Why GLP-1 Discontinuation Percentages Need Their Study Context

Headlines say most patients stop GLP-1 medicines within a year. The number is real — but it counts something more specific, and more interesting, than "people quit." Here is how to read a discontinuation percentage the way the researchers built it.

"Nearly two-thirds of patients stop within a year." That figure, from a large US study published in January 2025, travelled fast — cited as evidence that the medicines fail, that patients are intolerant, or that the whole category is overhyped. Each reading assumes the percentage measures something it does not. Discontinuation statistics are constructed objects: researchers choose a denominator, a definition and a window, and those choices decide what the number can honestly say.

One percentage, four questions

Before quoting any discontinuation figure, four questions decode it. First, the denominator: who entered the study — everyone prescribed, everyone who received a first fill, or only people still engaged with the health system? Second, the definition: does "discontinuation" mean a gap between prescription refills, a patient-reported decision, or a doctor's recorded stop? Third, the follow-up: over what period, with what censoring when people leave the data? Fourth, the aftermath: did anyone check what happened next — restarts, switches, or nothing? The three studies below answer these questions differently, which is why their numbers cannot simply be averaged or ranked.

What the 64.8% figure actually counts

The widely quoted number comes from Rodriguez and colleagues, who analysed electronic health records from a collective of US health care systems and identified 125,474 adults with overweight or obesity newly started on liraglutide, injectable semaglutide or tirzepatide between January 2018 and December 2023.^[1] One-year discontinuation was 64.8% (95% CI, 64.4%–65.2%) among patients without type 2 diabetes and 46.5% (95% CI, 46.2%–46.9%) among those with it.^[1]

The definition behind that figure is worth quoting precisely. In the study's methods, primary discontinuation was defined as the first date a patient was 60 days or more without any GLP-1 medicine on hand, computed from fill dates and days supplied, with leftover "carryover" medication from earlier fills accounted for — and later fills did not need to match the original brand.^[2] A 90-day version of the rule produced similar results in sensitivity analysis.^[2] So the statistic counts a refill-gap pattern in pharmacy records, not a surveyed intention to quit. The authors themselves flag the trade-offs: discontinuation definitions in the literature vary, some patients are likely misclassified either way, medication rationing — skipping doses to stretch a supply — was not counted, and drug shortages were not modelled.^[2] A recorded supply gap does not, by itself, explain the patient's intention or the role of shortages.

Discontinuation is not one decision

A second study shows why "stopped" must be unpacked. Abdel-Bary and colleagues followed 853 patients prescribed a GLP-1 medicine for class 2–3 obesity in one Midwestern health system between January 2020 and November 2022.^[3] Of those prescribed, 24.7% never received the drug at all, and 35.6% discontinued within two years. Among the discontinuers, the most common recorded reasons were loss to follow-up with the original prescriber (26.6%), side effects (23.7%) and cessation of insurance coverage (21.4%).^[3]

Note the shifting denominators: prescribed, received, discontinued, discontinued-for-a-reason. Note also what the reasons say. Side effects are real — in the Rodriguez cohort, incident moderate or severe gastrointestinal adverse events were associated with a higher hazard of discontinuation (hazard ratios 1.38 with type 2 diabetes, 1.19 without)^[1] — but they are one factor among income, insurance, follow-up access and weight change, not the hidden meaning of every stop. The Midwest cohort also measured outcomes: patients who stopped had two-year weight loss statistically similar to those who never started, while those who stayed on treatment averaged 9.22% total weight loss^[3] — a comparison between groups that stayed observable in one system, not a controlled test of what stopping does to an individual.

Stopping is not always the end

Both large datasets show movement after a stop. Among 41,792 Rodriguez-cohort patients who discontinued and had a weight measurement available, one-year reinitiation was 36.3% without type 2 diabetes and 47.3% with it, and regained weight was associated with a higher likelihood of restarting.^[1] A 2026 study by Gasoyan and colleagues examined 7,938 adults in an Ohio–Florida health system who had discontinued semaglutide or tirzepatide within 3–12 months of starting, using records from January 2021 to June 2025.^[4] In the year after stopping, 19.6% restarted the same medication and 35.2% received an alternative obesity treatment — another medication for 27.4%, a lifestyle-modification visit for 13.7%, bariatric surgery for 0.6%.^[4]

Study	Denominator and rule	Headline number
Rodriguez 2025, US records	New users, BMI 27+; 60-day refill gap	64.8% (no diabetes) vs 46.5% (diabetes) stopped in year 1
Abdel-Bary 2025, one Midwest system	Class 2–3 obesity prescriptions	24.7% never started; 35.6% stopped by year 2
Gasoyan 2026, Ohio–Florida	Stopped within 3–12 months	19.6% restarted; 35.2% tried an alternative

Average weight change in the post-discontinuation year was small — +0.5% for those treated for obesity, −1.3% for those treated for type 2 diabetes — but the authors stress considerable individual-level variability.^[4] That variability is the honest part: an average near zero can hide regain, stability and further loss in different people. And because these are patients who stopped within a defined window and remained observable, the numbers describe a selected group; they cannot support a causal claim that stopping is harmless, or beneficial, for anyone in particular.

Persistence, adherence and efficacy are different words

Three terms that headlines swap freely: persistence is whether a person remains on therapy at all; adherence is whether taken as directed while on it; efficacy is what the medicine does when taken. A patient can be persistent but under-dosed by rationing, fully adherent until an insurance change, or a discontinuer who restarts two months later. Clinical-trial "stopping" adds another wrinkle: trial discontinuation happens inside a protocol with defined visit schedules, while real-world discontinuation mixes biology, logistics and affordability. None of the studies reviewed here settles why any individual stopped, and none offers a basis for anyone to stop or restart on their own — those are decisions for the prescriber managing the treatment.

Reading the next discontinuation headline

A proposed checklist: find the denominator and the refill-gap rule; check whether "never received" is included or separate; look for reasons, restarts and switches; and ask whether the follow-up window matches the claim. For this article, the full text of the Rodriguez methods was reviewed in selected sections; the Rodriguez abstract and the abstracts of the other two studies were also reviewed, so details beyond those abstracts were not assessed. The evidence concerns prescribed medicines recorded in health systems. It does not establish clinical suitability for research-use materials, which should not be self-administered. The careful takeaway is narrower than the headlines: discontinuation percentages are measurement choices made visible, and they reward exactly the reading effort that a single number discourages.