Observational researchers can structure an analysis to resemble a trial. That improves clarity about the question, but it does not randomly assign the treatments people received. Primary study
What the researchers compared
Five US insurance-data cohorts included benchmarking exercises and comparative analyses involving tirzepatide and semaglutide. One benchmark used sitagliptin as a proxy for a placebo comparator. In the head-to-head analysis, the estimated one-year cardiovascular composite incidence was 1.3% in both groups; the hazard ratio was 1.06, with a 95% confidence interval of 0.95–1.18. Mean on-treatment follow-up was substantially shorter than one year. Study record
Our appraisal: three labels need care
“Placebo proxy” does not mean placebo. An active medication may identify a comparison group useful for a particular analysis, but its users and biological effects are not automatically equivalent to those in a randomized placebo arm.
“One-year risk” can be an estimated time-to-event quantity even when many participants contribute less than one year. It should not be paraphrased as complete one-year observation of every patient. Discontinuation and other censoring rules affect who continues contributing data.
“No clear difference” does not establish equivalence. A confidence interval may include potentially meaningful differences in either direction. A formal equivalence or noninferiority claim requires its own justified margin and analysis framework.
Benchmarking against a randomized trial is useful evidence about an observational design's behavior in a defined setting. Agreement does not guarantee that all confounding has been removed from a different comparison, population, or period. Prescribing decisions, access, disease severity, and incomplete clinical measurement can still matter.
This report adds real-world comparative evidence, not a replacement for randomization. Its interpretation should retain the claims-based population, outcome definitions, treatment-continuation rules, and follow-up distribution. It does not establish that the two treatments are interchangeable for every patient or that its results apply to people outside the studied eligibility criteria.
Frequently asked questions
Were treatments randomly assigned?
No.
Was sitagliptin literally placebo?
No.
Did everyone contribute a complete year on treatment?
No.
Were the estimated one-year risks similar?
Yes, in the reported head-to-head comparison.
Does that establish equivalence?
No.
Can benchmarking eliminate every source of confounding?
No.
What makes the result interpretable?
Explicit eligibility, comparator, endpoint, follow-up, and censoring definitions.