Do GLP-1 Drugs Cause Muscle Loss? What the Trials Actually Show

I believed the muscle story because it flattered me.

You know the version. These drugs work, sure — but they eat your muscle. You come out the other side lighter and weaker, a smaller copy of yourself. Meanwhile here I am at 54 doing it the long way: keto, fifty minutes under a bar three to five days a week, one meal on Fridays, low-intensity work on most of the days between. If the shortcut costs you the thing the long way builds, then the long way isn't just slower. It's correct.

That is a very comfortable story to hold if you are the person already doing the work — which is exactly the kind of story I have learned to distrust in myself.

So I went looking for the number, expecting it to take ten minutes and leave my position intact and properly sourced.

It took a weekend, and I came back having to correct myself on the central point. The number everyone repeats — the one that makes the whole argument land — is not in any of the trials. It is a calculation somebody ran across a table of group averages, and depending on how you run it you get 34%, or 39%, or 45%.

What is in the trials is more useful than the headline, and in one respect more worrying.

What the Trials Actually Measured

Start with the two body composition substudies sitting underneath the entire public argument, because almost nobody quoting them has read what they contain.

The first is an exploratory analysis of STEP 1, the phase 3 trial of semaglutide 2.4 mg. A subgroup of 140 participants got DXA scans at baseline and again at week 68. Over those 68 weeks, body weight fell 15.0% on semaglutide against 3.6% on placebo. Total fat mass fell 19.3%. Visceral fat — the fat that actually tracks with metabolic risk — fell 27.4%. Total lean body mass fell 9.7%.

Read those four numbers next to each other slowly, because their relationship is the finding. Fat fell faster than lean. Lean mass as a share of total body mass went up, from 52.7% to 53.9%. Those people ended lighter and, proportionally, leaner than they started.

The second is the SURMOUNT-1 body composition substudy of tirzepatide — 160 participants out of 2,539, DXA at baseline and week 72. Mean weight loss 21.3%. Of the weight lost, roughly 75% was fat mass and roughly 25% lean mass. Lean mass itself fell 10.9% on tirzepatide against 2.6% on placebo.

And here is the sentence that reframed the whole thing for me: that 75/25 split was approximately the same in the placebo group. It held across subgroups too — age, sex, and how much weight a person lost.

The placebo arm was losing weight the ordinary way, through a trial's lifestyle support, and their loss had the same composition. The drug moved the amount. It did not obviously move the ratio.

The drug moved how much weight came off. It did not obviously move what the weight was made of.

The Famous Number Isn't a Measurement

This is where I had to change my mind, so let me be precise.

You have seen the claim: roughly 40% of the weight people lose on these drugs is muscle. It is the load-bearing fact in every "but at what cost" piece written about this class of drug in the last two years.

It is not a trial endpoint. Nobody measured it. Someone took the mean percentage changes out of the STEP 1 table and did arithmetic on them. Do that one way and you get about 34%. Another way, 39%. Another, 45%. Those are not different studies disagreeing. That is one set of group averages being squeezed for a number it was never designed to produce.

The second problem is bigger, and it applies to every figure in this article including the ones I am relying on.

DXA does not measure muscle. It measures lean soft tissue, and lean soft tissue is protein plus glycogen plus soft-tissue minerals plus water. All four move during weight loss. Glycogen comes down and takes bound water with it. In one study, simply drinking 500 to 1,000 mL of water raised measured lean soft tissue — no tissue gained, just fluid in the room.

So when a headline says people lost 9.7% of their "muscle," what was observed is a 9.7% fall in a compartment that contains muscle, along with the water and glycogen living inside and around it. Some of that is real contractile tissue. Some is the tank draining. Nobody has separated the two cleanly at scale.

Anyone who has run a low-carb protocol already knows this in their body. The first two weeks of keto drop several pounds that are neither fat nor muscle, and you learn fast not to take that number personally. The same measurement problem sits underneath the GLP-1 argument, unacknowledged.

Losing Some Lean Mass Is What Weight Loss Is

Here is the context that almost never travels with the scary number.

When you lose weight by any method — diet, exercise, a drug, surgery, a bad flu — some of what comes off is fat-free mass. That is not a side effect. It is arithmetic. A smaller body needs less scaffolding: less structural tissue, less blood volume, less muscle to move a frame that is now carrying thirty fewer pounds.

The rule of thumb in the body composition literature is that roughly one quarter of the weight lost through ordinary dieting is fat-free mass — pooled estimates put it around 27% in men and 20% in women across dietary and behavioural interventions.

Now go back and look at the SURMOUNT-1 figure again. Twenty-five percent.

