Linking Provider Records and Classifying Healthcare Activity

The question

The company needed to connect physician and organization records and describe the clinical activity represented by claims.

The work

We developed an entity-resolution design covering individual and organizational NPIs, addresses, and site-of-service relationships, with confidence scoring for ambiguous matches. Reference-data work included procedure complexity, procedure type, and diagnosis classifications, along with validation and reproducible build scripts.

The matching design distinguished individual from organizational NPIs and site-of-service from rendering-provider relationships. Address-based rollups and confidence scores helped separate straightforward matches from ambiguous cases.

What the client gained

Versioned reference data and a repeatable approach to matching and classification. The implementation separated confident matches from cases requiring further review.

Volume-weighted procedure-complexity coverage increased from 65.7% to 99.5% across a reference table of approximately 102,000 codes. This measures the share of claims volume with an assigned complexity value; classification accuracy is a separate question.

Reference-data delivery

Procedure-type coverage reached 99.6%, and the procedural-code filter identified 79,020 codes for surgical and procedural analysis. The delivery included approximately 113,000 complexity rows, 128,000 procedure-type rows, national volumes, and build scripts in the client’s repository. These row counts include reference-table records and differ from the count of distinct codes.

Diagnosis classification used 21 body systems and 506 categories. Versioned tables and validation checks gave the client a way to refresh the reference data and investigate changes.

Limits

Reference coverage describes the share of records with a classification; it does not establish the accuracy of every classification or employer match.