PUBH5010 · Epidemiology Methods And Uses
Randomised Controlled Trials
The RCT is the design that earns a causal claim, because the investigator allocates the exposure — ideally at random — so that, on average, the groups differ only in the treatment. The exam tests four distinct safeguards that students routinely blur: randomisation (which balances known and unknown confounders across arms, the RCT’s core advantage over observational designs); allocation concealment (hiding the upcoming assignment from whoever enrols patients, so the randomisation cannot be subverted at entry); blinding / masking (keeping participants, carers and assessors unaware of arm, to prevent performance and ascertainment bias after entry); and a placebo (to isolate the specific treatment effect from expectation). Analysis carries its own marked distinction: intention-to-treat analyses everyone in the arm they were randomised to, preserving randomisation and giving a conservative, real-world estimate, whereas per-protocol analyses only compliant completers and can reintroduce confounding. Finally, efficacy (does it work under ideal trial conditions?) is not effectiveness (does it work in routine practice?). Name the safeguard, say what bias it blocks, and keep intention-to-treat as the default analysis — that is the whole chapter.
What this chapter covers
- 01Why randomisation earns causation: balancing known and unknown confounders
- 02Allocation concealment vs randomisation — protecting entry
- 03Blinding/masking: participants, carers, assessors — protecting after entry
- 04The role of a placebo and the placebo effect
- 05Intention-to-treat: preserving randomisation, the conservative estimate
- 06Per-protocol analysis and why it can reintroduce bias
- 07Efficacy vs effectiveness
Worked example: intention-to-treat vs per-protocol
- +1(a) Definitions. Intention-to-treat (ITT) analyses everyone in the arm they were randomised to, regardless of adherence. Per-protocol (PP) analyses only those who completed the assigned treatment as intended.
- +1(b) Preserves randomisation. ITT — because no one is excluded after randomisation, the original random balance of confounders is kept.
- +2(c) Overstates benefit. Per-protocol. Dropping the 20 non-compliers (who differ systematically — e.g. those with side effects) reintroduces selection/confounding and can flatter the drug. ITT gives the conservative, real-world estimate, so it is the primary analysis.
Key terms
- Randomisation
- Allocating participants to trial arms by chance so that, on average, both known and unknown confounders are balanced across groups. This balancing of unmeasured factors is the RCT's decisive advantage over observational designs and the basis of its causal strength.
- Allocation concealment
- Hiding the next treatment assignment from the person enrolling participants, so they cannot consciously or unconsciously steer who goes into which arm. It protects the integrity of randomisation at the point of entry and is distinct from blinding, which acts after entry.
- Blinding (masking)
- Keeping participants, treating clinicians and/or outcome assessors unaware of arm assignment, to prevent performance bias (different care or behaviour) and ascertainment bias (biased outcome judgement). 'Double-blind' means both participants and key staff are masked.
- Intention-to-treat (ITT)
- Analysing every participant in the group they were randomised to, regardless of adherence or crossover. It preserves the random balance, guards against bias from non-random dropout, and gives a conservative estimate that reflects real-world use; it is the standard primary analysis.
- Efficacy vs effectiveness
- Efficacy is whether an intervention works under ideal, controlled trial conditions; effectiveness is whether it works in routine practice with ordinary patients and adherence. A treatment can be efficacious yet less effective once real-world compliance and context apply.
Randomised Controlled Trials FAQ
Why does randomisation let an RCT claim causation?
Because allocating treatment by chance balances confounders — crucially including ones you never measured or thought of — across the arms, so any outcome difference can be attributed to the treatment rather than to pre-existing group differences. Observational designs can only adjust for confounders they measured; randomisation handles the unknown ones too, which is why a well-run RCT gives the strongest causal evidence.
What's the difference between allocation concealment and blinding?
They protect against bias at different moments. Allocation concealment hides the upcoming assignment from whoever enrols participants, so the randomisation sequence cannot be subverted at the point of entry. Blinding (masking) keeps participants, carers and assessors unaware of the assignment after entry, preventing biased behaviour and biased outcome assessment. A trial can be randomised yet have poor concealment, or concealed yet unblinded.
Why is intention-to-treat preferred over per-protocol?
Intention-to-treat keeps everyone in their randomised arm, so the random balance of confounders is preserved and the estimate reflects how the treatment performs in the real world, including dropouts and non-adherence — a conservative, honest figure. Per-protocol restricts to compliant completers, who differ systematically from non-compliers, which reintroduces selection and can exaggerate the benefit. Per-protocol is reported only as a secondary, sensitivity analysis.
What's the difference between efficacy and effectiveness?
Efficacy asks whether a treatment works under ideal trial conditions — selected patients, strict protocols, high adherence. Effectiveness asks whether it works in everyday practice, with unselected patients and ordinary compliance. An efficacious drug may prove less effective in routine use, which is why pragmatic trials and real-world studies complement explanatory RCTs.
Exam move
Keep the four safeguards distinct and pair each with the bias it blocks: randomisation (balances known and unknown confounders), allocation concealment (protects entry), blinding (protects after entry, prevents performance/ascertainment bias), placebo (isolates the specific effect). On analysis, default to intention-to-treat as primary — say it “preserves randomisation” and gives the conservative estimate — and treat per-protocol as secondary. Finish by separating efficacy (ideal conditions) from effectiveness (routine practice). Naming the safeguard and the bias it prevents is the marked skill.
