Describe experimental epidemiology and the randomised controlled trial.

The Randomised Controlled Trial

• Why randomisation IS the central idea — it is the only method that controls for unknown and unmeasured confounders. Matching, restriction and statistical adjustment can control only the confounders one knows about and has measured; randomisation distributes everything else, known and unknown, equally between the groups by chance
• This single property is what makes the RCT superior to every observational design, and stating it is what a good answer does
• The steps of a randomised controlled trial, in order:
• 1. Drawing up the protocol — stating the objectives, hypotheses, population, sample size, criteria, interventions, outcomes and analysis IN advance, which prevents post hoc selection of favourable results; and obtaining ethics approval and trial registration
• 2. Selecting the reference and experimental populations — with explicit inclusion and exclusion criteria
• 3. Randomisation — allocating participants to groups by chance, so that each has an equal and known probability of assignment
• Methods — simple (random numbers); block (ensuring balanced numbers throughout); stratified (randomising within strata such as age or centre); and cluster (randomising groups such as villages or schools rather than individuals)
• Allocation concealment is distinct from blinding and is frequently confused — allocation concealment means the person enrolling a participant cannot know or predict the next assignment, and it prevents selection bias at entry; it is achieved by sealed opaque sequentially numbered envelopes or a central randomisation service
• 4. Manipulation (the intervention) — applied to the experimental group, with a placebo or the standard treatment to the control group
• A placebo control is ethical only where NO effective treatment exists; otherwise the comparator must be the best available standard care
• 5. Follow-UP — identical in both groups, with attrition minimised and recorded
• 6. Assessment of outcome — using pre-specified, objective, clinically meaningful endpoints
• Blinding (masking) prevents information bias — single blind (the participant is unaware); double blind (participant and investigator are unaware); and triple blind (the analyst also unaware)
• Blinding is not always possible — in surgical, behavioural and health service trials — in which case blinded outcome assessment should still be used, which is achievable even where the intervention cannot be masked
• Intention-TO-treat analysis is the essential analytical principle: participants are analysed in the group to which they were randomised, regardless of whether they complied, crossed over or dropped out
• The reason must be given — it preserves the benefit of randomisation and reflects real-world effectiveness. Analysing only those who completed treatment (per-protocol) reintroduces selection bias, because non-compliers differ systematically from compliers — and per-protocol analysis characteristically exaggerates the benefit

Types of Trial, Bias and Limitations

• Types of experimental study:
• Clinical trials — conducted on patients, evaluating treatment
• Field trials — conducted on healthy people in the community, typically evaluating vaccines or preventive agents; the SALK polio vaccine trial of 1954, involving over a million children, is the classic
• Community trials — where the community rather than the individual is the unit of allocation, used for interventions that cannot be given individually, such as water fluoridation, salt iodisation or a mass media campaign
• The phases of a drug trial — phase I (safety and dose, in a small number of healthy volunteers); phase II (efficacy and dose-finding in a few hundred patients); phase III (the large comparative RCT establishing efficacy and safety); and phase IV (post-marketing surveillance, which detects rare and late adverse effects that no trial is large or long enough to find)
• Related designs — crossover trials, in which each participant receives both treatments in sequence, being their own control, suited to chronic stable conditions but requiring a washout period and vulnerable to carry-over effects; factorial designs, testing two interventions at once; and non-inferiority trials, which ask whether a new treatment is not appreciably worse than an existing one
• Ethical requirements are integral, not an afterthought — informed consent; ethics committee approval; equipoise, meaning genuine uncertainty in the expert community about which arm is better, without which randomisation cannot be justified; the right to withdraw without penalty; data and safety monitoring with pre-specified stopping rules; and trial registration and publication of results whether positive or negative
• Registration exists to combat publication bias — the tendency for positive results to be published and negative ones to disappear, which systematically distorts the literature and every meta-analysis drawn from it
• The advantages of the RCT — controls known and unknown confounders; establishes temporality beyond doubt; permits blinding; provides the strongest evidence of causation; and allows direct calculation of efficacy
• The limitations, which must be given for a balanced answer:
• Expensive and time-consuming
• Ethically impossible for harmful exposures — one cannot randomise people to smoke, which is why the evidence on smoking rests entirely on observational study
• Limited external validity (generalisability) — trial participants are selected, often younger, healthier, more compliant and with fewer comorbidities than real patients, so efficacy under trial conditions may exceed effectiveness in practice
• The efficacy-effectiveness distinction is worth stating explicitly — efficacy is the benefit under ideal, controlled conditions; effectiveness is the benefit in routine practice; efficiency relates the benefit to the resources used
• Unsuitable for rare outcomes or long latency, since the required size and duration become impossible
• May be underpowered, and a non-significant result is not evidence of no effect
• The hawthorne effect — participants change their behaviour simply because they are being observed
• The hierarchy OF evidence places systematic review and meta-analysis of RCTs at the top, then individual RCTs, cohort, case-control, cross-sectional, case series and finally expert opinion
• But the hierarchy should not be applied mechanically — a well-conducted cohort study is better evidence than a small, biased RCT, and for questions of prognosis, harm or rare outcomes, observational designs are the appropriate ones
• Randomisation alone controls unknown confounders, which is the single property making the trial superior to observation
• Analyse by intention to treat, or the selection bias randomisation removed is quietly reintroduced
• Per-protocol analysis exaggerates benefit, because non-compliers differ systematically from compliers
• Allocation concealment is not blinding; it prevents selection bias at entry rather than information bias afterwards
• Blind the outcome assessor even when the intervention cannot be, which is achievable in surgical and behavioural trials
• A placebo is ethical only where no treatment exists; otherwise the comparator is best standard care
• Equipoise justifies randomisation; without genuine uncertainty, allocating by chance cannot be defended
• Write the protocol before the study, which prevents post hoc selection of favourable outcomes
• Registration combats publication bias, which otherwise distorts every meta-analysis drawn from the literature
• Phase IV detects rare and late adverse effects that no premarketing trial is large or long enough to find
• Field trials study healthy people and clinical trials patients, which is the distinction to state
• Community trials allocate whole communities, for interventions such as fluoridation that cannot be given individually
• Cluster randomisation suits villages and schools, and requires larger numbers than individual randomisation
• Crossover designs need a washout period and are vulnerable to carry-over effects between treatments
• One cannot randomise people to smoke, which is why the smoking evidence rests entirely on observational work
• Efficacy is ideal and effectiveness is real, with efficiency relating benefit to the resources consumed
• Trial participants are healthier than real patients, which limits external validity however sound the internal design
• A non-significant result is not evidence of no effect, particularly in an underpowered trial
• The Hawthorne effect changes behaviour simply because participants know they are being observed
• Do not apply the evidence hierarchy mechanically; a good cohort study beats a small biased trial
• Observational designs suit harm and prognosis, where trials are unethical or impracticable
• The Salk polio trial enrolled over a million children, and remains the model field trial
• Stratified randomisation balances key variables such as centre and age, which simple randomisation may not in small trials
• Specify stopping rules in advance, so that a trial is not halted on the basis of a favourable interim glance