Data from randomized controlled trials (RCTs) are often used in economic evaluations when estimating cost-effectiveness. The data generated from RCTs represent ideal experimental conditions (efficacy) and the applicability of this data to real world settings (effectiveness) may be questionable. OBJECTIVES: 1) To conduct an economic evaluation of etanercept (a competitive inhibitor of TNF-a) use in rheumatoid arthritis patients in Canada, and 2) To compare the results of a cost-effectiveness analysis conducted with efficacy data obtained from a RCT, to results derived by using effectiveness data obtained from community-based clinical practice. METHODS: The data used to perform the analyses were obtained from a trans Canadian community-based cohort study conducted between 1999 and 2003. A cost-utility analysis was preformed and incremental cost-effectiveness ratio (ICER) was calculated as cost ($CDN) per QALY gained. QALYs were derived from the health assessment questionnaire (HAQ) scores collected prospectively on patients. RESULTS: With the effectiveness-based analysis, the QALYs gained during the 12-month monitoring period were estimated to be 0.45 and 0.35, for the treatment and control groups respectively. The resulting ICER was $167,282 (CDN) per QALY. Using boot-strapping techniques and cost-effectiveness acceptability curves the 95% confidence interval (CI) for the ICER was $119,500 to $285,000 per QALY. For the efficacy-based analysis, the incremental QALYs gained were 0.56 and 0.35, for the treatment and control groups respectively. This resulted in a substantially lower ICER, $81,213 per QALY [95% CI = 66,500, 103,430]. CONCLUSION: Depending on the data used for the analysis, the resulting ICER was very different ($167,282 versus $ 81,213 per QALY). This study highlights some the potential concerns with using RCT data for estimating cost-effectiveness. These results are consistent with, and help to explain, the difference in cost-effectiveness reported in 2 previous modeling studies, one based on efficacy data and the other based on effectiveness data.