It does take planning. A badly done systematic review (incomplete search, no criteria, no bias assessment) is just a literature review with pretensions. Before you commit, look at how long a systematic review takes and fit the scope to your calendar.

Structure of a thesis or dissertation built on a review

If you decide to frame your project as a review, here is a standard structure that fits most university guidelines. The literature review lives mostly in the introduction, but the methods section is what separates a systematic review from a simple compilation.

Introduction and literature review. Context of the problem, review of prior literature, justification and the gap you found.
Aims and research question. A general aim and specific aims, derived directly from the gap in the literature review.
Methods. The central section of a systematic review. It includes:
- Design and protocol (ideally registered in PROSPERO).
- Search strategy and databases searched.
- Inclusion and exclusion criteria.
- Screening and study selection process.
- Risk of bias assessment.
- PRISMA flow diagram with the count of studies at each phase.
Results. Synthesis of the included studies, usually with a table of characteristics and, where appropriate, a meta-analysis.
Discussion. Interpretation of the findings, comparison with prior literature, limitations of the work.
Conclusions. Answer to the aims and future lines of research.
References and appendices. In your program's citation style (APA, Vancouver, etc.).

The key is that the literature review (step 1) and the methods (step 3) line up. The gap you find in the literature must be exactly the one your systematic review sets out to fill.

Worked example

Picture a nursing master's dissertation on the effect of music therapy on preoperative anxiety. A well-built passage of its literature review could start like this.

"Several studies have evaluated non-pharmacological interventions to reduce anxiety before surgery. A number of clinical trials agree that music therapy lowers anxiety levels measured with the STAI scale (author A; author B), although other studies find no significant difference compared with standard care (author C). These discrepancies may be due to heterogeneity in the length of the intervention and in the measurement instruments. To date, no study has systematically synthesized these results in a Spanish-speaking adult population, which justifies the present review."

Why this systematic review example works for a student thesis:

It groups studies by idea (non-pharmacological interventions), not by author.
It shows the consensus (music therapy reduces anxiety) and the contradiction (not every study confirms it).
It explains a possible cause of the discrepancy (heterogeneity), which shows critical reading.
It closes with the specific gap the project is going to fill.

In four sentences it is clear what is known, what is debated and what is missing. That is the goal of any literature review, whether for an undergraduate thesis or a master's dissertation.

Common mistakes (and how to avoid them)

Writing by author instead of by theme. This is the most frequent mistake. To avoid it, organize the text in thematic blocks and make them talk to each other.
Summarizing without synthesizing. A string of summaries is not a literature review. Compare the studies with each other and take a position.
Using outdated or unreliable sources. Prioritize peer-reviewed articles from recent years (apart from essential classics). Avoid unsupported blogs or commercial websites.
Not documenting the search. If your project is a systematic review, a search that cannot be reproduced invalidates the method. Record databases, terms and filters.
Confusing the literature review with the theoretical framework. Remember the table above. One maps recent studies, the other defines the conceptual scaffolding.
Not linking the literature review to your aims. The gap you point out in the literature must match what your project investigates. Without that continuity, the justification falls apart.
Citing badly or not citing. Every claim that is not yours needs its reference. Plagiarism, even unintentional, can fail the project.

When to ask for methodological help

Doing a systematic review for an undergraduate thesis or master's dissertation on your own is entirely possible, especially if you follow the steps in this guide and spend time on planning. That said, there are moments when outside guidance saves weeks of work and quite a few corrections from the examiners.

It makes sense to seek help with a systematic review thesis, or a methods tutorial, if:

You cannot make your search strategy reproducible, or you retrieve thousands of unfiltered results.
You are unsure whether to do a systematic review, a scoping review or a narrative review.
You need to apply risk of bias tools or a meta-analysis and have no statistical training.
Your supervisor has asked for more methodological rigor and you do not know where to start.

Asking for methodological guidance takes no merit away from your work. The content, the analysis and the conclusions are still yours. What expert support adds is judgment and the confidence that the method will stand up to academic review. If you decide to do it alone, lean on reference sources such as the PRISMA statement, the Cochrane Handbook or the JBI Manual for Evidence Synthesis, and always start with a well-framed research question.

If your thesis or dissertation is a review, in revisia you carry the full method through and get the deliverables you need to defend it. Create your free account.

Frequently asked questions

What is the difference between the literature review and the theoretical framework?

The literature review sums up what has already been researched on the topic. The theoretical framework sets out the concepts and theories you use to interpret your work.

How many sources should a master's dissertation literature review have?

There is no fixed number. The relevance and currency of the sources matter more than the count, and so does covering the debate that matters.

Does the literature review go at the start or the end of a master's dissertation?

It usually goes near the start, after the introduction, because it places your work within what is already known and justifies the gap you set out to fill.

What is the literature review of an undergraduate thesis?

It is the section where you review prior research on your topic in a critical, organized way. Its job is to show that you know what has already been studied, to identify contradictions and gaps, and to justify why your thesis adds something new. It is not a summary of articles but an argued synthesis that leads to your research question.

State of the art or literature review?

In practice they are near synonyms. "State of the art" is more common in science, engineering and health, while "literature review" is the default label in the humanities and social sciences, where you may also see "state of the question". The difference comes from disciplinary tradition more than content. What matters is to check your program's guide, pick one term and use it consistently throughout the document.

Can I do my undergraduate thesis as a systematic review?

Yes. An undergraduate thesis or master's dissertation built as a systematic review is a valid and highly recommended option when you cannot collect primary data. It must follow the method, though, with a structured question, a documented and reproducible search, inclusion and exclusion criteria, screening, risk of bias assessment and its PRISMA flow diagram. Without those elements it would be a literature review, not a systematic one.

How do I cite in the literature review?

Use the citation style your program requires (APA is the most common for student theses; Vancouver in the health sciences), citing every claim you take from the literature. The practical way is to insert citations while you write with a reference manager such as Mendeley or Zotero, which generates the final bibliography in the chosen format. Always cite the source you actually read and avoid second-hand citations.

Inclusion and exclusion criteria: how to define them

2026-10-12T00:00:00Z

Inclusion and exclusion criteria are the explicit rules that decide which studies get into your systematic review and which are left out. Together they are called eligibility criteria. They set the "border" of your review before you start searching, so that study selection is reproducible and does not depend on the researcher's on-the-spot judgment. Inclusion criteria describe the features a study must have to be considered; exclusion criteria describe the specific reasons that rule it out even though, at first sight, it fits. Built well, they follow almost point by point from the framework that matches your question, such as PICO for an intervention, and they are then documented according to PRISMA. This guide walks through that process with examples by discipline.

This article is part of our guide on how to do a systematic review step by step.

What eligibility criteria are

Eligibility criteria are the set of predefined conditions that decide whether a study is admissible in your review. You write them in the protocol, before running the search, and they are one of the features that set a systematic review apart from a narrative one. Any reader should be able to take your list of criteria, apply it to the same search results and reach the same selection of studies.

Defining these criteria well does three jobs. First, it sets the scope by turning a broad question ("does exercise work for depression?") into a workable one with clear limits, and in doing so it governs how many studies the review will end up including. Second, it guarantees reproducibility because it prevents "eyeballed" decisions during screening. Finally, it controls selector bias. Two reviewers applying the same rules should agree on almost every decision, and disagreements are resolved by checking the study against the criterion, not against an opinion.

"Eligibility criteria" is the umbrella term. Under it sit the inclusion criteria and the exclusion criteria, which are neither redundant nor mirror images of each other.

