GLP-1 receptor agonists: what the trials actually measured

Few drug classes have been studied as heavily, or discussed as loosely, as the GLP-1 receptor agonists. The trial programmes behind them are enormous — tens of thousands of participants, follow-up measured in years, endpoints adjudicated by blinded committees. That rigour is why precision matters about what they set out to measure: a well-run trial answers its question and no others.

From an incretin effect to a drug class

The observation came first, and it was strange. In work reported in 1964 — separately, in The Lancet and in the Journal of Clinical Endocrinology and Metabolism — investigators found that glucose given by mouth produced a substantially larger insulin response than the same glucose given intravenously. Something in the gut was signalling the pancreas. That gap acquired a name, the incretin effect, roughly two decades before anyone identified the molecule responsible.

Glucagon-like peptide-1 was characterised in the 1980s, and it had an awkward property for a drug candidate: circulating half-life measured in a couple of minutes, thanks to rapid enzymatic degradation. The way around it arrived from an unexpected direction: a peptide isolated from Gila monster venom, described by John Eng and colleagues in 1992, similar enough to activate the same receptor and resistant enough to survive in circulation. A synthetic version of it, exenatide, was approved in 2005 — the first drug in the class.

What followed was engineering. Liraglutide, approved for type 2 diabetes in 2010, used fatty-acid acylation to bind albumin and extend duration. Semaglutide, approved for type 2 diabetes in 2017, extended it further and later reached an oral formulation. Tirzepatide, approved in 2022, activates the GIP receptor as well as the GLP-1 receptor. Approvals for chronic weight management, in specified populations, came separately and later for the products that hold them.

What the outcome programmes measured

The cardiovascular outcome trials exist for a regulatory reason. After earlier controversies over diabetes drugs and cardiovascular risk, the FDA issued 2008 guidance requiring sponsors to establish that new type 2 diabetes therapies did not carry unacceptable cardiovascular risk. The resulting trials were designed, in the first instance, to rule something out.

Their primary endpoint was typically three-point MACE: a composite of cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke. LEADER (liraglutide) and SUSTAIN-6 (semaglutide), both published in the New England Journal of Medicine in 2016, were built as non-inferiority trials against placebo on background standard care, with superiority testable if non-inferiority was met. Several met superiority. That is a meaningful result, and it is a result about that composite, in that enrolled population — largely people with type 2 diabetes at elevated cardiovascular risk — over that follow-up period.

The weight-outcome programmes asked a different question with a different endpoint. The STEP trials, beginning with the 2021 NEJM report, used percentage change in body weight from baseline as a co-primary endpoint. SURMOUNT-1, reported in 2022, did likewise for tirzepatide. Body weight is a measured quantity, not a clinical event; a change in it is not by itself a demonstration of reduced morbidity or mortality. SELECT, published in 2023, was the trial designed to test cardiovascular events directly in people with overweight or obesity and established cardiovascular disease but without diabetes — a genuinely different population, and a different question.

What blinding bought, and what the safety data show

These trials were randomised, double-blind, and placebo-controlled. Randomisation distributes unmeasured confounders across arms; blinding of participants and investigators keeps expectation from steering subjective assessments and treatment decisions; independent adjudication of endpoint events keeps the counting honest. Together, that design is what licenses causal language about the primary endpoint. Very little else in the discourse around this class has that backing.

Blinding in this class has a known limitation, and the trial reports acknowledge it: gastrointestinal adverse effects were common and considerably more frequent in the active arms, which gives participants and clinicians a plausible route to guessing assignment. For hard adjudicated events like death or stroke, that matters little. For subjective secondary measures, it matters more.

The adverse-event tables are worth reading directly rather than through summaries. Gastrointestinal effects — nausea, vomiting, diarrhoea, constipation — dominate. Discontinuation due to adverse events ran higher in active arms than placebo across the major trials. That figure is informative in a way the efficacy numbers aren't: it counts people who stopped, and it is reported per protocol rather than per press release.

What the trials did not study

Trial populations were defined by explicit inclusion and exclusion criteria, and results do not extend past them. Pregnancy, most paediatric populations, and several categories of severe comorbidity were excluded from the major programmes. Outcomes past the follow-up window are, definitionally, not in the data: durability of effect after stopping, and consequences over decades rather than years, are questions the trials were not built to close. Head-to-head comparisons between products were rare — most trials compared against placebo on background care, which makes cross-trial comparison unreliable.

None of this is a criticism of the programmes. Trials answer the questions they were powered to answer, and these ones did. The failure mode is on the reading end: a finding gets promoted from "measured in this population over this period" to "true in general," usually while passing through a headline. If you have a question about whether any of this applies to you, that question is for a clinician who can prescribe, who knows your history, and who can be held responsible for the answer. It is not one this publication can answer, and we would be doing you harm if we tried.