Randomised Controlled Trials Explained

The test that changed how we know what we know

Roughly 75 years ago, a tuberculosis researcher named Austin Bradford Hill ran what is widely considered the first modern randomised controlled trial. The question was whether streptomycin worked against TB. The method was simple and, at the time, genuinely radical: flip a coin (well, almost), give half the patients the drug and half standard care, and see what happens. It worked. The drug helped. More importantly, the method stuck.

That framework , the randomised controlled trial, or RCT , became the cornerstone of how we evaluate whether a treatment does anything at all. And yet, I'd argue it's also one of the most consistently misread pieces of evidence in public discourse. Headlines declare "scientists discover X cures Y" off the back of a 40-person pilot study. Supplement marketing lifts a single statistically significant finding from a paper with six limitations the size of a wall. We deserve better tools for reading this stuff.

So let's go through what an RCT actually is, what it can and can't tell you, and why understanding its limits matters as much as understanding its strengths , especially in areas like plant-based medicine, cannabinoids, and peptides, where the research is genuinely early-stage and the marketing frequently runs well ahead of the evidence.

What makes a trial "randomised" and "controlled"

Start with the two words in the name.

Randomised means participants are assigned to groups by chance, not by a researcher's preference, a doctor's hunch, or a patient's choice. That randomisation is doing a lot of heavy lifting. It distributes unknown confounding factors; age, baseline health, genetics, lifestyle, roughly evenly across groups, so that differences at the end are more likely attributable to the intervention than to something pre-existing.

Controlled means there's a comparison group. Usually a placebo group (receiving an inactive substitute), sometimes a group receiving the current standard of care, sometimes both. Without a control, you're essentially watching people get better over time and attributing it to whatever you gave them, which is a very easy mistake to make, because people often get better over time regardless.

Together, randomisation and control are designed to isolate the effect of one variable. That's the whole point.

Blinding: the detail that changes everything

A well-designed RCT also incorporates blinding. In a single-blind trial, participants don't know which group they're in. In a double-blind trial, neither participants nor researchers know until the data is unlocked at the end. This matters because both patients and clinicians behave differently when they think they're receiving (or administering) the "real" thing; and those behavioural differences can contaminate the results.

The placebo effect is real, measurable, and surprisingly robust. In trials involving subjective outcomes, things like perceived discomfort, mood, or energy, the placebo response can account for a substantial portion of the improvement in the active group. Double-blinding tries to neutralise that. But it's genuinely difficult to blind some interventions; you can't easily make a sham surgery look, smell, and feel identical to a real one, and some substances have distinctive effects that break blinding within hours.

This is relevant when you're reading cannabinoid research, for instance. Cannabidiol studies have struggled with blinding because many participants can identify the active compound from its subjective effects; meaning you're not always comparing drug versus placebo; you're also comparing "I know I got the drug" versus "I'm not sure." That's a subtler but real confound.

Sample size and statistical power

Here's where a lot of popular science reporting falls apart. A finding is "statistically significant" at p<0.05 when, under the assumption that the null hypothesis is true, results this extreme would occur by chance fewer than 5% of the time. That's it. It says nothing about whether the effect is large enough to matter clinically, and it says nothing about whether a small study was appropriately powered to detect a real effect in the first place.

An underpowered study, one too small to reliably detect the effect it's looking for, can produce both false negatives (missing a real effect) and, perversely, inflated effect sizes when it does find significance, because small studies are subject to more sampling noise. This is sometimes called the "winner's curse" in the methodology literature.

I'll admit I got this wrong for years. I used to scan for the p-value and move on. Now I look for the sample size, the power calculation, and the confidence interval. A wide confidence interval around a positive result is a flag worth noticing. It means the estimate is imprecise, and a different run of the same study might have landed anywhere in that range; including no effect at all.

Internal validity vs. external validity

A well-run RCT can have excellent internal validity, meaning the results within the trial are reliable, while having poor external validity, meaning they may not generalise to the real world or to different populations.

