What these medicines actually do

Start with the hormone, not the drug

Glucagon-like peptide 1 is a hormone you already make. It is released from L-cells in the lining of the lower small intestine and colon when food arrives, and it does several jobs at once: it increases insulin secretion in a glucose-dependent way, it suppresses glucagon, it slows how quickly the stomach empties, and it signals satiety to the brainstem and hypothalamus. Its natural half-life is a couple of minutes, because an enzyme called DPP-4 chops it up almost immediately.

The whole drug class is an answer to that last fact. If you modify the peptide so DPP-4 cannot degrade it and attach a fatty acid chain so it binds to albumin and circulates for days instead of minutes, you get a molecule that does what the hormone does, continuously. Liraglutide lasts about a day. Semaglutide lasts about a week, which is why it is injected weekly.

Almost everything people experience on these medicines follows from that paragraph. Slowed gastric emptying gives you early fullness, reflux, burping and constipation. Central appetite signalling is why the wanting changes. Glucose-dependent insulin release is why these drugs rarely cause low blood sugar on their own, and why the picture is different for someone also taking insulin or a sulfonylurea, where hypoglycaemia becomes a genuine risk to plan for.

Which drug is which

The names get used interchangeably in conversation and they should not be.

Molecule	Acts at	Dosing	Largest trials in our reading group
Liraglutide	GLP-1 receptor	Daily injection	Comparator arm in STEP 8 (JAMA, 2022)
Semaglutide	GLP-1 receptor	Weekly injection; an oral form also exists	STEP 1 (NEJM, 2021), STEP 4 (JAMA, 2021), SUSTAIN 6 (NEJM, 2016), SELECT (NEJM, 2023), FLOW (NEJM, 2024)
Tirzepatide	GIP and GLP-1 receptors	Weekly injection	SURPASS-2 (NEJM, 2021), SURMOUNT-1 (NEJM, 2022), SURMOUNT-4 (JAMA, 2024), SUMMIT (NEJM, 2025)
Retatrutide	GIP, GLP-1 and glucagon receptors	Weekly injection; investigational	Phase 2 (NEJM, 2023)

Two things follow. First, tirzepatide is not simply a stronger semaglutide; it acts on a second incretin receptor, and whether that explains the difference in average results is still argued about. Second, retatrutide is not available treatment. It is a phase 2 molecule with published results (Jastreboff, NEJM, 2023) and a great deal of internet enthusiasm, and Sunil’s reading group is blunt about the distance between those two things.

What the weight trials found

The headline numbers, with the caveats that matter.

STEP 1 (Wilding, NEJM, 2021): semaglutide 2.4 mg weekly plus lifestyle support in adults with overweight or obesity without diabetes. Mean weight change about −14.9 per cent over 68 weeks against about −2.4 per cent on placebo.

SURMOUNT-1 (Jastreboff, NEJM, 2022): tirzepatide at 5, 10 and 15 mg over 72 weeks, with mean reductions rising with dose from roughly 15 per cent to around 21 per cent, against about 3 per cent on placebo.

STEP 8 (Rubino, JAMA, 2022): the head-to-head that people quote. Semaglutide 2.4 mg weekly produced a substantially larger average reduction than liraglutide 3.0 mg daily over 68 weeks — roughly 16 per cent against roughly 6 per cent.

Sunil’s standing objection applies to all three. These are means in selected trial populations who had regular contact with a research team, and the individual spread was wide in every trial. Somebody in the STEP 1 semaglutide arm lost almost nothing. That person exists, is in the paper, and is never in the headline.

What happens when people stop

This is the part of the evidence base that is most often left out of the conversation, and it is the part members most need.

STEP 4 (Rubino, JAMA, 2021) ran everyone up to the maintenance dose over twenty weeks and then randomised half to continue and half to placebo. Those who continued lost roughly another 8 per cent over the following 48 weeks. Those switched to placebo regained roughly 7 per cent.

SURMOUNT-4 (Aronne, JAMA, 2024) did the same for tirzepatide after a 36-week lead-in. Continuing produced further loss; withdrawal produced substantial regain over the following year.

