How Long People Stay on GLP-1s
Persistence rates from every published study, from 30 days through 3 years — including why reported 12-month rates range from 10% to 63% depending on the source, and what's actually improved since 2021.
Last updated: April 14, 2026
Context
The number that every GLP-1 cost model depends on
Every ROI model, every budget projection, every coverage decision depends on how long members stay on GLP-1 therapy. If persistence is high, the cost per outcome may be justifiable. If it's low, most of the pharmacy spend generates no lasting clinical benefit — members discontinue before reaching meaningful weight loss and regain whatever they lost.
The published data on GLP-1 persistence is extensive but deeply conflicting. Reported 12-month rates range from 10% to 63% depending on the source. This isn't because some studies are wrong. It's because they measure different things, in different populations, at different time periods, using different definitions of what counts as "still on therapy." A study counting every prescription attempt — including those rejected by insurers — produces a fundamentally different number than one tracking only patients who filled at least one prescription during a post-shortage era.
This page compiles every published data point, explains why the numbers differ, and identifies which estimates are most relevant for different planning scenarios. For financial modeling, cross-reference with the costs page.
The data
Every published persistence rate, side by side
Click any column header to sort. Source quality: ★★★★ = peer-reviewed journal, ★★★ = PBM report with disclosed methods, ★★ = conference presentation, ★ = industry blog. Persistence is defined as no gap exceeding the threshold shown.
| Source ▼ | Quality ▼ | N ▼ | Population ▼ | Drug(s) ▼ | 3-mo ▼ | 6-mo ▼ | 12-mo ▼ | 24-mo ▼ | 36-mo ▼ | Gap threshold ▼ |
|---|---|---|---|---|---|---|---|---|---|---|
| Gleason/Prime (JMCP 2024) | ★★★★ | 4,066 | Comm. insured, no DM | All GLP-1s | — | 46.3% | 32.3% | — | — | ≥60-day |
| Prime 3-Year Study | ★★★ | 5,780 | Comm. insured, no DM | All GLP-1s | — | — | 32.3% | ~20% | 8.1% | ≥60-day |
| Gleason/Prime Trends (JMCP 2026) | ★★★★ | 33,607 | Comm. insured, no DM | Wegovy + Zepbound | — | — | 33%→63% (2021→2024) | — | — | ≥60-day |
| Rodriguez et al. (JAMA 2025) | ★★★★ | 48,950 | EHR/multi-payer (Truveta) | Lira/sema/tirz | — | — | 35.2% | 15.6% | — | ≥60-day |
| Do et al. (JAMA 2024)¹ | ★★★★ | 20,217 | All-payer (Komodo) | All GLP-1s | 64.2% | 55.2% | 49.7% | — | — | 135-day ¹ |
| Xie et al. (JAMA 2026) | ★★★★ | 126,984 | Comm. insured (MarketScan) | Sema/lira/tirz | — | — | 24.5% | — | — | ~60-day |
| Thomsen et al. (EASD 2025) | ★★ | 77,310 | Danish national, no DM | Semaglutide | 82% | 69% | ~48% | — | — | Rx refill |
| BHI/BCBSA (2024) | ★★★ | 169,250 | Comm. insured (BCBS) | Wegovy/Saxenda | 42% (≥12 wk) | — | — | — | — | 2× Rx duration |
| IQVIA (2025)² | ★ | Undisclosed | Broad US Rx data | Obesity GLP-1s | — | — | ~10% | — | — | Includes never-fillers ² |
| Gasoyan et al. (Obesity 2025) | ★★★★ | 7,881 | Cleveland Clinic, no DM | Sema/tirz | ~80% | — | ~48% | — | — | >90-day |
| MassHealth (JMCP 2026) | ★★★★ | ~Several K | Medicaid (MA) | Wegovy/Zepbound | — | 60.8% | — | — | — | ≤56-day |
| Samuels et al. (Diabetes Obes Metab 2025) | ★★★★ | 2,306 | Academic clinic, no cost | Sema/tirz | 86% | 76% | ~50% | — | — | ≥84-day |
Drug-specific hierarchy at 12 months: tirzepatide (Zepbound) ≈ 62–65% > semaglutide weekly (Wegovy) ≈ 33–63% (depending on cohort year) > liraglutide daily (Saxenda) ≈ 19%. The daily injection burden of liraglutide produces the lowest persistence across every study.
