The body composition data nobody quotes.

The number everyone quotes from SURMOUNT-1 is 20.9 percent — the mean body-weight reduction at 72 weeks on the highest dose of tirzepatide in Eli Lilly's landmark obesity trial, published in the New England Journal of Medicine in July 2022. The number almost nobody quotes is the one I find far more interesting: in the body-composition substudy nested inside that same trial, roughly one quarter of the weight participants lost was lean mass, not fat.

That second figure is the entire basis of the "GLP-1s waste your muscle" panic that now runs through every gym, podcast, and wellness feed. And here is my thesis: the substudy that produced it does not support the panic. Read in full, it points the other way — toward an improvement in body composition — and the loud version of the story survives only because people quote the one-quarter figure without the denominator, the comparison, or the definition of "lean."

I think this is the most misread dataset in the entire field. So let me read it properly.

The substudy almost nobody opened

SURMOUNT-1 enrolled 2,539 adults with obesity and no diabetes. The headline arms are familiar: about 15 percent weight loss at 5 mg, 19.5 percent at 10 mg, and 20.9 percent at 15 mg over 72 weeks. What got a single sentence in most coverage was a sub-analysis of around 160 participants who received DEXA scans — dual-energy X-ray absorptiometry, the imaging method that actually separates fat mass from lean mass instead of guessing from the scale.

In that subgroup, total fat mass fell roughly 33.9 percent while lean mass fell roughly 10.9 percent. Because the body started with far more fat than lean to lose proportionally, the ratio of total fat mass to lean mass improved — participants ended the trial leaner in composition, not just lighter on the scale. The "one quarter was lean" line is the same data expressed as a share of total kilograms lost. Same numbers. Opposite emotional valence depending on which way you say it.

"Lean mass" is not "muscle" — and the panic depends on pretending it is

This is the part that I think does the most damage. On a DEXA scan, "lean mass" is everything that is not fat and not bone mineral: skeletal muscle, yes, but also organ tissue, connective tissue, glycogen, and — crucially — water. When a person loses 20 percent of their body weight, blood volume, extracellular fluid, and glycogen stores all fall with it, and glycogen drags roughly three grams of water along for every gram stored.

So a meaningful fraction of the "lean mass" lost in any rapid weight-loss intervention is not contractile muscle being broken down. It is water and glycogen leaving a smaller body. The headline "you lose 25 percent muscle" silently swaps "lean mass" for "muscle," and the two are not the same thing. The trials measured the first. The internet reports the second.

The ratio is normal — that is the buried headline

Here is the comparison the panic never includes. When people lose weight through caloric restriction alone, lean mass has historically accounted for somewhere between 20 and 30 percent of total weight lost — a range documented for decades, long before anyone injected anything. Bariatric surgery, the most aggressive intervention we have, lands in a similar zone, with some studies reporting lean-mass fractions at or above the GLP-1 numbers.

Against that backdrop, tirzepatide's roughly one-quarter lean-mass share is not an anomaly. It is the expected physiology of losing a large amount of weight by any means. The honest reading of SURMOUNT-1's body-composition data is not "this drug melts muscle." It is "this drug produces the same body-composition pattern as every other large weight loss, while removing dramatically more fat in absolute terms." That is a far less alarming sentence, which is presumably why it travels less well.

Where the real concern actually lives

I do not want to flatten this into "nothing to see here," because there is something to see. The legitimate worry is not the ratio — it is the absolute lean mass lost, and who loses it. A 35-year-old with a large fat reserve can shed lean mass and remain functionally strong. A 72-year-old near the threshold of clinical sarcopenia is in a different situation entirely, and the SURMOUNT population skewed middle-aged, with a mean age around 45. We have far less long-term body-composition data in older patients, who are precisely the group for whom losing lean tissue carries the most functional risk.

The second real concern is the regain phase. SURMOUNT-4, the withdrawal trial, showed that stopping tirzepatide brought substantial weight back — and the worry the data cannot yet fully answer is whether weight regained after a drug holiday returns as proportionally more fat than the lean tissue that was lost. If the off-ramp rebuilds fat faster than muscle, the composition story gets worse on the way back up than it looked on the way down. That is the open question I would prioritize, and it is barely studied.

The industry already believes the concern is real

You can read the seriousness of the lean-mass question straight off the corporate balance sheets. In 2023 Eli Lilly acquired Versanis Bio for up to roughly $1.93 billion, largely for bimagrumab, an antibody that blocks activin type II receptors and, in mid-stage trials, increased lean mass while reducing fat — the explicit thesis being to pair it with a GLP-1 so the weight that comes off is disproportionately fat. Regeneron has pursued a parallel idea with myostatin-pathway antibodies, and Scholar Rock's apitegromab and similar muscle-preservation programs have drawn real money.

An entire sub-sector is now being built on the premise that the body-composition profile of GLP-1 drugs is worth improving. I think that is rational — but notice what it implies. The companies are not racing to fix a catastrophe; they are racing to optimize a ratio that is already roughly normal, because at a population of tens of millions even a modest composition gain is worth billions. The investment is a sign the concern is real at scale, not a sign the panic was right about the individual.

The counter-argument I take seriously

Let me argue against myself. The DEXA substudy was small — on the order of 160 people — and 72 weeks is not a lifetime. DEXA itself struggles to fully separate water shifts from true tissue change, which cuts both ways: it means the muscle-loss number may be overstated, but it also means we should hold any reassurance loosely. And "the ratio is normal" is cold comfort to an individual patient who needed the muscle they had. Population physiology and a single person's functional outcome are not the same claim, and I have been making the population one.

There is also a genuine unknown in the oldest and frailest patients, where the data is thinnest and the stakes are highest. So the defensible position is narrow: the broad "GLP-1s destroy muscle" claim is not supported by SURMOUNT's own body-composition data, but a targeted concern about older, sarcopenia-prone patients and about the regain phase is supported, and deserves the dedicated trials it is only now starting to get.

What the body-composition question actually decides

So watch the right numbers. Does lean-mass loss in older cohorts look like the middle-aged SURMOUNT figure, or worse? Does the regain phase return fat faster than lean? Do the muscle-preservation combinations — bimagrumab and its competitors — actually move the fat-to-lean ratio in a way that matters functionally, or only on a DEXA readout? Those three questions decide whether "muscle loss" is a manageable footnote of large weight loss or a real limit on who should be losing weight pharmacologically and how fast.

The 20.9 percent will keep getting the headlines. The body-composition data underneath it is where the more honest and more interesting argument lives — and it has been sitting in the supplementary appendix of one of the most-cited trials of the decade, mostly unread, the whole time.

Ozemback — June 2026