Background/Purpose:

Integration of registry information with administrative claims data may be used to conduct patient-centered outcomes research (PCOR), including comparative effectiveness and safety studies, and to improve the quality of clinical care. While a variety of methods exist to link data, the unique requirements for linking data provided by patients in a research registry to claims data held by health plans and payers may present unique obstacles. These challenges are even greater if 1) no patient identifiers can be directly shared, and 2) no unique identifiers (e.g. social security number) are acceptable for use given privacy concerns. We evaluated results of a method to link data from a patient registry to the clinical outcomes research subsidiary of a large commercial payer under these two constraints.

Methods:

A novel, preliminarily validated encryption algorithm using a secure HIPAA-compliant cryptographic one-way hash function was developed to convert a vector of non-unique patient identifiers (first name, last name, sex, date of birth) into unique hashed identifiers. Both the ArthritisPower registry and HealthCore, Anthem Inc.’s research subsidiary, utilized the hashing algorithm and exchanged only these identifiers; exact match on the hashed identifiers was required. The diagnoses self-reported by patients in the ArthritisPower registry were compared with ICD9/10-based diagnoses in the claims data for the same conditions, and similar autoimmune conditions, varying the amount of health plan coverage available (any, or >5 years), using both a sensitive (>=1 outpatient diagnosis) and a more specific (>=2 diagnoses from relevant specialist) claims-based definition, and considering whether the condition matched exactly vs. matched a broader set of inflammatory arthritis diagnosis codes (e.g. RA, PsA, ankylosing spondylitis).

Results:

Of 11,343 ArthritisPower registry participants enrolled at time of data integration with any health condition, 19.1% (n=2166) were linked to Anthem claims data with no minimum coverage duration requirement; 1600 were commercially insured. Of these, mean (SD) age was 49 (10.7), 93% women, and they resided in the Northeast (12%), Midwest (29%), South (37%) and West (22%). Among patients with more than 5 years of coverage and who met the ICD9/10 definition for the computable phenotypes of RA, PsA or psoriasis, confirmation rates varied modestly according to the various parameters permuted (Table).

Conclusion:

Information from a patient-led arthritis research registry where in-person visits are not required can be linked to data from a research subsidiary of a large commercial payer using a hashing algorithm that does not require unique identifiers nor sharing of individual patient information. Ongoing work is underway to maximize the accuracy of linkage and confirmation rates using various approaches.