Hastrup Review of Science

← All episodes

Deep divefrom Episode 14Fat loss21 min

Weight-Loss Drugs and the Lean Mass That Goes With the Fat

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Alhamad et al. 2026, Cureus: Incretin-based pharmacotherapies and lean body mass (open access)

MP3

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

How much of the loss is not fat

Host 1If someone loses a lot of weight on one of the new weight-loss injections, how much of what they lose is something other than fat?

Host 2In the placebo-controlled trials this review could actually use, people on the drugs lost between about three and a half and six kilograms more lean mass than people on placebo, over roughly a year to a year and a half. The catch is that lean mass on a scanner is not the same thing as muscle, and that distinction runs through the whole paper.

Host 1Welcome to the Hastrup Review of Science, where artificial intelligence reads real science, so you don't have to pretend you understood the paper.

Host 2Today it is one systematic review, published in October twenty twenty-six in the journal Cureus by a team of family doctors and a medical college in Saudi Arabia. It asks what the incretin drugs, the family that includes semaglutide and tirzepatide, do to lean body mass, and whether the trials allow a fair comparison between the drugs.

Where these drugs came from

Host 1Before we get to the numbers, where do these drugs come from, and why is everyone suddenly talking about muscle?

Host 2In the wider history of the field, the story starts with an observation from the nineteen sixties: sugar swallowed by mouth triggers far more insulin than the same sugar put straight into a vein. The gut hormones responsible were named incretins, and the most important one for these drugs is glucagon-like peptide one, usually shortened to GLP one.

Host 1So the gut is telling the pancreas that food is on the way, before the blood sugar has even risen?

Host 2That is the idea, and GLP one also slows the stomach and acts on the brain to reduce appetite. The natural hormone lasts only minutes in the body, so the first drug, exenatide, came from a longer-lasting version found in the saliva of a lizard, the Gila monster, and later drugs like liraglutide and semaglutide were engineered to last a day or a week.

Host 1And the newer names, tirzepatide and the others, how are they different from the originals?

Host 2They hit more than one receptor. Tirzepatide acts on GLP one and on a second gut hormone receptor called GIP, mazdutide combines GLP one with the glucagon receptor, and orforglipron is a pill rather than an injection; the review calls the whole group incretin-based therapies.

Host 1Why does the size of the weight loss change the question people ask?

Host 2The paper puts it plainly: these agents can produce substantially greater weight loss than drug treatment used to, and they are being used more widely and for longer. Once people lose fifteen or twenty percent of their body weight, the natural next question is what that weight was made of.

Host 1How big is the group of people this could apply to?

Host 2The authors cite about two and a half billion adults worldwide who were overweight in twenty twenty-two, more than eight hundred and ninety million of them living with obesity. They also cite an estimate of five hundred and eighty-nine million adults with diabetes in twenty twenty-four.

What was known before

Host 1What did people already know about weight loss and lean tissue before this review came along?

Host 2As a broad pattern in obesity research, any large weight loss, whether from diet, surgery or drugs, takes some lean tissue along with the fat, because a lighter body needs less supporting tissue. Earlier trials of these drugs had already reported lean-mass drops in their body-composition substudies, and other reviews had pooled them.

Host 1What exactly is lean body mass, in plain words, and how do they measure it?

Host 2In these trials it is mostly measured with a DXA scan, dual-energy X-ray absorptiometry, the same low-dose X-ray method used for bone density. It splits the body into fat, bone and everything else, and everything else is called lean mass.

Host 1So everything else would include muscle, but also what?

Host 2Water, organs, skin, connective tissue, any non-fat soft tissue. That is why the authors stress that a drop in DXA lean mass is not by itself muscle loss, not a measure of strength, and not evidence of sarcopenia, the medical term for loss of muscle mass and function that matters most in older people.

Host 1If the drop in lean mass was already known, what was still unsettled?

Host 2Two things. First, how the drugs compare with each other, which no trial has tested head to head for this outcome. Second, the trials report lean mass in so many different ways that it is easy to pool things that do not belong together.

The open question

Host 1What kind of different ways are we talking about, and why does it matter which one a trial picks?

Host 2The paper lists kilograms lost, percentage change from the starting value, lean mass as a share of body weight, fat-free mass, appendicular lean mass in the arms and legs, and tissue volumes from MRI. The authors say these answer related but non-equivalent questions.

Host 1Can you give an example where two of those would point in opposite directions?

