What a Placebo-Controlled Trial Can and Cannot Show

“Placebo-controlled, randomized, double-blind” is the phrase that makes a study sound settled. It is worth knowing what each of those three words actually buys you, because they buy different things, they fail in different ways, and a trial can carry all three labels and still answer a question nobody asked. The phrase is also conspicuously absent from most of the peptide literature, which is the other half of why it is worth understanding.

This guide explains how a placebo-controlled trial is built: what randomisation does, the step that is routinely confused with it, who can be blinded, what a control group is for, and how the analysis population changes the answer. It ends with the part most relevant to this catalogue — which research compounds actually have a randomised placebo-controlled human trial behind them, and what those trials really were.

What a controlled trial is for

The purpose of a control group is narrower and more precise than “proving the treatment works”. ICH E10, the international guideline on choosing one, puts it as a single job: control groups have one major purpose, to allow discrimination of patient outcomes caused by the test treatment from outcomes caused by other factors, such as the natural progression of the disease, observer or patient expectations, or other treatment.

Everything in a trial’s architecture follows from that sentence. If you measure one group of people before and after, every one of those other factors is still in your result and you cannot separate them. A concurrent control group experiences the same disease course, the same expectations and the same measurement over the same period, so subtracting one from the other leaves the part attributable to the treatment. Of the placebo design specifically, ICH E10 says subjects are randomly assigned to a test treatment or to an identical-appearing treatment and that it controls for all potential influences on the actual or apparent course of the disease, measuring the total pharmacologically mediated effect of treatment.

The design has a documented history worth one careful sentence. The 1948 Medical Research Council investigation of streptomycin treatment of pulmonary tuberculosis is widely treated as a landmark in the development of randomised trials: the surviving account of its method records that the allocations were contained in a set of sealed envelopes, with the details of the series unknown to any of the investigators or to the coordinator. Whether it was the first such trial is disputed among historians, including by the author who documented its allocation method, so this article calls it a methodological landmark and nothing stronger.

Randomisation, and the step confused with it

Randomisation’s job is comparability at baseline. The Cochrane Handbook states that if successfully accomplished, randomization avoids the influence of either known or unknown prognostic factors, so that on average each intervention group has the same prognosis before the start of intervention. That phrase — known or unknown — is the whole advantage over any other design: you cannot adjust for a confounder you have not thought of, but you can randomise against it.

For that to hold, the allocation must be unpredictable, and two separate steps are involved: generating the allocation sequence, and then concealing it. Conflating them is the most common error in reading trial methods, and CONSORT addresses it directly — allocation concealment should not be confused with blinding. The division of labour is clean once stated. Concealment protects the assignment sequence until allocation, preventing foreknowledge of treatment assignment so that enrolment decisions are made in ignorance of the next assignment in the sequence; it therefore seeks to prevent selection bias. Blinding, by contrast, seeks to prevent performance and ascertainment bias — bias arising after assignment. And there is an asymmetry that makes concealment the more fundamental safeguard: the Handbook states that allocation sequence concealment can always be successfully implemented, regardless of the study design or clinical area, while blinding cannot always be implemented.

This is also why some methods that look random are not. The Handbook names knowledge of a deterministic assignment rule, such as by alternation, date of birth or day of admission, as a route to broken concealment, and CONSORT notes that the word random is misused for trials allocating by alternation, hospital numbers or date of birth. Its verdict is blunt: trials based on non-random methods generally yield biased results. If a paper’s methods describe alternating assignment, the trial is not randomised however the abstract describes it.

Blinding: who, not how many

ICH E9 states that blinding or masking is intended to limit the occurrence of conscious and unconscious bias in the conduct and interpretation of a clinical trial, and gives the formal definitions: a double-blind trial is one in which neither the subject nor any of the investigator or sponsor staff are aware of the treatment received; in a single-blind trial the investigator and his staff are aware of the treatment but the subject is not, or vice versa; and in an open-label trial the identity of treatment is known to all. The guideline is candid that the ideal is not always reachable, noting that difficulties in achieving the double-blind ideal can arise and that in some cases only an open-label trial is practically or ethically possible.