Inclusion vs exclusion criteria (they are not the same)

It is tempting to think of exclusion as just "the opposite" of inclusion. It is not, and mixing them up is one of the most common mistakes.

Inclusion criteria define the universe of studies you care about: the population, intervention, comparator, outcome and design that would answer your question. They are positive statements that describe the ideal study.

Exclusion criteria are not the automatic negation of the inclusion criteria. They are specific reasons to drop a study that does meet the inclusion criteria but has a problem that makes it unsuitable. For example, a record that meets population, intervention and design but only exists as a conference abstract; a valid trial that has been retracted; or a study with a comorbidity that would distort the effect.

The practical test to avoid duplicated work is simple. If something is already out because it fails an inclusion criterion, you do not need to turn it into an exclusion criterion. If your inclusion requires "adults aged 18 or over", you do not need an exclusion criterion saying "under 18", because they are already out. An exclusion criterion is only justified when it adds a new reason that inclusion does not capture. This rule prevents bloated, contradictory lists, and it is what peer reviewers expect.

Criterion	Inclusion (what you are looking for)	Exclusion (what you drop even though it fits)
Population	Adults (≥18 years) with type 2 diabetes	Patients with type 2 diabetes and end-stage renal disease (a comorbidity that changes the outcome)
Intervention	Supervised aerobic exercise program	Interventions that combine exercise with drugs where the effect cannot be isolated
Comparison	Usual care or waiting list	No comparison group (single-arm studies)
Outcome	Glycated hemoglobin (HbA1c) measured	Studies that only report adherence, with no clinical outcome
Study design	Randomized controlled trials	Protocols, editorials, letters to the editor, conference abstracts
Language	English and Spanish	Languages with no option for a reliable full-text translation
Date	Published from 2010 onward	(not applicable: already ruled out by inclusion)

Look at the last row. When inclusion already sets the border, the exclusion cell is left empty on purpose. That is correct, not an oversight.

How to derive them from PICO

The rigorous way to build your criteria is not to invent them but to derive them directly from your research question. If you framed your question with PICO (Population, Intervention, Comparison, Outcome), each component translates almost literally into a block of inclusion criteria.

To understand where each element comes from, first review how to frame your PICO question. If your question falls short here, your criteria suffer. When we looked at how 2,368 published reviews write theirs, the ones that ended up retracted almost always fall short on the population (80% do not name it), which is exactly the first cell of this table. From there, the mapping is direct:

Population (P) defines the population criterion: age, sex, clinical or social condition, setting (hospital, community, school).
Intervention (I) defines which exposure or treatment the study must have examined, with its dose, format or duration if relevant.
Comparison (C) defines which control group you require (placebo, usual care, another intervention) or whether you accept studies without a comparator.
Outcome (O) defines which outcomes the study must measure to be useful for your question, and how they must be measured.

To these four blocks you almost always add the study design, the "S" in variants such as PICOS, which classic PICO does not include. For practical reasons you also add limits on language, dates and publication type. The golden rule is that no criterion should be foreign to your question. If you cannot justify it by pointing to PICO or to an explicit methodological reason, you are probably introducing an arbitrary bias.

Common criteria (design, population, language, dates)

Beyond PICO, there is a group of criteria that appear in almost every review. You should know them and, above all, justify them:

Study design. This is the most decisive methodological filter. A review of a treatment's efficacy is usually restricted to randomized controlled trials; one on prevalence or risk factors will accept observational studies (cohort, case-control). Ruling out editorials, letters, protocols and conference abstracts in advance is common and reasonable.
Population. Set limits on age, sex, condition and context. Watch out for criteria that are too narrow. The more restrictive the population, the fewer studies you will find and the less your conclusions will generalize.
Language. Limiting a review to English and Spanish for resource reasons is common, but it is not neutral. It introduces a possible language bias (positive results tend to be published in English). If you apply it, state it as a limitation, not as a harmless choice.
Publication dates. A time limit is justified when a relevant change (a new guideline, a technology, a diagnostic definition) makes earlier studies irrelevant. Writing "from 2010" for no reason is arbitrary; writing "from the new 2013 diagnostic criteria" has a basis.
Full text and publication type. It is legitimate to require an available full text and, depending on the case, peer-reviewed literature. If you exclude gray literature, accept that you may reinforce publication bias.

Every criterion should be able to answer "why is this here?". If the answer is "just because" or "to have less work", revise it.

Define your criteria once and in revisia screening applies them consistently, record by record. Create your free account.

Inclusion and exclusion criteria examples by discipline

The same principles take different shapes in each field. These examples are realistic and meant as a guide.

Nursing (wound care).

Inclusion. Hospitalized adults with pressure ulcers; foam dressings vs standard dressings; time-to-healing outcome; randomized trials; English or Spanish; from 2012.
Exclusion. Venous or diabetic ulcers (a different population even though they are wounds), and studies where the dressing is combined with negative pressure therapy and the effect cannot be isolated.

Medicine (cardiology).

Inclusion. Adults with heart failure with reduced ejection fraction; SGLT2 inhibitors; placebo or usual care comparator; hospitalization outcome; randomized trials.
Exclusion. Retracted trials, post hoc analyses of subgroups that were not prespecified, and studies whose only outcome is biochemical with no clinical endpoint.

Psychology (digital interventions).

Inclusion. Adults with depressive symptoms (validated scale); internet-delivered cognitive behavioral therapy; waiting-list or active control comparator; post-treatment depression severity; controlled trials.
Exclusion. Samples with bipolar disorder or psychosis (a comorbidity that changes management), and studies with no standardized outcome measure.

In all three cases, the exclusions add new reasons (comorbidity, no clinical outcome, retraction), not a mere negation of inclusion.

How to document them for PRISMA

Defining the criteria is not enough. You need to leave a trail of them in two places. First, in the protocol (ideally registered in PROSPERO), where they are published before the review is carried out so they cannot be adapted "as convenient". Second, in the final report, following the PRISMA statement.

PRISMA 2020 asks you to show eligibility in two places. In Methods, item 5 (eligibility criteria) asks you to "Specify the inclusion and exclusion criteria for the review and how studies were grouped for the syntheses" (the cleanest way is to structure them by PICOS). And in the PRISMA flow diagram, at the full-text screening stage, you must record how many reports you excluded and why, in the "Reports excluded" box, grouping the exclusions by the category they triggered (wrong population, ineligible design, missing outcome, and so on).

That is why it pays to give each exclusion criterion a short label from the start. It lets you count exclusions by category without going back through every article. To see how this fits into the visual layout, check our guide to the PRISMA flow diagram and its template. If someone reads your Methods and your diagram, they should be able to reconstruct why each excluded study was left out; if they cannot, your criteria are not properly documented.

In revisia you apply your inclusion and exclusion criteria traceably during screening. Create your free account.

Frequently asked questions

When do you apply the inclusion and exclusion criteria?

During study screening. First on title and abstract, then on the full text, always against the same criteria defined in advance.

Can I change the criteria halfway through the review?

Try to avoid it. If a justified adjustment is needed, document it and update the protocol to stay transparent.

How many criteria should I set?

As many as you need to define your question clearly. Not so few that screening becomes ambiguous, and not so many that you exclude relevant evidence.

What are inclusion and exclusion criteria?

They are the explicit rules that decide which studies get into a systematic review and which are left out. Together they are called eligibility criteria. Inclusion criteria describe the features a study must have (population, intervention, comparator, outcome and design); exclusion criteria give specific reasons to drop a study that fits but has a problem that makes it unsuitable. They are set before the search, and they are what makes study selection reproducible.