Host 2The paper gives exactly one: lean mass as a proportion of body weight can go up while the actual amount of lean tissue goes down, simply because fat falls faster. So a trial could report a rising lean share and a falling lean mass at the same time, and both would be true.

Host 1So the comparison they wanted was a network meta-analysis; what is that in plain words?

Host 2A meta-analysis pools trials that compare the same two things. A network meta-analysis also makes indirect comparisons: if drug A was tested against placebo in one trial and drug B against placebo in another, placebo acts as a bridge to estimate A against B.

Host 1And what has to be true for that bridge to be trustworthy?

Host 2The trials have to be similar enough in the things that change the result, which the paper calls transitivity: who was enrolled, how long, which measurement, which statistical question. The authors call the precise question an estimand, meaning the outcome, the population, the comparison, the time point and how dropouts or stopping treatment are handled.

What the team did

Host 1What did this team actually do, step by step?

Host 2They wrote a protocol in advance and registered it in PROSPERO, the international register for systematic reviews, in August twenty twenty-six. The main outcome was fixed there: the difference between drug and comparison in percentage change of total DXA lean mass, measured between forty-eight and eighty weeks, with kilograms lost as the key secondary outcome.

Host 1Why fix the outcome before looking, rather than picking the best measure afterwards?

Host 2Because choosing after seeing the data invites picking whichever outcome tells the nicest story. The paper says that eligibility criteria, outcome definitions, time windows and analysis plans were not changed after registration, and the one amendment only documented limits on database access.

Host 1Which trials were allowed in?

Host 2Randomized controlled trials in adults with overweight, obesity or type two diabetes, lasting at least twelve weeks, comparing an incretin drug with placebo, usual care or another incretin drug. Body-composition substudies counted only if the original randomization was preserved.

Host 1How did they find those trials?

Host 2They searched PubMed with two complementary search strategies on the twenty-fourth of August twenty twenty-six, and exported ClinicalTrials.gov the same day. That gave six thousand five hundred and twenty-six records, two hundred and fifty duplicates were removed, and six thousand two hundred and seventy-six were screened.

Host 1Is that the whole medical literature, or just part of it?

Host 2Only part, and they say so. The Cochrane trials register returned eight thousand two hundred and fifty records but could not be exported reproducibly, so those were never screened one by one, and Embase, Scopus and Web of Science were not searched at all because the team had no subscription access.

Host 1So what came out at the end of the screening funnel?

Host 2One hundred and seventeen reports went to a final eligibility check, sixty-two were excluded and two ongoing studies were set aside for monitoring. The remaining fifty-three reports and registry records covered fifty-two unique trials, which the paper calls trial families because one trial can have several publications.

Host 1Fifty-two trials sounds like plenty; how many of them could actually go into the numbers?

Host 2Very few, and that is the central lesson of the methods. Being eligible for the review only meant a trial had a relevant body-composition outcome; to enter a calculation, the outcome, the scanner type, the time window, the population and a measure of uncertainty all had to match.

Host 1What do they do when a trial reports the right outcome but leaves out the uncertainty?

Host 2They only reconstructed it when the arithmetic was exact, for example turning a ninety-five percent confidence interval back into a standard error. They refused to guess how strongly a person's before and after values are correlated, and they logged twenty requests to investigators and sponsors for missing data.

Host 1Did any of those requests pay off?

Host 2Two did. The authors of a trial called BELIEVE supplied the exact percentage-change contrasts for two semaglutide doses, and a Japanese tirzepatide trial, SURPASS J-mono, supplied bioimpedance data.

Host 1The paper also mentions artificial intelligence in its methods; what did it do there?

Host 2The authors disclose that supervised AI tools helped organize the literature, refine search terms, support preliminary screening, structure tables, check calculations and edit language. They state that no AI-generated eligibility decision, number, bias judgment or conclusion was accepted without checking it against the source.

Host 1And how did they judge whether each trial could be trusted?

Host 2With the Cochrane Risk of Bias two tool, applied to the lean-mass outcome in each trial rather than the trial as a whole. It looks at the randomization, deviations from the planned treatment, missing data, how the outcome was measured, and whether the reported result was selected.

A word from our sponsor

Host 1This episode is brought to you by Tupperware, the food container, now a rival to the vacuum flask that sponsored us before.

Host 2Where a flask only keeps soup warm, Tupperware preserves the lean body mass of your leftovers.

Host 1In a seventy-two week substudy, a lasagne kept in Tupperware lost all its fat and none of its pasta, confirmed by dual-energy X-ray.

Host 2Tupperware has not paid for this, was not consulted, and remains unaware of the lasagne.

