Identification of patients with Churg-Strauss syndrome (CSS) using automated data

Leslie R. Harrold, University of Massachusetts Medical School
Susan E. Andrade, University of Massachusetts Medical School
Mark Eisner, Kaiser Permanente
A. Sonia Buist, Oregon Health and Science University
Alan S. Go, Kaiser Permanente
William M. Vollmer
K. Arnold Chan, Harvard School of Public Health
E. Ann Frazier
Peter F. Weller, Harvard Medical School
Michael E. Wechsler, Harvard Medical School
Kourtney J. Davis
Richard Platt, Harvard Medical School


PURPOSE: Our aim was to identify individuals with Churg-Strauss syndrome (CSS) among asthma drug users, based on patterns of diagnostic and procedural codes (termed 'algorithms') contained in automated claims data. METHODS: A retrospective study was conducted among patients who had been dispensed asthma drugs at three HMOs. Individuals who received > or =3 dispensings of an asthma drug during any consecutive 12-month period beginning 1 January 1994 through 20 June 2000 were identified. Information on patient age, gender, enrollment status, asthma drugs dispensed, inpatient and outpatient diagnoses and procedures were obtained from the HMO automated databases. Twelve combinations of diagnostic and billing codes ('algorithms') were developed using the claims data to identify potential cases of CSS. Chart reviews blinded to drug exposure were performed using a standardized abstraction form. A rheumatologist reviewed abstracted information on all subjects, and those who met two or more American College of Rheumatology (ACR) criteria for CSS were further reviewed by two clinical experts. Cases were classified as unlikely, possible, or probable/definite CSS. Each clinical expert independently rated the cases; disagreements were resolved by consensus. RESULTS: A total of 185 604 patients who had been dispensed asthma drugs were identified. Three hundred fifty subjects were selected for chart review, and 15 were classified as having 'probable/definite' CSS. The algorithms that were most successful in identifying patients with CSS were as follows: (1) two or more codes for vasculitis (13 confirmed cases from 129 reviewed; positive predictive value 10%); (2) codes for both vasculitis and neurologic symptoms (6 confirmed cases from 15 reviewed; positive predictive value 40%) and (3) codes for both eosinophilia and vasculitis (4 confirmed cases from 5 reviewed; positive predictive value 80%). CONCLUSION: Automated claims data can be used to identify patients with CSS. This approach can facilitate better epidemiologic study of the risk factors for the condition.