Evidence

What we know, how we know it, and what we do not.

Every research figure on this site lives in one file with its source, its range and the study it came from. This page is that file, in full, including the parts that are inconvenient for us.

20 figures. 9 sources, every one with a DOI you can open. Last reviewed 18 September 2026.

Why this page exists

Most health sites tell you a number. Almost none tell you how wide the error bars are, how many people were studied, or how much of what you are reading is one small trial being repeated by everybody who quotes it. We would rather show our working, because if you are making decisions about your own body you need to know which of these figures is solid and which is a best guess.

Nothing here is medical advice, and nothing here mentions a dose. Anything about how you feel, what you are experiencing or what you should take goes to your prescriber, not to us.

On this page

How to read a figure on this page

Each one carries four things: the number, the range around it, what it rests on, and how much weight we think you should put on it. That last line is a judgement and we say so.

Strong means the figure is pooled from many randomised trials. Moderate means the studies agree on direction but disagree with each other on size, or the finding is indirect. Limited means one study, a small sample, or a result borrowed from people who were not on these medications. We publish limited findings, clearly labelled, rather than leaving a gap and letting you assume we know more than we do.

Two more things worth saying out loud. A range is not a decoration: when a figure carries a band either side of it, the band is part of the number. And an average across thousands of people in a trial is not a forecast for you. It is the shape of the problem, which is why the whole method here is to measure yourself every week rather than trust a figure from a study that was not measuring you.

What we know

Grouped by the question each figure answers. Every one links to its source below.

How much of the lost weight is lean mass

This is the figure everyone quotes and almost nobody puts a range on. The four rows below come from the same analysis, which is the only way the comparison means anything.

35.2%

Semaglutide: Ozempic and Wegovy

Across 20 randomised trials, an average of 35.2% of the weight lost on semaglutide (Ozempic, Wegovy) was lean mass.

Range
Confidence interval 31.5% to 38.9%
What it rests on
20 randomised controlled trials, 15,782 participants
How much weight to put on it
Strong. Pooled from many randomised trials. About as solid as this field gets, and unlikely to move much when the next study lands.

25.4%

Tirzepatide: Mounjaro and Zepbound

For tirzepatide (Mounjaro, Zepbound) the figure was 25.4%.

Range
Confidence interval 22.8% to 28.0%
What it rests on
20 randomised controlled trials, 15,782 participants
How much weight to put on it
Strong. What that means.

26.8%

Liraglutide: Saxenda and Victoza

For liraglutide (Saxenda, Victoza) the figure was 26.8%.

Range
Confidence interval 23.1% to 30.5%
What it rests on
20 randomised controlled trials, 15,782 participants
How much weight to put on it
Strong. What that means.

26.2%

Diet alone, no medication

Losing the same weight through diet alone cost 26.2% as lean mass, which was not significantly different from the medications (p = 0.42). The muscle comes off because the weight comes off, not because of the injection.

Range
Confidence interval 24.1% to 28.3%
What it rests on
20 randomised controlled trials, 15,782 participants
How much weight to put on it
Strong. What that means.

28.3%

The same question asked by a second review

Across 35 studies the median share of weight loss coming from muscle-related measures was 28.3%, with wide variation between studies.

Range
Interquartile range 15.9% to 39.9%. Half of the studies fell inside this band; the spread either side is the honest part.
What it rests on
35 studies, median duration 26 weeks
How much weight to put on it
Strong. What that means.

What changes the number

One thing in this evidence base moves the figure a long way, and it is not the medication you are on.

17.5%

Adding resistance training

Adding resistance training brought it down to 17.5%, the most favourable result of any approach studied.

Range
Confidence interval 14.2% to 20.8%
What it rests on
20 randomised controlled trials, 15,782 participants
How much weight to put on it
Strong. What that means.

What supervised training has produced

Supervised resistance training for longer than 10 weeks has produced gains of around 3 kg (6.6 lb) of lean mass and around 25% in strength in adults.

What it rests on
Narrative review
How much weight to put on it
Moderate. Consistent across studies, but the studies differ from one another or the finding is indirect. Trust the direction; treat the number as approximate.

What protein alone does, and does not do

In 191 older adults, higher protein was associated with more preserved lean tissue but not with better grip strength, walking speed or chair-stand scores. Protein looks after the material. Training is what turns it into function.

What it rests on
191 adults, mean age 65.1 years
How much weight to put on it
Limited. One study, a small sample, or borrowed from people who were not on these medications. We report it because it is the best there is, not because it is settled.

What a weekly strength test can and cannot tell you

We got this wrong until 18 September and the correction is logged below. A rep count you do at home measures what you can do. It does not measure how much muscle you have, and nothing you can do at home does.

What the 30-second sit-to-stand actually tracks

The number of stands you manage in 30 seconds tracks physical performance rather than muscle mass. In 129 older adults it correlated with the Short Physical Performance Battery (r = 0.52) and showed no significant relationship with knee extensor strength, grip strength or skeletal muscle index. It measures what you can do, which is worth knowing on its own.

