The Line Is Drawn by Surgery

Does losing more weight repair more liver damage? A meta-regression across six randomized trials found that study arms with greater average weight loss had higher rates of fatty liver disease resolving.^[1] The authors then say plainly that this cannot establish cause — and that the relationship is largely driven by a handful of bariatric surgery arms.

What was compared

Six randomized trials covering 1,379 people, testing tirzepatide, semaglutide, sleeve gastrectomy and gastric bypass against placebo or standard care, with liver biopsy outcomes: resolution of the disease, and improvement in fibrosis.

All four interventions beat their comparators for disease resolution. Beyond that, the reviewers note the network was weakly connected and largely placebo-anchored — almost every trial compared its drug against placebo rather than against another drug — so indirect comparisons between the treatments were imprecise.

The regression, and what it is a regression on

A worked example makes it concrete. Suppose surgical arms average 30% weight loss and 60% disease resolution, while drug arms average 12% and 30%. Plotting those four numbers gives a convincing line. But it is entirely possible that within each arm, the individuals who lost the most were not the ones whose livers improved most. Group-level data cannot tell you.

A line through four group averages is not a dose-response relationship.

And the line is being pulled by surgery

The paper states that the associations were strongly influenced by a small number of high-weight-loss surgical arms. Those arms sit far to the right of everything else on the weight axis, and points at the extreme of a range exert the most leverage on a fitted line.

Which means most of the apparent relationship is the gap between drug arms and surgical arms, not a gradient within either. And bariatric surgery differs from a drug in a great deal besides the amount of weight it removes — how fast the weight comes off, what happens to gut hormones, how eating changes permanently, who is selected for it.

How this sits beside the heart evidence

There is an instructive contrast here. For cardiovascular outcomes, a prespecified analysis of 17,604 individuals found the benefit largely independent of weight loss — no linear relationship between weight lost and later events, and only about a third mediated through waist circumference. That is in losing weight predicted more heart attacks.

For the liver, weight loss does appear to track improvement — at the group level, in an analysis whose authors call it exploratory.

Both can be true. The liver is directly loaded by fat stored in it, and removing that fat is a plausible mechanical route to improvement in a way it is not for an artery. But the strength of evidence behind those two statements is very different: 17,604 individuals analyzed prospectively against four to six arm-level averages, and the difference in what they can support is larger than the difference in what they found.

For what the liver evidence does establish, see GLP-1s and fatty liver; for a trial that measured the tests rather than the tissue, a trial about the measuring instrument.

What the authors ask for

Their conclusion is that the limited evidence base, sparse network and ecological nature of the analysis preclude causal inference, that the findings should be considered exploratory and hypothesis-generating, and that adequately powered head-to-head trials are needed.

That is a paper describing its own limits accurately. Every caveat in this article comes from the authors’ own text, which is worth saying — the problem is not the analysis but what happens to “more weight loss, more liver improvement” once it leaves the paper.