How do you define inclusion and exclusion criteria?

You derive them from the research question. If you framed it with PICO, each component (Population, Intervention, Comparison, Outcome) becomes a block of inclusion criteria, to which you add the study design and practical limits on language, dates and publication type. Exclusion criteria are only added when they bring a reason that inclusion does not capture. The key rule is that a criterion does not belong if you cannot explain why it is there.

Can language be a criterion?

Yes. Limiting a review to English and Spanish for resource reasons is common, but it is not neutral. It can introduce language bias, because studies with positive results tend to be published in English. If you apply it, state it explicitly as a limitation of your review and, if possible, justify it (for example, by the team's translation capacity).

Where do you report the inclusion and exclusion criteria?

They are documented in two places. Before you start, in the protocol (ideally registered in PROSPERO), and afterward, in the final report following PRISMA. The Methods section describes inclusion and exclusion by component, and the PRISMA flow diagram records the full-text exclusions grouped by reason. Labeling each exclusion criterion from the start makes that count easy.

How many systematic reviews are published each year?

2026-10-12T00:00:00Z

If you feel that a new systematic review on your topic appears every week, you are not imagining it. The answer to how many systematic reviews are published each year is a little dizzying. Where a decade ago a few thousand came out, today they number in the tens of thousands per year, and they pass one hundred thousand according to the broadest source. This post puts numbers on that feeling and explains where they come from and why the figure depends so much on how you decide to count.

How many systematic reviews are published each year

The short answer is more than one hundred thousand in 2025. According to OpenAlex, the broadest open bibliographic database, 105,815 works with "systematic review" in the title were published, compared with 17,228 in 2015. That is a 6.1-fold increase in a decade.

If we stick to PubMed and its stricter official publication type tag, Systematic Review, the 2025 figure is 43,514, compared with 13,804 in 2015. There the multiplier is ×3.2. Two sources, two different numbers, but they point the same way. Output is rising fast.

Put on a daily scale, those 105,815 works from 2025 come to about 290 new systematic reviews every day, holidays included. With PubMed's strict definition, about 120 a day. Keeping up with a field by hand, reading the reviews as they come out, stopped being realistic a long time ago.

Two ways of counting, and why both matter

The gap between 43,514 and 105,815 is not an error. They are two legitimate ways of counting the same thing.

PubMed, by publication type. The US National Library of Medicine tags each article with its type. Systematic Review[Publication Type] is a rigorous definition focused on biomedicine, but it has a catch. NLM introduced that tag in 2019 and applied it to earlier years retroactively and incompletely, so the older years are somewhat undercounted.
OpenAlex, by title. Counting the works that have "systematic review" in the title covers every discipline, not just health, and captures recent output better. In exchange, it lets in some false positives, such as methods guides, protocols or other kinds of review that use the phrase in the title.

Neither one is "the right one". The strict count tells you how many properly tagged biomedical systematic reviews exist; the broad one, how much weight the format carries across science as a whole. That is why, throughout this series, we always state which source each figure uses. If you want to go over exactly what we are counting, the guide on what a systematic review is explains it from scratch, and the comparison of the types of literature reviews helps you tell it apart from its relatives.

The COVID footprint (and an odd dip in 2023)

The curve does not rise perfectly smoothly. Between 2019 and 2021 there is a marked acceleration that coincides with the pandemic. In PubMed, the count went from 24,877 reviews in 2019 to 36,067 in 2021, a 45% jump in two years. The urgency to synthesize evidence on COVID-19, and to revisit almost everything else, sent output soaring.

And there is a detail that often goes unnoticed. 2023 was the first year in which PubMed recorded a drop from the year before (35,051 compared with 37,704 in 2022), before picking up the pace again in 2024 and 2025. A passing bump, not a change in trend, but a reminder that not even this curve rises in a straight line.

The numbers, year by year

For anyone who wants the full table, these are the two counts we used, queried in August 2026.

Year	PubMed (publication type)	OpenAlex (by title)
2015	13,804	17,228
2016	15,441	21,761
2017	17,803	26,062
2018	20,599	27,533
2019	24,877	33,554
2020	30,326	47,291
2021	36,067	57,056
2022	37,704	61,089
2023	35,051	67,059
2024	38,068	76,966
2025	43,514	105,815

In total, OpenAlex holds more than 700,000 works with "systematic review" in the title across its history. The rest of this series uses that corpus to study where those reviews come from: which countries and disciplines they cluster in, which ones are cited most, how many you can read for free, and what problems so much volume brings.

With hundreds of new reviews a day, finding the evidence that matters is the first bottleneck. In revisia AI does the first screening pass and you decide on the uncertain band, always with the quote from the abstract in front of you. Create your free account.

Why they are growing so fast

Several forces are behind the curve at once. Evidence-based medicine established the systematic review as the top rung of evidence synthesis, and journals and agencies ask for it more and more. The format has spread from health to psychology, the social sciences, education and computer science. And reference management and study screening tools have lowered the barrier to entry, so small teams, and even master's dissertations, can take one on today.

That growth has a less pleasant side. When hundreds are published every day, you get overlap, duplication and reviews that add little to what already existed. It is a serious debate within the field itself, and we devote a whole chapter of this series to whether too many systematic reviews are being published.

What this means for your work

If you are about to start your own, the practical lesson has two parts. First, something close to your question almost certainly exists already, so search for previous reviews first and register your protocol in PROSPERO to avoid repeating what has been done. Second, the volume of literature you will have to screen will be high, and that is the stage that takes the most hours; we look at it with data in how long a systematic review takes and in how many studies a systematic review needs.

Want to see the math with your own corpus? In revisia you screen with AI and verify with the quote alongside, from the PICO question to the manuscript. Create your free account and try it.

Frequently asked questions

How many systematic reviews are published worldwide each year?

It depends on how you count. With PubMed's strict definition (the Systematic Review publication type tag), there were 43,514 in 2025. Counting every work with "systematic review" in the title in OpenAlex, which covers all disciplines, there were 105,815 that same year. Both figures have grown sharply over the last decade.

Why are the PubMed and OpenAlex figures so different?

Because they measure different things. PubMed only counts what its indexers have tagged as a systematic review, mostly in biomedicine, and that tag was applied retroactively from 2019. OpenAlex counts any work with the phrase in the title in any field, which widens coverage but lets in some false positives. For the overall trend, the two agree.

How many systematic reviews have been published in total?

OpenAlex holds more than 700,000 works with "systematic review" in the title across its whole history. In PubMed, with the strict publication type tag, the cumulative total is around 341,000. Once again, the difference comes down to what each source counts.

Are they growing faster than the rest of the scientific literature?

Their volume has grown sixfold in a decade according to OpenAlex, a very fast pace. Claiming they grow faster than science as a whole would require comparing against the curve of all publications, which we have not measured here, so we prefer not to claim it. What is certain is that the format carries more and more weight and has spread far beyond medicine.

How many studies does a systematic review need? There is no minimum

2026-10-12T00:00:00Z

How many studies does a systematic review need? There is no minimum number, and you should be wary of anyone who gives you one. The validity of a systematic review doesn't depend on how many studies end up included. It depends on the search being comprehensive and on the selection criteria being defined before you start looking at results. The final number is a result of the method, not a decision you make. That said, there are published figures that serve as a benchmark. The typical Cochrane review includes six trials, there are legitimate reviews with zero included studies, and some searches return thousands of records only to end with a dozen. This guide goes through those figures, explains the funnel they come from, and gives a concrete plan for the two usual scares: coming up short and drowning in results.