Host 1Each lid ships with a matching placebo lid, and they share a control group, so never open both at once.

Host 2Mine has done resistance training on the bottom shelf since March, and this morning it carried me to work.

Host 1Tupperware. Your leftovers will outlive you, and lift their own lids.

The main results

Host 1Back to the trials: what did they find for the main outcome, the percentage of lean mass lost?

Host 2Only three comparisons had both an effect and a usable measure of uncertainty. In SURMOUNT-one, tirzepatide, with its three doses pooled, lowered lean mass by eight point three percentage points more than placebo at seventy-two weeks, with a confidence interval from minus ten point six to minus six point one.

Host 1Eight percentage points more; what were the two groups doing on their own?

Host 2Lean mass fell by ten point nine percent on tirzepatide and by two point six percent on placebo, so the difference is about eight points. In BELIEVE, at forty-eight weeks, semaglutide one milligram gave three point eight one points more loss than placebo, and semaglutide two point four milligrams gave six point oh six points more.

Host 1Which trials did not make it in, even though they measured the right thing?

Host 2The well-known semaglutide trial STEP one had a DXA substudy, but the placebo group's percentage change could not be recovered. The orforglipron trial ATTAIN-one showed a difference of about four point eight points, but no exact uncertainty was available, and the mazdutide trials and a liraglutide trial lacked the registered percentage outcome.

Host 1So could they then run the network and say which drug costs the most lean mass?

Host 2No, and the reason is subtle. The two BELIEVE comparisons share the same placebo group, so they are statistically linked, and the trial did not report how strongly the two contrasts are correlated.

Host 1Why would sharing one placebo group cause a problem in the maths?

Host 2If you treat the two comparisons as independent, you count the same placebo patients twice, and the result looks more certain than it is. The authors refused to make up a correlation, so instead of one network they ran two separate comparisons, each pairing one BELIEVE dose with the SURMOUNT-one result.

Host 1And what did those two indirect comparisons show?

Host 2Tirzepatide against semaglutide one milligram came out at minus four point four nine percentage points, with an interval from minus eight point oh five to minus zero point nine three. Against semaglutide two point four milligrams it was minus two point two four, with an interval from minus five point seven eight to plus one point three, which includes zero.

Host 1So if I have this right, tirzepatide looks like it costs somewhat more lean mass than semaglutide in percentage terms?

Host 2That is how the point estimates lean, but the authors grade both comparisons very low confidence and explicitly refuse to call one drug better or worse. They come from two different trials with different durations and different estimands, so the difference could just as well come from the trials as from the drugs.

Kilograms lost

Host 1What about the simpler number, kilograms of lean mass lost?

Host 2Three core trials qualified, all against placebo, between forty-eight and eighty weeks. Semaglutide two point four milligrams in STEP one gave three point four three kilograms more lean-mass loss, pooled tirzepatide in SURMOUNT-one four point four oh kilograms, and mazdutide nine milligrams in GLORY-two five point nine kilograms.

Host 1Are those differences between the drugs real, or could they be noise?

Host 2All three are clearly different from placebo, but comparing the drugs with each other is shakier. Indirectly, mazdutide against tirzepatide was minus one point five kilograms with an interval that crosses zero, semaglutide against tirzepatide was plus zero point nine seven with an interval crossing zero, and only mazdutide against semaglutide, minus two point four seven, excluded zero.

Host 1Should we read that last one as mazdutide taking the most lean mass?

Host 2The authors say no. The GLORY-two figure was calculated from raw arm-level changes in a selected DXA subgroup of Chinese adults, not from the adjusted estimate the other trials used, and each drug rests on a single trial, so they treat it as a hypothesis rather than a ranking.

Host 1Was there a fourth drug in that kilogram picture?

Host 2Liraglutide three milligrams in a trial called S-LiTE showed two point nine kilograms more loss at fifty-two weeks, but it only counted as sensitivity evidence. Its participants had already done an eight-week low-calorie diet before randomization, and the analysis only included people who stuck to the protocol.

Host 1Is losing three to six kilos of lean mass a lot, for someone who loses a lot of weight overall?

Host 2The paper deliberately does not translate it into harm. It points out that a drug that removes more total weight and more fat can remove more lean mass in kilograms without necessarily weakening anyone, and that there is no agreed minimum difference in lean mass that matters clinically.

Host 1What about the shorter trials and the other kinds of scanner?

