Accepted answer
Take the STEP 5 body-composition substudy, not the main paper, and note how few participants were in it. Composition substudies are typically a small imaged subset of the full randomisation, so the lean-mass fraction carries a confidence interval far wider than the weight figures beside it. Then check the method: DEXA reports lean soft tissue, which includes water, and a large early fluid shift is counted as lean loss whether or not any protein left. The fraction most of these programmes land near is around a quarter of total loss, which is also roughly what caloric restriction alone produces — the interesting question is not the fraction but whether it differs from the deficit-matched control.
Answer first: lean mass falls during any substantial weight loss, typically as twenty to thirty per cent of the total, and the levers that change that fraction are protein intake and resistance training rather than anything pharmacological.
The conventional figure is that lean tissue accounts for roughly a quarter of total weight lost in unstructured weight loss, falling towards ten to fifteen per cent with adequate protein and progressive resistance training, and rising above a third with very rapid loss and no training stimulus.
Scan-to-scan precision on lean mass is around one to two per cent for a good DXA under standardised conditions, so a change of half a kilogram is inside the noise.
The twenty-five per cent figure for lean loss in unstructured weight reduction is a long-standing result from body-composition studies across many interventions, including surgery.
The caveat is that lean mass on a scan is not muscle, and treating the two as identical over-reads the instrument.
Track absolute lean mass, not lean percentage, or the arithmetic will mislead you.
edited 28 Oct 2025 by Dr_Yusuf_Adeyemi — added the citation requested in comments