Where the number comes from (and why there is no minimum number of studies)

The number of articles in a systematic review is decided in three places, and none of them is up for negotiation at the end. The research question marks out the ground. The inclusion and exclusion criteria define what counts as an eligible study. And the search strategy determines how much of that ground gets covered. Whatever survives screening under those rules is your number. Fixing the number first and then adjusting the criteria until you reach it turns the method upside down, and peer reviewers catch it by comparing the published review with the registered protocol.

That is why PRISMA 2020 asks you to report the whole path in its flow diagram: how many records were identified, how many were screened, how many were excluded (with reasons, at the full-text stage) and how many studies remained. The diagram exists precisely because the final number can only be judged by seeing the path that leads to it.

How many studies a typical systematic review includes

Mallett and Clarke (2002) counted the trials in 1,000 reviews in the Cochrane Library. The typical review included six trials, and in a subsample of 258 reviews the median number of participants was 945, with half of the reviews between 313 and 2,511. Six studies may seem few for so much effort. They are what is left when the question is precise and the criteria are applied without exceptions.
Yaffe and colleagues (2012) found that 376 Cochrane reviews, 8.7% of active reviews, included no studies at all. These are known as empty reviews. Far from being a failure of the method, they are one of its possible answers. They document that, under the stated criteria, there is still no evidence to synthesize, and that also guides the research that comes next.

Both figures come from Cochrane, which means clinical reviews with the trial as the unit. Outside that territory, sizes change. Reviews in education, psychology or engineering tend to include more studies because they accept more designs, and an academic project with a focused question can land in almost any range. The useful benchmark is the funnel from the total identified to the total included, more than a universal number.

The funnel from thousands of records to the included studies

A serious health search easily returns thousands of records. In our guide on how long a systematic review takes we use 5,000 as a normal volume after removing duplicates, and overlaps between databases can inflate the initial raw total by another third. From there, each stage cuts. Title and abstract screening discards the vast majority, full-text reading filters out the candidates that looked like a fit and aren't, and the criteria make the final cut. If a search of thousands ends with six included studies, as in the typical Cochrane review, more than 99% of what was identified fell away along the way. That massive, record-by-record documented discard is exactly what the word systematic buys you.

It pays to absorb that proportion before you start, because the funnel's two usual scares are easier to handle if they don't catch you off guard. One is reaching the end with two studies. The other is opening your first export and finding 8,000 records. Neither one, on its own, means the review is badly designed.

If you get too few (or none)

Before accepting the result, audit the search. A tiny set of results sometimes just means the search string left out synonyms or that a relevant database for your field is missing. Go through the reference lists of the studies you do have, consider the gray literature and, if you can, ask a librarian or information specialist to review the strategy.

If the search was done well, you have three legitimate options and one trap. You can publish with few studies and say so. You can broaden the criteria through a declared amendment to the protocol, with its date and justification. Or you can turn the work into a scoping review if the field turns out to be more scattered than the question assumed. The trap is quietly loosening the criteria until enough studies get in, because that turns the review into a convenience selection dressed up as method.

And with zero? A well-conducted empty review gets published and cited. It records that the question is still unanswered, what evidence was available on the search date and which studies would be needed. For a thesis committee or a journal, a transparent funnel that ends at zero is worth more than an opaque one that ends at twenty.

If your goal was a meta-analysis, two comparable studies are technically enough to pool effects, although with so few the result is fragile and many reviews downgrade it to a supplementary analysis.

If you get thousands

The opposite scare has a plan too. First, diagnose whether you have too much noise or too much question. You tackle noise by refining the search string with more specific terms, validated methodological filters, and date or language limits that are justified and stated in the protocol. You tackle the question by narrowing it with the PICO template, for example with a better-defined population or one primary outcome instead of five. What is not legitimate is cutting by eye once you've seen the results, dropping years or journals until the number becomes manageable, because every cut made after registration smells of convenience.

And if the question is well framed and the volume is what it is, what's left is plain hard work. The numbers are in the guide on how long a review takes. At 45 seconds per abstract and with two reviewers, 5,000 records mean about 125 hours of first-pass screening alone. This is the stage where tools make the biggest difference. In a real review done in revisia, a search of 7,000 records ended with 657 reviewed by hand by the researcher, a pile 90.6% smaller.

Is your export scary? In revisia AI screens every record against your criteria and you review the uncertain band with the quote from the abstract in front of you. Create your free account.

What the examiners (or the peer reviewer) will ask

When your thesis committee or a journal's peer reviewer looks at your number of studies, the real question is about traceability. Why these and not others. The good answer exists before you start, it's called a protocol, and you show it with the PRISMA diagram and the table of excluded studies with their reasons. With that backing, 4 studies are as defensible as 40. Without it, 40 fall apart just like 4.

In revisia that funnel draws itself as you work. The PRISMA diagram fills in with every decision you make, from the total identified to the included studies. Create your free account and see for yourself.

If you are working on an academic project, the guide to the literature review for a thesis or dissertation helps you gauge how much depth your case calls for, and the comparison between systematic and narrative reviews helps you pick the format if the funnel gives you a scare in either direction.

Frequently asked questions

Is there a minimum number of articles for a systematic review?

No. No methods guidance (Cochrane, JBI, PRISMA) sets a minimum number of included studies. The review is valid if the search was comprehensive, the criteria were defined in advance and the process was documented in the PRISMA flow diagram. The final number is a result of the method. In fact, the typical Cochrane review includes six trials (Mallett and Clarke 2002).

How many articles should a systematic review for a thesis or dissertation have?

As many as come out of the funnel with a focused question. In practice that usually means screening hundreds of records and including anywhere from a few studies to a few dozen. More important than the number is being able to show the complete PRISMA flow diagram and justify every full-text exclusion. If the volume explodes, narrow the question in the protocol.

Is a systematic review with few articles, or none, valid?

Yes, if the process was sound. 8.7% of active Cochrane reviews included no studies at all (Yaffe 2012). An empty review documents that the question still has no evidence to answer it, points to the gap and keeps others from repeating the search blind. With few studies, a narrative synthesis and clearly stated limitations carry the publication.

Can I do a meta-analysis with only two or three studies?

Technically, yes. A pooled effect can be calculated from two comparable studies. Robustness is another matter. With so few, heterogeneity can barely be estimated and the interval comes out wide, so the usual approach is to pair the number with a study-by-study narrative synthesis, or to report only the narrative synthesis.

How many references are screened compared with how many end up included?

In health research, a search across several databases easily returns thousands of references, and duplicates can inflate the raw total by around a third. After screening, dozens of candidates remain for full-text review, and the final included studies usually number in the single digits or a few dozen. Having about 1% of what you identified survive, or less, is normal.

How long does a systematic review take?

2026-10-12T00:00:00Z

How long does a systematic review take? Longer than almost everyone estimates at the start. The studies that have measured it put the average at more than a year of work. The figure is not there to discourage anyone. It helps you plan realistically, and it hides some good news. Many of those hours go into mechanical, repetitive tasks, and those can now be compressed without touching the rigor. The published data is detailed enough to answer stage by stage, and to point out where you can shorten the timeline without the review ceasing to be systematic.

The data puts the average at over a year

At least three independent sources have measured it, each with its own method, and they reach a similar conclusion.

