Every health department eventually hits the same wall: the same person shows up as three, five, or a dozen different "patients" across immunization registries, case management systems, lab reporting pipelines, and surveillance platforms. A name gets misspelled at intake. A date of birth is transposed. Someone gets married and a new last name appears with no link back to their prior records. None of these are exotic edge cases — they're Tuesday.
The cost isn't abstract. Duplicate and fragmented patient records drive bad case counts, missed contact tracing links, redundant outreach, and — in clinical settings — real safety risk when a provider is working from an incomplete history. An Enterprise Master Patient Index (EMPI) is the infrastructure layer that's supposed to solve this: a system that assigns one durable identity to a person and links every downstream record back to it, regardless of which source system it came from.
Why this breaks specifically at scale
Simple deterministic matching — same name, same DOB, same SSN — works fine for a pilot with a few hundred records. It falls apart once you're running real volume, because:
- Data entry is inconsistent across systems. One system captures "Robert," another captures "Bob," a third truncates the field.
- Identifiers aren't universal. Not every source system captures SSN, and the ones that do don't always capture it correctly.
- People change. Names, addresses, and phone numbers all shift over time, and matching logic has to tolerate that without either over-merging distinct people or leaving true duplicates unlinked.
What's needed is probabilistic matching — algorithms that score the likelihood two records represent the same person using multiple weighted fields, not a single exact-match rule.
The open-source path: OpenCR and HAPI FHIR
Commercial EMPI platforms exist, and for large integrated delivery networks they're often the right call. But for public health agencies and mid-sized health systems, the licensing model — typically priced per managed record or per connected system — gets expensive fast, and the quote is rarely public until you're deep in procurement.
The open-source alternative built on OpenCR (Open Client Registry) paired with a HAPI FHIR server gives you the same core capability — probabilistic matching, a persistent Patient resource per unique identity, and a FHIR-native interface that every modern interoperability tool already knows how to talk to — without the per-record licensing model. You configure the matching algorithm's weights and thresholds to your own data quality, rather than accepting a vendor's black-box tuning.
What actually goes into building one
- Source system inventory — every system that will feed patient data in: registries, LIMS, case management, EHR feeds.
- Matching algorithm configuration — tuning field weights (name, DOB, address, phone) against a sample of your real data, not synthetic test data.
- Review queue workflow — defining who resolves the "possible match" tier and how often.
- FHIR interface layer — so downstream systems query the EMPI the same way they'd query any FHIR Patient resource.
- Ongoing monitoring — match rates drift as your data sources change, so this isn't a build-once system.
Done right, this isn't a multi-year enterprise procurement project. A properly scoped EMPI build for a health department or mid-sized system is a matter of weeks to a few months, not the 12–18 month timelines commercial vendors often quote — largely because you're not waiting on a vendor's implementation team's calendar.