OBJECTIVES: The purpose of this study is to establish a measure of the COPD severity using claims data. METHODS: The study sample was identified from a large claims database covering the period 1999-2002. Patients of age 18-65 with previous acute exacerbation of COPD (AECB) were included in the study sample (n=2068). Variables associated with COPD severity were extracted. Variables with low face validity, high endorsement rate (>97%), or low homogeneity (measured by Chronbach’s alpha) were excluded. Principal component analysis with orthogonal solution was conducted to identify the latent severity score. Scree test and eigenvalue-one criterion were used to determine the number of latent factors. The severity score was standardized (mean=50, SD=10). Construct validity was tested by comparing severe COPD patients to moderate/mild patients of 3-month AECB incidence rate and by comparing the failure rate (ER/hospitalization) of antibiotic treatment in AECB. RESULTS: Six variables were excluded from the original 18 potential variables due to low face validity or high endorsement rate. Principal component analysis based on the remaining 12 variables produced a single latent factor (eigenvalue=3.6), therefore no factor rotation was performed. The score loaded high on the use of oxygen therapy, corticosteroids, and bronchodilators, etc. The resulting factor loading agreed with the clinical recommendation in GOLD criteria of treating COPD by severity. Chronbach’s alpha test showed good homogeneity of the severity score (0.71) and no input variables were rejected. The construct validity tests showed that, compared with mild/moderate COPD patients, severe patients were about 3 times more likely to have AECB episodes and 60% more likely to have antibiotic treatment failure in AECB episodes. CONCLUSIONS: The COPD severity score developed in this study can be applied to a wide range of retrospective studies of COPD, where results from the usually accepted measure of COPD severity categorization, spirometry, are not available.