Skip to content
This GLP
← Research
Evidence

Body composition: measured by the scale in your bathroom

A 9.8% median weight reduction, split 7.5% fat and 2.3% muscle — measured by a household bioimpedance scale, in people who weigh themselves for years.

Dana Sullivan7 min read
Median weight change at 10 to 14 months: −9.8%fat 7.5%muscle 2.3%The two shares sum to the total, so both are shares of body weight.2.3 of 9.8 is 23% of the weight lost.What measured itA connected bathroom scale, reading bioimpedance through the legs.Not DXA. Everybody in both groups owned one and used it for years.

A study of body composition is only as good as the thing doing the composing. Here that thing is a bathroom scale that sends a small current through your legs and infers what is between them.

That is a real measurement and a useful one. It is not the measurement the phrase “body composition” usually implies, and the difference matters more than the percentages do. [1]

What the numbers say

Over months 10 to 14, the GLP-1 group showed a median weight reduction of 9.8% against matched non-users. The paper attributes 7.5% to fat mass and 2.3% to muscle mass.

Those two add to 9.8 exactly, which tells you both are expressed as shares of body weight rather than of each tissue. Work it through and muscle accounts for roughly a quarter of everything lost — close to the range the imaging studies keep landing in, which is at least a point in the method’s favor.

Systolic blood pressure fell 2.5 mm Hg, interval 1.5 to 2.9. Diastolic fell 1.5, interval 1.2 to 1.9. Those came from 148 exposed people rather than 396.

Who is in a study made of scale data

People who bought a connected scale, linked it to an app, weighed themselves regularly for years, and then answered a survey about their medications. Both groups.

That is a population already engaged with measuring itself. Matching happens inside it, so it controls for differences between the people who are there and tells you nothing about the people who are not — the same structural limit as a study whose sample defines its own conclusion, arriving by a different route.

The prediction that may be a floor

The strongest predictor of what the authors call weight-loss quality was the muscle-to-fat ratio somebody started with. People with less muscle relative to fat lost more fat and kept more muscle.

Which is worth sitting with. Somebody carrying little muscle has little muscle available to lose, so a measure of how much muscle came off will look favorable for them almost by construction. Whether that is a clinical insight or an artifact of the ratio is not something this design can separate, and what muscle loss does to strength is a different question again.

What it is good for

Scale data has one genuine advantage over a trial: it keeps going. Nobody schedules a DXA scan for a Tuesday in month thirteen, and daily home weights across years are a record no clinic produces.

So read this as a long, noisy, densely sampled look at a self-selected group, and not as a body-composition study. It agrees roughly with better instruments on the split, which is reassuring, and it says nothing about what comes back afterward.

Frequently asked

How much of the weight lost was muscle?
The paper reports 7.5% fat and 2.3% muscle against a 9.8% total, so muscle is about a quarter of what came off. That figure follows from the published numbers rather than being stated.
Is a bathroom scale accurate for body composition?
It estimates rather than measures. Bioimpedance infers tissue from electrical resistance and shifts with hydration and time of day, so it is not comparable to a DXA scan.
Why are the confidence intervals so narrow?
Because the quantity being estimated is a distributional one computed across a very large number of daily weigh-ins. The precision describes the arithmetic, not any individual result.
Did blood pressure improve?
Systolic fell 2.5 mm Hg and diastolic 1.5 mm Hg, sustained over months 10 to 14. That part of the analysis had 148 exposed people rather than 396.

Sources

  1. [1] Chabassier A, et al. (2026). Real-World Body Composition and Blood Pressure Changes Following Glucagon-Like Peptide 1 Receptor Agonist Initiation: A Causal Inference Study of Connected Device Data Mayo Clinic Proceedings: Digital Health. PMID 42746268

Where to get it

Best GLP-1 injections

Every injectable seller we can verify, with the price each one publishes and an honest read of what the trials measured.

Compare providers →

More in Evidence