¹ Do et al. gap threshold: Uses a 135-day gap — far more lenient than the 60-day standard. Applying a 60-day gap to this cohort would likely yield ~35%, consistent with other 2021 studies. This figure should not be cited without noting the non-standard threshold.
² IQVIA denominator: Includes every new prescription attempt — including the estimated 40% rejected by payers and patients who abandon at the pharmacy counter. All other studies require at least one filled prescription. This single difference likely accounts for 15–20 percentage points of the gap vs. other sources.
Model these persistence rates for your plan
Use the persistence planning tool ↓Why the numbers differ
The 10%-to-63% conflict, explained
The six-fold discrepancy in reported 12-month persistence is not contradictory — it reflects fundamentally different measurement approaches. Five factors explain most of the variance.
1. Intent-to-treat vs. as-treated — the biggest driver
IQVIA's ~10% captures every new prescription attempt, including those rejected by payers and patients who abandon at the pharmacy counter. All other studies require at least one filled prescription. This single difference likely accounts for 15–20 percentage points of the gap. For employer modeling, the as-treated denominator (members who actually started therapy) is the appropriate frame.
2. Cohort year and supply shortages
Wegovy was on the FDA shortage list from March 2022 through early 2024. Prime's data shows persistence nearly doubling from 33.2% (2021) to 62.6% (H1 2024) as supply normalized. Studies pooling 2021–2023 cohorts (Xie, Rodriguez) are mathematically anchored to the shortage era, producing lower averages that don't reflect current conditions.
3. Drug mix
Studies including all GLP-1s — with liraglutide/Saxenda at 19% persistence — show much lower rates than those restricted to high-potency weekly agents (Wegovy/Zepbound at 62–65%). For current employer planning, only Wegovy, Zepbound, and Foundayo are relevant. Class-average figures including older daily injectables understate current persistence.
4. Gap threshold definition
A 60-day gap is the most common standard, but Do et al. used a 135-day gap — far more lenient, which inflated their 12-month rate to ~50% versus ~35% had they used a 60-day threshold. Do et al. and Prime 2021 studied the same cohort year with claims data yet differ by 17 percentage points — explained almost entirely by this definitional choice.
5. Population selection
Merative required continuous enrollment, excluding members who left the plan — disproportionately non-persistent patients. Cleveland Clinic patients are in a structured academic obesity medicine program with higher engagement. Danish patients (Thomsen) face lower cost barriers (~€2,000/year vs. $12,000+/year in the US). Each selection criterion biases persistence upward relative to the general commercially insured population.
For a commercially insured employer covering Wegovy and/or Zepbound in 2025–2026, the Prime H1 2024 data (62.6%) is the most applicable benchmark — it uses the standard 60-day gap, restricts to high-potency drugs, and reflects the post-shortage era. A conservative planning range of 50–60% at 12 months is prudent, accounting for the possibility that Prime's Q1 2024 figure benefits from survivor bias in the cohort.
Indication comparison
Obesity vs. type 2 diabetes
Every study examining both indications finds obesity-only patients have substantially lower persistence. The gap is remarkably consistent.
