This is the shape of head-to-head evidence a buyer most often gets, and the shape that is hardest to spend. The winner is a product. The loser is a category. It is a cousin of a trial that proves one drug is not worse without ever showing it is better.
What happened
Adults with inflammatory bowel disease prescribed tirzepatide or a GLP-1 between May 2022 and January 2025 were pulled from a multi-institution database and matched one to one on demographics, other conditions and IBD medications. [1] That left 3,042 per arm, mean age 54.6, 71.3% women, followed for eighteen months.
Tirzepatide patients were less likely to need intravenous steroids, at an adjusted hazard ratio of 0.81, 95% CI 0.68 to 0.94, and did better on a composite of steroids and surgery at 0.86, 95% CI 0.73 to 0.97. In ulcerative colitis alone the steroid result held at 0.82.
What did not move
Hospitalization, emergency department visits and bowel surgery were the same in both arms. Adverse outcomes were the same too.
So the finding is narrow: fewer courses of a rescue drug, with no change in the events that send somebody to a hospital. That is worth something and it is not a different disease course.
The follow-up is lopsided
Both arms ran a median of 540 days. The spread differs. The comparator’s interquartile range is 540 to 540, meaning at least three quarters sat at the full window. Tirzepatide’s runs 478 to 540.
That is what a newer drug looks like in a database. It also means the two groups were watched over different calendar stretches, and 2022 and 2025 were not the same years for supply, formulation or price.
What a buyer does with it
Not much directly. Nobody should switch molecules on a database signal about a rescue drug, and the authors say prospective work is needed.
What it is good for is reading the next comparison you meet. Check whether the thing that lost has a name, and check who assembled the matchup, because the author of a comparison shapes it. The molecules do genuinely differ on weight — on that axis the evidence is direct — and a ranking built from mixed comparators is the weakest thing in any table.