| Registered phase | Obesity studies | Ongoing |
|---|---|---|
| Phase 1 (including early Phase 1) | 80 | 28 |
| Phase 2 (including Phase 1/2) | 105 | 53 |
| Phase 3 (including Phase 2/3) | 147 | 59 |
| Phase 4 | 133 | 58 |
| Not applicable / not stated | 169 | 99 |
Access and value of obesity drugs in Medicaid: coverage effect, eligible population, prescribers and budget impact
Study report, public aggregate data (CMS, NHANES, MEPS, KFF), 2018 to early 2026
Disclaimer. Public aggregate data; no company affiliation or endorsement. These are not patient-level claims. Medicaid cost figures are gross of rebates unless stated. Results for prescribers and payments are associations, and scenarios are not forecasts or effects of any company’s promotion.
The answer
When Medicaid programs began covering Wegovy and Zepbound for obesity, prescriptions rose by an estimated 12.2 per 1,000 enrollees per quarter on average (95% CI 8.4 to 16.0), comparing 10 covering states with 34 never-covering jurisdictions. The effect grew from 1.3 in the first quarter to 11.7 after four quarters and 17.2 after eight, partly because the national market grew at the same time. About 128.8 million U.S. adults meet the FDA label criteria (lower bound; 95% CI 117.2 to 140.3 million), including 17.8 million with Medicaid. For a state program of 1 million enrollees, the five-year net budget impact in the central case is $161.3 million, or $2.69 per enrollee per month, with a scenario range of $55.6 to $369.4 million that is driven mainly by two things the data cannot give: the rebate and the later years. At the announced $245 per monthly prescription, the central case would be $68.3 million ($1.14 per enrollee per month).
Every number in this report is read from analysis/outputs/key_numbers.csv, which links each one to the output file and row it came from.
A. Data and coding layer
A tested warehouse stands behind every number.
Public sources were loaded into PostgreSQL and transformed with dbt: Medicaid State Drug Utilization Data (SDUD) and enrollment, Medicare Part D Prescribers, NADAC drug prices, Open Payments, NPPES and NUCC taxonomy, NHANES, MEPS, ICD-10-CM value sets, FDA labels and the NDC directory, and state policy documents. A product map assigns each NDC to a drug, brand and label group using the FDA label indications text. The build has 54 dbt models (32 staging, 10 intermediate, 12 marts), 10 seeds and 126 tests, counted from the dbt manifest (analysis/outputs/build_counts.csv).
Pre-specified analysis plans were committed before any results of each module: Module C d46bd57 (2026-10-04 06:16 -0400; amendment f74da18, 2026-10-04 07:09 -0400, before any model run), Module B 7909c7a (2026-10-04 09:11 -0400), Module D 6d4234f (2026-10-04 09:28 -0400) and Module E 993e8c8 (2026-10-04 09:39 -0400). Commit hashes changed when history was rewritten to remove personal metadata and third-party copies; dates and order are unchanged. Departures from the plans are logged in deviations.md (items 1 to 17).
From NDC to label group
The product map holds 289 NDC codes. They resolve to 17 products (brands and generics) and 8 ingredients; 102 codes carry an obesity label and 187 a diabetes label. Counts are NDC codes, not prescriptions; the full count table is analysis/outputs/tables/moduleA_ndc_product_ingredient_label_counts.csv.
Market context: spending, timeline and pipeline
Gross spending on GLP-1 and GIP products rose from $7.0 billion in 2020 to $27.5 billion in 2024 in Medicare Part D and from $1.5 billion to $8.8 billion in Medicaid. These are CMS spending figures before manufacturer rebates, so they overstate what programs pay. Ozempic ($13.0 billion) and Mounjaro ($6.3 billion) lead in Part D in 2024, mostly diabetes use; in Medicaid, Wegovy ($1.2 billion) is the largest obesity-labeled brand.
The pipeline of registered studies (ClinicalTrials.gov, in-scope ingredients only: semaglutide, tirzepatide, liraglutide, dulaglutide, exenatide and orforglipron) has 634 studies listing obesity, of which 297 are ongoing and 59 are ongoing Phase 3. Oral programs account for 46 obesity studies (9 ongoing Phase 3; “oral” is matched on orforglipron or oral or tablet text and is approximate), and orforglipron alone has 23 obesity studies, 14 of them Phase 3. Programs of other ingredients are not counted.
The reporting heatmap shows that 81.0% of state-quarters have both fee-for-service and managed-care rows and 19.0% have fee-for-service rows only; 40.3% of Wegovy and Zepbound rows are suppressed (counts under 11), which is why Module C imputes suppressed cells.
B. Funnel and forecast
About half of U.S. adults meet the label criteria, and use is rising.