Clinical trials frequently exclude people who are pregnant, elderly, or taking multiple medications. They're often run in highly specific populations. A trial on a standardised plant extract (see our note on standardised extracts) run in a tightly controlled hospital setting may not map onto how a whole-plant preparation behaves differently when taken at home, with food, by a person with a different metabolism. Concepts like bioavailability and first-pass metabolism mean that even the same nominal dose can produce very different plasma concentrations across individuals; a variable most trials try to control for, but that the real world doesn't replicate neatly.

The short version is: an RCT tells you what happened under those specific conditions, with those specific people, using that specific preparation. Extrapolating beyond that takes more steps than a headline allows.

Where RCTs struggle in plant-based medicine research

RCTs were originally built around single-molecule pharmaceutical drugs with known doses and predictable pharmacology. Phytomedicine is structurally messier. A whole-plant extract contains dozens of compounds interacting in ways researchers are still mapping, the so-called entourage effect, for example, refers to the proposed interactions between cannabinoids, terpenes, and other plant constituents.

How do you blind for the smell of a terpene-rich extract? How do you standardise a botanical across multiple harvest batches? How do you design a control for a compound like THC whose psychoactive properties immediately inform the participant which group they're in? These aren't reasons to dismiss plant-medicine research. They're reasons to read it carefully and look at what the researchers actually controlled for, not just the headline finding.

Similarly, in the peptide space, many compounds being discussed publicly are either not approved or are the subject of small, early-phase trials with narrow populations. The regulatory situation matters here too: in Australia, the TGA's position is that supplying many peptides marketed for performance or body composition without appropriate registration is illegal, and the evidence base for many of them remains thin. A single positive Phase II trial is not a green light. It's an interesting early signal that warrants a Phase III. Those are different things. (Our glossary entries on the Special Access Scheme and Authorised Prescriber framework explain how access to unapproved substances works in Australia under TGA oversight.)

What systematic reviews and meta-analyses add, and where they fall short too

One RCT is a data point. A systematic review of multiple RCTs; pooling, critically appraising, and synthesising the results, sits higher on most evidence hierarchies. A meta-analysis goes further, statistically combining results across studies to produce a pooled effect estimate.

But a meta-analysis of poor-quality studies is still a meta-analysis of poor-quality studies. "Garbage in, garbage out" applies. And heterogeneity across studies, different populations, different doses, different outcome measures; can make pooling genuinely misleading, even when the math works out. Always check whether the included studies were similar enough to make combining them sensible. The I² statistic in a meta-analysis tells you how much of the variation across studies is due to true heterogeneity rather than chance; a high I² warrants caution.

Between sea swims this winter I've been working back through the Cochrane reviews in cannabinoid research, and the consistent refrain is "low to moderate quality evidence" with calls for larger, more rigorous trials. That's not a dismissal. It's an honest accounting of where the field is.

Reading an RCT: a quick checklist

• Was the randomisation method described clearly? (Coin flip vs. computer-generated sequence vs. quasi-random all have different properties.)
• Was allocation concealed? (Researchers shouldn't know which group the next enrollee will be in when they decide to enrol them.)
• Was the trial double-blinded, and was blinding verified?
• Was a power calculation reported before the trial started, and was the sample large enough to meet it?
• Were all enrolled participants accounted for at the end? (Dropouts can bias results if they're not handled with intention-to-treat analysis.)
• Are the outcome measures clinically meaningful, or are they surrogate endpoints?
• Who funded the study, and are there declared conflicts of interest?

None of these questions disqualify a study on their own. But they tell you how much weight to put on the finding before the media cycle inflates it into a certainty.

Sources

• Randomised Controlled Trials: An Introduction, National Library of Medicine, NCBI
• What is Cochrane Evidence?, Cochrane
• Australian Regulatory Guidelines for Prescription Medicines; TGA
• Understanding Statistical Power and Significance Testing, NCBI

, Hannah Bui, Evidence & Research-Literacy Writer