OBJECTIVES: Celecoxib and duloxetine are FDA approved medications to treat osteoarthritis (OA). The purpose of this study was to evaluate the impact of medication choice between duloxetine and celecoxib on medication adherence and direct health care costs among patients with OA. METHODS: This retrospective cohort study assessed commercially-insured Marketscan OA patients aged 18-64 years who initiated duloxetine or celecoxib in 2009. All selected patients had 12 months continuous enrollment in the pre- and post-index periods, and were grouped into the duloxetine or celecoxib cohort based on the index agent. The first dispense date, preceded by no coverage of study medication in the prior 90 days, was defined as the index date. Both cohorts were then matched via propensity score. The propensity scores were calculated based on a logistic regression model adjusting for demographics, comorbidities, prior health care utilization and costs, and prior medication history. Medication possession ratio (MPR), proportion of patients with MPR≥0.8 and health care costs over the 12-month post-index period were compared between cohorts. RESULTS: The matched sample included 2719 patients in each group. Both cohorts had a mean age of 53 years, and were mostly females. Over the 12-month post index period, duloxetine-treated patients had significantly higher MPR (0.53 vs. 0.37, p<0.001) and higher proportion of patients with MPR ≥0.8 (34.6% vs. 14.6%, p<0.001) than celecoxib-treated patients. While the total health care costs over the 12-month post-index period were similar between the two cohorts ($22,125 vs. $22,108, p=0.9834), duloxetine-treated patients had significantly lower OA-related inpatient ($1,607 vs. $2,694, p<0.001) and outpatient costs ($530 vs. $933, p<0.001), compared to celecoxib-treated patients. CONCLUSIONS: OA patients initiating duloxetine had significantly higher medication adherence and lower OA-related inpatient and outpatient costs. The results, however, should be interpreted with caution due to the observational nature of the claims database.