Appraising an RCT

Structured critical appraisal of a randomised controlled trial (the CASP RCT framework) asks three things in order. (1) Is it valid? — true randomisation, allocation concealment, blinding, intention-to-treat (ITT) analysis, complete follow-up, and groups similar at baseline. (2) What are the results? — the effect size and its precision (the confidence interval). (3) Will it help my patient? — external validity, whether all important outcomes and harms were measured, and whether benefits outweigh harms and costs. Only a trial that survives all three should change practice.

The fixed running order of appraisal. Validity is the first gate because a biased trial cannot give a trustworthy result no matter how large or how “significant”. Only after a trial is judged low-bias do you read the size and precision of its effect, and only then do you ask whether that effect transfers to the patient in front of you.

Work top to bottom. At gate 1, interrogate each bias domain — was allocation truly concealed, were patients and assessors blinded, was the analysis ITT, how much follow-up was lost? At gate 2, look past the p-value to the point estimate and its confidence interval (a narrow CI excluding the null is strong). At gate 3, ask whether your patient resembles the trial population and whether the measured outcomes are the ones that matter to them.

“It’s an RCT and p<0.05” is not appraisal. Randomisation only protects you if allocation was concealed and the analysis stayed ITT; an impressive effect is worthless if half the participants dropped out or the wrong outcome was measured. The three-question structure stops you accepting a headline result you have not actually scrutinised — the commonest examiner trap.

• VALID? → RESULTS? → APPLICABLE? (CASP order)
• Validity = randomisation · concealment · blinding · ITT · follow-up · baseline balance
• Results = effect size and its confidence interval, not the p-value alone

CRASH-2 (Lancet 2010) — 20,211 trauma patients with, or at risk of, significant haemorrhage; tranexamic acid vs placebo. Run it through the three gates: valid (concealed allocation, double-blind, intention-to-treat, near-complete follow-up); results — 28-day all-cause mortality 16.0% → 14.5%, RR 0.91 (95% CI 0.85–0.97), p=0.0035, a narrow CI excluding 1; applicable to undifferentiated ED trauma. A model trial to appraise. Applicability has limits: a pre-specified analysis showed TXA helps only when given early (≤3 h); given later it may increase bleeding deaths — so the benefit does not transfer to late presentations.

• Ignoring attrition or accepting a non-ITT (per-protocol) analysis as if it were unbiased.
• Overlooking baseline imbalance between arms despite randomisation (especially small trials).
• Assuming a valid result is automatically applicable — ignoring population, setting and outcome differences.