Host 2Two liraglutide trials at twenty-four to forty-eight weeks pooled to minus two point six two kilograms, though the authors say two trials are too few to judge consistency. Bioimpedance, which estimates body composition from a small electrical current, was kept separate, and in SURPASS J-mono tirzepatide against dulaglutide gave between minus zero point nine and minus two kilograms depending on dose.

Host 1How did the trials score on the risk-of-bias checks?

Host 2Of the fifty-two trial families, forty-five could be judged: forty had some concerns, five were at high risk of bias, and none is described as low risk overall. The other seven could not be assessed because the lean-mass result was not reported well enough.

Host 1What kept dragging the trials down?

Host 2Missing outcome data, the fact that only some participants were picked for body scans, open-label or partly unmasked treatment in some trials, and lean mass being reported as a secondary or selective result. The confidence grades follow from that: low for semaglutide and tirzepatide against placebo in kilograms, very low for mazdutide and for every drug-to-drug comparison.

Why it matters

Host 1So what is the principle here, the thing that stays true even if the exact numbers move?

Host 2That weight loss on these drugs comes with measurable loss of lean tissue in randomized trials, and that the current trials cannot tell you which drug spares it best. The paper is as much about the second point as the first: it shows how little of a large literature survives once you insist that like be compared with like.

Host 1What should an ordinary person on one of these drugs take from that?

Host 2The review does not give individual advice, and it says clearly that a lower lean-mass number is not proof of harmful muscle loss. It does say the meaning likely depends on age, frailty, protein intake, resistance exercise, other illnesses and physical reserve, and that most trials reported those factors inconsistently.

Host 1So it would matter more for, say, a frail eighty-year-old than for a fit forty-year-old?

Host 2That is the direction the authors point to, but the honest answer is that the trials they reviewed do not let them quantify it. Across the wider field, that is why many clinicians already encourage strength training and enough protein during large weight loss, though this paper did not test that.

Host 1And for the doctors and researchers, what changes?

Host 2Mostly how trials should report. The authors ask for arm-level before and after values in kilograms and percent, exact between-group differences with uncertainty, the correlation between contrasts that share a control group, and the reasons people were missing from scans.

Honest caveats

Host 1What are the weakest points of the review itself?

Host 2The search is the biggest one: only PubMed and ClinicalTrials.gov were screened record by record. The authors compared themselves with broader reviews, one of which found the same tirzepatide figure of minus four point four kilograms, and argue that missed trials would more likely change precision and the shape of the network than the overall direction.

Host 1Is there anything about the screening itself that a reader should know?

Host 2Yes, they state that not every one of the six thousand two hundred and seventy-six records was screened twice by independent reviewers, which is the usual standard. Key numbers were double-checked, but the full duplicate screening pass is not documented.

Host 1And the substudies, why do they keep coming up?

Host 2Because DXA scans were only done at some sites or in some participants, and only people with both a starting and a final scan could be counted. If those who dropped out or skipped scans differ in how they responded, the lean-mass figures could be biased in a direction nobody can check.

Host 1What can we not conclude from this paper at all?

Host 2We cannot conclude that these drugs cause harmful muscle loss, weakness or sarcopenia, because none of that was measured in the pooled outcomes. And we cannot rank the drugs, because there is no head-to-head trial, every drug rests on one trial, and the network has no closed loops to check itself.

What to watch next

Host 1What would actually settle the comparison between the drugs?

Host 2A direct head-to-head trial with body scans. The authors are monitoring one recruiting study, COMPARE-AT, which did not yet contribute any data.

Host 1And beyond which drug, what else would people want to see measured?

Host 2Strength and physical performance, the things that matter in daily life, alongside total and arm-and-leg lean mass. The authors say those functional outcomes need to appear in trials much more consistently before anyone can describe the drugs' effect on muscle health with confidence.

Sign-off

Host 1So in one sentence, where does this leave us?

Host 2The drugs take some lean mass along with the fat, roughly three to six kilograms more than placebo over a year or more, and nobody can yet say which drug takes the least or whether it matters for strength. It is a modest, careful answer, and the care is the point.

Host 1Every weekend this same feed carries the roundup, the week's new papers in one short episode, for those who like their lean mass already sorted.

More from this week

Get the next one

A roundup of new papers every week and a deep dive into one paper most nights. Free, no account needed: paste this address into any podcast app that can add a show by its address, or find “Hastrup Review of Science” by name in an app that searches Podcast Index, like Fountain or Podverse.

https://reviewofscience.org/podcast.xml

Only want the weekly roundup, or only the deep dives? There are two smaller feeds: see Listen and subscribe.