Randomised Controlled Trial (RCT)
Allocate participants by chance to intervention or control — the reference standard for answering a THERAPY question.
An experimental study in which participants are randomly allocated to an intervention or a control group, then followed for outcomes. Randomisation distributes both known and unknown prognostic factors evenly, so any outcome difference can be attributed to the intervention.
Chance decides the arm → groups start prognostically identical → the difference at the end is the treatment effect.
Patients enrolled, randomly allocated to intervention or control, and followed to outcome. Because the only systematic difference between the arms is the intervention, the trial isolates its causal effect.
Check the chain that protects the result: was allocation truly concealed (so clinicians couldn’t steer sicker patients to one arm)? Were patients, carers and assessors blinded? Was the analysis intention-to-treat? Was attrition low? Each weak link lets bias back in.
Randomisation is the only design feature that balances unmeasured confounders, so a well-run RCT gives the least biased estimate of a treatment effect — the top of the hierarchy for therapy questions, below only a systematic review of several such trials.
- Best design for:
therapy / interventionquestions - Randomisation balances
known + unknownconfounders - Safeguards:
concealment + blinding + ITT
CRASH-2 (Lancet 2010) — 20,211 trauma patients with, or at risk of, significant haemorrhage randomised to tranexamic acid vs placebo. 28-day all-cause mortality fell 16.0% → 14.5% (RR 0.91, 95% CI 0.85–0.97). Randomisation across 274 hospitals in 40 countries made the balanced arms — and the result — credible. Benefit is time-critical: a pre-specified analysis showed TXA reduces bleeding deaths only when given early (≤3 h); given later it gave no benefit and possibly harm.
- Efficacy vs effectiveness — a result under ideal trial conditions may not hold in messy routine practice.
- Poor external validity — narrow inclusion criteria mean the trial population may not match your patient.
- Per-protocol vs intention-to-treat — switching to per-protocol breaks randomisation and can exaggerate the effect.
Quick check
What does randomisation achieve that statistical adjustment cannot?
Answer: It balances unknown / unmeasured confounders too — adjustment can only correct for factors you have actually measured.
Ready to build your plan? EMF Premium gives you all 40,000+ questions, 20 mocks and 1,215 OSCE stations from £29/month — or a one-off 3- or 6-month pass.