Reference

Systematic review glossary

45 terms · Updated September 5, 2026

Definitions for the methodological vocabulary used across this e-book. Each entry links to the chapter where the concept is covered in depth.

Review types and frameworks

Systematic review: A review that answers a defined research question using prespecified, systematic, and reproducible methods to identify, appraise, and synthesize all relevant studies. Its transparency and prespecification are what place it at the top of the evidence hierarchy. Read the Overview chapter.
Rapid review: A review that compresses or single-reviews stages such as screening and extraction to deliver findings faster. Appropriate for time-critical decisions but generally considered less methodologically rigorous than a full systematic review with dual independent review. Read the Research question chapter.
PICO(Population, Intervention, Comparison, Outcome): The standard framework for framing quantitative clinical review questions. Each element maps directly onto search concepts and onto the eligibility criteria applied during screening. Read the Research question chapter.
SPIDER(Sample, Phenomenon of Interest, Design, Evaluation, Research type): A question framework suited to qualitative and mixed-methods reviews, where a discrete intervention and comparator may not exist in the way PICO assumes. Read the Research question chapter.
PEO(Population, Exposure, Outcome): A question framework used for reviews about exposures and lived experiences rather than deliberate interventions. Read the Research question chapter.
SPICE(Setting, Perspective, Intervention, Comparison, Evaluation): A question framework common in the social sciences, which makes setting and perspective explicit parts of the question. Read the Research question chapter.

Searching

Search strategy: The documented plan for finding evidence: which databases and interfaces are searched, on what dates, with which queries, plus any supplementary methods such as citation chasing or grey-literature searching. Read the Searching for articles chapter.
Search query: The expression of a search strategy in one database's native syntax, combining synonyms within a concept using OR, concepts using AND, plus field limits, phrase and proximity operators, and controlled vocabulary. Read the Searching for articles chapter.
Controlled vocabulary: A database's standardized indexing terms, such as MeSH in PubMed or Emtree in Embase. Because these vocabularies differ, a query cannot be reused verbatim across databases and must be translated for each one. Read the Searching for articles chapter.
MeSH(Medical Subject Headings): The controlled vocabulary used to index records in MEDLINE and PubMed. Searching MeSH terms retrieves records by assigned subject rather than by the words that happen to appear in the title or abstract. Read the Searching for articles chapter.
Sensitivity (of a search): The proportion of genuinely relevant studies a search retrieves. Systematic review searches deliberately favour sensitivity over precision, which is why synonym expansion matters and why search results contain a large share of irrelevant records. Read the Searching for articles chapter.
Grey literature: Research outputs not published in peer-reviewed journals, including preprints, theses, conference abstracts, and institutional reports. Including it can reduce publication bias and matters most in fast-moving fields where formal publication lags. Read the Searching for articles chapter.
Deduplication: Removing records retrieved more than once because an article is indexed in several databases. Matching on unique identifiers such as DOI and PMID resolves most duplicates; residual duplicates arising from inconsistent metadata are typically removed during screening. Read the Searching for articles chapter.

Screening and extraction

Eligibility criteria: The explicit inclusion and exclusion rules every retrieved record is judged against, covering population, intervention or exposure, comparator, outcomes, eligible study designs, and any limits on language, date, or setting. Read the Screening chapter.
Title and abstract screening: The first screening phase, in which reviewers decide include, exclude, or unclear using only citation metadata. It filters the bulk of irrelevant records before anyone reads full text. Read the Screening chapter.
Full-text screening: The second screening phase, in which every record marked include or unclear is read in full and judged against the same eligibility criteria. Reasons for exclusion at this stage are reported in the PRISMA flow diagram. Read the Screening chapter.
Dual independent review: Having two reviewers assess each record without seeing each other's decisions, with disagreements resolved by a predefined rule. It is the mechanism that makes screening and extraction decisions traceable and defensible. Read the Screening chapter.
Adjudication: Resolution of a screening or extraction conflict, conventionally by a third senior reviewer who was not part of the disagreeing pair. The rule should be set before screening begins. Read the Screening chapter.
Extraction fields: The structured items captured from each included study, such as population characteristics, intervention details, outcome definitions, numeric results with measures of precision, and risk-of-bias items. They determine what the results section can support. Read the Extraction chapter.
Evidence table: A table summarizing the characteristics and findings of included studies, keyed to the extraction fields. The narrative should highlight patterns and outliers rather than restate every row. Read the Extraction chapter.
Risk of bias: An assessment of how far a study's design and conduct may distort its results. Reviews name the tool used, such as RoB 2 for randomized trials or the Newcastle–Ottawa Scale for some observational designs, and report how many reviewers applied it. Read the Writing chapter.