The problem is that the labels do not travel well. A 2020 methodology paper reports that agreement about which groups are blinded is poor, with a concrete illustration: in one study, 91 physicians reported 17 unique combinations of groups. Its authors propose eliminating the use of adjectives that modify “blinded” altogether, so a trial would be described as either blinded or unblinded with the specifics spelled out. CONSORT takes the same approach, asking authors to report who was blinded after assignment to interventions, for example participants, care providers, those assessing outcomes, and how. The practical reading rule follows: if a paper says “double-blind” and nothing more, you do not actually know who was kept in the dark. Blinding the outcome assessor is often achievable where blinding participants is not, so the useful question is never how many blinds a trial claims but which parties were blinded.

Choosing the control group

ICH E10 categorises control groups into five types, and the choice is a design decision with consequences rather than a formality.

Two properties of the placebo design pull against each other. ICH E10 notes that, like other superiority trials, a placebo-controlled trial contains internal evidence of assay sensitivity — the trial itself demonstrates it was capable of detecting a difference, which an active-controlled trial showing no difference between arms cannot. That is the methodological argument for placebo arms. Against it sits the ethical constraint: the Declaration of Helsinki states that new interventions must be tested against those of the best proven intervention or interventions, and permits placebo only where no proven intervention exists, or where compelling methodological reasons require it and those receiving placebo face no additional risk of serious or irreversible harm.

Why the placebo arm improves

People in the placebo arm usually get better, at least somewhat, and the near-universal interpretation of that is “the placebo effect”. It is the wrong interpretation, and the correction is published rather than a matter of opinion. A meta-analysis of three-armed trials comparing no treatment, placebo and active intervention states it directly: it is wrong to describe the effect that is observed in a placebo arm of a randomised trial as the effect of placebo, as it includes the spontaneous improvement that would also have occurred without administration of a placebo. The same authors treated regression to the mean as part of that spontaneous improvement, noting the statistical reason it occurs — where a measurement is extreme at one time, for statistical reasons the value will then likely be lower at a later time. Since trials usually enrol people when their measurements are at their worst, regression to the mean is built into the design and appears in every arm.

ICH E10 says the same from the design side, listing what a placebo control absorbs: spontaneous change, which it glosses as the natural history of the disease and regression to the mean, alongside subject or investigator expectations. So the placebo arm is not a measurement of the placebo effect; it is a measurement of everything that is not the test substance, and isolating a placebo effect requires a third, no-treatment arm. The same asymmetry applies to side effects: a nocebo effect is adverse outcomes secondary to negative patient expectations rather than the pharmacologic activity of an intervention, but adverse events in a placebo arm are not all nocebo either.

Intention to treat, and its exception

Who gets counted changes the answer, which is why the analysis population is specified in advance. ICH E9 defines the intention-to-treat principle as asserting that the effect of a treatment policy can best be assessed by evaluating subjects allocated to a treatment group as members of that group, followed up and analysed there. It defines the full analysis set as the set of subjects that is as close as possible to the ideal implied by the intention-to-treat principle, and says the primary analysis should include all randomised subjects. The per-protocol set instead defines a subset of those subjects who are more compliant with the protocol. The full analysis set is the default because it is usually the harder test: in many clinical trials its use provides a conservative strategy, in ICH E9’s words, since counting only those who completed the intervention as intended tends to flatter it.

There is one important exception. ICH E9 states that the equivalence or non-inferiority trial is not conservative in nature, and that using the full analysis set in one may be biased toward demonstrating equivalence — in those trials the usual safeguard points the wrong way. The guideline also specifies how to read their results: equivalence is inferred when the entire confidence interval falls within the equivalence margins, and for non-inferiority trials a one-sided interval should be used. A terminology note worth carrying: CONSORT deliberately stopped using the phrase, recording that it replaced mention of “intention to treat” analysis, a widely misused term, with a more explicit request for information about retaining participants in their original assigned groups. When a paper says “ITT analysis” without saying who was counted, that is the gap CONSORT was trying to close.

What randomisation does not fix

Randomisation settles who ends up in which group. Everything downstream of that remains open, and four failure modes are worth recognising by name.

