Three forces are at work. First, the headline and the setting bars use different denominators: 19.3% is the share of Part D lives, while setting bars are shares of all patients the setting serves, and only 81.9% of Medicare patients carry Part D. The apples-to-apples baseline is 15.8%, and settings cluster right around it. Second, the covered-use carve-out removes patients whose GLP-1 use is already covered (diabetes, OSA, MASH, cardiovascular), and it removes the most patients exactly where populations are sickest: the county-level correlation between diabetes prevalence and carve-out share is 0.995. Third, Bridge eligibility is BMI-anchored, and obesity falls with age: the oldest fifth of counties runs 16.6% eligibility against 22.6% in the youngest fifth. Care settings serve the oldest patients, so they inherit that gradient.
The Care Settings bars measure eligibility per patient served, including the 18.1% of Medicare patients with no Part D coverage, who cannot use a Part D Bridge at all. Once the frames match, home health at 15.0% is not below the market, it is at the all-patient baseline with a positive lift, exactly as the Eligibility Lift column reports. The gate math behind 19.3%: BMI 35+ contributes 12.1% of Part D lives, BMI 30 to 34 plus a condition adds 3.7%, BMI 27 to 29 plus prediabetes or vascular history adds 9.1%, for 24.9% meeting clinical criteria, and the carve-out then removes 22.8% of them.
The Bridge covers weight management uses that Part D does not already cover. That design decision means the sickest patients, the ones already on covered GLP-1 therapy for diabetes or cardiovascular disease, are excluded by construction. Across counties the correlation between diabetes prevalence and the share of clinical criteria removed by the carve-out is 0.995, as close to mechanical as public data gets.
| County diabetes quintile | Bridge eligible, % of Part D | Carve-out, % of clinical criteria removed |
|---|---|---|
| Q1, least diabetic | 18.0% | 19.0% |
| Q2 | 19.0% | 22.1% |
| Q3 | 20.6% | 24.1% |
| Q4 | 20.3% | 26.4% |
| Q5, most diabetic | 22.1% | 30.2% |
Sicker markets still net out slightly higher eligibility because obesity travels with diabetes and the BMI gates dominate, but nearly a third of their clinical pool is carved out, versus a fifth in the healthiest markets. Sickness raises the gross pool and shrinks the net one at the same time.
County Bridge rates correlate with obesity at 0.958, with diabetes at only 0.442, and with the 65-plus share at minus 0.173. The program is BMI-gated, and BMI declines in the oldest populations. That is exactly the population care settings serve.
| County age quintile, share 65+ | Bridge eligible, % of Part D | Carve-out share | Median obesity |
|---|---|---|---|
| Q1, youngest markets | 22.6% | 27.5% | 41.0% |
| Q2 | 21.7% | 24.9% | 39.2% |
| Q3 | 20.6% | 22.8% | 37.6% |
| Q4 | 19.4% | 21.8% | 36.4% |
| Q5, oldest markets | 16.6% | 20.8% | 34.4% |
Six points of eligibility separate the youngest and oldest county quintiles, driven by the obesity gradient. County distribution overall: p10 18.2%, median 21.7%, p90 24.1%, extremes 9.6% to 27.6% of Part D lives.
Each setting rate is Part D capture × local Bridge rate × setting enrichment × local driver, burden, and age adjustments. Nationally that means a 15.8% all-patient baseline before enrichment. Home health enriches at 1.18 (high Medicare exposure and cardiometabolic burden that is screenable, not yet carved out), landing near 18.6% before local adjustment. Skilled nursing enriches at 0.94: its patients are sicker, but sicker here means frailer, older, lower BMI, and more already-covered diagnoses, so the eligible share compresses. Palliative and hospice fall further on program fit, not eligibility: goals-of-care filters remove most of what the gates would admit. Sick is not the same as Bridge-eligible; the gates want heavy, aging-into-risk, not-yet-covered patients, and those concentrate in home health and primary care panels.
Running all 462,078 GLP-1-relevant clinician panels through the national gates, using each panel's real CMS condition mix, produces the sharpest version of the paradox. Endocrinology has the most diabetic panels in medicine, median 56% panel diabetes, and the lowest modeled Bridge share, because the carve-out strips its covered patients. Cardiology and nephrology rank highest: their ischemic and vascular patients qualify through the BMI 27 to 29 vascular gate while losing less to the diabetes carve-out.
| Specialty | Clinicians | Median Bridge share of panel | P90 | Median panel diabetes | Median HCC risk |
|---|---|---|---|---|---|
| Cardiology | 19,006 | 17.8% | 20.5% | 39% | 1.72 |
| Nephrology | 8,963 | 17.4% | 20.7% | 61% | 3.97 |
| Geriatric Medicine | 1,765 | 16.8% | 21.2% | 35% | 1.84 |
| Internal Medicine | 85,471 | 15.9% | 21.2% | 41% | 1.77 |
| Nurse Practitioner | 167,941 | 15.3% | 20.3% | 35% | 1.44 |
| Preventive Medicine | 211 | 15.0% | 17.6% | 35% | 1.25 |
| Physician Assistant | 95,708 | 14.9% | 19.2% | 32% | 1.29 |
| General Practice | 2,537 | 14.6% | 18.4% | 39% | 1.30 |
| Family Practice | 74,425 | 14.1% | 17.0% | 31% | 1.13 |
| Endocrinology | 6,051 | 12.1% | 14.9% | 56% | 1.46 |
Across all panels, higher diabetes still nets modestly higher Bridge share, Q1 14.4% to Q5 17.9%, because diabetes co-travels with obesity, heart failure, and risk in the condition gates. The relationship is real but flattened, and it inverts within diabetes-dominant specialties. Commercial takeaway: the largest Bridge pools sit in cardiology, nephrology, and high-risk primary care panels, not in endocrinology, and home health concentrates more eligible patients per patient served than any other care setting.
County eligibility, gates, and carve-out come from the atlas dataset: CMS Medicare Monthly Enrollment (March 2026) joined to CDC PLACES prevalence on real Census geography, with the CMS Final Rule gate logic (BMI 35+; BMI 30 to 34 plus comorbidity; BMI 27 to 29 plus prediabetes or vascular history; covered-use carve-out). Physician panels use the CMS Medicare Physician and Other Practitioners provider file, panels of 30 or more, each panel passed through the national gates with its own diabetes, heart failure, ischemic heart, stroke, and HCC risk mix relative to the national panel average, ratios clamped to 0.35 to 2.5. Quintiles are county-count or clinician-count weighted as labeled. Correlations are Pearson across 3,142 counties. Analysis script ships with the atlas pipeline.