The eating disorder score predicted GLP-1 use six months later, the BMI did not
- Prospective online cohort of community women in the United States, recruited through Prolific and surveyed twice six months apart. 431 women provided baseline data and 266 of them (61.72%) returned at follow-up. Published open access in Eating and Weight Disorders on 6 March 2026, volume 31, article 29, single author, funded from internal Penn State Abington funds, no competing interests declared.
- The baseline predictor was the global score of the seven-item Eating Disorder Examination Questionnaire, mean 3.43 (SD 1.95). The outcome was self-reported current use of a product for the purposes of weight loss at six months, coded four ways: anything at all, herbal or over-the-counter supplement, any prescription drug, and GLP-1 specifically, meaning semaglutide, tirzepatide or liraglutide.
- Each one-point rise in the baseline score raised the odds of GLP-1 use at six months to 2.00 (95% CI 1.27 to 3.15), p = 0.003, with baseline BMI and baseline use of anything for weight loss held constant. Baseline BMI in the same model did not reach significance: 1.05 (0.99 to 1.11), p = 0.11.
- The score predicted the other three outcomes as well: 1.67 (1.29 to 2.15) for any use, 1.49 (1.11 to 2.00) for over-the-counter supplements, 1.86 (1.30 to 2.68) for any prescription weight-loss drug. BMI did predict overall use, 1.06 (1.02 to 1.11), and prescription use, 1.06 (1.01 to 1.12).
- Of the 12 women taking a GLP-1 at baseline, 5 had a body mass index below 27. Use of anything for weight loss rose from 29 women to 45 over the six months, and GLP-1 use from 12 to 18. The author concludes that eating disorder screening is warranted before and during weight-loss medication, with particular attention to atypical anorexia nervosa.
A community sample that was never asked about drugs
Twelve of the 431 women in this cohort were taking a GLP-1 medication for weight loss when they filled in the baseline questionnaire, and five of those twelve had a body mass index below 27. The study reached them through Prolific and was advertised as a survey about women's health, with no mention of weight or medication, which is how it picked up people that a weight-loss clinic would never see. Participants lived in the United States, were not pregnant and read English. At baseline all 431 completed the seven-item Eating Disorder Examination Questionnaire, whose global score averaged 3.43 (SD 1.95), and reported their own height and weight. 266 of them, 61.72%, answered again six months later. Both waves put the same plain question, which of a listed set of products the respondent was currently using for the purposes of weight loss, and semaglutide, tirzepatide and liraglutide were coded together as GLP-1 use.
The score moved the odds and the weight did not
Use rose over the six months. Anything at all went from 29 women to 45, prescription products from 18 to 27, GLP-1 medications from 12 to 18. Four logistic models then asked what at baseline predicted use at follow-up, each carrying the same two controls, baseline BMI and baseline use of anything for weight loss.
The questionnaire score predicted every one of the four. A one-point rise in the global score raised the odds of using anything at six months to 1.67 (95% CI 1.29 to 2.15), of an over-the-counter supplement to 1.49 (1.11 to 2.00), of any prescription weight-loss drug to 1.86 (1.30 to 2.68), and of a GLP-1 in particular to 2.00 (1.27 to 3.15), p = 0.003.
Baseline BMI behaved differently from one model to the next. It predicted overall use, 1.06 (1.02 to 1.11), and prescription use, 1.06 (1.01 to 1.12). In the GLP-1 model it did not, 1.05 (0.99 to 1.11), p = 0.11, and the heaviest term there was simply already taking something at baseline, 7.85 (2.56 to 24.06). Exploratory models broke the score into its parts, and dietary restraint, over-evaluation of shape and weight, and body dissatisfaction each predicted GLP-1 use on their own.
Where this lands in the consultation
The usable part is small and specific. What a woman answered on seven questions at one visit carried information about what she would be taking half a year later that her body mass index, sitting in the same model, did not carry. Whatever else that score represents, it is available before the prescription is written and it takes two minutes to collect.
Rosenbaum states the clinical reading plainly: eating disorder symptoms are worth assessing before weight-loss medication starts and while it continues, with attention to presentations that do not look like anorexia because the patient's weight is normal or high. The five women under a BMI of 27 who were already on a GLP-1 are the concrete form of that sentence. What the data cannot separate is motive. The study never asked why anyone was taking anything, so a GLP-1 prescribed to reduce binge eating and a GLP-1 that supports restriction are the same entry in this dataset. Causation is not on the table either, and the paper says so.
A one-point rise in a seven-item questionnaire score doubled the odds that a woman would be taking a GLP-1 six months later, 2.00 (95% CI 1.27 to 3.15), while her body mass index, in the same model with the same controls, did not reach significance, 1.05 (0.99 to 1.11).
The GLP-1 model rests on 18 events. Only 18 of the 266 women who completed follow-up reported GLP-1 use, which is why the confidence interval around the odds ratio of 2.00 runs from 1.27 to 3.15, and a handful of cases either way would move it. All use, height and weight were self-reported, and BMI was calculated from those reports. 38.28% of the baseline sample did not return, although the author reports no significant demographic difference between those who did and did not, and Little's test indicated the missing values were missing completely at random. The sample was women only, all in the United States, drawn from an online panel, which limits generalisation. Use before baseline was not measured and so could not be controlled for. The study did not record whether a clinician recommended any product, how prescription drugs were obtained, why they were being taken, or whether the participant had ever been diagnosed with or treated for an eating disorder. The author states that causal inferences cannot be drawn, and the journal rates the work Level IV evidence.