Of 253.8 million adults, 128.8 million (55.0% of adults with the measure) are label-eligible by BMI of 30 or more, or BMI 27 to 29.9 with a measurable weight-related condition. This is a lower bound, because some qualifying conditions are not measured in NHANES. Among adults with Medicaid (34.3 million), 17.8 million (52.0%) are eligible. The measurable part of the Medicare GLP-1 Bridge criteria covers 10.8 million adults aged 65 and over. KFF’s poll puts current GLP-1 use at 30.4 million adults (95% CI 23.5 to 37.5), of whom about 17.4 million do not have diagnosed diabetes (a weight-management proxy, modeled).
The forecast of GLP-1 users without diagnosed diabetes reaches a median of 33.8 million by end-2030 (90% interval 17.4 to 51.7) from 20.6 million in 2026; the downside, base and upside parameter sets give 16.5, 35.1 and 65.6 million. The ranges are assumptions about uptake, retention and the Medicare Bridge, listed in the assumptions table.
BMI class and diabetes by age and sex
| Age | Men BMI 30+, % | Men diagnosed diabetes, % (95% CI) | Women BMI 30+, % | Women diagnosed diabetes, % (95% CI) |
|---|---|---|---|---|
| Survey-weighted prevalence among adults with the measure available; BMI 30+ is the sum of classes 1 to 3. Full table with every BMI class: moduleB_prevalence_age_sex.csv. | ||||
Obesity (BMI 30 or more) peaks at ages 40 to 59, reaching 47.5% in the highest group (women aged 40 to 59). Diagnosed diabetes rises with age to 24.2% of men and 20.4% of women at 65 and older.
C. Coverage study
Coverage was associated with more prescriptions, quickly at first and more over time.
The map shows the 10 states in the primary analysis; 7 more with uncertain start quarters appear only in a sensitivity analysis. Prior authorization is documented in 9 of 10 of the covering states (California’s is not found in sourced documents), and BMI thresholds, comorbidity rules and step therapy are in the criteria table below.
| Utilization-management criteria in the 17 covering states | ||||||
| Compact view; 'Not found' means no sourced document states it; criteria differ by period | ||||||
| State | Delivery system | Prior authorization | BMI rule | Comorbidity | Step therapy | Source |
|---|---|---|---|---|---|---|
| Sources: state Medicaid agency documents; South Carolina's criteria are as reported by a news outlet quoting the state agency. | ||||||
Full criteria text for all 17 states
| State | Delivery system | Prior authorization | BMI threshold | Comorbidity requirement | Step therapy | Criteria version (document) |
|---|---|---|---|---|---|---|
The full table is coverage_um_criteria.csv: prior authorization is documented for 16 of 17 states, BMI thresholds for 12, comorbidity rules for 13 and step therapy for 13. Tennessee requires prior authorization (TennCare notice, August 2025: BMI 30, or 27 with a weight-related condition). In 2025 Q3, states with active coverage filled 37.0 Wegovy and Zepbound prescriptions per 1,000 enrollees against 0.7 elsewhere:
The overall effect is 12.2 per 1,000 enrollees per quarter (95% CI 8.4 to 16.0). The estimate holds in all 15 estimable alternative analyses (fee-for-service only, specification 12, is not estimable). A different estimator (Sun and Abraham) gives 9.3 (95% CI 8.2 to 10.4), a wild cluster bootstrap by state gives an interval of 4.5 to 14.2, and the sensitivity analyses range from 10.1 to 14.7. A managed-care-only outcome (sensitivity 14, per managed-care enrollee, so a different denominator) gives 12.5 (95% CI 6.0 to 19.0). A placebo with coverage moved four quarters earlier gives 0.4 (95% CI -0.1 to 0.8). The result tolerates a violation of parallel trends up to 1.4 times the largest pre-period violation (HonestDiD). Saxenda (0.2, 95% CI -0.3 to 0.7) and diabetes GLP-1 products (-0.8, 95% CI -4.2 to 2.7) show no effect.
Managed care is where coverage operates in South Carolina and Rhode Island, so a fee-for-service-only estimate is not reliable (deviation 6). After coverage ended, prescriptions per enrollee fell by -67.2% in California and -90.8% in Pennsylvania from 2025 Q4 to 2026 Q1 (preliminary, descriptive).
D. Prescribers
Primary care writes most Part D incretin claims, which reflect diabetes and other covered uses.
Part D reflects diabetes and other covered uses, not obesity-brand adoption.
Part D incretin claims grew from 2.6 million in 2018 to 19.6 million in 2024, and prescribers from 56,306 to 199,822. In 2024, primary care physicians wrote 55.8% of claims, nurse practitioners and physician assistants 27.7%, endocrinology 12.7% and cardiology 1.0%; the NPPES taxonomy agrees with the specialty grouping for 95.4% of prescribers. The top 10% of prescribers wrote 43.2% of claims in 2018 and 41.4% in 2024 (Gini 0.551 in 2024).