| Source | Timepoint | Obesity only | T2D only | T2D + obesity | Difference |
|---|---|---|---|---|---|
| Do et al. 2024 | 3 months | 64.2% | 73.8% | 76.1% | −9.6 pp |
| Do et al. 2024 | 6 months | 55.2% | 69.6% | 71.9% | −14.4 pp |
| Do et al. 2024 | 12 months | 49.7% | 64.2% | 65.9% | −14.5 pp |
| Rodriguez et al. 2025 | 12 months | 35.2% | 53.5% | — | −18.3 pp |
| IQVIA 2025 | 12 months | ~10% | ~24% | — | ~−14 pp |
Obesity patients are 1.5–1.8× more likely to discontinue than T2D patients. Adjusted OR for discontinuation in obesity-only vs. T2D-only: 1.79 (95% CI 1.74–1.85). Authors attribute the gap to: narrower payer coverage for obesity, absence of an objective biomarker like HbA1c that reinforces treatment necessity, patient perception of weight loss as a temporary project, and weight loss plateaus reducing perceived efficacy. Patients with both T2D and obesity show the highest persistence (65.9%) — likely reflecting dual clinical motivation and broader insurance coverage.
Sources: Do et al., JAMA Network Open, 2024 (n=195,915); Rodriguez et al., JAMA Network Open, 2025 (n=125,474).
The efficacy-effectiveness gap
Clinical trial vs. real-world persistence
The gap between trial completion rates (85–92%) and real-world persistence (25–63%) is one of the central challenges for GLP-1 benefit design. The gap has narrowed substantially for 2024 cohorts.
| Trial | Drug | Duration | N | Trial completion | Best RW comparator | RW persistence | Gap |
|---|---|---|---|---|---|---|---|
| STEP 1 | Semaglutide 2.4mg | 68 wk | 1,961 | ~90% | Prime 2021 | 33% | 57 pp |
| STEP 1 | Semaglutide 2.4mg | 68 wk | 1,961 | ~90% | Prime H1 2024 | 63% | 27 pp |
| STEP 5 | Semaglutide 2.4mg | 104 wk | 304 | 79.9% | Prime 24-mo | ~20% | 60 pp |
| SURMOUNT-1 | Tirzepatide 5/10/15mg | 72 wk | 2,539 | ~85% | Prime 2024 (Zepbound) | 63% | 22 pp |
| SCALE | Liraglutide 3.0mg daily | 56 wk | 3,731 | ~73% | Prime 2021 (Saxenda) | 19% | 54 pp |
| SELECT | Semaglutide 2.4mg | ~40 mo | 17,604 | 73.3% | Prime 3-yr | 8% | 65 pp |
Even in Samuels et al.'s no-cost academic obesity clinic, 12-month persistence reached only ~50%, suggesting that cost removal alone cannot close the gap entirely. Clinical trial infrastructure — frequent monitoring visits, dose titration support, patient engagement protocols — accounts for a substantial portion of the remaining difference. Trial discontinuation is driven primarily by adverse events (4–17%), while real-world discontinuation is driven by cost/insurance (48%), side effects (15–23%), and supply shortages (8–12%).
Trend data
Is persistence improving?
Prime Therapeutics provides the only year-by-year cohort analysis, showing 12-month persistence nearly doubling for high-potency, obesity-indicated GLP-1s (Wegovy and Zepbound only).
| Cohort year | 12-mo persistence | 12-mo adherence (PDC ≥80%) | N (approx.) |
|---|---|---|---|
| 2021 | 33.2% | 30.2% | ~2,000 |
| 2022 | 34.1% | — | ~3,500 |
| 2023 | 40.4% | — | ~7,000 |
| H1 2024 | 62.6% | 55.5% | ~10,500 |
Three factors likely drive the improvement. First, supply shortage resolution is the primary driver — Wegovy entered the FDA's shortage list in March 2022 and semaglutide shortages persisted through 2023. Second, tirzepatide (Zepbound) availability from November 2023 introduced a second high-potency option with high initial persistence (~64%) and a switching target for patients struggling with semaglutide tolerability. Third, improved clinical management — better dose titration protocols, GI side-effect management, and care programs — may contribute, though this is harder to quantify from claims data alone.