Borah and colleagues (2017, BMJ Open) followed 195 systematic reviews registered in PROSPERO through to publication and measured an average of 67.3 weeks from protocol registration to published article, about fifteen and a half months, with some projects finishing in under six months and others taking more than two years.
Michelson and Reuter (2019) estimated the total effort at 1.72 person-years per review. The figure rests on three earlier estimates: a minimum of six months at 10 to 20 hours a week, an average of sixteen months for teams of about five co-authors, and one to two years for a Cochrane review. The authors' calculation averages the first two.
The Galter Health Sciences Library guide (Northwestern University) adds data from the field. Teams working with a librarian take an average of 12.5 months from protocol to completed review.

Be careful when you quote these figures, because they do not measure the same thing. Borah measures calendar time (how many months pass between registering the protocol and publishing) and Michelson and Reuter measure effort (how many hours of work the whole team adds up). You need both to plan. Calendar time tells you when you will have results, and effort tells you how many hours you will have to carve out of your schedule.

How long a systematic review takes, stage by stage

The total looks daunting as a single block, but split by stage it is clear where it comes from. These are the published or approximate times for each stage of the method (if any stage is unfamiliar, they are explained one by one in the guide on what a systematic review is).

Stage	Typical time	Source
Designing the search strategy	2-4 weeks	Galter guide, Northwestern
Running the searches in the databases	2-4 weeks	Galter guide, Northwestern
Title and abstract screening	30-60 seconds per abstract and reviewer (average ≈ 45 s)	Studies that have timed screening
Retrieving full texts	7-10 days per 50 citations	Galter guide, Northwestern
Full-text screening	4-20 minutes per article	Published estimates (Polanin and colleagues, 2019, assume 20 minutes)
Data extraction	36-107 minutes per included study	Timing studies published in 2025
Synthesis, writing and peer review	Several months	Borah 2017 (by difference from the total time)

The table leaves out the protocol, which has no published standard time but rarely takes less than a couple of well-spent weeks (and PROSPERO registration adds its own wait). It also mixes units on purpose, because some rows give calendar time (weeks, days) and others give effort per record (seconds, minutes) that you will have to convert into hours based on your volume. The math comes right below. And the search timeline hides highly specialized work. Librarians who take part in systematic reviews spend an average of 30.7 hours of their own time on each one (Bullers 2018), mostly designing the search string and translating it across databases.

Screening is the bottleneck

People who have timed the process agree that title and abstract screening is the stage that takes the most hours in the whole review. Picture a search that returns 5,000 records after removing duplicates, a normal volume for a health question.

5,000 records × 45 seconds per abstract × 2 independent reviewers ≈ 125 hours for the first screening round alone.

That is more than three weeks of full-time work saying yes or no to abstracts. Add the deduplication beforehand, chasing PDFs (7 to 10 days per 50 citations) and full-text reading of the candidates that pass the filter. None of those hours go into thinking. It is pure grunt work, but someone has to do it.

It depends on volume, of course. With 500 records, the first double screening round takes about 12.5 hours, and with 10,000 it climbs to 250, a whole quarter at half time. To gauge what volume to expect and how many studies usually survive the funnel, see the guide on how many studies a systematic review needs. That is why it pays to invest in a precise search strategy and well-defined inclusion and exclusion criteria before you open the first database.

Screening is the most daunting part. In revisia AI does the first pass and you decide on the uncertain band, always with the quote from the abstract in front of you. Create your free account.

The work that is still yours

There is a part of the work that no tool will do for you, and you should be clear about it before looking for shortcuts. A systematic review is, above all, an exercise in judgment. You have to frame a question worth asking and decide which evidence counts and which does not. Then you interpret results that sometimes contradict each other, write an honest discussion and defend the work to the journal's reviewers. That intellectual work takes months of calendar time, and it is exactly your contribution as a researcher. No tool replaces it. Be wary of any tool that promises to.

The data itself says so. A good share of the 1.72 person-years estimated by Michelson and Reuter goes into thinking, team discussion and writing rather than repetitive tasks.

What can be compressed (and by how much)

The rest of the time is mechanical work, and that can be compressed: deduplicating, screening thousands of abstracts, chasing full texts, moving data into tables and keeping the PRISMA flow diagram counts consistent through to the end.

In revisia, AI screens each record against your criteria and separates the clear yeses and clear noes from the uncertain band. You review that band and the included records, with the verbatim quote from the abstract alongside so you can check each decision in seconds. The doubts are not resolved on their own or in the background. They are put in front of you with the evidence, and you decide them.

There are numbers measured in real use. In one review done in revisia, out of 7,000 records retrieved, the researcher only had to review 657 by hand, 90.6% fewer. In hours of work, the savings in the mechanical stages range from 30 to 120 hours depending on the size of the corpus. And that estimate is conservative, because it uses the fastest published screening rate, 30 seconds per abstract. At the average published rate, about 45, it comes out even higher. The calendar feels it too. The first pass over thousands of records is no longer measured in weeks, and the pace is set by your review of the uncertain band instead of the whole pile.

How to shorten the timeline without losing rigor

None of what follows is a trick. It is standard practice done well, and each point targets a known source of delay.

Protocol before search. Changing the criteria halfway through screening means screening again. A protocol registered in PROSPERO fixes the rules while they are still easy to change.
Put into the search what you want to save in screening. The sharper the search string, the less noise you will have to screen later. If your university offers support from a librarian, use it, because it is expert work measured in tens of hours. Our guides on databases and the search strategy give you the basics.
Before you dive into the thousands, have both reviewers screen the same 50-100 records and measure the disagreement. An ambiguous criterion caught in the sample saves arguments and rescreening later.
Deduplicate before screening. Overlap between databases can inflate the corpus by a third. A reference manager or the screening tool itself sorts it out in minutes.
Use dedicated tools for the mechanical work. For classic blinded double screening, Rayyan and Covidence are the veterans, and revisia adds a verifiable AI first pass and the complete workflow.
Keep traceability up to date. Recording exclusion reasons as you go saves you from reconstructing them at the end for the PRISMA flow diagram.
And if the deadline rules, change the format, not the rigor. A rapid review or a scoping review relaxes steps of the method in a declared, legitimate way. Dressing up an express review as systematic, on the other hand, costs you in peer review.

Want to see the math with your own corpus? In revisia you screen with AI and verify with the quote alongside, from the PICO question to the manuscript. Create your free account and try it.

Frequently asked questions

Can you do a systematic review in a month?

A full systematic review, with a protocol, a search across several databases, double screening, extraction and synthesis, will hardly fit in one calendar month when the published average is over a year (Borah 2017). What does fit in weeks is the mechanical part, if you lean on tools, and there are shorter formats (rapid reviews, scoping reviews) that state their limitations. If the deadline cannot move, narrow the question and the number of databases from the protocol on.

How long does a systematic review take for an undergraduate thesis or a master's dissertation?

The academic calendar sets a different scale, with a narrow question, two or three databases and a corpus of hundreds of records, not thousands. At that size, three to six calendar months of steady work is a reasonable range if you plan from the start. Remember to state the limitations (for example, if you screen alone) and take a look at the guide on the literature review for a dissertation. If you are still torn between formats, first compare systematic review vs narrative review.

How long does screening alone take?

At published rates (30 to 60 seconds per abstract and reviewer), double screening 1,000 records takes 17 to 33 hours of human work for the title and abstract stage alone, plus full-text reading of the candidates (4 to 20 minutes per article). It is the part of the process that takes the most hours and the first place where tools are worth using.