The share of prescribers with tirzepatide claims rose from 1.6% in 2022 to 25.8% in 2023 and 48.6% in 2024. Wegovy had 76,176 Part D claims from 3,956 prescribers in 2024; cardiology wrote 14.5% of them against 1.2% of Ozempic claims, consistent with the cardiovascular indication added in March 2024.
Segments (exploratory). The pre-specified rule gave a two-segment split of new versus continuing prescribers. A supplementary five-segment solution, shown with its stability beside each segment, is exploratory:
| Decile | Prescribers | Claims | Median claims | Share of claims, % | Cumulative from the top, % |
|---|---|---|---|---|---|
| Decile 10 = highest-volume tenth of the 199,822 prescribers. | |||||
Volume is concentrated: the top decile of prescribers wrote 41.4% of 2024 claims and the bottom five deciles together 13.9%.
Open Payments (association only; technical report, not for headlines). In 2023, 36.6% of 2024 prescribers had in-scope payments. The negative binomial model uses log(1 + dollars) as the predictor, so a “per $1,000” statement would be wrong; the clear contrasts are incidence rate ratios for 2024 claims: any 2023 payment versus none at the median payee amount, 1.30 (95% CI 1.29 to 1.31), and 1.05 (95% CI 1.04 to 1.05) per doubling of the amount among payees. Payments go to prescribers who already write more; these are associations, not effects of payments, and no individual or company is identified.
E. Budget impact
About $161 million over five years for 1 million enrollees, with the rebate the biggest unknown.
The model scales Module C’s extra prescriptions per 1,000 enrollees to a program of 1 million enrollees and prices them. Module C measures filled prescriptions in states that already applied their own prior authorization rules, so the central case uses the effect as observed (multiplier 1.0, “PA as observed in the 10 covering states”); tight and loose prior authorization are assumption scenarios. The central rebate is 51.2% (midpoint of 23.1% and 79.3%), where 23.1% is the statutory minimum (42 U.S.C. 1396r-8) and 79.3% is the rebate implied by the announced $245 price; actual net prices are confidential. No medical cost offsets are included because no published source supporting a five-year offset was retrieved.
In the central case the five-year net cost is $161.3 million ($2.69 per enrollee per month; gross $330.7 million, $5.51 per month), with $11.4, $30.3 and $39.9 million in years 1 to 3. About 15,873 members a year are users at the plateau, out of 300,948 eligible Medicaid adults in the plan. Two per-member costs answer different questions: net cost per user per year (4.3 fills a year, MEPS, sourced) is $2,511, and net cost per member-year of continuous treatment (12 fills a year, assumption) is $6,943. The gross reimbursement per prescription is $1,186 (range 1,164 to 1,252), measured in SDUD and consistent with NADAC within 3%.
The probabilistic range (10,000 draws) has a median of $145.4 million (90% interval 58.1 to 319.9 million), or $50.3 to 347.3 million if the effects at different event times move together. Prior authorization scenarios (five-year net, central rebate): tight at multiplier 0.5, $80.7 million; tight at 0.75, $121.0 million; loose at 1.25, $201.6 million.
Value context: trial efficacy next to net cost (context only)
| Trial weight loss and Medicaid net cost per user per year | |||||||
| Context only: different populations, durations and fills; no cost-effectiveness claim | |||||||
| Trial | Arm | Dose | Weeks | Weight change, % (95% CI) | Medicaid gross reimbursement per prescription, covering states, 2025, USD | Net cost per user per year, USD (4.3 fills, midpoint rebate; blended Wegovy and Zepbound) | Net cost per member-year of continuous treatment, USD (12 fills, assumption) |
|---|---|---|---|---|---|---|---|
| STEP 1 | semaglutide | 2.4 mg once weekly (subcutaneous) | 68 | -14.9 (arm CI not in abstract) | 1,303 | 2,511 | 6,943 |
| SURMOUNT-1 | tirzepatide 5 mg | 5 mg once weekly (subcutaneous) | 72 | -15.0 (-15.9 to -14.2) | 1,048 | 2,511 | 6,943 |
| SURMOUNT-1 | tirzepatide 10 mg | 10 mg once weekly | 72 | -19.5 (-20.4 to -18.5) | 1,048 | 2,511 | 6,943 |
| SURMOUNT-1 | tirzepatide 15 mg | 15 mg once weekly | 72 | -20.9 (-21.8 to -19.9) | 1,048 | 2,511 | 6,943 |
| SURMOUNT-5 | tirzepatide | maximum tolerated dose (10 mg or 15 mg) once weekly | 72 | -20.2 (-21.4 to -19.1) | 1,048 | 2,511 | 6,943 |
| SURMOUNT-5 | semaglutide | maximum tolerated dose (1.7 mg or 2.4 mg) once weekly | 72 | -13.7 (-14.9 to -12.6) | 1,303 | 2,511 | 6,943 |
