Most organizations funding GLP-1 therapy right now believe their cost problem is the people taking the drug. Yet the published data points somewhere else entirely.
The expensive population is the one that starts, absorbs the most expensive part of the therapy, and stops without a plan. Weight comes back. Metabolic gains reverse. And the person is left with a sense of disappointment and lost control that makes the next attempt harder than the first. That is the opposite of what the spend was authorized to produce. One of the most common reasons people stop is the coverage design itself.
That is an uncomfortable finding, because it means the benefit is generating a meaningful share of its own waste.
Half to two-thirds stop inside a year
In a cohort of 125,474 US adults with overweight or obesity who started a GLP-1, 64.8% of those without type 2 diabetes had discontinued within one year, compared with 46.5% of those with diabetes. Within that same year, 36.3% of the group without diabetes started again (Rodriguez et al., JAMA Network Open, 2025).
A separate health-system cohort of 7,881 adults without diabetes found a similar shape: 20.4% stopped within three months, another 32.0% between three and twelve months, and 47.6% did not stop at all (Gasoyan et al., Obesity, 2025).
Those are not outlier populations or bad programs. That is roughly what the therapy looks like in practice, most places, and it is far above what the phase 3 trials reported.
What a stop actually costs, which is not obvious from a claims line
Here is where clinical reality and the finance view diverge, and it is the part that tends to get missed in a benefits meeting.
The beginning of GLP-1 therapy is the expensive part in every sense. Titration is where the side effects cluster, where the visits are most frequent, and where the clinical work is heaviest. The benefit, meanwhile, is back-loaded. It accrues over months, at maintenance dose, after the difficult stretch is behind the patient.
So a member who starts and stops at month three has consumed the early expensive phase and collected almost none of the return. In the health-system cohort above, that shows up clearly. At one year, people who stopped early had lost a mean of 3.6% of body weight. Those who stopped later, 6.8%. Those who stayed on therapy, 11.9%.
The pattern in glycemic terms is starker. Among members who began with prediabetes, 67.9% of those who stayed on therapy converted to normal glucose at one year. Among those who stopped early, 33.1%.
Note: a restart is not just resuming where someone left off. Clinically, it is a reversal of whatever metabolic benefit had accrued, plus weight regain, plus a return to titration: the same slow ramp, the same side-effect window, the same intensity of support, sometimes more. With more than a third of the group without diabetes restarting inside twelve months, an organization with no long-term plan is not paying for one course of therapy per member. It is paying for the front of the course, repeatedly.
One caution about how to read this, because it is regularly overstated. The STEP 1 trial extension followed people for a year after the drug was withdrawn. Peak loss had been 17.3% of body weight. Twelve months later the same group sat 5.6% below where they started (Wilding et al., Diabetes Obesity and Metabolism, 2022). That is trial data, in people who reached and held a full dose. The real-world health-system cohort found something different: weight trajectories stayed relatively stable after people stopped.
Both can be true, and reconciling them is the point of this piece. Trial participants had reached full dose and had something substantial to lose. Most people who stop in practice never got that far, either because the return did not justify the effort and cost, or because nobody ever built them a long-term plan. Those are different situations, and only one of them is a problem with the medication.
The reason they stop is not okay
When the same investigators went back through the records to find out why people discontinued, the leading reason was not side effects and was not disappointing results.
It was cost and insurance. Nearly half, 47.6%, stopped over coverage or out-of-pocket issues. Intolerance accounted for 14.6%, and supply shortages 11.8% (Gasoyan et al., Obesity, 2025).
Think about that from the plan's side of the table. The design decides who starts. The design, more than the medicine, then decides who stops. And the members who stop are the ones on whom the financial investment produced the least. A benefit built to control cost at the front door is, through the same mechanism, manufacturing the least efficient version of its own spend.
Which is why the obvious lever is the weak one
Faced with a line item growing faster than the rest of the plan, the reflex is to tighten eligibility. Raise the BMI threshold, add criteria, slow the intake.
It is the weakest available move, and it is the most expensive in goodwill. It touches only new starts, leaves the entire existing population untouched, does nothing about restarts, and it is the change employees notice and resent. Organizations that have already used it once have paid that price and generally cannot spend it twice.
What almost nobody has built is the other end.
Go look at the plan language, or the practice protocol, and you will usually find a well-specified front door and then nothing for the long term. No definition of a good response. No criteria for continuation. No written pathway for what happens when someone reaches their goal. No maintenance protocol, no re-authorization point, no standing review.
That single absence is what makes the spend impossible to forecast. It is not the price of the drug, which is knowable, but the fact that the program has no defined shape after admission.
The same gap, in three different buildings
For an employer, it shows up as a pharmacy line growing faster than the plan and a benefits team with no defensible answer for leadership.
For a health system, it shows up as a clinic that fills, then never turns over, because there is no point at which a stable patient moves into maintenance and frees the slot.
For a private practice adding this as a service line, it shows up personally, as a follow-up burden that grows every month and lands on the physician who started it.
Same design gap with a different presenting symptom. In each case, the missing piece is a back end, and in each case, building one is not a restriction. Announced correctly, it is better care: a patient who is appropriately educated, reaches a goal and is handed a written maintenance plan has been given something, not had something taken away. That framing is available because it happens to be true, which is what makes it durable under challenge.
The programs that are successful, across all three settings, are the ones that decided early what happens after the beginning.
Four questions worth asking this quarter
None of these requires a consultant, and the answers will tell you most of what you need to know.
What share of our members started and stopped within twelve months? And can your reporting distinguish a discontinuation from someone who simply left the plan? If it cannot, everything downstream of that number is guesswork.
Of those who stopped, how many started again? That figure tells you what you are paying for titration twice.
What is written down about what happens when someone reaches their goal? Not what people assume happens, but what is documented.
When did anyone last look at this? If the answer is measured in quarters, the program is being managed on a lag longer than the therapy takes to come apart, and the financial risk grows in that gap.
An organization that can answer all four has a program. An organization that cannot has a front door, a growing bill, and a conversation coming that it is not ready for.