This is a study with an unusual virtue and one blind spot, and they are the same feature seen from two sides.
What it found
Adults with metabolic dysfunction-associated steatohepatitis were pulled from a large real-world health database and split by whether they had a documented GLP-1 prescription. [1] Propensity matching on demographics, coexisting conditions and concomitant drugs left 168,733 patients on each side, out of 168,740 exposed.
Three outcomes were reported, all as absolute risk differences. Ascites was lower among GLP-1 users by 0.013, a 1.3-point difference, 95% CI 0.014 to 0.012 lower, p < 0.001. All-cause death was lower by 0.022, or 2.2 points, 95% CI 0.023 to 0.021 lower, p < 0.001. Liver cancer went the other way: higher among users, p = 0.025, risk difference 0.000, 95% CI 0.000 to 0.001.
The good part
Reporting all three on one scale is better practice than most papers manage. A relative figure would have made the cancer result sound alarming and the mortality result sound heroic, and both would have been the same numbers dressed differently — which is how a statistical result usually gets oversold.
The authors then looked at the cancer signal, saw a difference of zero at three decimal places, and declined to treat a p-value of 0.025 as a clinical finding. In a matched sample this size, tiny differences clear significance easily. That judgment is correct.
Nobody said over how long
The indexed abstract gives no follow-up duration. A risk difference without a window is not a rate. Two point two points of mortality across one year and the same figure across eight years describe very different drugs, and a reader cannot tell which this is.
Prescribed is not randomized
Everyone here chose, or was offered, their treatment. Being prescribed a GLP-1 for MASH and filling it repeatedly selects for people healthy enough to be considered, insured enough to afford it, and engaged enough to keep going.
Matching on codes cannot see any of that. It is the standing weakness of every database comparison on this site, and it pushes in the direction that flatters the treated group — the same shape as a conclusion stated more confidently than its design allows.
What to do with it
Read it as reassurance about liver safety rather than as a reason to buy. The liver-cancer result is the kind of signal that would matter if it were larger, and it is not. The survival result points the right way and is small.
If liver disease is the reason someone is considering one of these drugs, the relevant page is the trial evidence in that population, not a claims cohort — and which events a study counts decides most of what its numbers can mean.