Headache Vault Research Program
Methodology and open research questions.
How each dataset is built, what assumptions it makes, and what questions it is designed — but not yet able — to answer.
← Research overviewDataset 1 — Population Estimates
How the ZCTA-level population file was built.
Population base: U.S. Census ACS 2023 Table B01001 — sex- and age-specific population data for all 33,772 ZIP code tabulation areas in the United States, including all 50 states, the District of Columbia, and Puerto Rico.
Prevalence rates: Burch et al. (2021) calibrated rates — 21% in women, 10.7% in men — with age-band distributions from Burch et al. (2015). Migraine-specific rates only. Episodic tension-type headache excluded. All-headache prevalence rates are not used.
Preventive-eligible subpopulation: Buse et al. (2020) AMPP coefficient of 0.273 applied to the ZCTA-level migraine prevalence pool. Represents patients with moderate-to-high frequency episodic migraine (MFEM, 8–14 MHD, 17.7%) and high-frequency episodic migraine with disability (HFEM, 7.8%), normalized within the episodic subpopulation (93.2% of total migraine).
Known limitations: Prevalence rates are applied uniformly across ZCTAs and do not adjust for variation by race, ethnicity, or socioeconomic status. Published data suggest prevalence varies in ways that age/sex adjustment alone does not capture. An equity-adjusted analysis is planned for a subsequent release.
Proposed Framework — Patient Complexity Levels
A starting point for workforce planning discussion.
The headache field has characterized the specialist shortage quantitatively — provider counts, patient-to-specialist ratios, geographic deserts — but has not produced a patient complexity stratification framework that matches patients to the appropriate level of care. The epilepsy field has such a framework, endorsed by the National Association of Epilepsy Centers and cited by the Institute of Medicine. Headache medicine does not.
The five-level framework below is proposed as a basis for clinical and policy discussion. Criteria are illustrative. The framework has not been formally validated. Input from clinicians, workforce researchers, and policymakers is explicitly invited. Send responses to alex@headachevault.com or via commentary on the preprint.
CM = chronic migraine. MHD = monthly headache days. MOH = medication overuse headache. CGRP = CGRP-targeting therapy. Criteria are proposed, not validated. 90:9:1 caseload distribution from Steiner et al. (2021) — international data, U.S.-specific distribution unpublished.
Provider Classification
How providers are classified and why it matters.
UCNS-certified headache specialists: Providers holding active UCNS Headache Medicine diplomate certification. Approximately 900 nationwide as of 2024. Counted and reported separately from all other provider types. Never combined with advanced headache prescribers in any ratio or access metric.
Advanced headache prescribers: PCPs and neurologists identified by CGRP or Botox prescribing volume in CMS Medicare Part D data. A proxy for headache exposure — not a credentialing standard. Approximately 7,285 nationally. Reported separately from UCNS-certified specialists.
Combining specialist and prescriber counts produces a misleading ratio. A patient who needs specialist-level care cannot be served by a high-volume PCP. The Vault never combines these two populations in a single access metric.
Dataset 2 — Payer Policy Database
How payer policies are sourced and maintained.
Primary source requirement: Every record is sourced from a primary document — formulary PDFs, medical policy bulletins, coverage determination letters, or state legislative records. Secondary sources are not used.
Confidence classification: Each record carries a confidence classification based on source recency, document type, and verification status. Gemini Deep Research identifies candidates; human review confirms before any record is added or changed.
Update cadence: January primary refresh. July mid-year sweep. October ICD-10 update check. Ad hoc updates when significant policy changes are identified between scheduled refreshes.
Prospective Research Program
A longitudinal dataset the field doesn't have yet.
The Vault is building a prospective longitudinal dataset from enrolled patients — daily functional diary data, validated MIDAS and HIT-6 scores, treatment initiation dates, and PA outcomes — linked to the payer and geographic data above.
The research questions this dataset is designed to answer include: how much do patients systematically underestimate their own disability on standard retrospective instruments compared to prospective daily tracking? What does CGRP treatment response look like in the real-world patient populations excluded from pivotal trials? Does better clinical documentation predict higher PA approval rates? None of these questions are answerable from existing datasets.
The infrastructure is built — research consent at enrollment, validated instrument collection, PA outcome tracking, demographic fields collected prospectively. IRB routing is in active consultation with clinical advisors at major academic headache centers.
For academic collaboration, data use, or research partnership inquiries: alex@headachevault.com
Open Questions
What the Vault is designed to answer.
These questions are not yet answerable from existing datasets. The Vault's infrastructure is designed to make them answerable over time.
How much do patients systematically underestimate their own disability on standard retrospective instruments (MIDAS, HIT-6) compared to prospective daily functional tracking?
What does CGRP treatment response look like in the real-world patient populations excluded from pivotal trials — older patients, patients with significant comorbidities, patients with shorter treatment windows?
Does better clinical documentation of headache disorder burden predict higher PA approval rates — and if so, by how much?
What is the actual U.S. caseload distribution across complexity levels? The Steiner et al. 90:9:1 figure is from international headache service data and has not been validated in the U.S. system.
What proportion of patients currently occupying specialist waitlists are complexity Level 4–5 and would be appropriate for primary care management with support?
Does geographic access to a headache specialist correlate with CGRP prescribing rates at the ZIP code level — and if so, what is the magnitude of the effect?
Research Governance
IRB routing and collaboration.
Retrospective research using patient data collected with research consent at time of enrollment does not require IRB review per current guidance. Prospective research with pre-defined intervention arms requires IRB submission. IRB routing is in active consultation with clinical advisors at Jefferson Headache Center and Mayo Clinic.
The Vault's research infrastructure is built to support academic collaboration — research consent at enrollment, validated instrument collection, PA outcome tracking, and demographic fields collected prospectively. All data CC BY 4.0.
For academic collaboration, data use agreements, or research partnership inquiries: alex@headachevault.com