Photos vs Self-Report: GLP-1 Accuracy

Researchers at the University of Michigan and the VA Ann Arbor Healthcare System asked 322 adults who said they had used semaglutide or tirzepatide for at least three months to submit a photo of their medication, then checked what the photo actually showed against what each person had reported on a survey. The two methods agreed 82.61% of the time^[1]. The disagreements were not random noise: photo review identified 83 participants, 25.78% of the sample, as using a compounded product, not an FDA-approved branded one, and 54 of those 83, nearly two out of three, had told researchers on the survey that they were using a branded drug^[1].

What the study actually did

The study, published in Obesity Science & Practice and led by Jacqueline Katz, Ashley Gearhardt and Dina Griauzde, recruited participants through the research platform Prolific between October 2025 and January 2026^[1]. Eligibility required at least three months of once-weekly injectable semaglutide or tirzepatide use, and the prescreening filter was strict: of more than 300,000 people on the platform, only 1,209 qualified, and 322 completed the full protocol^[1]. Each participant submitted a photo of their medication during screening and separately answered a survey question asking which drug they used. Two coders, working independently of each other and of the survey answers, classified each photo as one of the four FDA-approved branded formulations, Ozempic and Wegovy (semaglutide), or Mounjaro and Zepbound (tirzepatide), or as a compounded product prepared outside that branded set^[1].

This design matters because it separates two things that a survey alone cannot: what a participant believes or is willing to report about their medication, and what the medication actually is. A photo does not remove human error entirely, since a coder can misread a label or a participant can photograph the wrong box, but it gives researchers a second, independent read on the same question instead of relying on the survey answer alone.

The numbers behind the mismatch

Agreement between the self-reported drug and the photo-coded drug was 82.61%, with a Cohen's kappa of 0.78 (95% CI 0.73 to 0.83), which the authors describe as substantial agreement^[1]. That still left roughly one participant in six with a mismatch between what they reported and what the photo showed. Among the 83 participants whose photo showed a compounded product, 54 (65.06%) had self-reported using one of the four branded drugs, and the remaining 29 (34.94%) had selected a "none of the above" option on the survey^[1]. The authors note the survey did not offer an explicit compounded-product answer choice, which may have pushed some compounded users toward the closest branded option rather than an accurate one^[1].

Branded and compounded are not interchangeable

The distinction the researchers were trying to capture matters because branded and compounded products sit in different regulatory categories. Mounjaro, Eli Lilly's tirzepatide brand, carries FDA-approved prescribing information covering manufacturing standards, dosing and a boxed warning^[2], and Wegovy, Novo Nordisk's semaglutide brand, carries its own FDA-approved label^[3]. Compounded versions are prepared under a separate legal framework, sections 503A and 503B of the Food, Drug and Cosmetic Act, which let a pharmacy or a registered outsourcing facility prepare a drug formulation without going through the FDA's standard new-drug approval process, subject to their own ingredient and labeling rules^[4]. A compounded product can differ from the branded version in concentration, formulation or inactive ingredients, and it has not been through the same trial-based approval pathway. Lumping the two together under one drug name, which is exactly what happens when a study relies on self-report alone, mixes products that are not guaranteed to behave the same way.

What this study does not show

This is a single feasibility study, not a validated diagnostic. The sample was recruited entirely from one platform, Prolific, and the aggressive prescreening that produced 322 eligible participants out of over 300,000 users means this group is not representative of everyone taking semaglutide or tirzepatide. Photo classification itself was never checked against an outside standard, such as a pharmacy dispensing record, so the study is a comparison between two imperfect methods, not a comparison against ground truth. A kappa of 0.78 means the two methods usually agree, not that either one is definitively correct when they disagree.

The authors frame their result as a call for future obesity and diabetes research to add a verification step instead of relying on a single survey question, and they note that self-report-only designs are common practice across the field, not a flaw unique to this dataset. That framing is honest about the scope: the study demonstrates that photo review is feasible and that it catches real disagreement, but it does not tell researchers how often self-report misclassification has already distorted a published finding, because most existing GLP-1RA studies were never designed to check.

Why this matters for readers of GLP-1 research

If you read a study or a news summary that reports outcomes for "semaglutide users" or "tirzepatide users" based on what participants said on a survey, this study is a reason to ask how the researchers confirmed that. A self-report-only design can quietly fold compounded and branded users into the same exposure group, and if the two products behave differently, that could blur the safety or effectiveness numbers attributed to the drug itself. The classification step is a real source of error in self-report studies, worth checking the same way you would check a trial's sample size or its funding source.

This is one reason PeptideMethods treats compounded and branded tirzepatide or semaglutide as separate topics rather than one interchangeable drug name. For the regulatory difference between the two, see the tirzepatide regulation page and the semaglutide regulation page. Neither page is medical advice; if you are trying to work out what you or someone you know is actually taking, that conversation belongs with a pharmacist or prescriber who can check the product directly.