Methodology Hub

The standards your peer reviewers actually read.

Plain-English explainers for the frameworks, risk-of-bias instruments, evidence-rating systems, and statistical diagnostics that make a systematic review defensible. New entries publish weekly.

Evidence rating

How to translate a body of evidence into a single, structured rating of certainty.

Quality assessment

Picking the methodologically appropriate risk-of-bias instrument for the study design at hand.

  • RoB 2: the five domains explainedComing

    The randomisation, deviations, missing-outcome, measurement, and selection-of-result domains, and what counts as some-concerns vs. high.

  • ROBINS-I: assessing confounding in non-randomised studiesComing

    When ROBINS-I is the right tool, the seven domains, and the role of pre-specified confounders.

  • QUADAS-2: the four domains for diagnostic-accuracy reviewsComing

    Patient selection, index test, reference standard, and flow/timing — the QUADAS-2 framework end-to-end.

  • QUIPS: prognostic-factor study qualityComing

    When to use QUIPS over RoB 2, the six domains, and the place of attrition in prognostic-factor reviews.

Frameworks

PICOS and its variants — the structures that turn a research question into an answerable, searchable, and analysable plan.

  • PICOS vs. PCC vs. PIRD: choosing the frameworkComing

    Intervention reviews use PICOS, scoping reviews use PCC, diagnostic-accuracy reviews use PIRD — and the framework drives every downstream choice.

  • PRISMA 2020 and its extensionsComing

    Where PRISMA-ScR, PRISMA-DTA, and PRISMA-NMA take over from the base 2020 checklist.

Statistics

The estimands and diagnostics that determine whether a meta-analysis can be trusted.

  • Heterogeneity: I², τ², and when high heterogeneity mattersComing

    Distinguishing statistical heterogeneity from clinical heterogeneity, and when subgroup analysis is the right answer.

  • Effect measures: OR, RR, RD, SMD, and MDComing

    Picking the right scale for binary, continuous, and time-to-event outcomes, with worked examples.

  • Prediction intervals — the under-reported sibling of CIsComing

    Why a 95% CI on a random-effects pooled estimate isn't the whole story, and what a prediction interval adds.