The 3-year persistence rate of 8.1% reflects 2021 initiators exclusively. If persistence for 2024 cohorts starts at ~63% and follows a similar decay curve, projected 3-year persistence might reach 20–25% — but this remains speculative. No multi-year data exists for post-shortage cohorts. The Prime trend data is the most important finding on this page, but it is unreplicated — no other source has published year-by-year cohort trends.
Source: Gleason et al., JMCP, Mar 2026 (DOI: 10.18553/jmcp.2026.32.3.281; n=33,607). Corroborating: MassHealth, JMCP, Mar 2026 (DOI: 10.18553/jmcp.2026.32.3.271) — 60.8% 6-month persistence among Jul–Dec 2024 initiators.
Discontinuation drivers
Why people stop
Four studies provide direct data on reasons for discontinuation. The variation between studies reflects different data capture methods, not different realities.
| Reason | Gasoyan 2025 ★★★★ Chart review, n=288, US | Truveta 2025 ★★ NLP-EHR, n=6,939 | Almohaileb 2025 ★★★★ Chart review, n=83, Ireland | KFF 2025 ★★ Survey, US gen pop |
|---|---|---|---|---|
| Cost / insurance | 47.6% | 13.7% | 23% | 14% |
| Side effects | 14.6% | 22.5% | 36% | 13% |
| Supply shortages | 11.8% | 7.6% | 11% | — |
| Ineffective / unsatisfactory weight loss | 1.7% | 4.5% | 7% | — |
| Therapy completed / goal reached | ~1.4% | 11.4% | — | 5% |
| Logistical challenges | — | — | 24% | — |
Why cost appears lower in the Truveta study: Chart reviewers (Gasoyan) actively search pharmacy messages, prior authorization documents, and patient portal messages where cost discussions occur. NLP algorithms (Truveta) underdetect cost-related language in clinical notes. For employer audiences, the Gasoyan figure (47.6%) is more credible because it was purpose-built to capture insurance and cost barriers.
Gasoyan found cost rose to 54% among late discontinuers (3–12 months) and 76.5% among Medicaid patients. Within side effects: nausea (31%), abdominal pain (19%), vomiting (17%), diarrhea (14%), and depression (10%).
Sources: Gasoyan et al., Obesity, 2025 (DOI: 10.1002/oby.70058); Cartwright et al., Truveta/ISPOR 2025, poster EPH85; Almohaileb et al., Diabetes Obes Metab, 2025 (DOI: 10.1111/dom.16531); KFF, Nov 2025.
Cost/insurance is the dominant modifiable driver of discontinuation. Xu et al. found a dose-response relationship: discontinuation rising from 41% to 51% across copay quintiles, with 33% higher hazard at the highest tier. MassHealth's 60.8% 6-month persistence under Medicaid (minimal cost-sharing) versus ~46% in commercial plans during the same era further supports cost as a structural driver. Employers who impose high copays may be systematically selecting for discontinuation.
For context: higher persistence means higher total pharmacy spend. Whether sustained use generates medical cost offsets that justify the investment has not yet been demonstrated — Prime Therapeutics found no medical cost offsets through 2 years of follow-up for obesity-only GLP-1 users.
After discontinuation
Reinitiation and cycling
Among non-T2D patients who discontinued, 36.3% reinitiated within 1 year (Rodriguez et al., 95% CI: 35.6–37.0%). For T2D patients, the reinitiation rate was higher at 47.3%. Weight regain was the strongest driver: each 1% of weight regained increased the hazard of reinitiation by 2.8%.
Gasoyan et al. found that among discontinuers, 19.6% restarted the same medication within 1 year while 35.2% received some alternative obesity treatment (27.4% a different medication, 13.7% lifestyle visits, 0.6% bariatric surgery). Among obesity-only patients, the restart rate was lower at 14.2% — roughly half the diabetes restart rate (23.5%), consistent with insurance coverage barriers.
