01Principles and purposeThe professional or clinical skill and the decisions it supports.
Evidence-based medicine connects research evidence with clinical expertise and the patient's circumstances and values. It is a decision process, not a hierarchy recitation. Begin with a focused uncertainty: who is the patient or population, what intervention, exposure or test is considered, compared with what, and which benefits, harms and burdens matter over what time? A question about diagnosis needs accuracy and clinical utility; prognosis needs an inception cohort and adequate follow-up; treatment needs comparative effects; experience or acceptability may require qualitative evidence.
Efficient searching usually starts with current guidance or a high-quality systematic review, then moves to newer primary studies or gaps. Currency matters because an apparently authoritative review may predate a new intervention or safety signal. Relevance matters equally: a review can be rigorous yet answer a different dose, severity group or outcome. Search methods, eligibility criteria and excluded studies should be transparent enough to judge whether selection was systematic rather than convenient.
Critical appraisal asks how results could be wrong. In randomised trials consider sequence generation, allocation concealment, deviations from intended intervention, missing outcome data, measurement and selective reporting. In observational studies add confounding and selection. Blinding may protect some outcomes more than others. Bias is outcome-specific and direction may be uncertain. Precision, inconsistency, indirectness and publication bias affect certainty across a body of evidence. A large effect does not automatically erase serious design flaws.
Moving from evidence to action requires explicit translation. Present absolute risks with a common time horizon, clinically meaningful outcomes and plausible uncertainty. Ask whether participants, care pathway and comparator resemble the current context. Incorporate burdens, preferences, inequalities and resource effects. A guideline recommendation represents a committee's evidence-to-decision judgement at publication; it still requires patient-level application and surveillance for updates. When evidence is weak, a transparent conditional plan with review may be safer than either therapeutic inertia or false certainty.
Key points
- Translate uncertainty into a structured question with population, intervention or exposure, comparator, outcomes, setting and relevant time horizon.
- Search for the highest-level current evidence that can answer that question, then check that its included studies and dates match the decision.
- Appraise validity by bias domain rather than awarding a study a single quality label based only on design name.
- Separate magnitude and precision of each important outcome from certainty in the evidence and from the eventual recommendation.
- Relative effects require applicable baseline risks to estimate absolute benefit and harm for the patient or population in front of you.
- Recommendations also depend on values, acceptability, feasibility, equity, resources and alternatives; strong language is not produced by a p value alone.
- Apply through shared decision making, document the rationale, monitor outcomes and update when new evidence or patient priorities change.
02Situations and prioritiesThe context, relevant information and actions that matter most.
A structured question prevents a search from drifting toward convenient evidence that addresses a different population, comparator, outcome or time horizon.
Bias domains identify systematic departures from the intended effect estimate; study-design labels alone do not determine validity.
Confidence can differ for mortality, symptoms and adverse events within one review because evidence quantity, bias and indirectness differ.
Biology, baseline risk, service setting, adherence, comparator and co-interventions determine whether study effects transport to current practice.
Benefits, harms, certainty, values, equity, acceptability, feasibility and resources jointly shape recommendation strength and wording.
Follow-up outcomes, safety alerts, guideline updates and changed patient priorities can reopen a previously reasonable evidence-based plan.
03Assessment and interpretationHow to gather information, assess the situation and recognise uncertainty.
Consider the information, its meaning and its limitations before deciding what follows.
- 01
Structured question formulation - Why
- Define what evidence would be relevant before searching.
- Interpretation and limitations
- Specify population, intervention or exposure, comparator, outcomes, time and setting; omitting harms or patient-important outcomes biases the question itself.
- 02
Systematic search and selection review - Why
- Assess whether the evidence base was identified comprehensively and reproducibly.
- Interpretation and limitations
- Check databases, dates, language or publication restrictions, duplicate screening and exclusions; database count alone does not prove completeness.
- 03
Domain-based critical appraisal - Why
- Identify mechanisms by which the estimate could systematically differ from the target effect.
- Interpretation and limitations
- Judge each important outcome and likely direction where possible; avoid converting a checklist score into a false numerical measure of validity.
- 04
Effect and certainty table - Why
- Present relative and absolute effects, precision and confidence for each critical outcome.
- Interpretation and limitations
- Use applicable baseline risks and consistent horizons; a precise relative effect can still produce uncertain absolute benefit when baseline risk varies.
- 05
Applicability assessment - Why
- Compare the study's target population and care pathway with the current patient or service.
- Interpretation and limitations
- Identify effect modifiers, excluded groups, adherence and comparator differences; demographic similarity alone is insufficient.
