Session Information
Date: Tuesday, November 7, 2017
Title: Epidemiology and Public Health Poster III: Rheumatic Disease Risk and Outcomes
Session Type: ACR Poster Session C
Session Time: 9:00AM-11:00AM
Background/Purpose: Registries and claims data can complement each other to facilitate observational studies. However, methods to link on multiple non-unique identifiers (MNUI) are limited and therefore, there is a need to develop and test an approach to link registry data to Medicare claims using multiple non-unique identifiers.
Methods: Social security numbers (SSN) from participants in the Corrona registry with at least one registry visit in 2014 or prior were linked by CMS to Medicare fee-for-service (FFS) data. Using CMS created crosswalk file between Corrona ID (CID) and Medicare ID (MID), birth date (DOB) and sex, we established the SSN linkage as a gold standard. Using full DOB, sex, visit dates in registry and CMS data, and Corrona physician NPI, we developed an approach for linkage using MNUI based on 1) sex; 2) DOB elements (day, month and year, with year at most +-1 year), 3) number of visit dates matching exactly between Corrona and Medicare data, where the Medicare provider NPI matched the Corrona physician. These features were each included in a logistic regression model to evaluate the likelihood of a successful match using SSN+sex+DOB as the gold standard.
Results: The SSN linkage with sex and DOB confirmation resulted in 2,527 linked patients with any type of Medicare coverage. Among these, 1,854 had at least one month of Medicare FFS coverage in which a Corrona visit occurred. The initial match result in 565,856 potential pairings of CID and MID. The C-index for the model was 0.996. Choosing 0.06 as cut-points of predicted probability to achieve a PPV greater than 0.90, the algorithm predicted 1,858 matches; among then 1,732 were consistent with the SSN linkage. Keeping only the pairs with highest predicted probability result in 1,846 matches; among these, 1,731 were correct matches. Sensitivity of the approach was 0.93 (1,731/1,854), 95% CI: 0.92-0.95. The Positive predicted value (PPV) was 0.94 (1,731/1846), 95%CI: 0.93-0.95.
Conclusion: Linkage of an outpatient registry with administrative claims data using multiple non-unique identifiers is both technically feasible and accurate and yields high sensitivity and PPV
To cite this abstract in AMA style:
Xie F, Chen L, Yun H, Greenberg JD, Curtis JR. An Approach to Linkage of Registry Data to Medicare Claims Using Multiple Non-Unique Identifiers [abstract]. Arthritis Rheumatol. 2017; 69 (suppl 10). https://acrabstracts.org/abstract/an-approach-to-linkage-of-registry-data-to-medicare-claims-using-multiple-non-unique-identifiers/. Accessed .« Back to 2017 ACR/ARHP Annual Meeting
ACR Meeting Abstracts - https://acrabstracts.org/abstract/an-approach-to-linkage-of-registry-data-to-medicare-claims-using-multiple-non-unique-identifiers/