Glossary
The terms used on this site, in plain language.
- Adjudication
- Settling every disagreement between two independent data extractions by going back to the source. Each ruling is logged so that the choice can be checked.
- Arm
- One group of a randomized trial, for example the group given semaglutide 1 mg once weekly.
- Baseline characteristic
- A feature of the participants when the trial started, such as age, HbA1c or body mass index, reported as an average or a percent per group.
- Bayesian analysis
- A way of estimating effects that combines the trial data with prior assumptions and describes uncertainty as probabilities. The primary analysis here is Bayesian (multinma, Stan).
- Comparator
- The treatment a medicine is compared with: placebo, standard care or another medicine.
- Confidence interval (CI)
- In the frequentist analysis, a range built so that 95% of such ranges would contain the true effect if the study were repeated many times.
- Consistency
- The assumption that direct evidence (trials of A versus B) and indirect evidence (A versus C and B versus C trials) agree. Its violation is called inconsistency or incoherence.
- Credible interval (CrI)
- The range in which the true average effect lies with 95% probability, given the data and the model. The Bayesian counterpart of a confidence interval.
- Direct evidence
- Evidence from trials that randomized participants between the two treatments being compared.
- Effect modifier
- A characteristic that changes how well a treatment works relative to another, for example baseline HbA1c for glucose lowering. Uneven effect modifiers across comparisons threaten transitivity.
- Estimand
- The precise question a trial result answers, including how events after randomization (such as stopping the drug or starting rescue medication) are handled. Treatment policy and hypothetical estimands are analyzed separately.
- Frequentist analysis
- The classical statistical approach based on confidence intervals and P values. Here netmeta provides a frequentist check of every Bayesian result.
- GLP-1 receptor agonist
- A medicine that acts like the gut hormone glucagon-like peptide 1: it raises insulin when glucose is high, slows stomach emptying and reduces appetite. Semaglutide, liraglutide and dulaglutide are examples; tirzepatide also acts on the GIP receptor.
- Hazard ratio (HR)
- How fast an event happens with one treatment relative to another over time. 0.80 means a 20% lower rate.
- HbA1c
- Glycated hemoglobin: the average blood glucose over about three months, in percent or mmol/mol. A fall of 1 percentage point is clinically large.
- Heterogeneity (tau)
- Differences in the true effect between trials of the same comparison, beyond chance. Tau is its standard deviation on the analysis scale.
- I squared
- The share of the variation between trial results that is due to heterogeneity rather than chance (frequentist).
- Indirect evidence
- Evidence about A versus B obtained through a common comparator C, from A versus C and B versus C trials.
- League table
- A grid comparing every treatment with every other treatment in one network.
- Living systematic review
- A systematic review that is continually updated as new evidence appears.
- Mean difference (MD)
- The difference in average outcome between two treatments, in the outcome’s own units (for example kilograms or percentage points).
- Network
- All trials that compare any of a set of treatments for one outcome, joined through shared comparators.
- Network meta-analysis (NMA)
- A statistical method that combines direct and indirect evidence from a network of trials to compare all treatments at once.
- Node
- One treatment in a network. Here a node is a medicine at one dose, route and frequency; drug level analyses pool the doses of each medicine.
- Node splitting
- A check of inconsistency that estimates each comparison separately from its direct and its indirect evidence and compares the two.
- Odds ratio (OR)
- The odds of an event with one treatment divided by the odds with another. Values below 1 mean fewer events.
- P-score
- The frequentist counterpart of SUCRA: the mean certainty that a treatment is better than the others, from 0 to 1.
- Placebo
- A dummy treatment that looks like the real one but has no active ingredient, used so that participants and staff do not know who gets what.
- Prediction interval
- The range in which the effect of a new trial or setting would be expected to fall, taking heterogeneity into account. It is wider than the credible interval.
- Prognostic factor
- A characteristic that predicts the outcome whatever the treatment, such as age for cardiovascular events.
- R-hat
- A convergence check of Bayesian sampling: values close to 1 (below 1.01) mean the chains agree.
- Randomized controlled trial
- A study that assigns participants to treatments by chance, so that the groups differ only by the treatment received.
- Reference treatment
- The treatment all others are compared with in the main plot, usually placebo.
- SUCRA
- Surface under the cumulative ranking curve: a 0 to 1 summary of how likely a treatment is to rank near the top. It ignores how large the differences are.
- Transitivity
- The assumption that the trials for different comparisons are similar enough in effect modifiers that their results can be combined indirectly.