Design: Prospective diagnostic study within a phase 2b RCT, with external validation
Population: 160 adults with biopsy-proven MASH F2–F3 (validation n = 48)
Primary outcome: Accuracy of ≥30% LSM decline for fibrosis regression without worsening MASH
Effect: AUC 0.68 (95% CI 0.58–0.77); validation AUC 0.62; adjusted OR 4.23

AASLD guidance says a decline in liver stiffness on vibration-controlled transient elastography (VCTE) can serve as a non-invasive marker of treatment response in MASH. This study tested the commonly used threshold of a 30% relative decline against paired biopsies in 160 adults with biopsy-proven MASH and F2–F3 fibrosis from a phase 2b trial of pegozafermin, with a separate validation cohort of 48 from the US and Singapore.

A 30% or greater fall in liver stiffness detected fibrosis regression without worsening MASH with an AUC of 0.68 (95% CI 0.58–0.77), and 0.62 in the validation cohort. It was independently associated with regression after adjustment (adjusted OR 4.23, 1.79–10.38).

So a large stiffness fall does point in the right direction, but it misclassifies many patients. As drugs for MASH reach practice, clinicians will want a non-invasive way to judge response. This study says liver stiffness alone is not yet good enough.

Treat a ≥30% fall in liver stiffness as supportive of response, not proof of fibrosis regression.
Do not stop or switch MASH therapy on a single VCTE result; combine with trends in ALT, weight and other markers.
Measure liver stiffness under consistent conditions — fasting, same probe, same operator where possible.
Remember that falls in liver inflammation and steatosis can lower stiffness without fibrosis change.

Why it matters

It tempers reliance on elastography as the response test just as MASH drugs arrive.

The statistics, in plain English

An AUC of 0.68 means that if you took one patient whose fibrosis regressed and one whose did not, the stiffness test would rank them correctly about two times in three — modest. An adjusted odds ratio of 4.23 shows a strong association, but a strong association does not make a good individual-level test.