OBJECTIVES: High health care costs are incurred by Chronic Obstructive Pulmonary Disease (COPD) patients. Identifying high cost patients prospectively benefits health care providers and insurance companies. The purpose of the current study was to compare the performance of generic and COPD specific diagnosis based comorbidity measures in predicting high medical expenditures in COPD patients. METHODS: A longitudinal study was conducted using nationally representative data from the Medical Expenditure Panel Survey. Prediction models were developed using cohorts initiated in 2008-2009 and validated using cohorts initiated in 2006-2007. Logistic regression was used to predict being in the upper expenditure decile in year 2, using data from year 1. Three comorbidity measures were compared: D’Hoore Charlson comorbidity index, Elixhauser comorbidity index and COPD specific comorbidity Test (COTE). Models were evaluated using c-statistics, Hosmer-Lemeshow statistic, correlation coefficient and positive prediction value. RESULTS: Comorbidity indices improved prediction of high cost COPD patients beyond age and gender, with the Elixhauser index providing the best prediction (c-statistic: 0.82), followed by COTE (c-statistic: 0.74). The Hosmer-Lemeshow test for both was also non-significant, indicating high accuracy. The COTE index had the best external validation properties (c-statistics: 0.66, Hosmer-Lemeshow: 7.4), followed by Elixhauser index (c-statistics: 0.61, Hosmer-Lemeshow: 12.1). In both learning and testing models, the difference in c-statistics of Elixhauser and COTE models was non-significant (learning, z=1.53 and testing, z=0.12), indicating that both measures do not differ in their discriminative power. CONCLUSIONS: Both, Elixhauser index and the COPD specific COTE index can best predict high costs in COPD patients and inclusion of either of them in risk adjustment models will be valuable in managing high cost COPD patients.