That is not a drug doing something unusual to muscle. That is weight loss behaving the way weight loss has always behaved, at an amount that used to require surgery. The alarm was aimed at the ratio, and the ratio is ordinary. What is new is the magnitude — 15 to 21% of body weight gone in fourteen months — and the absolute quantity of lean tissue a normal ratio takes with it at that scale.

Which matters far more at 54 than at 34, and that is the part the coverage gets backwards. The question was never "does this drug attack muscle." It is whether someone already on the wrong side of the sarcopenia curve can afford a normal-ratio loss of that size without doing anything to defend it. Because the baseline is already moving: muscle mass declines roughly 3 to 8% per decade after about 30, accelerating after 60. That is a population estimate from the sarcopenia review literature, with enormous individual variation, not a constant that describes you. But here is the part that gets left out — strength is lost two to five times faster than mass. The tissue goes slowly. The ability to use it goes quickly.

The alarm was aimed at the ratio, and the ratio is ordinary. The thing that's new is the scale.

What This Evidence Cannot Tell You

I am putting this before the practical section rather than burying it at the end, because it is the section I would want first.

It cannot tell you whether the lean mass people lost mattered. This is the honest centre of the question and it is still open. Reviews of the GLP-1 and skeletal muscle literature report that the available evidence does not show impaired muscle strength — handgrip and sit-to-stand performance hold up. One semaglutide trial reported improved self-reported physical function despite lean mass loss. A liraglutide trial beat placebo on the five-times sit-to-stand test. Reassuring as far as it goes, and it goes about as far as "we measured grip strength in people who were mostly middle-aged and mostly not frail, for about a year."

It cannot tell you what happens in the people you would actually worry about. The reviews say this plainly: studies in older adults are limited, and the clinical effect on physical function and fragility fracture risk in that group remains unclear. Nobody has established whose lean mass loss is adaptive and self-limiting and whose translates into functional decline. A 72-year-old with borderline sarcopenia and a 41-year-old athlete are both "lean mass loss" in a spreadsheet.

It cannot tell you about ten years. Every number in this article comes from 68 to 72 weeks of follow-up. Obesity is a chronic condition and these are, increasingly, chronic medications. The body composition trajectory over a decade of continuous use is unknown, because nobody has watched it yet.

And the reassuring mouse study is a mouse study. A 2026 paper in Cell Reports Medicine travelled with exactly the conclusion I wanted to be true — GLP-1 medicines reduce absolute muscle mass mildly while improving relative muscle mass, strength relatively preserved, running performance improved. It is four preclinical studies plus one proof-of-concept clinical study. The running mice are the strongest part of it. The relative-versus-absolute distinction is genuinely useful; a proof-of-concept study in humans is a beginning, not an answer.

The Deficit Is the Variable. Not the Drug.

Here is the finding that changed what I think, and it comes from the strongest design in this article.

The BELIEVE trial was a randomised phase 2b study pairing semaglutide with bimagrumab, an investigational antibody that blocks a pathway limiting muscle growth. The body composition results separate cleanly:

• Semaglutide alone: weight −15.7%, lean mass −7.4%. About 71.8% of the weight lost was fat.
• Bimagrumab alone: weight −10.8%, lean mass +2.5%.
• The combination: weight −22.1%, lean mass only −2.9%. About 92.8% of the weight lost was fat.

More total weight off, and almost all of it fat. Which demonstrates the thing I needed to know: the lean mass share of weight loss is not fixed. It is not an unavoidable consequence of appetite suppression. It is a variable, and it responds to inputs.

Bimagrumab is investigational and not approved, and I am not recommending it — I am citing it as proof of concept. Because there is a second, unglamorous, already-available way to move that same variable.

In a 2018 systematic review and meta-analysis of older adults with obesity in caloric restriction, resistance training prevented 93.5% of the lean body mass loss the restriction caused — 0.819 kg of it. Three sessions a week, twelve to twenty-four weeks. That is not a GLP-1 trial and I will not pretend it is one. But the mechanism it defends against is the same mechanism.

And there is a randomised trial in the drug context. In a one-year trial in the New England Journal of Medicine, 195 adults with obesity were randomised after an eight-week low-calorie diet to placebo, exercise, liraglutide, or both. The combination cut body fat percentage by 3.9 percentage points — roughly twice exercise alone (1.7) or liraglutide alone (1.9).

Same drug. Same appetite suppression. Different training. Different body at the end.

The lean mass share isn't fixed. It's a variable — and lifting is the input nobody is selling you.

The Protein Problem Nobody Puts on the Label

This is the part I would put in the box if I designed the packaging.

