Enterprise Master Patient Index (EMPI)

The core algorithmic framework that evaluates the statistical likelihood that two disparate records belong to the same patient. Unlike deterministic matching, which requires exact field agreement, probabilistic engines assign weighted scores to different attributes—name, date of birth, gender, and address—and tolerate real-world data entropy such as typos, transpositions, and phonetic variations. The engine computes a composite match confidence score between 0 and 100, where scores above a defined auto-link threshold trigger automatic merging and scores below a clerical review threshold are routed to a human-in-the-loop exception queue for manual adjudication.

A high-precision matching mode that links records using exact, rule-driven logic on highly reliable identifiers. This capability is critical for integrating with systems that use a known, shared key such as a Medical Record Number (MRN) or a National Health Identifier. Rules can be configured to perform exact comparisons on composite keys—for instance, linking records only when MRN + Facility Code are identical. This method eliminates false positives in controlled environments and is often used as a first-pass filter before invoking computationally heavier probabilistic algorithms.

The process of constructing a single, non-redundant golden record from a cluster of linked source records. Survivorship rules define which source system is the system of record for each demographic attribute. For example, the system may be configured to always trust the name from the state immunization registry over a clinic's scheduling system, while preferring the address from the most recently updated billing transaction. The EMPI maintains a full lineage, preserving all source contributions while presenting a unified, cleansed view to downstream consuming applications.

A proactive hygiene function that operates at the point of registration to prevent the creation of new duplicate records. When a registrar enters a new patient, the EMPI performs a real-time search against the existing index. If a potential match is found, the system alerts the user before the record is committed, allowing them to either select the existing identity or override the warning. This capability directly reduces the Duplicate Record Rate (DRR), a key performance indicator that measures the percentage of a database occupied by redundant patient entries.

The EMPI functions as a universal translator by maintaining a persistent cross-reference table that maps its own enterprise-wide unique identifier to the local identifiers from every connected source system. When a consuming application queries the EMPI with a local MRN, the index resolves it to the global Enterprise ID, and then translates it back to the corresponding local IDs in all other federated systems. This capability enables interoperability without requiring any source system to abandon its native identification scheme.

The transactional logic that governs the correction of identity errors over time. When a false negative is identified, the system executes a merge, collapsing two previously separate records into a single golden identity and cascading the update to all subscribing systems. Conversely, if a false positive is discovered—where records from two distinct patients were erroneously linked—the system performs a surgical unmerge, separating the commingled data and restoring the integrity of both individual records. All operations are logged immutably for audit trail compliance.

The computational logic that determines whether two records belong to the same individual. Algorithms range from deterministic (exact match on name and date of birth) to probabilistic (statistical weighting of multiple attributes).

• Deterministic Matching: Requires exact or normalized agreement on a predefined set of identifiers
• Probabilistic Matching: Assigns a likelihood score (e.g., 98.7%) based on agreement and disagreement across fields
• Machine Learning Matching: Uses trained classifiers to evaluate record pairs, learning from historical merge decisions

A statistical approach that calculates the likelihood that two records represent the same patient, even when demographic data contains typographical errors, name variations, or missing fields.

• Uses Fellegi-Sunter mathematical framework to compute match weights
• Assigns positive weight for agreement (e.g., matching SSN) and negative weight for disagreement
• Handles real-world messiness: 'Robert' vs. 'Bob', transposed digits in birth dates, hyphenated last names

A rule-based approach that links records only when specific identifiers match exactly or within narrowly defined tolerances. Often used as a first-pass filter before probabilistic methods.

• Requires exact match on a composite key (e.g., Name + DOB + Gender)
• Highly precise but low recall—misses records with even minor discrepancies
• Common in MPI cleanup projects where high-confidence matches are processed automatically

The process of identifying multiple records that refer to the same patient within a single system or across an enterprise. Unchecked duplicates lead to fragmented clinical histories and patient safety risks.

• Duplicate Rate: Industry benchmarks range from 8-20% in large health systems
• Overlay Detection: Identifies when one patient's data is incorrectly merged into another's record
• Requires ongoing stewardship, not just a one-time cleanup project

The broader discipline of connecting records across disparate data sources. In healthcare, this extends beyond patient identity to link clinical, claims, and social determinants data.

• Blocking: Reduces computational load by grouping records into blocks (e.g., by ZIP code) before comparison
• Clerical Review: Manual adjudication of ambiguous pairs that fall into a 'gray zone' of match probability
• Essential for population health analytics and value-based care reporting

The overarching discipline of creating and maintaining a single, trusted view of critical business entities—of which patient is the most vital in healthcare. EMPI is the patient-specific implementation of MDM.

• Golden Record: The curated, best-known version of a patient's demographics, synthesized from multiple source systems
• Survivorship Rules: Logic that determines which source system's value 'wins' when conflicts exist (e.g., most recent vs. most trusted)
• Governs the full lifecycle: create, read, update, merge, unmerge