Terms
The glossary covers 73 terms.
A
- Absolute risk reduction
- The difference between the risk of an outcome in the control group and in the intervention group. It shows how many fewer events occur per person treated, and its reciprocal is the number needed to treat.
- Adjusted estimate
- An effect estimate from an observational study that has been statistically adjusted for other variables, such as age and sex, to reduce confounding. Adjusted and unadjusted estimates should usually be pooled separately.
- AMSTAR 2
- A critical appraisal tool for systematic reviews with 16 items, used to rate confidence in the results of a review. See umbrella review.
- Arm-based model
- A model for network meta-analysis that analyzes the outcome in each arm of each trial directly, as opposed to the contrasts between arms. It is used less often than contrast-based models.
B
- Bayesian meta-analysis
- Meta-analysis in which unknown quantities have probability distributions, combining prior information with the data to give posterior distributions. See Bayesian meta-analysis.
- Bias
- A systematic error that makes an estimate differ from the truth, as distinct from random error. In reviews, it arises in the studies and in the review process itself.
- Bivariate model
- A model that pools sensitivity and specificity together, allowing for their correlation across studies. See diagnostic accuracy meta-analysis.
C
- Certainty of evidence
- The confidence that an estimate reflects the true effect, rated from high to very low. See GRADE.
- Cluster-randomized trial
- A trial in which groups of people, such as clinics or schools, are randomized together. Analyses must allow for the similarity of people within clusters, or the standard errors will be too small.
- Cochrane
- An international organization that produces systematic reviews of health interventions and the Cochrane Handbook of methods. See Cochrane Handbook.
- Confidence interval
- A range of values calculated from the data so that, in repeated sampling, a stated proportion of such intervals would contain the true value. A 95 percent interval is the usual choice.
- Continuity correction
- A small number, often one half, added to the cells of a two-by-two table so that a ratio can be calculated when a cell is zero. It can bias results, and alternatives exist. See zero-event studies.
- Credible interval
- In Bayesian analysis, an interval that contains the unknown quantity with a stated probability, given the model and prior. It differs in interpretation from a confidence interval.
- Cumulative meta-analysis
- Meta-analysis in which studies are added in order, usually by date, and the pooled estimate is recalculated each time, showing how evidence accumulated. See cumulative meta-analysis.
D
- Dose-response
- A relationship between the amount of an exposure and the size of an outcome. See dose-response meta-analysis.
E
- Effect modifier
- A characteristic, such as age or severity, that changes the size of a treatment effect. Effect modifiers are central to the transitivity assumption in network meta-analysis.
- Effect size
- A quantity that measures the size of an effect or association, such as a ratio, a difference or a correlation. See effect sizes.
- Egger's test
- A regression test for funnel plot asymmetry, used to look for small-study effects. See Egger's test.
- Evidence synthesis
- A general term for methods that bring together the results of multiple studies, including systematic reviews, meta-analysis, scoping reviews and qualitative synthesis.
F
- Fixed-effect model
- A meta-analysis model that assumes one common true effect for all studies. See fixed-effect meta-analysis.
- Forest plot
- The standard figure of a meta-analysis, showing each study and the pooled estimate. See how to read a forest plot.
- Funnel plot
- A scatter plot of study effect sizes against their precision, used to look for small-study effects and publication bias. See funnel plots.
G
- GRADE
- A system for rating the certainty of evidence and the strength of recommendations. See GRADE.
- Grey literature
- Material not published in conventional commercial channels, such as reports, theses and conference abstracts. See grey literature.
H
- Hazard ratio
- The ratio of the hazard rates in two groups in a time-to-event analysis. See pooling hazard ratios.
- Hedges' g
- A standardized mean difference with a correction for bias in small samples. It is the most commonly used form of standardized mean difference in meta-analysis.
- Heterogeneity
- Variation in true effects between studies beyond chance. See heterogeneity.
I
- I-squared
- The percentage of variability in estimates attributable to heterogeneity and not chance. See I-squared.
- Indirect comparison
- An estimate of the difference between two treatments obtained through a common comparator, when no head-to-head trial exists. The basis of network meta-analysis.
- Individual participant data
- Participant-level data from studies, as opposed to published summaries. See individual participant data meta-analysis.
- Intention to treat
- An analysis that includes participants in the groups to which they were randomized, whether or not they received the treatment, which preserves the benefit of randomization.
- Inverse-variance method
- The general method of pooling in which each study is weighted by the reciprocal of the variance of its estimate, so that more precise studies count more.
L
- Living systematic review
- A systematic review that is updated continually as new evidence appears. See living systematic review.
M
- Mantel-Haenszel method
- A method of pooling binary outcomes using the counts directly, with good properties when events are few or studies are small.
- Mean difference
- The difference between the average outcome in two groups, in the original units of measurement. Used when all studies measure the outcome on the same scale.
- Meta-analysis
- The statistical combination of results from separate studies. See what is meta-analysis?
- Meta-regression
- A regression of effect sizes on study characteristics. See meta-regression.
- MOOSE
- The reporting guideline for meta-analyses of observational studies. See MOOSE.
N
- Network meta-analysis
- Meta-analysis comparing several interventions using direct and indirect evidence. See network meta-analysis.