Does working as a team shorten the time?

The calendar time, yes. The total hours, no, because independent double screening doubles them in exchange for controlling bias, although more hands let you split the full-text reading and the extraction. The sixteen-month average reported by Michelson and Reuter (2019) refers to teams of about five authors. Working alone is possible (and common in academic work), but you need to state it as a limitation.

The best databases for systematic reviews

2026-10-12T00:00:00Z

A systematic review is a synthesis of evidence that follows an explicit, reproducible method. Its search cannot depend on a single source. The main methods guidance (PRISMA and the Cochrane Handbook) recommends searching several databases for systematic reviews, usually at least three to five, to reduce study selection bias. In practice the core almost always includes PubMed/MEDLINE, Scopus or Web of Science and, in the health sciences, the Cochrane Library, supplemented with regional and non-English databases (Dialnet, SciELO, LILACS) and subject-specific sources. Each section below explains what the database is, what it covers, how you get access and how to search it.

How many databases for a systematic review, and which ones (PRISMA/Cochrane guidance)

Before discussing which database is best, decide how many you need and which ones. No database indexes all the literature. Each has its own subject focus, journal selection policy and geographic coverage. If you search only one, you will miss relevant studies and your review will lose rigor.

The recommendations in the methods guidance can be summarized like this:

PRISMA 2020 requires you to report every source you searched (databases, registers and other methods) with the date of the last search, so anyone can reproduce it. It doesn't set an exact number, but it assumes you search several, and those counts are the first thing the PRISMA flow diagram asks for.
For reviews of health interventions, the Cochrane Handbook recommends searching at least MEDLINE (through PubMed or another platform), Embase and the Cochrane Central Register of Controlled Trials (CENTRAL). From there you add subject or regional databases depending on the question.
For reviews outside clinical topics (education, psychology, social sciences), the usual approach is to combine a multidisciplinary database (Scopus or Web of Science) with specialized databases (PsycINFO, ERIC) and sources in the language of the region you are studying.

As a rule of thumb, search at least three complementary databases (one multidisciplinary, one specific to your field and one regional or in your language), document the full strategy for each, and record the date and number of results. Before you start searching, work out your literature search strategy with Boolean operators and MeSH terms, because each database requires you to adapt the string to its own syntax.

PubMed / MEDLINE (what it is, non-English literature, how to search)

What PubMed is

PubMed is the free search engine of the US National Library of Medicine (NLM). Its core is MEDLINE, the reference bibliographic database in biomedicine, but PubMed also includes records in the process of being indexed and content from PubMed Central. It is by far the most used source in health reviews. It is free, its biomedical coverage is very broad, and its MeSH controlled vocabulary allows very precise searches.

It helps to keep the terms apart:

MEDLINE. The database (the records indexed with MeSH terms).
PubMed. The free interface that gives access to MEDLINE and a bit more.
MeSH (Medical Subject Headings). The thesaurus the NLM uses to tag each article by topic.

Non-English literature in PubMed

PubMed has no full official interface in other languages. The platform and the MeSH headings are in English. You can still work with literature in other languages, such as Spanish, in several ways:

It indexes many non-English journals, so many articles in those languages show up in the results (with their original abstract).
To search for concepts that come to you in another language, translate your terms into English MeSH vocabulary (the descriptors structure the search) and, in parallel, use the DeCS equivalent (Health Sciences Descriptors), the trilingual version of MeSH in Spanish, Portuguese and English, which is very useful for searching LILACS or SciELO afterward.
Your browser can translate the interface automatically, but this doesn't change the indexing. An effective search is still built with English MeSH terms.

How to search PubMed

A reproducible PubMed search combines free text and controlled vocabulary:

Identify the key concepts in your question (ideally framed with the PICO model).
For each concept, find the matching MeSH term in the MeSH Database.
Combine the MeSH terms with free-text synonyms using Boolean operators (OR within each concept, AND between concepts).
Use filters with caution. Limiting by language or date can introduce bias, so justify it in the protocol.
Save the full strategy and the number of results for the PRISMA flow diagram.

Scopus (what it is, Scopus vs Web of Science)

What Scopus is

Scopus is a multidisciplinary bibliographic and citation database from the publisher Elsevier. Unlike PubMed, it is not free. Access usually comes through your university's or library's institutional subscription. Its coverage spans every field (science, engineering, social sciences, arts and humanities), with thousands of peer-reviewed journals plus conference proceedings and books. Because of its size and its citation analysis, it is one of the preferred multidisciplinary databases outside purely biomedical topics.

Scopus vs Web of Science

Scopus and Web of Science are the two big multidisciplinary citation databases. Which one to use is a very common question, but the methods guidance doesn't frame it as a choice. In a systematic review you should search both, because they only partly overlap and each indexes journals the other doesn't cover. Even so, it helps to know how they differ.

Feature	Scopus	Web of Science
Publisher	Elsevier	Clarivate
Coverage	Very broad, multidisciplinary	Multidisciplinary, with a more selective core (Core Collection)
Historical focus	Broader on recent and regional journals	More selective, with deep citation history
Access	Institutional subscription	Institutional subscription
Typical use	Broad coverage and metrics	Core of high-impact journals and citation analysis

In practice, if you have access to both, search both. If you only have one, state the limitation in your review protocol and back it up with other sources.

Web of Science

Web of Science (WoS), from Clarivate, is the other big multidisciplinary citation database. Its best-known component is the Core Collection, a selective set of indexes (including the Science Citation Index and the Social Sciences Citation Index) that brings together high-impact journals in every field. As with Scopus, access is through an institutional subscription, with no general open access.

Its value for a systematic review is twofold:

Multidisciplinary coverage that complements PubMed (which is biomedical) and Scopus.
Citation analysis with deep historical reach, useful when you need to trace how a study has been cited over time or to run backward and forward citation tracking (snowballing) from the references.

As with Scopus, you have to adapt your search string to WoS syntax and document the search fields, filters, date and number of results.

Cochrane Library (reviews and trials)

The Cochrane Library is a collection of databases focused on evidence for health decision-making. For a clinical systematic review it is close to mandatory. The reason lies in two of its components:

The Cochrane Database of Systematic Reviews (CDSR), which holds Cochrane systematic reviews, known for their methodological quality. It is useful for checking whether a review on your question already exists and avoiding duplicated effort. Its reviews often include meta-analyses built with RevMan, its free software.
The Cochrane Central Register of Controlled Trials (CENTRAL), a very comprehensive register of controlled clinical trials gathered from many sources. This is why the Cochrane Handbook includes it among the minimum recommended searches.

As for access, part of the content is free, and in many countries national or institutional agreements provide full access. The interface is in English, although some plain language summaries are translated. For intervention reviews, searching CENTRAL helps you capture trials that are not always clearly indexed in MEDLINE or Embase.

In revisia you search several sources at once and duplicates are removed automatically before you start screening. Create your free account.

Dialnet, the main Spanish-language database

Limiting the search to English-language databases introduces language bias and leaves out relevant Ibero-American literature. For reviews carried out in Spain or Latin America, or on topics from that region, you should add Spanish-language databases. The largest is Dialnet.

Dialnet is one of the largest Spanish-language bibliographic portals, run by the University of La Rioja. Its coverage is multidisciplinary, with a strong weight in the social sciences and humanities and also in the health sciences, and it includes journal articles, book chapters, theses and conference proceedings from the Spanish-speaking world. Access to the references is free (with more options if you register for Dialnet Plus), although full-text availability depends on each document. It is especially useful for undergraduate theses, master's dissertations and reviews in fields where much of the research is published in Spanish.

