Critical appraisal and applying evidence
Introduit| English | Sens |
|---|---|
| surrogate outcomes | surrogate outcomes |
| applicability | applicability |
The trial patient who is not your patient
- A trial shows a drug works in 60-year-olds with severe disease. Your patient is 82, with mild disease and five other medicines. Does the result apply?
- Appraisal asks whether a study is valid, whether its effect is important, and whether it fits the patient in front of you.
PICO
- PICO: population, intervention, comparison, outcome. It turns a clinical worry into a question you can search and answer.
- Example: in adults with high blood pressure (P), does drug X (I) compared with placebo (C) reduce strokes (O)?
Match each part of this question to its PICO letter. "In children with asthma, does an inhaled steroid, compared with placebo, reduce hospital visits?"
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
PICO turns a worry into a searchable question.
Validity
- Were the results biased? Check randomisation, allocation concealment, blinding, loss to follow-up and intention to treat.
- A large effect from a flawed study can be less trustworthy than a small effect from a sound one.
Size and precision
- How big is the effect, in absolute terms (ARR, NNT) as well as relative ones? How wide is the interval?
- Prefer patient-relevant outcomes (strokes, deaths) to surrogate outcomes 替代终点 (blood test values), which may not translate.
Which is a surrogate outcome?
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
A lab value may not translate into patient benefit.
GRADE certainty
- GRADE starts randomised evidence at high certainty and observational evidence at low.
- Downgrade for risk of bias, inconsistency, indirectness, imprecision or publication bias; upgrade for a large effect or a dose-response gradient.
Where does GRADE start a body of observational evidence?
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
Randomised evidence starts high, observational evidence low.
A randomised trial always deserves more trust than any cohort study.
Difficulté 1/5 · 100 pièces + un petit bonus lors de la première réussite aujourd'hui
A small, unblinded, badly analysed trial can be weaker than a large, careful cohort.
Applying it to your patient
- Applicability 适用性: would your patient have been enrolled? Check age, severity and other illnesses.
- Relative effects often travel between settings; absolute benefits depend on your patient's own baseline risk, so recompute them.
Put the appraisal steps in order.
Difficulté 3/5 · 200 pièces + un petit bonus lors de la première réussite aujourd'hui
Validity first; a biased result is not worth sizing.
Worked example: one abstract, five questions
- (Fictional) "Randomised, double-blind, n = 400; RR 0.75 (0.60–0.94)." Validity: randomised and blinded. Size: RRR 25%, and the interval excludes 1.
- Applicability: were patients like mine enrolled? Outcome: patient-relevant or surrogate? Certainty: any reason to downgrade?
A trial reports RR 0.75. What is the relative risk reduction in %?
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
RRR = 1 − 0.75 = 0.25, so 25%.
Worked example: transplanting the NNT
- (Fictional) Trial: baseline 20%, ARR 8%, NNT 13. Your clinic's baseline is 5%, with the same RRR of 40%.
- ARR = 5% × 0.40 = 2%, so NNT = 50. The relative effect travelled; the workload did not.
(Fictional) RR 0.75 in a trial. Your patient's baseline risk is 8%. What is the NNT for your patient?
Difficulté 2/5 · 140 pièces + un petit bonus lors de la première réussite aujourd'hui
ARR = 8% × 0.25 = 2%, so NNT = 1 ÷ 0.02 = 50.
Three common slips
- Appraising validity only. A valid but tiny effect may not matter.
- Trusting the design label alone. GRADE starts there and then moves with the flaws.
- Applying averages to an excluded patient. If your patient could not have joined the trial, the result is indirect.
Recap
- PICO → validity → size and precision → applicability.
- GRADE: randomised starts high, observational low; downgrade for bias, inconsistency, indirectness, imprecision, publication bias.
- Relative effects travel; recompute ARR and NNT from your patient's baseline risk.