- Number needed to treat
- The number of people who need to be treated for one additional person to benefit, calculated as the reciprocal of the absolute risk reduction.
O
- Odds ratio
- The ratio of the odds of an outcome in two groups. See odds ratio versus risk ratio.
P
- Peto method
- A fixed-effect method for pooling odds ratios from observed and expected events. It suits rare events with small effects and balanced groups.
- PICO
- A framework for framing questions by population, intervention, comparator and outcome. See PICO frameworks.
- Pooled estimate
- The combined effect estimate from a meta-analysis, a weighted average of the study estimates.
- Prediction interval
- The range in which the true effect in a new study is expected to fall. See prediction intervals.
- Prior distribution
- In Bayesian analysis, a probability distribution describing what is believed about an unknown quantity before seeing the data.
- PRISMA
- The reporting guideline for systematic reviews. See PRISMA 2020.
- PROSPERO
- An international register of systematic review protocols. See PROSPERO registration.
- Publication bias
- The tendency for studies with certain results, usually positive or significant ones, to be published more readily. See publication bias.
Q
- Q statistic
- A measure of the weighted squared differences between study estimates and the pooled estimate, used to test heterogeneity. It is compared with a chi-squared distribution.
R
- Random-effects model
- A model that allows true effects to vary between studies and estimates their mean and variance. See random-effects meta-analysis.
- Reference standard
- In diagnostic accuracy studies, the best available method of determining whether a person truly has the condition, against which a test is compared.
- Risk difference
- The absolute difference between the risks in two groups. It is the same as the absolute risk reduction when the intervention reduces risk.
- Risk of bias
- The likelihood that features of a study design or conduct distort its result. See risk-of-bias tools.
- Risk ratio
- The ratio of the risk of an outcome in two groups, also called relative risk. A value of 1 means no difference.
- RoB 2
- The revised Cochrane tool for assessing risk of bias in randomized trials, which has five domains.
- ROBINS-I
- A tool for assessing risk of bias in non-randomized studies of interventions, with seven domains, comparing each study with an ideal target trial.
S
- Scoping review
- A review that maps the extent and nature of evidence on a topic. See scoping review.
- Selective reporting
- Reporting only some of the outcomes or analyses that were carried out, usually the favorable ones. Comparing publications with protocols can reveal it.
- Sensitivity (of a test)
- The proportion of people with a condition whom a test correctly identifies as positive. It is a measure of accuracy, not to be confused with sensitivity analysis.
- Sensitivity analysis
- An analysis repeated with different assumptions or data to test whether the conclusion changes. See sensitivity analysis.
- Small-study effects
- The tendency of smaller studies to show different, often larger, effects than larger ones, of which publication bias is one cause. See small-study effects.
- Specificity
- The proportion of people without a condition whom a test correctly identifies as negative.
- Standardized mean difference
- A difference in means expressed in standard deviation units, allowing studies with different scales to be combined. See standardized mean difference.
- Subgroup analysis
- A comparison of effects across categories of a study or participant characteristic. See subgroup analysis.
- Summary of findings table
- A table presenting, for each important outcome, the number of studies and participants, relative and absolute effects and the certainty of evidence.
- Systematic review
- A review that uses explicit, reproducible methods to find, appraise and synthesize all relevant studies. See systematic review.
T
- Tau-squared
- The between-study variance of true effects in a random-effects model. See tau-squared.
- Transitivity
- The assumption in network meta-analysis that trials making different comparisons are similar enough in effect modifiers for indirect comparisons to be valid. See transitivity and inconsistency.
- Trim-and-fill
- A method that estimates the number of missing studies from funnel plot asymmetry and adjusts the pooled estimate. Used as a sensitivity analysis. See trim-and-fill.
U
- Umbrella review
- A review that synthesizes existing systematic reviews. See umbrella review.
- Unit of analysis
- The entity that is treated as one independent observation, such as a participant, a cluster or a study. Errors arise when it differs from the unit of randomization.
W
- Weight
- The share of the total influence that a study has on the pooled estimate, usually based on the precision of its estimate.
Frequently asked questions
What is the difference between sensitivity and sensitivity analysis?
Sensitivity of a test is the proportion of people with a condition whom it identifies as positive. A sensitivity analysis repeats a meta-analysis with different assumptions or data to see whether the conclusion changes.
What is the difference between a risk ratio and an odds ratio?
A risk ratio compares probabilities of an outcome in two groups, while an odds ratio compares odds. They are similar when the outcome is rare and differ when it is common.
What is the difference between a confidence interval and a credible interval?
A confidence interval is defined through repeated sampling. A credible interval, from a Bayesian analysis, contains the true value with a stated probability given the model and prior.
What is the difference between a systematic review and a meta-analysis?
A systematic review is the structured process of finding and appraising studies, and a meta-analysis is the optional statistical combination of their results.
References
- Higgins JPT, Thomas J, Chandler J, et al., editors. Cochrane Handbook for Systematic Reviews of Interventions. Cochrane; current version available at training.cochrane.org/handbook. See its glossary.
- Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-Analysis. Wiley; 2009.
- Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71
- Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924-926.