How to search Dialnet. Combine your key concepts in the document search box. Dialnet doesn't support PubMed's advanced Boolean syntax, so run several searches with synonyms and also browse the journals and tables of contents in your field. You can filter by document type (article, thesis, book), year and language, and browse by author or journal to find related literature.

Limits for a systematic review. Dialnet is the best way to avoid missing Spanish-language research, but its quality control is uneven and its search is less reproducible than that of an indexed database like PubMed or Scopus. Document your strategy carefully and don't use it as your only source. Exporting. Each record lets you download the citation (RIS) to bring it into your reference manager and deduplicate it against the other databases.

Other Spanish- and Portuguese-language databases: SciELO and LILACS

SciELO

SciELO (Scientific Electronic Library Online) is an open-access electronic library that brings together quality scientific journals from Latin America, Spain, Portugal and other countries. Its coverage is multidisciplinary, and its big advantage is free full text. For a review, it adds peer-reviewed regional literature that is often missing from Scopus or Web of Science.

LILACS

LILACS (Latin American and Caribbean Health Sciences Literature) is the reference database for health sciences in the region, coordinated by BIREME/PAHO/WHO. Access is free and you search it with DeCS descriptors (the Spanish and Portuguese counterpart of MeSH). It is the main regional source for biomedical reviews with an Ibero-American focus and complements PubMed very well.

Subject databases: CINAHL, PsycINFO and Embase

General databases are not enough. Almost every review also needs a source specialized in its field, and these are the three most common.

CINAHL (nursing)

CINAHL (Cumulative Index to Nursing and Allied Health Literature) is the reference database for nursing and allied health (physical therapy, occupational therapy, nutrition, and so on). It indexes journals that are not always in MEDLINE and uses its own subject headings. It requires a subscription, usually available through your university library. If your review is about care, skipping CINAHL leaves out important literature.

PsycINFO (psychology)

PsycINFO, from the American Psychological Association, is the specialized database for psychology and the behavioral sciences. It covers clinical, educational and social psychology and related fields, and includes articles, book chapters and dissertations. Access is by subscription. For reviews in psychology or mental health it is the reference subject source, complementing PubMed and the multidisciplinary databases.

Embase

Embase, from Elsevier, is a large biomedical database that complements MEDLINE. Its coverage is especially strong in pharmacology, medical devices and European literature, and its own controlled vocabulary (Emtree) is more detailed for drugs. It requires a subscription. The Cochrane Handbook considers it part of the minimum search for intervention reviews precisely because it captures studies (mainly European and pharmacological ones) that MEDLINE doesn't index.

Another source that shows up more and more in protocols is Epistemonikos, a free collaborative database focused on systematic reviews and secondary evidence, useful for finding existing reviews on your question.

Comparison table (coverage, access, language)

The table below summarizes the main databases for systematic reviews. Use it as a starting point for choosing your combination, not as a substitute for the methodological decision. What matters is that the sources you choose complement each other and cover your field and your region.

Database	Coverage	Access	Language
PubMed / MEDLINE	Biomedicine and health sciences	Free	English interface (includes articles in several languages)
Scopus	Multidisciplinary, with citation analysis	Subscription	English
Web of Science	Multidisciplinary, selective citation core	Subscription	English
Cochrane Library	Systematic reviews and clinical trials (health)	Partly free, depending on country and institution	English (some summaries translated)
Dialnet	Multidisciplinary, Spanish-language focus (humanities and social sciences)	Free (references)	Spanish
SciELO	Multidisciplinary, Ibero-American open access	Free (full text)	Spanish, Portuguese and English
LILACS	Health sciences, Latin America and the Caribbean	Free	Spanish, Portuguese and English
CINAHL	Nursing and allied health	Subscription	English
PsycINFO	Psychology and behavioral sciences	Subscription	English
Embase	Biomedicine, strong in pharmacology and Europe	Subscription	English

Once you have the references from all these sources, you need a system to manage them and remove duplicates. That is where choosing a reference manager such as Mendeley or Zotero comes in: it centralizes all the results and helps you document the screening of the articles.

Remember that this guide is one piece of the whole process. Read the main guide on what a systematic review is and how to do one step by step to place the database search within the stages of the method.

Bring the results from several databases into one place. In revisia you import, deduplicate and screen without switching tools. Create your free account.

Frequently asked questions

Is PubMed the same as MEDLINE?

Not exactly. PubMed is the free interface that gives access to MEDLINE and to other additional records of biomedical literature.

What is the Cochrane Library?

A collection of evidence-focused databases that includes Cochrane systematic reviews and a register of clinical trials.

Can I do a review with Google Scholar alone?

It is not recommended as your only source. Its coverage is neither transparent nor reproducible, although it can work as a supplement.

What is PubMed?

PubMed is the free search engine of the US National Library of Medicine that gives access to MEDLINE, the reference bibliographic database in biomedicine. Its MeSH controlled vocabulary allows precise searches, and it is the most used source in health reviews because it is free and has very broad coverage.

Scopus or Web of Science?

In a systematic review you should use both rather than pick one, because they index different journals and only partly overlap. If you have access to both (both need an institutional subscription), search both. If you only have one, state it as a limitation in the protocol and strengthen the search with other databases.

What is Dialnet?

Dialnet is a large Spanish-language bibliographic portal run by the University of La Rioja, with multidisciplinary coverage and a strong presence in the humanities and social sciences. Access to the references is free, and it is especially useful for reviews, theses and dissertations in fields where much of the research is published in Spanish.

How many databases should you search in a systematic review?

There is no fixed number, but methods guidance recommends searching several complementary sources, usually at least three. For reviews of health interventions, the Cochrane Handbook suggests searching at least MEDLINE, Embase and CENTRAL, and adding subject or regional databases depending on the question. The key is to combine a multidisciplinary database, one specific to your field and one regional or in your language, and to document the strategy for each one.

Data extraction in a systematic review: table, process and common errors

2026-10-12T00:00:00Z

Data extraction in a systematic review is the stage where you turn each included study into structured, comparable information. You record which population it studied, which intervention it tested, with how many participants, what it measured and what results it got. Everything goes into a form designed in advance. It is the raw material for the synthesis and the meta-analysis, and also the quietest stage of the method. Screening gets talked about a lot because it hurts; extraction hardly at all. Yet this is where a review stakes the reliability of its conclusions. They rest either on faithful data or on typos. This guide covers which fields go in the table, the process the standard asks for, how long it really takes and the documented errors that have spoiled the most syntheses.

Where it falls in the method (and why it matters so much)

In the workflow of a systematic review, extraction comes once screening is over and you know which studies are in. Everything that follows is built on what you extract. The risk of bias assessment uses design details that come from here, and the synthesis or meta-analysis combines the numbers you entered. A transcription error at this stage carries through to the pooled effect, the forest plot and the conclusion.

That is why PRISMA 2020 devotes two full items of its checklist to it, items 9 and 10. Item 9 asks you to "specify the methods used to collect data from reports, including how many reviewers collected data from each report, whether they worked independently" and how any disagreements were handled, and item 10 asks which variables you sought, including those that did not appear in every study. If the extraction method cannot be described, the review cannot be reproduced, and we saw in the comparison with the narrative review that reproducibility is exactly what the effort of the systematic method buys.