Chance imbalance, which is not bias. The Cochrane Handbook is careful in both directions: chance imbalances are not a source of systematic bias, and it is important that baseline imbalances consistent with chance are not interpreted as evidence of risk of bias. A baseline difference in a properly randomised trial is not evidence of malpractice, though it is a reason to read the result less confidently.

Being too small to answer the question. ICH E9 notes the probability of type II error is conventionally set at 10% to 20%, and that sample size should be set against the minimal effect which has clinical relevance in the management of patients. It also flags a trap specific to subgroup claims: the statistical test of interaction may lack power to detect an interaction if the sample size was calculated based on the test for main effects. A trial powered for an overall effect is not powered for the subgroup finding in its discussion.

Reporting outcomes other than the pre-specified ones. A prospective cohort study that tracked and corrected misreported trials found that trial reports in academic journals routinely fail to report pre-specified outcomes, and calls outcome misreporting an important and widespread source of bias in clinical trials. CONSORT asks for any changes to trial outcomes after the trial commenced, with reasons; switching is acceptable where openly declared, and the problem is the undeclared kind. This is why a registry entry and a published paper are worth reading side by side.

Confusing significance with importance, and surrogates with outcomes. The American Statistical Association’s statement on p-values says it plainly: a p-value, or statistical significance, does not measure the size of an effect or the importance of a result. A surrogate endpoint, in the FDA’s definition, is a clinical trial endpoint used as a substitute for a direct measure of how a patient feels, functions, or survives — one that does not measure the clinical benefit of primary interest in and of itself, but rather is expected to predict that clinical benefit. ICH E9 adds that a surrogate may not be a true predictor of the clinical outcome of interest, and that relationships between clinical and surrogate variables for one product do not necessarily apply to another. And CONSORT states that its items do not contain any judgments on how and what the authors should have done, so CONSORT 2010 is not intended as an instrument to evaluate the quality of a trial: complete reporting and good conduct are different properties, and the first is much easier to see.

Which research peptides have one

This is where everything above gets applied, and the honest answer differs sharply by compound. The strongest sources are the FDA’s own July 2026 substance reviews, which examined the human literature for several of these compounds directly, plus two 2026 reviews and one 2025 review. Nothing in this table says a compound works, is safe or is effective; it records only whether a study design exists and what it was.

Three reading lessons generalise from that table, which is why it sits in a methods article rather than a compound guide. A conference abstract is not a trial report — for BPC-157 and GHK-Cu alike the strongest-sounding evidence is unpublished meeting material, and an abstract that cannot state its own arm sizes cannot support a conclusion, which is exactly what the FDA said. An analogue is not the compound — CB4211 is not MOTS-c and the norleucine-substituted analogue is not sermorelin, and the same trap catches a well-known 2012 randomised, double-blind, placebo-controlled cognition trial whose title says only “Growth Hormone-Releasing Hormone” but whose study drug the paper identifies as tesamorelin, a stabilized analog of human GHRH. Measuring the body’s own molecule is not testing the compound — every human MOTS-c dataset describes levels of the peptide people already make, and a correlation between an endogenous molecule and an outcome says nothing about what administering it would do.

What’s established and what isn’t

Reading the evidence before you order

Everything above turns into a short set of habits when you are assessing a compound rather than a trial. None of this is about whether to buy anything; it is about not over-reading what the literature says.

• Check the molecule, not the family. An analogue’s trial is not the compound’s trial. Our guide to reading a peptide study covers what cell, animal and human work can each support.
• Check the publication type. A conference abstract, a retrospective chart review and a randomised trial are three different things. Clinical trial phases explained sets out what each phase is designed to test.
• Check whether anyone pooled the results. Systematic review vs meta-analysis covers how evidence gets synthesised, and how a paper can be titled “systematic review” without being one.
• Check what regulators have said. The FDA’s 2026 peptide reviews go through the human evidence for several of these compounds directly, and BPC-157 safety covers that compound’s record in more depth.

BPC-157Its one randomised placebo-controlled study exists as a meeting abstract the FDA calls inadequateView product GHK-CuReviews disagree on whether randomised human trials of it exist; the identifiable studies are small and topicalView product KPVThe FDA states it identified no clinical studies on the use of KPV in humansView product MOTS-cHuman data describe the body’s own endogenous peptide, not administration of itView product

Frequently asked questions