Linkage of a clinical surgical registry with Medicare inpatient claims data using indirect identifiers

A variety of data sources are available for measuring the quality of health care. Linking records from different sources can create unique and powerful databases that can be used to evaluate clinically relevant questions and direct health care policy. The objective of this study was to develop and validate a deterministic linkage algorithm that uses indirect patient identifiers to reliably match records from a surgical clinical registry with Medicare inpatient claims data.

Patient records from the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP), years 2005-2008, were linked to claims data in the Medicare Provider Analysis and Review file (MedPAR) by the use of a deterministic linkage algorithm and the following indirect patient identifiers: hospital, age, sex, diagnosis, procedure and dates of admission, discharge, and procedure. We validated the linkage procedure by systematically reviewing subsets of matched and unmatched records and by determining agreement on patient-level coding of inpatient mortality.

Of the 150,454 records in ACS-NSQIP eligible for matching, 80.5% were linked to a MedPAR record. This percentage is within the expected match range given the estimated percentage of ACS-NSQIP patients likely to be Medicare beneficiaries. Systematic checks revealed no evidence of bias in the linkage procedure and there was excellent agreement on patient-level coding of mortality (kappa 0.969). The final linked database contained 121,070 patient records from 217 hospitals.

This study demonstrates the feasibility and validity of a method for linking 2 data sources without direct personal identifiers. As clinical registries and other data sources continue to proliferate, linkage algorithms such as described here will be critical for quality measurement purposes.