Observational studies can show how treatment and outcomes are associated in ordinary care. They can also reveal why a clinical trial average is not the experience of everyone who starts a medicine. Their principal limitation is that the treatment groups were formed by clinical decisions and circumstances rather than random assignment.
This article is written for Serum Research, a commercial supplier. Its research analysis neither authenticates nor establishes the clinical suitability of the supplier’s catalog products.
A comparison built from health records
Rodriguez and colleagues used linked electronic health records and dispensing data for adults with overweight or obesity starting diabetes-labeled tirzepatide or semaglutide. Propensity matching produced a cohort of 18,386. The estimated difference in on-treatment weight change at 12 months favored tirzepatide by 6.9 percentage points, with a 95% confidence interval of 5.8 to 7.9 points in that direction. JAMA Internal Medicine primary report, Abstract.
Matching addresses recorded differences selected for the analysis. It does not guarantee equality in unrecorded factors, such as the reason a clinician preferred one treatment or details of support outside the health record. The appropriate wording is an association in this cohort, even when the comparison is carefully designed and statistically precise. Precision describes uncertainty under the analysis; it cannot eliminate every source of bias.
The meaning of on treatment
Follow-up ended because of discontinuation for 55.9% of tirzepatide recipients and 52.5% of semaglutide recipients. The analysis of weight at a time point focused on those still at risk under its follow-up rules. Primary report, Results and Statistical Analysis.
This changes the question being answered. A result among people remaining on treatment is not automatically the expected outcome for every new starter. If people with poorer responses or greater burdens stop earlier, later observations can describe a selected group. That is a possible source of distortion to examine, not proof that it explains the observed treatment difference. A useful sensitivity analysis asks whether the conclusion changes when available outcomes after stopping are included.
Missing weight measurements create another interpretive issue. A recorded clinic weight depends on a visit and documentation as well as the person’s underlying weight. Statistical handling of missing observations therefore rests on assumptions about people whose measurements are unavailable. A reader should ask whether those assumptions are plausible for the actual health care setting before generalizing the estimate.
Similar recorded harms are not identical tolerability
The study reported similar gastrointestinal adverse event rates, but its safety methods excluded mild nausea and vomiting because health records were expected to capture them inconsistently. Primary report, Safety Outcomes.
That finding cannot establish equal day-to-day tolerability. An event too mild to generate a coded encounter may still matter to a patient, while a serious coded event is measured through a different process. A null comparison also should not be rewritten as proof that the treatments have identical risk. Outcome definitions determine what the comparison can detect.
Use observational and randomized evidence together
The value of this cohort is its defined question about recorded care, persistence, and on-treatment outcomes. It complements randomized research but does not become randomized through matching. The existing tirzepatide overview and trial emulation discussion are relevant navigation points. This article does not claim to audit every later observational paper or infer a population success rate from selected records.
Published by Serum Research. Sources checked 2026-10-01. How we review evidence · Send a correction.