Why we measure lean mass,
and what the evidence says.

The headline numbers come from DEXA substudies, and the table is what they actually reported. Read the last column first: what each study could and could not see.

What the MRI studies add

The SURPASS-3 MRI analysis scanned 246 adults with type 2 diabetes at week 52 on tirzepatide or insulin. Thigh muscle volume fell, by about 0.64 litres pooled across doses, and the fall matched what a population model predicts for that much weight loss. Fat infiltration inside the muscle also fell, which is a good sign. Secondary reviews that say the muscle was stable overstate it: the volume went down, by the expected amount.

A 2026 Cell Reports Medicine paper ran the question in mice and in a twelve-week pilot of ten adults with type 2 diabetes on semaglutide 1 mg. In the people, the loss was 70 percent fat and 30 percent lean, quadriceps size fell, and knee-extension and handgrip strength did not change. The authors say plainly that the pilot is too small and too short to be conclusive, and the mouse figure of 20 to 32 percent lean is a mouse figure.

The reviews, and the range they actually print

A 2026 systematic review of 35 randomised trials put the median share of weight loss from muscle-based indices at 28.3 percent (interquartile range 15.9 to 39.9) and found no included study that measured physical function. The pooled-by-molecule analysis behind the 25–39% span found the same partition in untrained lifestyle arms. The Lancet Diabetes and Endocrinology comment arguing muscle should be a treatment goal prints 25 to 39 percent. The widest honest span, 15 percent or less up to 40 to 60, comes from a review that attributes the spread to method and population, not drug.

The textbook rule is that about a quarter of any weight loss is fat-free mass. The critical review of that rule traced it to a 1984 cohort of women whose regression gave 22 percent, rounded up, and concluded it is an approximation with a limited mechanistic basis: the proportion varies with sex, starting fat, rate of loss and what you eat. Forbes showed the direction of the main effect decades ago. The fatter you start, the smaller the lean fraction of what you lose, and the harder you cut intake, the larger it gets. Hall later showed Forbes's original equation underestimates lean loss for the very large losses these medications now produce.

So the split is real, expected, and personal, and a weight line cannot show it: two people losing the same ten pounds can lose different things. That is the case for two lines, fat and lean, and for reading the lean one as a lean line, since water, glycogen and organ mass move it too.

The deficit evidence is a controlled-feeding study of 39 adults held at a 40 percent energy deficit for three weeks on 0.8, 1.6 or 2.4 grams of protein per kilogram a day. Both higher intakes lost proportionally less fat-free mass than the reference intake, and tripling it bought nothing over doubling. That settles that protein matters. The question is what to divide by.

Body weight is the denominator in almost every published figure because it is what every trial participant could measure. But fat needs almost no protein, and body weight falls exactly as the medication works, so a per-kilogram-of-weight target shrinks while the tissue it exists to protect does not. The one published guideline written on fat-free mass, 2.3 to 3.1 grams per kilogram for lean athletes dieting hard, is a different population from a GLP-1 user; it is the reason to sit at the bottom of that range, not to copy it. The shipped coefficient, its taper, and the floors are stated on the methodology page, and the protein calculator runs the same rule with the method shown. One useful property: a fat-free-mass error of a couple of kilograms moves the daily target by only a few grams, below the noise of anyone's adherence.

In the 20-trial analysis, the arm with the lowest lean share was not a drug arm. It was lifestyle plus a structured resistance-training programme, at 17.5%. In the strongest randomised trial of exercise alongside a GLP-1 medication, one year of exercise plus liraglutide roughly doubled the drop in body-fat percentage compared with either alone and was associated with increased lean mass; the exercise there was aerobic-dominant and the drug was liraglutide, so it argues for exercise changing what the loss is made of, not for lifting specifically. Rate matters too: 24 elite athletes assigned slow or fast loss, all lifting, gained 2.1 percent lean on the slow rate and held flat on the fast one, on diets the study set for them.

