There are numerous methods for identifying patients with depression from claims databases. Different indicators for disease may lead to different results in both population and health resource utilization. OBJECTIVES: To evaluate the economic impact of depression in a chronic non-malignant pain (CNMP) population using two different indicators for depression. METHODS: The 1995-1997 MarketScan claims database was used to identify a CNMP cohort. Patients were labeled with CNMP if two or more prescriptions for a long-acting opioid were given in any one-year period. Depression was categorized in this cohort using two different methods: ICD-9-CM diagnosis or antidepressant claim. Antidepressant usage (1997 numbers) was categorized as patients having a claim for TCAs (n=157), SSRIs (n=143), other antidepressants (n=36), combination of antidepressants (n=236), or no claims (n=233). Resource utilization was measured using outpatient, inpatient, and prescription costs. RESULTS: For the 1997 CNMP cohort, 805 subjects were identified. Using ICD-9-CM diagnoses (depression n=159, non-depression n=646), demographics between the groups were similar, with the non-depression group represented by a higher percentage of males (46% versus 35%). Average inpatient ($19,303 versus $11,419), outpatient ($12,665 versus $7738), prescription ($6486 versus $4548), and total ($38,454 versus $23,705) costs were all higher for the CNMP group diagnosed with depression. Similar values were also derived from the 1995 (n=154) and 1996 (n=323) cohorts. Using the second methodology (categorizing depression by antidepressant usage) showed that the group not prescribed antidepressants had the lowest overall cost utilizing the same 1997 cohort. However, in 1995 and 1996, the group taking no antidepressants did not have the lowest costs in either of these years. Categorizing patients through antidepressant usage resulted in conflicting differences between the groups. CONCLUSION: Characterization and assessment of depression using different methodologies in a claims database led to different results in both the population identified as well as total costs.