GLP-1 Drugs and Muscle Loss: What the DXA Data Show

The objection is now reflexive: GLP-1 drugs make you lose muscle, not just fat. It is a serious question and it has been measured — but almost never with the numbers people quote, and almost never against the right comparison. The honest version of the answer has three parts: roughly a quarter of the weight lost on these drugs is lean tissue; that fraction is close to what non-drug weight loss produces; and essentially nobody has measured whether it costs you anything you can feel.

What "muscle loss" means in these trials, and what it does not

Every figure below comes from dual-energy X-ray absorptiometry — a DXA scan, which sorts body mass into fat, bone mineral, and everything else. That third bucket is reported as "lean mass," "fat-free mass," or "lean soft tissue" depending on the paper, and it is not muscle. It includes organs, connective tissue, glycogen and the water bound to it. A person who loses 10 kg quickly loses glycogen and its associated water early, and a DXA scan books that as lean mass gone.

That matters because the alternative measurements disagree with DXA by a predictable margin. The 2026 systematic review in Annals of Internal Medicine, the broadest synthesis currently available, applied two different benchmarks for exactly that reason: about 25% of total weight loss for fat-free mass measured by DXA or bioelectrical impedance, but about 15% for skeletal muscle measured by CT or MRI³. Same people, same weight loss, different tissue definitions, ten percentage points apart. When a headline says "40% of the loss is muscle," the first question is which instrument produced it.

The best-measured single answer

That is the SURMOUNT-1 body-composition substudy: 160 of the trial's 2,539 participants had DXA scans at baseline and at week 72, 124 of them on pooled tirzepatide doses and 36 on placebo¹. The substudy participants were 73% female with a mean weight of 102.5 kg and a mean BMI of 38.0.

The result the paper leads with is the one everyone quotes: "Of the body weight lost, approximately 75% was fat mass and 25% was lean mass¹." The clause that follows in the same sentence is the one nobody quotes: "for both tirzepatide and placebo." The placebo arm lost 5.3% of body weight and split it the same way the drug arm split its 21.3%. On this evidence, the ratio is a property of losing weight, not a property of the drug — what the drug changes is how much total weight comes off, which is four times as much.

The paper carries a published correction, and it is worth saying exactly what it is rather than letting the word "correction" do unearned work: the erratum fixes the y-axis label on Figure 4, which read "Change from baseline (%)" where the plotted data were kilograms². No reported result changed. It is a units label on one figure.

Semaglutide's own DXA data says two opposite-sounding things at once

Tirzepatide is not semaglutide, and semaglutide's own randomized DXA evidence comes mainly from the SUSTAIN 8 substudy, which scanned 178 participants with type 2 diabetes and analyzed observed end-of-treatment data from 114 of them at 52 weeks⁴. Read its two headline sentences together:

Total lean mass, from a baseline of 51.3 kg, "was also reduced by 2.3 kg" on semaglutide. And the proportion of lean mass, from a baseline of 59.4%, "increased by 1.2%-point⁴." Both are the same participants over the same 52 weeks. Absolute lean mass fell; the share of the body made of lean tissue rose, because fat fell faster.

Neither sentence is a spin on the other, and which one you lead with determines the entire tone of an article. Someone arguing these drugs waste muscle quotes the first. Someone selling them quotes the second. This site's position is that a reader needs both, plus the substudy authors' own caveat, which is unusually blunt: SUSTAIN 8 compared semaglutide against canagliflozin with no placebo arm, and the authors wrote that "the specific impact of both treatments on body composition in the absence of a placebo arm is speculative at this stage⁴."

The independent network meta-analysis of 22 randomized trials and 2,258 participants reaches the identical two-sided conclusion: GLP-1 receptor agonists significantly reduced lean mass (mean difference -0.86 kg, 95% CI -1.30 to -0.42), "with lean mass loss comprising approximately 25% of the total weight loss. However, the relative lean mass, defined as percentage change from baseline, was unaffected⁵."

Where the syntheses genuinely disagree

They do not all land in the same place, and pretending otherwise would be the easy move here.

The network meta-analysis singles out the two most potent agents unfavorably: "Tirzepatide (15 mg weekly) and semaglutide (2.4 mg weekly) were the most effective for weight and fat mass reduction but were among the least effective in preserving lean mass⁵." That directly cuts against SURMOUNT-1's substudy finding that the 75/25 split held across dose and subgroup, and against its finding that placebo split the same way. The most likely reconciliation is that a drug producing four times the total weight loss produces roughly four times the absolute lean-mass loss even at an unchanged ratio — "least effective at preserving lean mass" in absolute kilograms and "same proportional split" can both be true — but the two papers do state it in opposite registers, and readers meeting only one of them will come away with opposite impressions.

