# Master data and patient identity resolution
Maria Gonzalez walks into the emergency department at 2am with chest pain. She is scared, in pain, and gives her name as "Maria Gonzalez." The registration clerk, working fast, spells it "Marie Gonzales" and types her birth year as 1974 instead of 1947. A new medical record number (MRN) is created.
Three hours later, the lab runs her troponin test. Their interface cannot match the misspelled name to her existing chart, so it spawns a second record. Two weeks after, billing sends a statement to an old address on file under yet a third record.
One patient. Three identities. This is the everyday reality of patient identity chaos, and it is one of the most expensive, dangerous data problems in healthcare.
Master data is the core reference data that everything else depends on: who the patient is, who the provider is, what the facility is. Unlike transactional data (a single lab result, one claim), master data is supposed to be stable and shared across systems.
Master Data Management (MDM) is the discipline of keeping one trusted version of that core data. In hospitals, the two big MDMMDMMaster Data Management (MDM) is the discipline of creating and maintaining a single, consistent, trusted version of an organization's core business entities like customers, products, and suppliers. domains are:
When identity breaks, clinical safety breaks. A duplicate record can hide an allergy, a prior diagnosis, or a medication list. The ECRI Institute has repeatedly listed patient identification failures among top health technology hazards.
Follow Maria's data trail:
1. ED registration: typo in name and date of birth. New MRN issued.
2. Lab system (LIS): cannot deterministically match the typo to the existing patient, creates a shadow record.
3. Billing system: pulls from an outdated demographic feed, uses a stale address, generates a third identity.
Each system technically did its job. The failure is that no shared identity layer reconciled them. That layer is the EMPI.
An EMPI (Enterprise Master Patient Index) is a system that assigns each real person one enterprise identifier and links all their source records (ED, lab, radiology, billing) to it. Think of it as the switchboard that says "these three MRNs are all Maria."
The EMPI does not delete the source records. It cross-references them. Maria keeps her ED MRN and her lab record number, but the EMPI knows they belong to the same enterprise ID.
Two core matching strategies:
Deterministic matching requires exact agreement on chosen fields. If first name, last name, date of birth, and gender all match exactly, it is the same person. Fast, but brittle: "Marie" will never equal "Maria."
Probabilistic matching scores partial agreement. It assigns weights to each field, adds them up, and compares the total to a threshold. This catches typos, nicknames, and transposed digits.
Here is a simplified scoring illustration (weights are illustrative, not a real vendor algorithm):
Field Agreement Weight
Last name partial (Levenshtein 1) +4.0
First name partial (Marie/Maria) +3.5
Date of birth digits transposed +2.0
SSN (last 4) exact +8.0
--------------------------------------------------
Total score 17.5
Auto-match threshold : >= 15.0 -> LINK
Manual review band : 8.0 to 14.9
No match : < 8.0Records that fall in the manual review band go to a data stewarddata stewardA business-side owner responsible for the quality, consistency and appropriate use of data in their domain.View full definition →, a trained person who decides whether to link or keep separate. Modern EMPIs use referential matching too: they compare hospital records against a large external reference database of identities (built from credit, address, and public records) to confirm that "Marie" and "Maria" are one human.
For a solid vendor-neutral primer, see HIMSS on patient identity and matching.
You might ask: why not just give everyone one number? In the United States, a longstanding provision in federal appropriations law blocked federal funding for a unique patient identifier (UPI). That restriction shaped the whole industry: hospitals had to build EMPIs precisely because there is no government-issued patient ID. The debate continues in 2026, but as of this writing no national UPI exists.
Contrast with parts of Europe. Several countries use national health or citizen numbers (for example, systems tied to national IDs) that dramatically simplify matching. This is a real structural difference: European hospitals often start from a stronger identity anchor, while US hospitals lean harder on probabilistic EMPI logic.
Patients are not the only master data. Provider MDM manages the identity of clinicians and billing entities.
Key reference: the NPI (National Provider Identifier), a 10-digit number every US provider needs to bill, published in the free CMS NPPES registry. But NPI alone is not enough. A single doctor may have:
Provider MDMMDMMaster Data Management (MDM) is the discipline of creating and maintaining a single, consistent, trusted version of an organization's core business entities like customers, products, and suppliers.View full definition → stitches these into one golden record (the single authoritative version). Bad provider data causes claim denials, broken referral networks, and inaccurate "find a doctor" directories. Directory accuracy is regulated: US federal rules require plans to maintain accurate provider directories, and inaccuracies carry penalties.
🎬 [VIDEO: "What is an Enterprise Master Patient Index (EMPI)?" - youtube.com - a short explainer walking through how EMPIs link duplicate patient records across systems]
You cannot manage what you do not measure. Here are the core data-quality metrics hospitals track.
The headline metric. It is the share of records that are duplicates of an existing patient.
$$\text{Duplicate Rate} = \frac{\text{Duplicate records}}{\text{Total records}} \times 100$$
Worked example: A hospital has 900,000 records in its EMPI. An audit flags 63,000 as duplicates.
Duplicate rate = 63,000 / 900,000 = 7%.
Industry commentary (for example, from AHIMA, the American Health Information Management Association) has long cited duplicate rates commonly in the range of roughly 5% to 10% within a single system, and higher when systems merge. Treat those as widely cited estimates, not audited universal figures. Best-in-class organizations target duplicate rates under 2%.
An overlay is the dangerous opposite of a duplicate: two different people merged into one record. If Maria's chart absorbs another patient's allergy list, that is an overlay, and it is a direct patient-safety event. Target: as close to zero as possible. Every overlay should trigger investigation.
Knowledge check
1. What fundamentally distinguishes master data from transactional data?
2. In Maria's case, each system 'technically did its job' yet three identities were created. What does this best illustrate about identity resolution?
3. Why does a broken patient identity create a clinical safety risk, not just an administrative one?
4. Select ALL correct answers about why deterministic matching failed to link Maria's records.
Select all the correct answers.
5. Select ALL correct answers about Master Data Management (MDM) in a hospital setting.
Select all the correct answers.
Metrics need owners. Data governanceData governanceData governance is the set of policies, roles, and processes that ensure data is accurate, secure, well-defined, and used responsibly across an organization.View full definition → is the set of roles, policies, and accountability for data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition →.
Typical structure:
Governance also intersects with privacy law. In the US, HIPAA (the Health Insurance Portability and Accountability Act) governs how patient identity data is handled and shared. In Europe, GDPR (the General Data Protection Regulation) classifies health data as a special category requiring extra protection. Any referential matching against external databases must respect these rules.
Back to Maria. With a functioning EMPI:
1. Her ED typo record scores 17.5 against her existing chart. Auto-linked.
2. The lab record links to the same enterprise ID, so troponin results attach to her real history, showing a prior cardiac note.
3. Billing pulls the golden demographic record, mailing to her correct address.
One patient, one enterprise identity, three source records correctly linked. The clinician sees the full picture, and the safety risk disappears.