- 06
Evidence-to-decision record - Why
- Make the route from evidence to recommendation or shared plan explicit.
- Interpretation and limitations
- Document benefits, harms, certainty, values, resources, feasibility and review conditions; disagreement may reflect different value judgements rather than factual error.
04Worked approachesCases with ordered reasoning, an action and a check of the outcome.
01Worked caseTranslate a relative effect for shared decision makingInputs: a review estimates RR 0.75 for a harmful event over five years; applicable baseline risk is 20%; evidence certainty is moderate and treatment adds a 3% absolute risk of a troublesome adverse effect.+
- 1Calculate treated event risk under a constant relative effect: 0.20 multiplied by 0.75 equals 0.15, or 15% over five years.
- 2Calculate absolute benefit: 20% minus 15% equals 5 percentage points, equivalent to about 50 fewer harmful events per 1,000 treated over five years.
- 3Place benefit beside harm: about 30 additional troublesome adverse effects per 1,000, while noting that the outcomes differ in severity and cannot be netted arithmetically without patient values.
- 4Explain moderate certainty as a real possibility that the effect estimate could change, then discuss burden, alternatives and how much the patient values avoiding each outcome.
- 5Agree an individual plan and review point rather than presenting RR 0.75 alone as the decision.
- 6Verify arithmetic with NNT for benefit: one divided by 0.05 equals 20 over five years; 1,000 divided by 20 gives 50 events prevented, matching the absolute calculation.
02Rapid appraisalAssess a new treatment trialA headline reports a large statistically significant benefit from a newly published randomised trial.+
- 1Retrieve the full report and protocol or registration, confirm population, comparator, prespecified outcomes and follow-up.
- 2Assess allocation, deviations, missing data, outcome measurement and selective reporting for the outcome driving the headline.
- 3Extract absolute event counts, effect estimate, confidence interval, harms and subgroup interaction tests.
- 4Compare with existing evidence and current guidance before deciding whether practice or a patient discussion should change.
03Guideline applicationIndividualise a recommendation transparentlyA current guideline recommendation broadly applies, but the patient has a different baseline risk and strongly values avoiding treatment burden.+
- 1Confirm recommendation wording, population, evidence date and any stated exceptions or conditions.
- 2Estimate absolute benefit and harm using the patient's baseline risk where defensible.
- 3Discuss certainty, alternatives, burden and the patient's goals in understandable frequencies.
- 4Document the agreed decision, reasoning, safety-net and trigger for reassessment.
05Feedback, follow-up and evidenceReview outcomes, seek feedback and identify what to improve.
- Record search date and source version so that a later reviewer can judge whether the evidence is still current.
- Monitor sources for guideline updates, corrections, retractions and safety communications when the decision has lasting consequences.
- Measure outcomes and adverse effects that mattered to the decision, using an agreed interval and threshold for changing course.
- Revisit the plan when adherence, baseline risk, diagnosis or patient preference changes because the original absolute effects may no longer apply.
- Audit whether local implementation reproduces the intended population, comparator, dose or process before attributing different outcomes to evidence failure.
06Special situationsVariants, exceptions and circumstances that change the usual approach.
Evidence hierarchies are question-specific
Randomised trials suit many intervention effects, while harms, prognosis, diagnosis, service delivery and lived experience may require different designs or combined evidence.
Certainty is not recommendation strength
High-certainty small benefit may support a conditional choice, while low-certainty evidence may still support urgent action when consequences and alternatives justify it.
Absolute effects travel poorly
A stable relative effect produces different absolute benefits across baseline risks; both effect modification and risk calibration must be considered.
Checklists support, not replace, judgement
Reporting and appraisal tools prompt important domains, but numerical totals can hide a single fatal source of bias.
Absence of evidence has several causes
No eligible study, imprecise studies and evidence of no meaningful effect are different findings and lead to different decisions.
07Common pitfallsFrequent interpretation and management errors.
- 01
Searching for evidence before defining the patient-important outcome and accepting the first paper that supports a preferred answer.
- 02
Calling all randomised evidence high certainty or all observational evidence unusable without outcome-level appraisal.
- 03
Using statistical significance as a substitute for effect magnitude, precision, bias assessment and clinical relevance.
- 04
Applying a relative effect without showing baseline risk, absolute consequences and the same time horizon.
- 05
Treating a current guideline as self-executing without checking eligibility, exceptions, feasibility or patient values.
- 06
Documenting shared decision making as a phrase while omitting the options, material outcomes and preference that shaped the choice.