Comparative treatment studies can appear contradictory when their populations and outcomes differ. STEER addresses a specific group without diabetes, and its observational design is essential to its interpretation. Primary study
Population and reported association
US claims data identified adults aged at least 45 with overweight or obesity and established atherosclerotic cardiovascular disease, but no diabetes. Matching produced 10,625 participants per treatment group. The reported cardiovascular composites favored semaglutide over tirzepatide. The three-component outcome used all-cause death, rather than cardiovascular death. A per-protocol analysis applied additional treatment-continuation rules and reported larger associations. Primary report
Our appraisal: matching is not a treatment lottery
Matching can align recorded baseline characteristics. It cannot ensure balance in factors that were missing, poorly measured, or absent from the database. Therefore, a difference between matched groups should be described as an association rather than an unqualified causal treatment effect.
The population boundary is important. A comparison in people without diabetes is not the same question as a trial restricted to people with type 2 diabetes. Differences in baseline risk, treatment selection, and care patterns may affect both absolute outcomes and the credibility of comparisons.
Endpoint labels also need unpacking. A composite containing all-cause death is not identical to one containing cardiovascular death. Differences in event ascertainment and component definitions can limit direct comparison even when both reports use the term MACE.
Per-protocol analyses can introduce additional selection: people who remain on a treatment may differ from those who stop. A stronger association after censoring treatment gaps does not, by itself, prove that adherence causes a larger benefit.
This study should be read alongside, not substituted for, randomized evidence and other observational cohorts. A useful synthesis asks why eligibility, follow-up, coding, treatment discontinuation, and comparator selection differ before declaring one report the final answer. This page does not recommend switching treatment on the basis of claims-data associations.
Frequently asked questions
Did this cohort include diabetes?
The specified population excluded diabetes.
Was it randomized?
No.
Were treatment groups matched?
Yes.
Does matching control unmeasured factors automatically?
No.
Did the composite use all-cause death?
Yes.
Does a stronger per-protocol result prove an adherence effect?
No.
Why can comparative studies yield different findings?
They may differ in populations, endpoints, measurement, treatment selection, and follow-up.