The alarm is strength, not the lean line. The sarcopenia consensus definitions diagnose on mass and function together, never mass alone, and the field's own reviews call DEXA lean a surrogate for muscle. When a creatine-dilution measure of muscle was compared with DEXA lean in 1,382 older men, the creatine measure predicted injurious falls and new mobility limitation, and DEXA lean did not. Power, the speed at which force is produced, declines earlier than strength with age and predicts function better than mass. So a lean number that dips while your lifts progress is mostly water and organ; a lean number that dips while your lifts stall is the signal to act.

Healthy body-fat ranges are not an opinion chart. The reference paper mapped body-fat percentage onto the BMI cutoffs the WHO and NIH already use, by sex and age, on four-compartment measurements, so the ranges inherit BMI's evidence base. On its combined table, the healthy span for men runs from about 8 to 20 percent in the twenties and thirties to 13 to 25 past sixty; for women, 21 to 33 rising to 24 to 36. A second paper repeated the method on nationally representative DEXA data and found the equivalent cutoffs sit a few points higher, because a DEXA number and a four-compartment number are different numbers for the same body.

That instrument dependence is why the app's bands are DEXA-referenced, why a body-composition scale reading of the same body lands lower, and why the target is set within a band rather than at a point. The bands themselves are stated on the methodology page.

The largest study of human water turnover measured 5,604 people from 23 countries with isotope tracking. Turnover tracked fat-free mass, physical activity and climate, and once fat-free mass was in the model the effect of sex was not significant. Body weight over-prescribes for fat, which turns over little water, so the app scales the target to lean tissue, not to weight, and does not carry a separate figure for men and women beyond the clinical floors.

The one randomised trial built to measure it enrolled 64 adults at raised fracture risk, 86 percent of them postmenopausal women, on semaglutide 1.0 mg or placebo for a year. Its primary endpoint, a bone-formation marker, did not move. Its secondaries did: bone resorption rose, and total-hip and lumbar-spine density fell relative to placebo by 2.6 and 2.1 percent, with 6.8 kg more weight lost. A retrospective cohort of 255 semaglutide and tirzepatide users matched to 255 non-users found faster annual hip bone loss in users without diabetes. And in the trial that added exercise to liraglutide, the combination held hip and spine density where liraglutide alone reduced it, despite similar weight loss.

Loading and protein protect bone as well as muscle, every imported DEXA scan already carries a bone density and a T-score, and the women's page works through why this matters most from perimenopause on.

A page that only lists what is known is not worth trusting. Nobody has the data on these yet.

• Whether strength and function hold in the people at real risk. Older, sarcopenic, or already low-muscle users. The reassurance so far is mass-based, from younger and heavier cohorts, or from uncontrolled studies. The 35-trial review found no included study that measured physical function.
• Whether tirzepatide partitions worse than semaglutide. One preprint of 7,965 paired measurements in routine care, about 85 percent of them from body-composition scales, reports one to two points more lean loss on tirzepatide alongside more weight loss. It is not peer reviewed, its authors call it descriptive, and it may change on review.
• Whether the drug adds lean loss beyond matched weight loss. No pair-fed human trial exists. The SURMOUNT-1 placebo arm is the closest evidence and says no.
• What happens after 72 weeks, and on stopping. The substudies end there. Regain in most weight-cycling data is disproportionately fat, which is its own argument for measuring through the whole arc.
• What a muscle-specific trial will show. LEAN-PREP, a 232-person trial of home resistance training, protein at 1.6 grams per kilogram, both, or neither alongside semaglutide or tirzepatide, has MRI quadriceps area as its primary outcome at six months. It is a protocol; there are no results. No creatine-dilution study in a GLP-1 cohort has been published.

Each of these is a sentence we, or a source we read, once got wrong. The corrections are the reason to trust the rest of the page.

A Senior Scientist at Tufts' Jean Mayer USDA Human Nutrition Research Center on Aging and Professor of Nutrition and Medicine, Roger Fielding has spent three decades studying why muscle declines with age, what that costs in function, and how exercise, protein, and nutrition can slow the loss. He reads what we ship against his field's literature. Where our numbers and that literature disagree, we say which, and ask him to arbitrate rather than pretend the literature is unanimous. Our targets go to him as proposals, not as settled fact, and several of the sources the app rests on are his.

Each principle runs somewhere in the product; each constant is stated in one place.