Accuracy of Veterans Affairs Databases for Diagnoses of Chronic Diseases

INTRODUCTION: Epidemiologic studies usually use database diagnoses or patient self-report to identify disease cohorts, but no previous research has examined the extent to which self-report of chronic disease agrees with database diagnoses in a Veterans Affairs (VA) health care setting. METHODS: All veterans who had a medical care visit from October 1, 1996, through May 31, 1998, at any of the Veterans Integrated Service Network 13 facilities were surveyed about physician diagnosis of chronic obstructive pulmonary disease (COPD)/asthma, depression, diabetes, and heart disease. Four administrative case definitions (data from VA databases) consisting of combinations of International Classification of Diseases, Ninth Revision, codes and disease-specific medication data were compared with self-report of each disease to assess sensitivity, specificity, positive and negative predictive values, area under receiver operating characteristics curve, and κ statistic. RESULTS: Sensitivity for administrative definitions compared with self-report of physician diagnosis was 24% to 54% for COPD/asthma, 25% to 47% for depression, 27% to 59% for heart disease, and 64% to 78% for diabetes. Specificity was 88% to 100% for all diseases. The κ statistic showed fair agreement for COPD/asthma, depression, and heart disease and substantial agreement for diabetes. CONCLUSION: Diagnoses identified from databases agree with self-report for diabetes but not COPD/asthma, depression, or heart disease in a VA health care setting.