Estimand

An estimand is a precise description of the treatment effect a trial is trying to estimate. The framework comes from the ICH E9(R1) addendum, adopted by drug regulators in November 2019. Its central problem is what the guideline calls intercurrent events: things that happen after randomisation, such as stopping the drug, switching to another treatment or needing rescue therapy, which change what the final number means.

The addendum lists five attributes that define an estimand: the treatment, the population, the variable (the endpoint), how intercurrent events are handled, and the summary measure, such as a difference in means.

The two you see in GLP-1 trials

• Treatment policy, or treatment regimen. Counts everyone randomised, whatever happened after. If someone stopped the drug at month four, their week 68 weight still counts. This answers: what happens to people who are prescribed this drug?
• Trial product, or efficacy. A hypothetical strategy. It estimates the result had everyone kept taking the drug as intended. This answers: what does the drug do in people who take it?

The efficacy numbers are always larger, because people who stopped early and lost less are modelled as if they had stayed on. The SURMOUNT-1 efficacy figures above come from Lilly's April 2022 announcement; the NEJM paper's headline is the treatment-regimen estimand.

Other strategies

The addendum describes five strategies for handling intercurrent events: treatment policy, hypothetical, composite variable, while on treatment and principal stratum. Obesity trials mostly report the first two. A composite strategy would count stopping the drug as a failure outcome; a while-on-treatment strategy would look only at the period before stopping. Each answers a slightly different question, and a protocol has to say which one it uses before the data are unblinded. A different choice can move the headline by a couple of points without the drug changing at all.

Which number to use

Neither is wrong. They answer different questions. Treatment policy is closer to real life, where some people stop. The efficacy estimand is closer to what the drug does when taken. The mistake is comparing one trial's efficacy number with another trial's treatment-policy number, which makes the first drug look better by a point or two for reasons that have nothing to do with the drug. Press releases tend to lead with the larger figure; the published paper usually leads with treatment policy. Check which one you are reading. The drug comparison page and the clinical trials page label each figure.

Your own record is a treatment-policy estimand of one: missed doses and breaks included. The dose log captures those alongside weight, so the trend you see reflects how you actually took the drug.

Related terms: placebo-controlled trial, confidence interval, clinical trial phases, hazard ratio. Back to the glossary.