Post-discontinuation weight trajectories are less catastrophic than trial withdrawal data suggests. While the STEP 1 extension showed participants regained two-thirds of lost weight within 1 year of stopping, the Cleveland Clinic real-world data found obesity patients lost an average of 8.4% before stopping but regained only 0.5% at 1 year post-discontinuation. The explanation: real-world patients stop earlier and at lower doses, losing less weight to begin with and potentially adopting compensatory lifestyle behaviors. (For the full weight regain evidence, see What Happens When People Stop GLP-1s.)
No study tracks outcomes beyond the first reinitiation event. The dominant real-world utilization pattern appears to be cycling (start → stop → regain → restart), but its cost-effectiveness is unmeasured. Each restart involves dose re-escalation (~16 weeks), new prior authorizations, and repeated clinical engagement costs.
Two different measures
Adherence vs. persistence
Persistence asks: "Is the patient still on therapy?" — defined by no gap exceeding a threshold (typically 60 days). Adherence asks: "What proportion of days did the patient have medication on hand?" — measured as PDC (proportion of days covered), with ≥80% considered adherent.
| Metric | 1 year (n=4,066) | 2 years (n=3,364) | 3 years (n=5,780) |
|---|---|---|---|
| Persistence (no ≥60-day gap) | 32.3% | 14.8% | 8.1% |
| Adherent (PDC ≥80%) | 27.2% | 16.6% | 12.5% |
| Mean PDC | 51.0% | 40.7% | 37.5% |
The 37.5% mean PDC at 3 years means the average patient had medication on hand for roughly 410 of 1,095 days — just over one-third of the time. By 3 years, adherence (12.5%) actually exceeds persistence (8.1%) because PDC captures partial medication coverage from patients who eventually stopped, while persistence is an all-or-nothing measure. These figures paint a picture of widespread intermittent use rather than clean binary adherence or discontinuation.
Source: Prime Therapeutics, JMCP, 2024/2026; Gleason et al. Corroborating: Do et al. mean PDC for obesity-only = ~54% at 12 months; MassHealth 6-month PDC ≥80% = 60.1% among 2024 initiators; Xie et al. switchers mean PDC 63% vs. non-switchers 52%.
Who stays on
Predictors of persistence and discontinuation
| Predictor | Direction | Key effect size | Sources |
|---|---|---|---|
| Out-of-pocket cost | Higher copay → higher discontinuation | 33% higher hazard at highest vs. lowest quintile | Xu 2025 ★★★★, Gasoyan 2025 ★★★★ |
| Age | U-shaped: 18–29 and ≥65 most likely to stop | RR 1.48 for age 18–29 vs. 45–59 | Thomsen 2025 ★★, Xu 2025 ★★★★ |
| T2D diagnosis | Strongly protective | OR 1.79 for discontinuation in obesity-only vs. T2D | Do 2024 ★★★★, Rodriguez 2025 ★★★★ |
| Prescriber type | Specialists → better persistence | 50.2% vs. 44.2% at 12 weeks | BHI 2024 ★★★ |
| Income / SES | Lower income → higher discontinuation | RR 1.14 in low-income areas | Thomsen 2025 ★★, Xu 2025 ★★★★ |
| Early weight loss | Early responders less likely to stop | OR 0.81 (95% CI 0.70–0.94) | Durden 2019 ★★★★ |
| Drug formulation | Weekly injectable >> daily injectable | 47.1% vs. 19.2% at 12 months | Gleason 2024 ★★★★ |
| Switching | Switchers have higher persistence | PDC 63% vs. 52% (switchers vs. non) | Xie 2026 ★★★★ |
| Comorbidity burden | Paradoxically, higher comorbidity → higher discontinuation | CCI ≥3: 14% less likely to persist 12 wk | BHI 2024 ★★★ |
| Provider visit frequency | More visits → better persistence | Each visit → 60% ↑ likelihood of 12-wk persistence | BHI 2024 ★★★ |
Out-of-pocket cost is both the strongest predictor of discontinuation and the most modifiable lever. Xu et al.'s dose-response relationship — discontinuation rising from 41% to 51% across copay quintiles — provides direct evidence that benefit design choices affect persistence. The comorbidity paradox (sicker patients are less likely to persist) is counterintuitive and may reflect polypharmacy burden, competing health priorities, and greater vulnerability to GI side effects.
