OBJECTIVES: During 2001, Kentucky Medicaid covered 530,184 prescriptions for Cox-2 inhibitors and 2,710 Cox-2 related inpatient and emergency room claims. The purpose of this study was to determine whether the Kentucky Medicaid Services should expand an existing Cox-2 inhibitor prior authorization (PA) requirement. METHODS: We used claims data to construct a series of decision trees to model this policy decision. The model compared two strategies "prescribe non-selective NSAID" and "prescribe Cox-2 inhibitor" under conditions of uncertainty regarding gastrointestinal (GI) event. Consequences of interest were total 2001 Medicaid costs associated with prescription, inpatient, outpatient, and emergency room claims. We refined the model to include use of prophylactic proton pump inhibitors (PPI) in high-risk patients and completed sensitivity analyses on probability of GI event and cost of GI event. We used expected value of perfect information methodology to calculate the expected value of PA. RESULTS: "Prescribe Cox-2 inhibitor" had the lowest expected cost ($4999). This result was very sensitive to the changes in the cost of treatment and less sensitive to the probability of GI problems. The maximum amount of money Medicaid should spend for a perfectly predictive PA is $14 per case. CONCLUSIONS: To minimize total Medicaid costs, prescribe Cox-2 inhibitors. This result contradicts a priori assumptions. Possible explanations include: the Medicaid populations is a sub-group for whom the Cox-2 is the overall least cost option, the results are clouded by a "selection bias" effect, or the majority of costs for the Cox-2 patients are covered by Medicare. If the Cox-2 PA can operate under the absolute cap of $14 per case, we recommend further analyses to incorporate the actual accuracy of the PA system.