Key facts
- Plain meaning: the estimand is what the trial is trying to measure, fixed before the data are seen.
- Five parts: population, treatment condition, outcome (endpoint), handling of intercurrent events, and the population-level summary.
- Treatment-policy: counts every randomised participant regardless of adherence — closest to real-world use.
- Trial-product: estimates the effect assuming participants stayed on the assigned drug — usually a larger number.
- Why it matters: the same trial can report two valid percentages; comparing across mismatched estimands is misleading.
- Golden rule: compare treatment-policy with treatment-policy, and trial-product with trial-product.
What an estimand actually is
An estimand is the precise definition of what a clinical trial is trying to measure — the specific treatment effect of interest, written down before anyone looks at the results.
It sounds like statistical bureaucracy, but it answers a surprisingly slippery question: when a trial says a drug "works," what exactly does that mean? The effect of taking the drug perfectly for the whole study? The effect of being prescribed it in the messy real world where some people stop? Those are different questions with different answers, and the estimand is the discipline of choosing one on purpose rather than drifting into whichever number looks best later.
The idea was formalised so that a trial's headline effect could not quietly shift after the data arrived. Clinical trials are the stage of drug development where a candidate is tested for efficacy and safety in people, and specifying the estimand up front keeps everyone honest about the question being asked (see the general primers on drug development and drug trials).
The five parts of an estimand
A fully specified estimand has five moving parts. You do not need to memorise them, but recognising them helps you see why two summaries of the same trial can legitimately differ:
- Population. Who the effect applies to — for example, adults with obesity and no diabetes.
- Treatment condition. What is being compared with what — the drug at a given dose versus placebo or another drug.
- Outcome (endpoint). The variable being measured, such as percentage change in body weight at a fixed week. This is closely tied to how a trial defines a responder rate — the share of people who hit a threshold like 5% or 10% loss.
- Intercurrent events. The crucial one: how the analysis treats things that happen after randomisation and complicate the picture — stopping the drug, starting a rescue medication, or dropping out. This is where treatment-policy and trial-product estimands part ways.
- Population-level summary. How the individual results are boiled down into one number — a mean difference, a hazard ratio, an odds ratio.
Why one trial reports two numbers
Imagine a weight-loss trial. Everyone is randomised to the drug or placebo, but real life intervenes: some participants stop the drug because of side effects, some start a different medication, some drift away from the study. What do you do with their data when you calculate the average weight loss?
There are two defensible answers, and they correspond to the two estimands in this page's title. You can count everyone at their final measured weight regardless of what they did after randomisation — that is the treatment-policy view. Or you can estimate what the average weight loss would have been if everyone had stayed on the assigned drug as intended — that is the trial-product view. Same participants, same trial, two questions, two numbers. The trial-product figure is typically the larger one, because it strips out the dilution from people who stopped taking the drug.
This is not a loophole or a trick. Both numbers are legitimate; they simply answer different questions. The failure mode is reporting one without saying which — or, worse, comparing one drug's trial-product figure against another drug's treatment-policy figure.
The treatment-policy estimand
The treatment-policy estimand measures the effect of a treatment strategy: being assigned to the drug, with all the imperfect adherence that comes with real-world use. Intercurrent events like stopping the drug are essentially ignored for the purpose of the analysis — the participant's outcome still counts, wherever they ended up.
Its strength is realism. It answers, "If we adopt a policy of prescribing this drug to people like the ones enrolled, what happens on average?" That is often the question a payer, a health system, or a cautious reader cares about most, because it bakes in the fact that not everyone tolerates or continues a medication. Its cost is that the headline number can look more modest, precisely because it includes the people for whom the drug did not stick.
The trial-product estimand
The trial-product estimand — sometimes described as an on-treatment or hypothetical estimand — estimates the effect of the drug itself, under the hypothetical condition that participants stayed on it as intended. It asks, "How well does this molecule work when actually taken?" rather than "What happens when we hand it out in the real world?"
Its strength is that it isolates the pharmacology from adherence and behaviour, which is useful for understanding what a drug is biologically capable of. Its cost is that it flatters real-world expectations: a person reading a trial-product percentage may not achieve it if they, like many trial participants, cannot stay on the drug for the full duration. This is one reason the same molecule can be described with a bigger number in a marketing headline and a smaller one in a formulary review — and why understanding how trials are designed across phases helps you judge which figure to trust.
Why this matters when comparing drugs
The estimand is where a lot of casual drug comparisons quietly go wrong. Two weight-loss drugs might each report an impressive percentage, but if one figure is a trial-product estimand and the other is a treatment-policy estimand, the comparison is rigged before it starts — not by dishonesty, but by mismatched questions.
The rule is simple: compare like with like. Line up a treatment-policy figure against another treatment-policy figure, and a trial-product figure against another trial-product figure. Head-to-head trials that randomise both drugs in the same study and pre-specify one estimand for both — such as the design used in REDEFINE 1 — sidestep the problem entirely, which is exactly why they carry more weight than stitching separate trials together. For the wider map of which weight-loss peptides have this caliber of evidence, and how to read those numbers, see our pillar guide on peptides for weight loss.
When you encounter a percentage in a headline, the single most useful question is: which estimand is this? If the source cannot tell you, the number is not yet ready to be compared with anything. Browse the rest of the PepMate research library for more of these trial-reading tools.
Frequently asked questions
What is an estimand in simple terms?
An estimand is the precise definition of what a clinical trial is measuring — the exact treatment effect it is trying to estimate. It nails down the population, the treatments being compared, the outcome, how the trial deals with people who stop the drug or start a rescue therapy, and how the numbers are summarised. Pinning down the estimand before a trial starts stops teams from quietly changing the question after they see the data.
What is the difference between a treatment-policy and a trial-product estimand?
The treatment-policy estimand measures the effect of being assigned to a treatment and counts every randomised participant, even those who stopped the drug or started something else — it reflects real-world use where people do not always adhere. The trial-product estimand (sometimes called the on-treatment or hypothetical estimand) estimates the effect if participants had stayed on the assigned drug as intended, which usually looks larger because it sets aside the dilution from people who stopped.
Why does the same trial report two different weight-loss percentages?
Because the two figures answer two different questions. A treatment-policy number includes everyone as randomised, so it reflects the average result across people who did and did not keep taking the drug. A trial-product number estimates what would have happened if everyone had stayed on treatment, which is typically a bigger figure. Neither is wrong; they are two lenses on the same trial, and reputable reports label which estimand each percentage belongs to.
Which estimand should I look at when comparing two drugs?
Compare like with like: a treatment-policy figure for one drug should be set against a treatment-policy figure for the other, and the same for trial-product figures. Comparing a trial-product number for one drug against a treatment-policy number for another can make one look better purely because of the estimand, not the biology. If a source does not say which estimand a percentage came from, treat the comparison with caution.
Is this page medical advice?
No. This is an educational glossary entry that explains a clinical-trial concept so you can read published research more critically. It does not recommend, prescribe, or provide dosing for any peptide or medication, and it is not a substitute for guidance from a licensed clinician.
Sources
This glossary entry draws on general references describing how drugs are developed and tested:
- Drug Development 101: A Primer — International Journal of Toxicology, 2020; PubMed 32762387.
- Drug Trials — StatPearls; PubMed 31536202.