Method glossary

The terms that appear in our appraisals and verdicts. Click a term to open its definition.

• 95% CI

  95% confidence interval — the range where the true value probably lies.

• absolute risk

  The real difference in risk between the groups (e.g. 2 in 100 vs 1 in 100).

• AMSTAR-2

  Tool for appraising the quality of systematic reviews (16 items, 7 critical).

• Bradford Hill

  A set of viewpoints (temporality, dose-response, consistency…) for weighing cause and effect in observational data.

• CKD

  Chronic kidney disease.

• clinical outcome

  Concrete event experienced by the patient (e.g. heart attack, death), not a laboratory marker.

• cohort

  Study that follows a group over time to see who develops the outcome.

• comparator

  The group or condition the intervention is compared against (e.g. placebo, another diet).

• CONSORT

  Checklist for how to REPORT a trial — transparency of reporting, not quality.

• cross-sectional

  Study that measures everything at a single moment — it shows association, not cause.

• effectiveness

  Effect in the real world, with imperfect adherence and varied patients.

• efficacy

  Effect measured under ideal, controlled conditions (inside the trial).

• eGFR

  Estimated glomerular filtration rate — how much the kidneys filter, which is ONE of the kidney's functions.

• FFQ

  Food frequency questionnaire: it estimates habitual diet from memory, and is prone to error.

• fixed effects

  Meta-analysis model that assumes a single true effect common to the studies.

• GFR

  Glomerular filtration rate — how much the kidneys filter the blood.

• GRADE

  System that grades the certainty of the body of evidence, outcome by outcome.

• hard outcome

  Event that matters to the patient (heart attack, death, fracture) — not an intermediate marker.

• healthy-user

  Someone who takes up a habit seen as healthy tends to have other healthy habits alongside it — it confounds the analysis.

• HR

  Hazard ratio — risk over time (>1 higher, <1 lower).

• intention-to-treat

  Analyses each person in the group they were randomized to, even if they dropped out — it avoids inflating the effect.

• I²

  Measures how much of the difference between studies is real (not chance): 0% = none; high = a lot.

• Mendelian randomization

  Uses genetic variants as a 'natural lottery' to test cause and effect.

• meta-analysis

  Statistically combines the results of several studies into a single estimate.

• narrative review

  Opinion or context piece, without a systematic search or a reproducible appraisal of the studies.

• Newcastle-Ottawa

  Scale for appraising the quality of cohort and case-control studies.

• NNT

  Number of people who need to be treated to prevent one event — it turns the effect into something practical.

• normative leap

  Jumping from weak evidence straight to 'change practice or guidelines' — a common overreach.

• OR

  Odds ratio — how many times more (>1) or less (<1) likely the outcome is.

• p-value

  Probability of an effect equal to or larger than the one observed showing up if there were no real effect; it measures neither size nor importance.

• PICO

  The outline of the question: Population, Intervention (or exposure), Comparison and Outcome.

• placebo

  Inert treatment used as a comparison to isolate the real effect of the intervention.

• power

  The study's ability to detect an effect that is there; a small sample has little power.

• prevalence ratio

  Measure of association typical of cross-sectional studies (it compares the proportion who have the outcome).

• PRISMA

  Checklist for how to report a systematic review.

• publication bias

  Studies with a 'positive' result get published more, and faster, which distorts the whole.

• QUADAS-2

  Tool for appraising diagnostic accuracy studies.

• random effects

  Meta-analysis model that allows the effect to vary between studies.

• regression to the mean

  Natural tendency of extreme values to come back closer to the average at the next measurement.

• relative risk

  How many times the risk changes between the groups (e.g. 'it doubled') — it can look large while being little in absolute numbers.

• reverse causation

  When it is the disease that changes the measured factor — and not the other way around.

• RoB 2

  Standard tool for assessing risk of bias in randomized trials (5 domains).

• ROBINS-E

  Tool for risk of bias in exposure studies (e.g. diet → outcome).

• ROBINS-I

  Tool for risk of bias in intervention studies without randomization.

• scoping review

  Maps what exists on a topic, without judging the quality of the studies.

• SMD

  Standardized mean difference — effect size in standard deviations (0 = no difference).

• STROBE

  Checklist for how to report observational studies.

• surrogate outcome

  Intermediate marker (e.g. GFR, cholesterol), not the event that matters to the patient.

• systematic review

  Searches and appraises, in a structured way, every study on one question.

• vote-counting

  Counting how many studies fell on each side, without combining the effect sizes — a weak synthesis.