The reading of this that we think is honest: these medicines treat an ongoing condition while they are being taken, in the way a blood pressure medicine does. When they stop, the physiology they were acting on returns. Benedikt, who holds coming-off, insists that this is a fact about the drug and not a verdict on a person, and he will have that argument with anyone. Coming off, what to expect is the guide that takes it further.

The outcome trials, which are the interesting part

Weight is a surrogate. The trials that changed clinical thinking measured events.

SELECT (Lincoff, NEJM, 2023) enrolled people with established cardiovascular disease and overweight or obesity, without diabetes, and followed them for years. Major adverse cardiovascular events occurred in about 6.5 per cent on semaglutide against about 8.0 per cent on placebo — a hazard ratio around 0.80. That is a real reduction in events in a high-risk group, and it is not the same claim as "this drug is good for your heart" in a person at low risk, which SELECT did not test.

FLOW (Perkovic, NEJM, 2024) enrolled people with type 2 diabetes and chronic kidney disease and found roughly a 24 per cent relative reduction in a composite kidney outcome. Our summary of it is here.

SUSTAIN 6 (Marso, NEJM, 2016) was the cardiovascular safety trial in type 2 diabetes, and it is also where the retinopathy signal comes from: complications of diabetic retinopathy were reported in 3.0 per cent on semaglutide against 1.8 per cent on placebo. The suspected explanation is rapid improvement in glucose control in people who already had retinopathy, which is a known phenomenon with any effective glucose treatment, but it remains a reason to mention your eyes to your clinician.

SUMMIT (Packer, NEJM, 2025) looked at tirzepatide in heart failure with preserved ejection fraction and obesity — a different population again, and a sign of where this class is heading.

The known harms, stated plainly

Nobody is served by a guide that lists benefits and hand-waves risks.

• Gastrointestinal effects are the main event. In STEP 1 nausea affected roughly four in ten on semaglutide against roughly a quarter on placebo, with vomiting, diarrhoea and constipation all more common than placebo. Mostly early, mostly mild to moderate, and the reason a small percentage of trial participants stopped.
• Gallbladder disease. Gallstones appear in the adverse-event tables more often on drug than placebo, and rapid weight loss by any means is an established risk factor for them. We go through this properly here.
• Pancreatitis. The trials have not shown a clear excess, but cases occur, the product labels carry a warning, and the symptoms are ones to know. That guide is short and worth reading once.
• Thyroid C-cell tumours in rodents. Seen in rodent studies with this class; human relevance is unknown. The labels contraindicate use in people with medullary thyroid carcinoma or MEN2 in themselves or their family. More here.
• Hypoglycaemia in combination. Low risk alone; a real risk with insulin or a sulfonylurea, and a reason those doses often need reviewing.
• Loss of lean tissue. Any substantial weight loss includes some muscle. Resistance training and adequate protein are the standard mitigations, and that evidence comes from the general weight-loss literature rather than from these trials.

Where the evidence runs out

Sunil writes the "and here is what this does not tell us" section for every paper the reading group covers, on the grounds that it is the section that keeps people out of trouble. Here is the version for the class as a whole.

The trials are short relative to a life. Sixty-eight weeks, seventy-two weeks, a few years in the outcome trials. Members in their fourth year are past the map.

The populations were selected. Trial participants are not a cross-section of the people now taking these drugs, and almost none of the enormous number of people currently using compounded or unlicensed material were represented anywhere in the evidence base.

Most of the internet’s enthusiasm concerns things nobody has trialled: peptides outside this class, stacks, research-use-only material used in humans. There is no trial. There is no safety data. A certificate of analysis, however clean, describes a sample and not a supply chain, and Sunil says that sentence in quality-and-testing roughly once a fortnight. The honest-risks guide is where that conversation lives.

And the mechanism itself is not fully understood. We can describe what these drugs bind to and what happens; why appetite changes so differently between two people on the same dose is not something anyone can currently explain.