Data Pipeline & Patient Matching
Data from every system lands in one normalized, de-duplicated model, mapped to FHIR, coded with standard terminologies, quality-checked and loaded into a warehouse that powers dashboards, models and the agent. It runs on any major cloud.
Nightly + streaming pipeline
HL7 streams in real time; bulk sources run nightly
EMR (FHIR + HL7), PM/billing, clearinghouse 835/837, fax/Document AI, phone/SMS, PT system
MLLP listener, FHIR $export (bulk), SFTP/X12 drops, webhooks → durable queue
Map to FHIR R4, code systems (ICD-10, CPT, LOINC), patient matching, dedupe, data-quality checks
Operational FHIR store + analytics warehouse (star schema), immutable audit log
APIs for apps & agent, dashboards, ML features, scheduled reports
Patient matching (MPI)
Same person from two systems? Edit either record to see the score change
Accent-insensitive Jaro-Winkler on names, compound-surname handling, exact DOB and phone matching. Borderline pairs go to a review queue, never an automatic merge.
Data-quality checks
Failing records are quarantined with a reason, not dropped
- Required fields present (PID-3, PID-5, PID-7)99.6%182 quarantined
- Valid ICD-10-CM codes (no retired M54.5)99.1%41 quarantined
- Appointment ↔ encounter reconciliation98.4%77 quarantined
- Duplicate patients resolved by MPI99.8%12 quarantined
- 835 payments matched to 837 claims97.9%64 quarantined
Terminology normalization
Local codes and free text → standard code systems
| Source | Raw value | System | Standard code |
|---|---|---|---|
| eCW local | LBP w/ sciatica R | ICD-10-CM | M54.41 · Lumbago with sciatica, right side |
| athena order | MRI L-SPINE WO | CPT | 72148 · MRI lumbar spine w/o contrast |
| PT system | Pain NRS | LOINC | 72514-3 · Pain severity 0-10 (reported) |
| Intake form | Smoker? No | LOINC | 72166-2 · Tobacco smoking status = Never |
| Fax OCR | TFESI L4-5 RT | CPT | 64483-RT · Transforaminal ESI, lumbar, single level |
Warehouse query
Star schema: fact_authorization, fact_claim, fact_appointment + dimensions
-- Median authorization turnaround by payer, last 90 days
SELECT p.payer_name,
COUNT(*) AS requests,
PERCENTILE_CONT(0.5) WITHIN GROUP
(ORDER BY a.decided_at - a.submitted_at) AS median_turnaround,
AVG(CASE WHEN a.decision = 'denied' THEN 1 ELSE 0 END) AS denial_rate
FROM fact_authorization a
JOIN dim_payer p ON p.payer_key = a.payer_key
WHERE a.submitted_at >= CURRENT_DATE - INTERVAL '90 days'
GROUP BY p.payer_name
ORDER BY median_turnaround DESC;| payer_name | requests | median | denial_rate |
|---|---|---|---|
| UnitedHealthcare | 214 | 3.4 d | 7.9% |
| Aetna | 163 | 3.1 d | 6.1% |
| Empire BCBS | 188 | 2.7 d | 5.3% |
| Humana MA | 97 | 1.9 d | 4.1% |
| Cigna | 72 | 2.2 d | 3.8% |