What we don't know
Evidence gaps and limitations
No multi-year data exists for post-shortage cohorts. The encouraging 62.6% 12-month persistence from 2024 initiators has no 24- or 36-month follow-up. Whether the steep attrition seen in 2021 cohorts will repeat or whether newer cohorts sustain gains is unknown.
Cycling outcomes are unmeasured. The dominant real-world pattern — start, stop, regain, restart — has no published cost-effectiveness or health outcomes data. No study tracks what happens after the second or third course of treatment.
Compounded semaglutide creates a blind spot. All claims-based studies cannot capture compounded GLP-1 use or cash-pay purchases. Patients appearing "non-persistent" in claims data may have continued therapy through these channels.
No randomized evidence compares benefit design strategies. Whether low copays, mandatory counseling, step therapy, or coverage duration limits optimize persistence-adjusted ROI has not been studied in controlled settings.
The Prime trend data is unreplicated. The most important finding on this page — persistence doubling from 2021 to 2024 — comes from a single PBM's commercially insured book. No other source has published year-by-year cohort trends.
Optimal treatment duration is undefined. The question "how long must a patient stay on therapy to achieve durable clinical benefit?" has no definitive answer.
Clinical trial populations don't match employer populations. Trial participants were screened, motivated, and received free medication with frequent monitoring. Extrapolating trial efficacy to a broad commercial population requires discounting by persistence, adherence, dose attainment, and patient selection — a compound discount that is rarely quantified.
Model it
Persistence planning tool
Estimate how many members will remain on treatment and what the total spend looks like under different persistence scenarios.
GLP-1 persistence cost projector
Select a persistence scenario, set your population and cost assumptions, and see the 3-year outlook.
Conservative scenario uses Prime 2021 data (Gleason et al., JMCP 2024). Current best estimate uses Prime H1 2024 12-month data (62.6%, rounded to 60%) with projected decay extrapolated from earlier cohort patterns — 24-month and 36-month figures are estimates, not observed data. Optimistic scenario uses clinical trial completion rates. Total spend assumes members on treatment pay the annual cost proportional to months on treatment, with persistence declining linearly between each timepoint. This is a simplified model — actual utilization involves cycling, partial adherence, and dose variation.
Methodology
How this page was built
Persistence means continuous treatment without a gap exceeding a defined threshold — most commonly 60 days. A patient who fills their prescription every 4 weeks with no gaps longer than 60 days is "persistent." The moment they go 60+ days without a fill, they are classified as having discontinued, even if they restart later.
Adherence is measured as proportion of days covered (PDC) — the number of days with medication on hand divided by the observation period. PDC ≥80% is the standard threshold for "adherent." Unlike persistence, adherence is a continuous measure (0–100%) and captures partial use.
The 60-day gap is the most common standard in published GLP-1 persistence research. Some studies use different thresholds (Do et al.: 135-day; BHI: 2× expected Rx duration; Gasoyan: 90-day). These definitional differences explain much of the variance in reported rates and are flagged throughout this page.
Switching between GLP-1s should generally count as persistence for employer modeling purposes — switching from Wegovy to Zepbound is not discontinuation from the employer's perspective. Most studies in the master table allow switching; those that don't are noted.
Source quality hierarchy: ★★★★ = peer-reviewed journal with disclosed methods; ★★★ = PBM report or white paper with methods disclosed; ★★ = conference presentation or poster; ★ = industry blog or news article with undisclosed methodology. Where sources conflict, higher-quality sources are given greater weight in summary statements.
All data is drawn from publicly available sources. Conflicts between sources are flagged rather than resolved. Gaps are noted as gaps. This page is reviewed and updated as new data becomes available. Corrections can be submitted via the contact page.
References
Sources
Questions about this data? Corrections or updates?
Get in touch