What it rests on
129 community-dwelling older adults, mean age 71.8
How much weight to put on it
Moderate. What that means.

The measure in that same test that does track muscle

Peak power during that same sit-to-stand does track muscle: it correlated with skeletal muscle index at r = 0.70, and with knee extensor and grip strength at r = 0.64. Power cannot be counted by eye, which is why a rep count is the wrong instrument for the question most people are really asking.

What it rests on
129 community-dwelling older adults, mean age 71.8
How much weight to put on it
Moderate. What that means.

Why the scale, the scan and your strength can disagree

Three findings that explain why somebody can lose real muscle and have every number they are shown look fine.

Lean mass as a share of body weight

Lean mass as a percentage of body weight was unchanged in pooled trials, which is why people can lose real muscle and still look like the numbers are fine.

What it rests on
22 randomised controlled trials, 2,258 participants
How much weight to put on it
Strong. What that means.

The most effective medications are not the kindest to muscle

The medications that take off the most weight and the most fat, the stronger tirzepatide and semaglutide arms, were among the least effective at holding on to lean mass. More weight lost is not the same as a better result.

What it rests on
22 randomised controlled trials, 2,258 participants
How much weight to put on it
Moderate. What that means.

How much lean mass that is, in plain terms

The lean mass lost on the strongest of these medications, around 6 kg (13 lb), has been described as comparable to a decade or more of ageing.

What it rests on
Narrative review
How much weight to put on it
Moderate. What that means.

Protein

Targets from the nutrition literature. Read the first method note below before you use them: we deliberately calculate from goal weight and the papers do not.

1.2 g/kg

The floor

Guideline-informed protein targets during this kind of weight loss start at 1.2 g per kg of body weight a day (0.55 g per lb).

What it rests on
Narrative review, guideline-informed targets
How much weight to put on it
Moderate. What that means.

1.6 g/kg

The ceiling

Up to 1.6 g per kg a day (0.73 g per lb) in adults without chronic kidney disease.

What it rests on
Narrative review, guideline-informed targets
How much weight to put on it
Moderate. What that means.

Per meal, not per day

Spread across meals at roughly 0.3 to 0.4 g per kg per meal, rather than taken mostly at dinner.

What it rests on
Narrative review, guideline-informed targets
How much weight to put on it
Moderate. What that means.

Leucine per meal

Each of those meals wants roughly 2.5 to 3 g of leucine, the amino acid that actually switches muscle building on. That is about 30 g of whey, 120 g of chicken breast, or four eggs.

What it rests on
Narrative review, guideline-informed targets
How much weight to put on it
Moderate. What that means.

The markers this literature watches

Vitamin D, B12, iron studies, folate, zinc and thiamine are the markers this literature suggests watching. Ask your prescriber whether they are worth checking for you; we do not order or interpret bloods.

What it rests on
Narrative review, guideline-informed targets
How much weight to put on it
Moderate. What that means.

After the medication

Thin evidence, reported as thin.

What comes back if you stop

Weight regained after stopping appears to come back disproportionately as fat, so repeated cycles can leave someone heavier in fat and lighter in muscle than when they started.

What it rests on
GRADE-assessed review
How much weight to put on it
Limited. What that means.

What nobody has measured

The most important thing on this page is a hole.

Function was not measured, in any of it

In a 2026 review of 35 studies in the Annals of Internal Medicine, not one reported whether people could actually do more or less afterwards. Scans were measured. Getting out of a chair was not.

What it rests on
35 studies, median duration 26 weeks
How much weight to put on it
Strong. What that means.

How we work out the numbers

Three places where what we publish is not a straight copy of what the papers say. Each one is a deliberate choice, and here is the reasoning so you can disagree with it.

We calculate protein from goal weight, not current weight

The research states protein targets per kg of body weight. We use goal weight instead. The reason is that fat tissue has a much smaller protein requirement than muscle does, so for someone carrying a lot of weight, a target set on current body weight can produce a number nobody with a suppressed appetite will ever eat. Goal weight gives a figure that is achievable and still above the floor. If your weight is close to your goal the two numbers nearly match. If you want the literature's version, multiply your current weight in kg by 1.2.

We report ranges, not single numbers

Every pooled figure here carries a confidence interval, and the studies behind them disagree with each other more than most health articles admit. Heterogeneity in the main meta-analysis was moderate (I-squared 68%). Treat these as the shape of the problem, not as your personal forecast.

What we do not know

Nobody has shown what any of this does to the things you would actually notice: stairs, shopping bags, getting off the floor. The scans have been measured and the function has not. That is the honest state of the evidence, and it is why we ask you to test yourself every Sunday rather than trust a number from a trial that was not measuring you.

Every source in full

The complete reference list, with what each paper is doing for us. Every DOI opens the paper itself, not a summary of it and not another health site quoting it.

