The randomised controlled trial is one of the most important inventions in medicine. Knowing how trials are built makes it much easier to judge whether a result is believable.
A short history
In 1747 the naval surgeon James Lind compared six treatments for scurvy among sailors and found citrus fruit worked. It's often called the first controlled trial. In 1948 the UK Medical Research Council's trial of streptomycin for tuberculosis, designed with the statistician Austin Bradford Hill, is widely regarded as the first properly randomised controlled trial.
The key ingredients
1. A control group
Without a comparison group, it's impossible to separate the effect of a treatment from natural recovery, regression to the mean or the placebo effect. Controls may get a placebo, the current standard treatment (an active comparator), or no treatment.
2. Randomisation
Participants are assigned to groups by chance, so known and unknown factors such as age, severity and lifestyle are balanced on average. Allocation concealment stops researchers from knowing or influencing the next assignment.
3. Blinding
In a double-blind trial, neither participants nor the people assessing outcomes know who got what. Blinding prevents expectations from shaping reported symptoms or how results are judged.
4. A pre-specified primary endpoint
The main outcome is chosen before the trial starts. Without that, researchers could measure dozens of outcomes and report whichever looks best. Endpoints can be clinical (events, survival, symptoms) or surrogate (lab values or imaging), and surrogates don't always predict real benefit.
5. An adequate sample size
Trials are designed with enough participants to have a high chance, typically 80–90% power, of detecting a meaningful effect if one exists. Small trials are prone to both missing real effects and exaggerating chance ones.
Common designs
| Design | How it works | Good for |
|---|---|---|
| Parallel group | Each participant receives one treatment | Most trials |
| Crossover | Each participant receives every treatment in sequence | Stable, chronic conditions |
| Non-inferiority | Shows a new treatment is not meaningfully worse than an existing one | Treatments with other advantages, e.g. fewer side effects |
| Adaptive | Pre-planned changes based on interim results | Testing several doses or treatments efficiently |
Analysing the results
- Intention-to-treat analysis counts everyone in the group they were randomised to, even if they stopped treatment. That keeps the benefit of randomisation.
- Per-protocol analysis includes only those who followed the plan. It's useful, but more prone to bias.
- Multiple comparisons: the more outcomes and subgroups tested, the more likely a "significant" result appears by chance.
Transparency
Trials should be registered before they start, on registries such as ISRCTN or ClinicalTrials.gov, and reported following the CONSORT guidelines. Registration makes it possible to check that the published outcomes match the planned ones and that negative trials aren't quietly left unpublished.
For how trials fit into medicine development, see From lab to licence. For judging claims in general, see How to evaluate peptide research claims.
Related articles
- How to evaluate peptide research claims
- What "research use only" actually means
- How laboratories validate analytical methods
Sources and further reading
- Medical Research Council. Streptomycin treatment of pulmonary tuberculosis. BMJ 1948;2:769–782. doi:10.1136/bmj.2.4582.769 · PMID: 18890300
- Schulz KF, Altman DG, Moher D. CONSORT 2010 Statement: updated guidelines for reporting parallel group randomised trials. BMJ 2010;340:c332. doi:10.1136/bmj.c332 · PMID: 20332509
- Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics 2019;20:273–286. doi:10.1093/biostatistics/kxx069 · PMID: 29394327
- ISRCTN registry. www.isrctn.com
- ClinicalTrials.gov. clinicaltrials.gov