| ATTAIN-1 | orforglipron 6 mg | 6 mg once daily (oral) | 72 | -7.5 (-8.2 to -6.8) | not a Medicaid-covered obesity product in the SDUD window | 2,511 | 6,943 |
| ATTAIN-1 | orforglipron 12 mg | 12 mg once daily | 72 | -8.4 (-9.1 to -7.7) | not a Medicaid-covered obesity product in the SDUD window | 2,511 | 6,943 |
| ATTAIN-1 | orforglipron 36 mg | 36 mg once daily | 72 | -11.2 (-12.0 to -10.4) | not a Medicaid-covered obesity product in the SDUD window | 2,511 | 6,943 |
| Trials: published abstracts via PubMed (STEP 1, SURMOUNT-1, SURMOUNT-5, ATTAIN-1; citations and DOIs in data/reference/trial_inputs.csv). Costs: Module E, net of an assumed rebate of 51.2%; one blended Wegovy and Zepbound net cost is shown for every row. | |||||||
Published primary results range from -7.5% (orforglipron 6 mg, 72 weeks) to -20.9% (tirzepatide 15 mg, 72 weeks) mean body-weight change; semaglutide 2.4 mg gave -14.9% at 68 weeks (STEP 1) and head-to-head tirzepatide gave -20.2% against -13.7% for semaglutide (SURMOUNT-5). Against these, the Module E central net cost is $2,511 per user per year (4.3 fills, MEPS) and $6,943 per member-year of continuous treatment (12 fills, assumption). The trials, populations, durations and Medicaid fills differ, so this table supports no cost-effectiveness, cost-per-kilogram or QALY statement.
The interactive model is a hosted decision tool at https://erickyegon.github.io/incretin-access-value/budget-model.html (also runnable locally with shiny::runApp("app")); a static PDF of the scenarios is in analysis/outputs/budget_impact_scenarios.pdf.
F. Research pack
A design for primary research, with no fieldwork done.
Module F is a design pack, rendered as research_pack/research_pack.pdf: a one-page brief from a fictional access team, prescriber and payer screeners with quotas, a 60-minute discussion guide with probes tied to the findings above, fictional stimulus boards, a discrete choice experiment (five attributes, three levels, 24 tasks in two blocks of 12, an efficient design from idefix with a zero prior), sample size by the Johnson and Orme rule of thumb and by simulation, and an analysis framework. No respondents exist and no interviews have been fielded; the recommendation slides are marked illustrative and carry only secondary-data numbers.
Exploration notebooks
One short notebook per module, built only from the existing outputs (no new analysis): A, data layer, B, eligible population, C, coverage study, D, prescribers and E, budget impact. The data dictionary (marts, units, suppression, NDC rather than HCPCS) is docs/data_dictionary.md.
Limitations
- Gross, not net. SDUD amounts are gross of rebates; the rebate range is an assumption between a statutory floor and a price-implied ceiling.
- Ten treated states. The effect comes from 10 states and may not carry over to other states or later years; later event times fall in later calendar quarters, so part of the growth is market growth.
- Eligibility is a lower bound. NHANES does not measure every qualifying condition; modeled use steps rely on KFF poll estimates.
- Part D is not obesity use. Part D claims reflect diabetes and other covered uses; prescriber findings are not about obesity-brand adoption.
- Suppressed cells. SDUD hides counts under 11; the analysis imputes them and combines 20 imputations.
- No offsets, no outcomes. The budget model has no medical cost offsets and does not estimate health outcomes.
- Associations, scenarios. Payments results are associations; the budget model and forecast are scenarios, not forecasts.
- Preliminary data. SDUD 2026 Q1 is preliminary; the withdrawal figures are descriptive.
Reproducibility
The code, the dbt project and the scripts are in the repository; raw and interim data are not committed and are re-downloaded by the scripts in scripts/fetch. In order: load (scripts/load), build (dbt build in dbt/), analyze (analysis/run_all.R, then scripts/build/build_counts.py and analysis/scripts/50_key_numbers.R), render (quarto render report, deck, research_pack). Every source is listed with its URL, access date and checksum in docs/sources_index.csv; U.S. government documents and open-license manuals are kept in docs/sources/, and copies of copyrighted third-party pages stay local only, and the R packages are pinned in analysis/renv.lock.
AI-use statement
I used AI tools to help write code and documentation. The study design, methods and conclusions are my own, and I verified all results.