Which fields to include in the data extraction table

There is no universal data extraction template, because the fields depend on your research question. There is, however, broad agreement on the blocks a data extraction sheet should cover. This table sums them up with examples.

Block	Typical fields	Example
Identification	Authors, year, country, journal, funding and conflicts of interest	Martín 2023, Spain, public funding
Design and methods	Study type, duration, follow-up, unit of randomization (individual or cluster)	Randomized trial, 12 weeks of follow-up
Participants	Sample size, eligibility criteria, age, sex, setting	n = 84, adults with chronic insomnia
Intervention and comparator	What was done in each arm, dose, frequency, adherence	Melatonin 2 mg vs placebo
Outcomes	Measures used, time points, numerical data per arm	Sleep latency at 8 weeks, mean and SD
Notes for risk of bias	Sequence generation, blinding, losses	12% losses reported per arm
Source of each value	Page, table or sentence the value comes from	Mean in table 2, page 6

Two tips that save a lot of grief. First, define each field in writing (which format, which units, what to do if the value is missing), because "follow-up" can mean three different things to two extractors. Second, take the last block seriously: the source of each value. When the time comes to check a disagreement, that record turns an afternoon of rereading PDFs into a one-minute check.

The extraction process the standard asks for

Extraction has its rituals, and they are not red tape. The documented errors in the next section are the bill for skipping them.

Design the form from the protocol. The variables to extract are declared in the protocol before you see the results, just like the inclusion criteria. Choosing what to extract after reading the studies opens the door to cherry-picking results, even unintentionally.
Pilot it on a few studies. Before you tackle the full corpus, run the form on a handful of varied studies. The pilot exposes ambiguous fields, units that do not fit and data you assumed would be there and are not. Fixing the form after 5 extracted studies is cheap. After 40, it is not.
Extract outcome data in duplicate. The Cochrane standard requires two independent extractors for outcome data, no shortcuts. For the other fields, such as study characteristics, the acceptable minimum is that one person extracts and another checks against the article. The logic is the same as in paired screening. Individual errors are unavoidable, and cross-checking catches them.
Record disagreements and how they were resolved. Consensus, a third person or going back to the text. The mechanism does not matter as long as it is written down, because PRISMA will ask you to report it, just as it asks for the numbers in your PRISMA flow diagram.
Contact the authors when a key value is missing. It is standard practice (PRISMA even asks you to describe how data were confirmed with study investigators), and it shows the gap was not laziness.

If you screen and extract alone, as in many undergraduate theses and master's dissertations, the honest advice is the same as for the rest of the method. State the limitation and offset it where you can, for example by checking your own extraction a few days later, with a blank form, on a sample of studies.

In revisia, extraction comes with verification built in. Every value the AI extracts comes with its verbatim quote from the article next to it, or with a flag if it has none, so confirming a cell takes seconds, not another read of the PDF. Create your free account.

The extraction errors you pay for in the synthesis

The evidence on extraction errors exists, and it is uncomfortable. The classic audit is the one by Gøtzsche and colleagues (JAMA, 2007), who recalculated published meta-analyses by comparing their data with the original articles and found errors in 17 of the 27 they checked. One was retracted after the check. These are the classics:

Transcription errors. The n = 84 of Martín 2023 becomes 48, the standard deviation is entered as a standard error, or the control group's value ends up in the intervention column. They are the silliest and the most frequent, and duplicate cross-checking exists to catch them.
Double counting in multi-arm studies. If a trial compares two doses against the same placebo and you put both comparisons into the meta-analysis, the control group counts twice and the study's weight is inflated.
The same trial, counted twice. A study published in two articles (the interim and the final analysis, or two different journals) can slip into the synthesis as if it were two studies. Compare registration numbers, sample sizes and authors before you accept the list of included studies.
Mixing adjusted and unadjusted effects. One study reports the crude effect and another the effect adjusted for age and sex. Combining them without an explicit rule adds noise that is hard to detect later.
Medians where you needed means. Many studies report the median and interquartile range. Converting them to a mean and standard deviation is possible with published formulas, such as those of Wan and colleagues (2014), but you have to declare it and apply it the same way to every study.
Extracting only the striking result. If the study measured the outcome at three time points and you only enter the favorable one, you have inherited its reporting bias. The form must fix in advance which time point and which measure are extracted.

The reader of the review cannot spot any of these errors. Only a process with a piloted form, duplicate extraction and the source recorded for each value prevents them.

How much extraction costs (and which part can be compressed)

Time studies published in 2025 put extraction at 36 to 107 minutes per included study, the range we already used in the guide to how long a systematic review takes. The numbers add up fast. With 30 included studies and duplicate extraction you get 36 to 107 hours of human work, and that is not a typo. 30 studies times two extractors is 60 extractions, exactly the factor that turns minutes into hours. With the extract-and-check model for fields that are not outcomes, somewhat less. It is one of the densest stages of the review, though not the longest, because unlike screening it does not allow a cruising pace. Every value needs careful reading.

Which part can be compressed? Locating the value, not the judgment. Finding where the article says how many participants completed follow-up is mechanical search work. Deciding whether that value answers your variable, in which unit and what to do if it is ambiguous is reviewer judgment, and you should not hand it off. The tools that really help keep the two apart. They bring you the value with its context and leave the decision to you.

Tools for extracting without losing control

A well-built spreadsheet. For small corpora it is still perfectly respectable. One row per study, the fields defined in writing on a separate tab, and the source column. Its limit is that nothing forces you to be disciplined, and slips do not announce themselves.
Covidence and EPPI-Reviewer. The established review platforms offer extraction modules with structured templates and comparison between extractors. We covered Covidence in the guide to Rayyan and Covidence.
revisia. AI-assisted extraction fills in the form and attaches to each cell the verbatim quote from the article the value comes from. If the value is not in the text or no sentence supports it, the cell says so instead of making it up. Your role shifts from transcribing to verifying, which is exactly what the method asks for.

For the synthesis that follows, the extracted data feed into your meta-analysis software (JASP, jamovi, RevMan or R), so extract with the format that program will ask for in mind.

Want to see your extraction table fill up with the quotes alongside? In revisia the AI extracts and you confirm, from screening to a table ready for synthesis. Create your free account and try it.

Frequently asked questions

What is data extraction in a systematic review?

It is the stage in which the studies included after screening are turned into structured information. From each article, identification details, methods, participants, interventions and results are entered into a pre-designed form. That table is the basis of the synthesis, of the meta-analysis if there is one, and of the risk of bias assessment.

Is double data extraction mandatory?

For outcome data, yes. The Cochrane standard requires two people extracting independently. For the other fields, such as study characteristics, the acceptable minimum is that one person extracts and another checks against the original article. In academic work done alone, it is accepted to state the limitation and offset it, for example by checking your own extraction again after some time has passed.

How do you pilot a data extraction form?

Pick a handful of varied studies from the corpus, extract them with the first version of the form and review which fields turned out to be ambiguous, which units did not fit and which data were systematically missing. Use those lessons to fix the form before extracting the rest, and document the pilot for the methods section.

How long does it take to extract data from one study?

The studies that have timed this stage give a range of 36 to 107 minutes per included study, depending on how complex the article and the form are. With double extraction, a corpus of 30 studies means between 36 and 107 hours of human work (60 extractions, exactly the factor that turns the minutes into hours).

Can I do data extraction in Excel?

For small corpora, yes. One row per study, the fields defined in writing and a column that records which page or table each value comes from. Dedicated platforms add structure, automatic comparison between extractors and traceability, which is what Excel does not enforce and the method does.