Data Inventory & Insights: building the menopause data substrate

EHR clinical data

FHIR R5 via JupyterHealth Exchange + SMART-on-FHIR

• Volume / coverageStructured + unstructured visit notes for ~250M US patients across Epic, Cerner — accessed through design-partner connections, not held by Pause.
• Representative signalsVitals, problem list, medications, lab panels (FSH, estradiol, TSH), encounter notes
• Why it mattersThe longitudinal clinical truth set. Where we measure outcomes and where guidelines are applied. Federated access first; Pause does not centralize PHI by default (see /proposal/integration).

Claims data

X12 837/835, plus partner data lakes

• Volume / coverageVisit, procedure, prescription, and ER utilization across commercial + MA plans — accessed via payer design-partner contracts.
• Representative signalsICD-10 N95.x, CPT for endometrial biopsy / DEXA, RX fills for HT/SSRIs/SNRIs
• Why it mattersQuantifies avoidable utilization and the economic case for the payer-side product. Required for the PMPM contracting motion on /proposal/customers.

Patient-generated wearables

HealthKit / Health Connect, Oura, Whoop, Garmin

• Volume / coverageContinuous HRV, sleep, heart rate, skin temperature, cycle data — exposed today through Salesforce Data Cloud Calculated Insights (HRV z-score, vasomotor burden, sleep disruption) on the trailsignup tenant; demo-cohort values are seeded CIs until first-party ingestion lands.
• Representative signalsSleep fragmentation, resting HR drift, nocturnal heat events, HRV decline patterns
• Why it mattersEarliest and most sensitive signal for perimenopause onset — usually invisible to clinicians. Phase 2 grounding is LIVE: the endpoint reports 'Phase 2: SOQL (Health Cloud) + Data Cloud Calculated Insights' on every call, and each insight falls back to its intake baseline independently if a DC call fails. Phase 2-bis swaps the demo-cohort seeded CIs for real JHE/DBDP wearable math via the same client + token flow (see PHASE_2_ACTIVATION_CHECKLIST.md).

Provider directory (NPPES-derived)

Public-domain · NPPES + Census + state sanction overlays

• Volume / coverage2,015 menopause-relevant providers across all 50 states + DC and 930 ZIP-3 prefixes (15 MSCP-certified, 2,000 menopause-relevant non-certified). A coverage-aware selection round-robins the non-certified budget across ZIP-3 prefixes (and the pipeline gates out non-US/garbage postals), so the same 2,000 rows give far more ZIPs a placeable local result. Refreshed by the provider_ingest pipeline against the monthly CMS NPPES bulk file (~9.6M rows) in ~1m50s.
• Representative signalsDistance from patient ZIP, MSCP/NCMP credential, six NPPES board-cert / multi-specialty signals, telehealth + accepting-new-patients flags, license disposition (CA Medi-Cal + NY OPMC + TX TMB checked at build), insurance acceptance
• Why it mattersSanctioned providers (1,720 dropped in the June 2026 build) cannot surface in any agent recommendation, /api/mulesoft/providers response, or the /provider UI — the patient-safety filter is verifiable per response under provenance.dataset.sanctionedFilteredBySource. The Care Router consumes the same query function the agent and /provider use, so triage and the directory stay in lockstep.

Patient-reported outcomes (PROs)

Validated instruments + adaptive intake

• Volume / coverageMRS (Menopause Rating Scale), GCS, PHQ-9, GAD-7, plus structured symptom diaries — collected via the intake experience in production deployments.
• Representative signalsVasomotor severity, sleep quality, mood, cognition, sexual function, urogenital
• Why it mattersCaptures the symptoms that drive lived experience but rarely make it into the EHR. The intake page (/demo/intake) is the surface this lands on.

Public research corpora

Open / licensed

• Volume / coverageSWAN, UK Biobank menopause cohort, NHS, PubMed, ClinicalTrials.gov
• Representative signalsTrajectory data, hormone-therapy outcomes, cardiovascular risk modeling
• Why it mattersPretraining and clinical evaluation; provides population-level priors. Surfaced today as grounding sources in the federated retrieval layer.

Specialty guideline corpus

Structured + retrievable

• Volume / coverageNAMS / Menopause Society, ACOG, IMS, AACE position statements; menopause hormone-therapy guidance — curated in the prototype grounding store.
• Representative signalsEvidence levels, contraindications, dosing, monitoring intervals
• Why it mattersGrounding source for every recommendation — explainability requires it. Live in the prototype: each Care Router decision cites it (see /demo/agent-fabric).

FHIR-native by default

All ingestion conforms to FHIR R5 resources via JupyterHealth Exchange. No bespoke data formats. SMART-on-FHIR auth means we plug into Epic and Cerner without bespoke integration projects (see /proposal/integration).

Patient-side first, EHR-side second

We begin collection on the patient side (wearables + PROs) before the visit, then merge in EHR context. This produces a fuller picture than starting from the EHR alone.

De-identified product telemetry

Every AI recommendation is logged in the Agent Fabric trace plane with inputs, outputs, clinician acceptance, and downstream outcome. This dataset compounds in value monthly (live today at /demo/agent-fabric).

Outcomes registry

With each design-partner contract, we co-build a clinical outcomes registry — diagnostic time, symptom resolution, utilization deltas, HT adoption and adherence. Operationalized via the Agent Fabric trace plane (/proposal/agent-fabric).

Federated where required

For health systems unwilling to move data, our inference layer runs federated against an in-VPC deployment. Model weights leave; PHI does not. This is the default posture, not the exception.

Acceptance + outcomes telemetry

Every recommendation accepted, edited, or rejected by a clinician — paired with the eventual outcome — becomes training signal. Competitors without deployments can't replicate this.

Multi-modal patient timeline

Wearable + PRO + EHR merged on a per-patient timeline is rare and expensive to assemble. The longer a patient is on Pause, the more valuable this representation becomes.

Specialty guideline grounding library

A curated, regularly updated, retrievable corpus of menopause clinical guidelines mapped to structured concepts. Hard to build, harder to keep current. Live today; see /demo/intake for citations in action.

Provider relationships and EHR install base

Each Epic / Cerner deployment takes time and trust. The N-th install is dramatically faster than the first, while remaining a meaningful barrier for newcomers.

Sanction-filtered, menopause-scored provider graph

Vendor provider graphs (Definitive Healthcare, IQVIA OneKey, etc.) are excellent for general healthcare but don't score for menopause and don't run a build-time sanction filter against state license registries. Pause's directory unions CMS NPPES + The Menopause Society MSCP overlay (synthetic + self-reported NPPES today; licensed feed once partnership lands) + three state license-sanction overlays (CA Medi-Cal, NY OPMC, TX TMB — 1,720 sanctioned candidates dropped in the June 2026 build) + Census ZCTA centroids for distance ranking + six NPPES service-line signals. Live today (see /provider). Each refresh recompounds the moat without requiring a paid feed.