These drugs work by making you not want food, and the mechanism is not selective: it does not suppress your appetite for the doughnut and leave your appetite for the ribeye intact. Total intake falls — and protein, the most effortful and least appealing macronutrient to eat when nothing appeals, tends to fall with it or faster.

So the person on the drug is frequently running a large deficit and an under-supplied protein intake at once, in their fifties, without lifting. Then we blame the molecule for the body composition.

The joint advisory from the American College of Lifestyle Medicine, the American Society for Nutrition, the Obesity Medicine Association and The Obesity Society names this directly. To preserve lean mass on GLP-1 therapy: protein above 1.2 g per kg per day, distributed evenly across meals, alongside resistance training and baseline assessment of muscle strength, function and body composition. The wider guidance literature puts the range during active weight reduction at roughly 1.2 to 1.6 g/kg/day on adjusted body weight. The ACSM's standing recommendation — resistance training two to three times a week, moderate to vigorous, all the major muscle groups — is the other half.

Distributed evenly is the phrase to hold onto, because it is harder than the total. Someone eating 40 grams at dinner and almost none before it hits a daily number on paper and still feeds the muscle badly. On a drug that makes early-day eating unappealing, that is the default failure, not the edge case.

And the obvious thing should be said: this is a recommendation about deliberate, structured eating, given to people whose reason for taking the medication is often that deliberate, structured eating has not worked for them. That tension is real, and pretending otherwise is how a lot of well-meaning advice fails.

Bone Is the Quieter Signal

If I were worried about something in this literature, it would not be muscle.

A 2026 study in the Journal of Clinical Endocrinology & Metabolism matched 255 people using semaglutide or tirzepatide for at least six months — all with DXA scans before and after — against 255 non-users matched on age, sex, BMI and diabetes status. Median follow-up 17 months.

Among users without diabetes, the annualised rate of total hip bone loss was significantly greater than in matched controls. Among users with diabetes, it was comparable to controls. Within the GLP-1 group, more weight loss went with more bone loss at the total hip and femoral neck.

Two honest caveats. This is a retrospective matched cohort, not a randomised trial — it cannot establish that the drug caused the bone loss, and weight loss by any means reduces bone density, because bone responds to the load it carries. And the diabetes split is odd enough that I would want it replicated before building anything on it.

But it points the same direction as everything else here. Rapid, large weight loss unloads the skeleton at the same time it shrinks the muscle pulling on it, in a population where the fracture that ends independence is already the thing to fear. Resistance training is the only intervention that addresses both at once. It is also the one thing nobody can sell you a subscription to.

The Regain Is the Part I'd Plan For

One more number, and this is the one I would want on the wall.

In the STEP 1 trial extension, participants lost 17.3% of their body weight over 68 weeks on semaglutide. Treatment stopped. One year later they had regained two-thirds of it, with cardiometabolic markers drifting back in step.

That is not a failure of the drug. It is the clearest statement anyone has made that obesity behaves like a chronic condition — remove the treatment, the condition returns. Blood pressure medication works the same way and nobody calls it a scandal.

But set it beside the body composition question and something uncomfortable appears. If a quarter of what comes off is lean tissue, and the weight comes back — what comes back? Whether regained weight returns preferentially as fat is widely asserted and, as far as I can find, not established: the extension reported weight and cardiometabolic variables, not the composition of the regain.

I flag it because the arithmetic is easy to do in your head and the conclusion is easy to reach without evidence. A cycle that removes lean mass and returns fat would be a serious problem. Whether that is what happens is unknown, and if you are making a decision here you are making it without that number.

What you can control is the state you are in when the weight comes back. Someone who spent those fourteen months lifting is a different person on the far side of a regain than someone who spent them only losing.

What This Has to Do With Leading Anything

I run platform engineering teams, and I have made the mistake in this article about twenty times in a conference room.

The mistake is not being wrong. It is that the belief I was wrong about was the one that made me look good — the long-way-is-the-only-way story, held by a person visibly doing it the long way. Comfortable conclusions get less scrutiny than uncomfortable ones, and the failure is invisible from the inside, because being right feels identical to being flattered. The tell, when I catch it, is always the same: I am repeating a number I never traced. In engineering it is the latency figure everyone quotes and nobody has pulled since the migration.

So the defence is to check the claims that flatter me first. The ones that annoy me already get checked, with enthusiasm. It is the ones making my practice look correct that slip through.

This site has argued for years that Eat, Train and Lead are one system. Here is what that looks like, and it isn't tidy: a nutrition question turned out to be a training question, the training question turned out to be a protein question, and the reason I had all three wrong was a leadership failure — I trusted a number because of who it made me.

Check the claims that flatter you first. The ones that annoy you already get checked.