Synthesis and meta-analysis

Evidence synthesis: The stage that decides how extracted data are brought together so readers can see what the body of evidence shows, and whether that happens through statistical pooling or a structured narrative. Read the Evidence synthesis chapter.
Meta-analysis: Statistical pooling of effect estimates across studies into a summary effect with a confidence interval. Appropriate when studies are similar enough across PICO elements for a pooled estimate to be meaningful and heterogeneity is not so large that an average would mislead. Read the Meta-analysis chapter.
Narrative synthesis: Structured description and comparison of findings without statistical pooling, used when studies differ too much in design, population, or intervention, or when outcome reporting is too incomplete to pool defensibly. Read the Evidence synthesis chapter.
SWiM(Synthesis Without Meta-analysis): Reporting guidance for systematic reviews that synthesize results without statistical pooling, published by Campbell and colleagues in 2020. It exists to keep narrative synthesis rigorous, transparent, and reproducible rather than subjective. Read the Evidence synthesis chapter.
Effect size: A standardized quantity expressing the magnitude of a result, which allows studies to be placed on a common scale before pooling. Without harmonization to a shared metric, numbers from different studies are not commensurable. Read the Meta-analysis chapter.
SMD(Standardized mean difference): An effect size for continuous outcomes measured with different instruments or scales, such as different depression questionnaires. Hedges' g is a variant that corrects for small-sample bias. Read the Meta-analysis chapter.
Odds ratio(OR): An effect measure for binary outcomes expressing the ratio of the odds of an event between groups. The null value is 1.0. Reviews should state explicitly which contrast was modelled. Read the Meta-analysis chapter.
Relative risk(RR): An effect measure for binary outcomes expressing the ratio of event risk between groups. Like the odds ratio, its null value is 1.0, but it is often easier to interpret clinically. Read the Meta-analysis chapter.
Heterogeneity: The extent to which effect estimates scatter across studies beyond what sampling error alone would predict. High heterogeneity means studies may effectively be answering different questions, so a single pooled average can mislead. Read the Meta-analysis chapter.
I² statistic: An estimate of the percentage of total variation across studies attributable to genuine heterogeneity rather than within-study error, calculated as ((Q − df) / Q) × 100%. Roughly 0–40% is often treated as low and above about 75% as considerable, but these are rules of thumb rather than hard rules. Read the Meta-analysis chapter.
Cochran's Q: A test for whether between-study dispersion exceeds what chance would predict under a fixed-effect framing. It is useful but has low power when few studies are available. Read the Meta-analysis chapter.
Random-effects model: A pooling model that assumes true effects genuinely differ across studies and estimates both within-study and between-study variance. It is the default for most contemporary reviews because each study estimates its own context-specific effect. Read the Meta-analysis chapter.
Fixed-effect model: A pooling model that assumes one shared true effect underlies all studies, with observed differences reflecting sampling error only. Rarely plausible outside highly standardized settings, so it is uncommon in contemporary reviews. Read the Meta-analysis chapter.
Forest plot: The central figure of a meta-analysis, showing each study's effect estimate and confidence interval on a common axis with a diamond summarizing the pooled estimate. The null line sits at 1.0 for ratio measures and 0 for mean differences. Read the Meta-analysis chapter.
Funnel plot: A scatter of effect size against precision used to look for asymmetry that may indicate missing small or negative studies. Asymmetry is suggestive of publication bias but has other possible causes, so it warrants cautious interpretation. Read the Meta-analysis chapter.
Publication bias: Distortion of the evidence base arising because studies with statistically significant or positive results are more likely to be published. Often probed with funnel plots and formal tests such as Egger's regression. Read the Meta-analysis chapter.
Meta-regression: A method for examining whether study-level characteristics such as dose, population, or risk-of-bias stratum explain variation in effect sizes. Best prespecified, and underpowered when few studies are available. Read the Meta-analysis chapter.
Subgroup analysis: Pooling within predefined strata to explore whether effects differ by population, setting, or design. Prespecify subgroups wherever possible, since post-hoc subgrouping inflates the chance of spurious findings. Read the Meta-analysis chapter.

Reporting and registration

PRISMA 2020(Preferred Reporting Items for Systematic Reviews and Meta-Analyses): The current reporting standard for systematic reviews, published by Page and colleagues in 2021. It specifies what must be reported, not how the review must be conducted. A PRISMA-AI extension has been registered with EQUATOR since 2022 but remains under development. Read the Writing chapter.
PRISMA flow diagram: The figure accounting for the path of every record from identification through deduplication, each screening stage, full-text decisions with exclusion reasons, and final inclusion. Its counts must reconcile with the text and tables. Read the Writing chapter.
PRISMA-S: The PRISMA extension for reporting literature searches, published by Rethlefsen and colleagues in 2021. It structures how databases, interfaces, dates, queries, and supplementary search methods are described. Read the Writing chapter.
PROSPERO: The most widely used prospective register for health-related systematic review protocols. Registration should occur before screening begins, and the record should be cited in the manuscript. Read the Writing chapter.
Protocol: The prespecified plan for a review, covering question, eligibility criteria, search strategy, screening and extraction procedures, and synthesis approach. Material deviations from it must be reported with reasons. Read the Overview chapter.
Audit trail: The record linking every reported claim back to the decision or source that produced it. Current expectations are that another team could understand, and often approximate, the process from the paper and its supplements alone. Read the Writing chapter.