One detail in that meta-analysis should be flagged rather than repeated: its abstract describes "Liraglutide, at 3.0 mg weekly or 1.8 mg daily," as the only agent achieving significant weight reduction without significantly reducing lean mass. The 3.0 mg liraglutide product is Saxenda, and its current FDA label instructs "Inject SAXENDA subcutaneously once daily⁷" — daily, not weekly. The finding may well stand; the dosing frequency printed next to it does not. (Our standalone liraglutide review covers that drug's own approval history and forms.)

The Annals review is the one that puts incretin weight loss on the same axis as non-drug weight loss, and its numbers are the most useful in this whole literature. Across 35 studies with a median duration of 26 weeks and a median of 78 participants, the median proportion of total weight loss attributable to muscle-based indices in the incretin groups was 28.3% (IQR 15.9% to 39.9%), with 65% of those arms exceeding the roughly 25% benchmark³. But among the 13 studies that reported weight loss in their lifestyle or placebo comparator groups, 38% also exceeded the respective benchmark — and the review's own conclusion phrases that same comparison as "nearly half of nonpharmacologic interventions that produced weight loss." The abstract does not agree with itself by roughly ten points on that figure; both versions are printed here rather than choosing the more convenient one.

The measurement nobody made

The Annals review states it in one clause that ought to end most arguments on this subject: "no study reported objective physical function outcomes³."

Thirty-five studies, thousands of participants, DXA and BIA and CT and MRI — and not one measured grip strength, gait speed, chair-rise time, or anything else that would tell you whether the lean tissue that went missing was tissue anyone needed. The review's own conclusion is a call for exactly that: "Future trials are needed to better understand mechanisms underlying these changes and their clinical implications." Only 10 of the 35 studies (28.6%) had even prespecified body composition as a primary outcome, and fewer than half were rated at low risk of bias.

So the current honest state of the evidence is: the lean-mass loss is real and quantified; whether it produces weakness, frailty, or a lower resting metabolic rate that matters is, at the level of randomized evidence, unmeasured. That is a different sentence from "it's fine," and a different sentence from "it causes sarcopenia."

One provider on this roster has built its product around that gap. HumeCare+ ships a body-composition analyzer with every plan and prescribes a weekly scan, so a patient watches fat mass and lean mass move separately instead of watching one number on a bathroom scale. Measured against everything above, that is the right variable — it is the same quantity the DXA substudies report, and tracking it weekly at home is more than any trial asked of its participants. It is worth being exact about what it still does not answer: a home impedance scan estimates what your body is made of, not what it can do, and the measurement this literature is actually missing is grip strength and gait speed rather than another composition number. A weekly reading tells you the split is moving. It does not tell you the lean tissue you kept is tissue that works.

What this does and does not imply about scale

The composition question only matters in proportion to how much weight is coming off, and on the most potent agents that is a lot. STEP 1 reported a mean weight change of -14.9% on semaglutide 2.4 mg against -2.4% on placebo at 68 weeks, or -15.3 kg against -2.6 kg⁶. Apply a 25% lean fraction to 15.3 kg and you get roughly 3.8 kg of lean tissue — against roughly 0.65 kg for the placebo group's much smaller loss. The ratio is unchanged; the absolute number is nearly six times larger. Anyone dismissing the concern by pointing only at the ratio is answering a question nobody asked.

This is also why the compounds marketed as muscle-sparing add-ons are being studied at all. Our review of bimagrumab — a monoclonal antibody, not a peptide, now developed by Eli Lilly — covers the one serious pharmacologic attempt to change the split rather than the total, and what its own trial data actually support. Nothing on this site's provider boards is sold for that purpose, and no roster provider sells bimagrumab.

For the other visible consequences of losing weight at this speed, our article on loose skin and body changes after GLP-1 weight loss covers the tissue changes people actually see in the mirror, which are driven by the same underlying fat-volume loss rather than by anything specific to the drug class. If you are choosing between the two most potent agents, our head-to-head on tirzepatide and semaglutide covers the efficacy comparison directly, and our tirzepatide provider rankings carry the price research on the drug with the largest body-composition substudy behind it.

The bottom line

Across the randomized body-composition substudies, roughly 25% of the weight lost on a GLP-1 or GLP-1/GIP drug is lean tissue by DXA, and roughly 15% is skeletal muscle when measured by CT or MRI instead. In SURMOUNT-1's DXA substudy the placebo arm split its own smaller weight loss the same way, which makes the ratio a feature of weight loss rather than of the drug. Absolute lean mass falls; the proportion of the body that is lean tissue goes up. And across 35 studies, not one measured whether any of it costs you strength or function — which is the question everyone is actually asking, and the one the evidence cannot yet answer.