OBJECTIVES: A new disease modifying agent for the treatment of MS, natalizumab (Tysabri), was introduced to the market at the end of 2004 and withdrawn in early 2005 because of two cases (one fatal) of progressive multifocal leukoencephalopathy (PML). In the event that natalizumab is reintroduced to the market, the present study was conducted to assess the cost-effectiveness of natalizumab compared to interferon beta-1a (Avonex) and no treatment. Expected value of perfect information (EVPI) and partial EVPI (PEVPI) analyses were conducted to characterize the existing uncertainty in the model parameters. METHODS: The main analytical technique used in this study was incremental cost-effectiveness analysis using a Markov model. Two-level Monte Carlo simulations were performed to obtain the EVPI and PEVPI estimates. Health care costs were derived from the literature. The Disability Status Scale (DSS) was used as the measure of disability; utility values were assigned to the 10 DSS disability states based on data from the literature. Cost valuations were based on the direct health-care costs associated with disease relapse and medical care in each disability state expressed in 2005 US dollars. RESULTS: The Markov cohort analysis returned the following costs and QALYs: No Treatment - $175,790 and 30.971 QALYs; interferon beta-1a - $830,861 and 34.391 QALYs; natalizumab - $1,076,327 and 34.497 QALYs. The incremental cost-effectiveness ratios for interferon beta-1a and natalizumab compared to no treatment were: interferon beta-1a - $191,541 per QALY gained; natalizumab - $255,399 per QALY gained. CONCLUSION: Model inputs were based on a limited number of available studies and the results should be interpreted with caution. Nevertheless, the results of this preliminary analysis suggest that treatment with interferon beta-1a is somewhat more cost-effective than natalizumab. The value of information results indicate that more information about the transition probabilities and QALY parameters are necessary to reduce the uncertainty in the model.