Curriculum/Method/Weighing evidence
Method · how we weigh evidence

Hierarchy is
only one axis.

A meta-analysis is not automatically more useful than a direct trial, and a guideline is not a study design. This is the method every module uses to decide whether a source should change what you do—so the clinical pages can stay clinical.

Three visible layers

Always separate what is settled from what is not.

Every clinical page keeps these three layers visibly apart, so a fresh signal is never mistaken for a recommendation, and our own reasoning is never dressed up as a society's.

The convention

New does not automatically mean practice-changing.

A trial, an approval, or a correction can matter enormously—or not yet. Labeling the layer is how the reader keeps the distinction.

  • Established guidanceWhat a current guideline, regulator, or consensus recommends, with its class or certainty where available.
  • New evidenceTrials, safety signals, approvals, and corrections that postdate the guideline and are not yet incorporated.
  • Expert synthesisOur reasoned integration, labeled as inference—never presented as a society recommendation.

Two axes, not one ladder

Study design and recommendation authority are different questions.

For therapy, a high-quality synthesis of direct trials often sits above a single trial, then comparative observation, then uncontrolled signals. But guidelines, consensus, and labels are implementation authorities, not study designs—they belong on their own axis.

Evidence design
  1. SynthesisSystematic review / meta-analysis
  2. ExperimentRandomized trial
  3. ObservationCohort / case-control
  4. SignalsCase series / reports / pharmacovigilance
  5. MechanismHuman physiology / translational evidence
Recommendation authority
  1. GuidelineMethods, evidence review, classes/levels
  2. ConsensusExpert integration where evidence is incomplete
  3. RegulatoryLabel, safety communication, approval review
  4. InstitutionalLocal formulary, workflow, resources
  5. Our synthesisExplicit inference, never disguised as guidance

What actually decides usefulness

Judge each source on seven questions, then whether it moves the bedside.

Design rank is a starting point, not a verdict. A source earns its place—or loses it—on the questions below, and a claim only changes practice when directness and applicability hold for the patient in front of you.

01Question

Is this a therapy, diagnosis, prognosis, harm, or dosing question?

Each prioritizes a different design—rare harms lean on pharmacovigilance; dosing on PK/PD and organ-impairment studies.

02Risk of bias

How was it conducted, and by whom?

Randomization, blinding, allocation, and conflicts shape how much the estimate can be trusted.

03Directness

Does it answer this exact question in this population?

Surrogates, selected samples, and off-label extrapolations widen the gap between the study and the patient.

04Precision

How wide is the uncertainty?

Read the confidence interval, not just the point estimate; a nonsignificant result is not proof of no effect.

05Applicability

Does it transfer to your setting and patient?

Exclusions, era, and local practice decide whether a result belongs at your bedside.

06Recency

Has it been superseded?

A newer guideline, pivotal trial, label change, or safety communication can reset the answer.

07Corrections

Has the record itself changed?

Errata and retractions must propagate to every dependent claim, figure, case, and assessment—an original PDF is not a safe vault.

The bedside test

Does it actually change the move?

Absolute effect, harm, and the transfer boundary decide practice impact—hierarchy alone never does.

Where the records live

The method is here; the sources stay with the module.

Every module keeps its own verified source register—citations, identifiers, review dates, and corrections—next to the claims it supports. This page is the shared method; the traceable evidence for a topic lives on the topic's page.

See the arrhythmias evidence register