Development and Validation of Algorithms to Identify Individuals With Cutaneous Lupus From Healthcare Databases.

Abstract

Background: There are no validated methods to identify individuals with cutaneous lupus erythematosus (CLE) from large databases including claims data and electronic health records, severely limiting the study of the epidemiology of this disease.

Objectives: To develop and validate accurate algorithms to identify individuals with CLE from healthcare records.

Methods: Twelve case-finding algorithms were developed based on the International Classification of Diseases (ICD)-10 diagnosis codes, provider specialty, and medication prescription data. To validate performance, algorithms were applied to a test cohort of 300 individuals drawn from a clinical data repository of a multi-institutional healthcare network in Boston, MA. Documentation of a CLE diagnosis by a dermatologist or rheumatologist determined from chart review or supportive biopsy findings was used as the case definition standard. Performance was evaluated based on calculated positive predictive values (PPVs), specificities, and sensitivities of each algorithm.

Results: PPVs ranged from 58.0% to 92.9%. The use of a single diagnosis code for CLE from any provider had poor PPV. The algorithm with the highest PPV (89.0%) while maintaining sensitivity required at least 1 ICD-10 CLE diagnosis code recorded by a dermatologist.

Conclusions: Utilizing CLE diagnosis codes and dermatology as the coding provider specialty is a valid method for identifying CLE patients from electronic health records.