We cite the paper, never a press release, a supplement company's own data, or another content site. Where a figure is pooled from many trials we say how many trials and how many people. Where it rests on one study, we say that too, and we mark it limited.

What would change our mind

A position you cannot describe the disproof of is not a position. So, specifically:

  1. A trial showing resistance training does not protect lean mass on these medications

    The training figure is the whole argument for what we sell. If a well-run randomised trial put a supervised strength programme against usual care on a GLP-1 and found no difference in lean mass or in strength, we would say so on this page within a week, and the product would have to change or go.

  2. Function data that goes the other way

    The gap we point at, that scans have been measured and function has not, cuts both ways. If the trials start measuring grip strength, walking speed and getting out of a chair, and people on these medications hold their function while their scans drop, then the scan loss matters far less than we have been saying and we would have to rewrite a great deal.

  3. A protein target that turns out to be wrong at the top end

    We publish a range with a ceiling, not "eat as much protein as you can". If the evidence in older adults moved the useful ceiling down, our calculator numbers would move with it the same day.

  4. A larger or better synthesis of the lean mass share

    This already happened once, on 18 September 2026, and it is logged below. The next one will be treated the same way: new figure in, old figure named, correction dated, nothing quietly edited.

What would not change our mind: a single dramatic study, a preprint, a conference abstract, a press release, or an expert with a strong opinion and no data. We wait for the synthesis.

Corrections log

Dated, public, and newest first. When a figure on this site changes, the old one is named here rather than deleted. A site that publishes its own corrections is worth more than one that edits quietly and hopes nobody kept a screenshot.

  • 18 September 2026

    We were overstating what the weekly strength test can tell you

    This site has been telling people that if their weekly sit-to-stand score holds, their muscle is intact. That was a stronger claim than the evidence supports, and we have corrected it everywhere.

    In 129 community-dwelling older adults, mean age 71.8, the rep count of a 30-second sit-to-stand correlated with physical performance (Short Physical Performance Battery, r = 0.52) but showed no significant relationship with muscle mass or muscle strength (Hwang W et al, Journal of Cachexia, Sarcopenia and Muscle 2026). The measure in that test that does track muscle is peak power, which correlated with skeletal muscle index at r = 0.70. Power is not something you can count by eye.

    So the honest version is this. The strength test measures what you can do, not how much muscle you have. Nothing you can do at home measures how much muscle you have. We think what you can do is the more useful of the two, and it is the thing not one of the 35 studies in the Annals review bothered to measure, so we are not backing away from the test. We are describing it accurately.

    One practical note that follows from the same literature. The smallest change in this test that can be told apart from measurement error is roughly two to four repetitions, which is why we have always said to look at a trend across three or four Sundays rather than react to one week. That guidance was right and it has not changed.

    The coaching does not change either: protein first, two sessions a week against resistance, and keep measuring. Only the claim about what the number proves has changed.

  • 17 September 2026

    We replaced the 40% semaglutide figure with 35.2%, and added the comparison we had been leaving out

    Until today this site led with a figure of around 40% of the weight lost on semaglutide being lean mass. That came from the DEXA sub-study inside STEP 1, Wilding JPH et al, New England Journal of Medicine 2021. It has been replaced with 35.2%, pooled across 20 randomised controlled trials, 15,782 participants, Eisa N et al, Diabetes, Obesity and Metabolism 2026.

    Two things were wrong with the old number. It was a single sub-study, kept in place after a much larger pooled estimate existed. And standing on its own it implied that the medication does something uniquely bad to muscle. That was not supported. In the same analysis, losing the same weight by diet alone cost 26.2% as lean mass, and the difference between the medications and lifestyle was not statistically significant (p = 0.42). The muscle comes off because the weight comes off.

    So we have added the diet-only comparison everywhere the figure appears, and the number that actually helps you: with resistance training added, the share drops to 17.5%.

    We were not wrong about the direction, and the coaching does not change. We were wrong to lead with the scariest defensible number and leave out the context that makes it usable. The old figure and its source stay on this page rather than disappearing from it.

  • 17 September 2026

    This page exists

    Every research figure on the site moved into one file with its source, its range and its study design, and the pages now render from that file instead of quoting numbers in prose. Before today, figures were typed into individual pages and could drift. This is the first version of the public record; the changelog runs from here.

If a figure here looks wrong, tell us

We would rather be corrected than be quoted incorrectly. If you think a number on this page is out of date, misread, or attached to the wrong study, write to hello@everstrong.online with the paper you think we have missed. A real person reads it.

If we agree, the figure changes, the change is dated in the log above, and you get a reply saying what moved. If we disagree, you get a reply explaining why, which is a fair trade for the two minutes it took you to write.

How we work: sources, money and corrections

EverStrong provides nutrition and strength coaching and education. It is not medical advice and does not replace your prescribing doctor. Nothing on this site gives a dose, a taper, or tells you to change your medication. Last reviewed 18 September 2026.