Topic

Evidence synthesis and appraisal

A systematic review finds and appraises the relevant studies on a question, and meta-analysis pools their results into one estimate. Network meta-analysis and other indirect comparisons extend this to treatments never tested head to head. Random effects models, I-squared and the Egger test deal with heterogeneity and possible publication bias.

Concepts in this topic

• Aggregate Data Meta-AnalysisAggregate data meta-analysis pools study-level summary results, such as effect estimates and standard errors, from several studies into one estimate.
• Analysis ErrorAnalysis error is a mistake in specifying, coding or reporting a statistical or economic analysis, such as a model slip that alters an ICER.
• Applicability AssessmentApplicability assessment judges whether an economic evaluation or model fits a decision problem's population, comparators, care setting and methods.
• Bayesian Meta-AnalysisBayesian meta-analysis combines prior distributions with results from several studies to give posterior and predictive distributions for an effect.
• Begg TestThe Begg test is a rank correlation test (Kendall's tau) for small-study effects in a meta-analysis, a possible sign of publication bias.
• Between-Study HeterogeneityBetween-study heterogeneity is variation in effects across trials in a meta-analysis beyond chance, so one pooled effect may not fit every HTA setting.
• Construct ValidityConstruct validity is how well a health measure's scores fit hypotheses about what it measures, such as links to other measures or group differences.
• Convergent ValidityConvergent validity is evidence that a health measure correlates, as predicted in advance, with other measures of the same or a closely related construct.
• Copas ModelThe Copas model is a selection model for publication bias in meta-analysis that re-estimates the pooled effect under assumed rates of non-publication.
• Criterion ValidityCriterion validity is the degree to which a measure's scores agree with a gold standard for the same concept, assessed at the same time or later.
• Discriminant ValidityDiscriminant validity is evidence that a health measure correlates less with measures of other constructs than with its own, as predicted in advance.
• Egger TestA statistical test for funnel plot asymmetry, and by extension potential publication bias, using regression of effect estimates against standard errors.
• EvidenceEvidence is the information from studies, routine data and expert judgement used in health economics to estimate costs and outcomes of decision options.
• Feasibility AssessmentAn evaluation of whether a proposed study or intervention can practically be conducted given available resources, infrastructure, and population characteristics.
• Fixed Effect ModelA meta-analysis model assuming all included studies estimate the same true effect, with observed variation attributed entirely to sampling error.
• Fixed Effect SynthesisAn approach to combining evidence assuming all studies estimate a single common true effect, weighting purely by statistical precision.
• Frequentist Meta-AnalysisAn approach combining evidence using classical frequentist methods, producing a point estimate and confidence interval without formally incorporating prior beliefs.
• Heterogeneity AssessmentHeterogeneity assessment evaluates the nature, extent, and possible causes of differences in effects or outcomes across studies, populations, or settings.
• I-SquaredA statistic quantifying the percentage of total variation across studies in a meta-analysis attributable to genuine heterogeneity rather than chance.
• Indirect ComparisonA method estimating the relative effect of two treatments never studied against each other directly, using their separate effects versus a shared comparator.
• Indirect Treatment ComparisonIndirect treatment comparison (ITC) is the umbrella term for methods, such as NMA or MAIC, that compare treatments no trial has tested head to head.
• Individual Patient Data Meta-AnalysisIndividual patient data meta-analysis systematically obtains, checks, and jointly synthesises participant-level records from eligible studies to estimate effects and investigate variation across studies or patients.
• Information BiasA category of bias from errors in how exposure, outcome, or covariate data are measured or recorded, including recall bias and measurement error.
• Inter-Rater ReliabilityThe degree of agreement between two or more independent observers assessing the same phenomenon, commonly quantified using a statistic such as Cohen's kappa.
• Internal Consistency ReliabilityA measure of how well different items within an instrument, all meant to assess the same construct, produce correlated results.
• Intra-Rater ReliabilityThe degree of agreement between repeated assessments made by the same rater evaluating the same phenomenon on separate occasions.
• Measurement ErrorMeasurement error is the discrepancy between an observed or recorded value and the intended value of the quantity being measured, arising from the measurement process rather than sampling alone.
• Meta-AnalysisMeta-analysis statistically combines effect estimates from multiple studies addressing a sufficiently similar question.
• Multivariate Meta-AnalysisA meta-analysis method jointly analysing two or more related outcomes simultaneously, accounting for the correlation between them.
• Network Meta-AnalysisNetwork meta-analysis (NMA) compares three or more treatments in one analysis, pooling direct and indirect trial evidence to inform HTA decisions.
• Network Meta-Analysis HTAThe application of network meta-analysis specifically within health technology assessment, generating comparative effectiveness estimates when treatments were never studied together.
• Node-SplittingA technique testing for inconsistency in a network meta-analysis by separately estimating direct and indirect evidence for one comparison and comparing them.
• One-Stage Meta-AnalysisAn individual patient data meta-analysis analysing combined patient-level data from all studies within a single unified statistical model.
• P-ScoreA frequentist network meta-analysis measure ranking treatments by the extent one is expected to outperform every other treatment in the network.
• Pairwise Meta-AnalysisPairwise meta-analysis is the statistical pooling of trials of the same two treatments, such as head-to-head RCTs, into one estimate of relative effect.
• Quality of EvidenceA judgement about how much confidence can be placed in an effect estimate, based on study design, risk of bias, and precision.
• Random Effects ModelA random effects model represents variation in a parameter across defined units or settings by treating their unit-specific effects as draws from a shared distribution.
• Reliability AssessmentAn evaluation of the consistency and reproducibility of an instrument's results, covering internal consistency, inter-rater, and test-retest reliability.
• Simpson ParadoxA statistical phenomenon where a trend seen within subgroups reverses or disappears once those subgroups are combined into a single aggregated analysis.
• SUCRAAn abbreviation for the surface under the cumulative ranking curve, ranking treatments in a Bayesian network meta-analysis by probability of being most effective.
• Systematic ReviewA structured, comprehensive, reproducible approach to identifying, evaluating, and synthesising all available evidence relevant to a clearly defined question.
• Tau-SquaredA statistical parameter representing the estimated variance of the true treatment effect across studies in a random effects meta-analysis.
• Test-Retest ReliabilityThe degree of agreement between scores from the same instrument given to the same people on two separate occasions, when no real change is expected.
• Two-Stage Meta-AnalysisAn individual patient data meta-analysis first analysing each study separately, then combining the resulting study-level estimates using standard methods.