Some results are too good to believe. This is one. The finding is not wrong because it is large. It is doubtful because of how large it is, and where it came from.
What the study did
Adults with hidradenitis suppurativa and either type 2 diabetes or obesity were pulled from a US record network. [1] Those on a GLP-1 drug or tirzepatide were matched one-to-one against those on other metabolic medicines. Outcomes were followed two years.
The skin disease defines who is in the study. It is not an outcome. Nothing here measured whether anyone’s hidradenitis got better.
The number that stops you
All-cause mortality came out at 0.24, interval 0.145 to 0.395. Three quarters of deaths, gone.
Compare that to a randomized trial. SELECT moved all-cause death by roughly a fifth over four years, and that is the largest properly randomized read we have — the same body of evidence that a Cochrane review of eighteen trials graded. A four-fold larger effect, in two years, in unrandomized records, is not a discovery. It is what we should expect when the people who get a drug differ from the people who do not.
The rest of the table
Major cardiac events came out at 0.61. Stroke and heart failure both fell. Depression came out at 0.78, substance use disorder at 0.58, suicidal ideation at 0.38.
These cannot be separated from the mortality figure. They are the same people, matched the same way, over the same two years. If the design can produce 0.24 for death, it can produce 0.38 for suicidal ideation, and nobody gets to keep the reassuring rows and discard the implausible one — the problem every large matched-record study carries.
Three things went up: hyperlipidemia at 1.25, fatty liver disease at 1.37, nausea and vomiting at 1.31. Those come with the same caveat and deserve the same weight.
Why healthy people get prescriptions
Doctors do not start an expensive injectable in somebody who is dying. They start it in people well enough to benefit, who show up, and who can pay.
Everyone in the comparison group was on some other metabolic drug, which helps. It does not fix it. Matching handles what got recorded, and how sick somebody looks in clinic is mostly not recorded — the same gap behind who ends up meeting an indication.
What to do with it
Read it as a hypothesis about a population nobody studies much. That is a real contribution.
Do not read 0.24 as a mortality benefit. A ratio needs its absolute counts before it means anything, and the counts are usually smaller than the ratio sounds. An effect this size showing up only outside a trial is the oldest warning sign in observational medicine, and the reverse error is just as common.