OBJECTIVES: Progressive multifocal leukoencephalopathy (PML) is an infrequent but serious adverse event associated with natalizumab in the treatment of patients with relapsing-remitting multiple sclerosis (RRMS). The objective is to evaluate the risks and benefits of natalizumab treatment across three PML risk sub-groups against comparator agent interferon-β and no treatment. METHODS: The risk-benefit analysis was conducted using an adaptation of the original natalizumab model submitted to the National Institute for Health and Clinical Excellence that simulates natural disease progression and potential treatment impact on outcomes. A PML risk stratification component was incorporated to define PML risk by anti-JC virus (JCV) antibody status, previous immunosuppressant (IS) history, and duration of exposure to natalizumab therapy. A disutility score was applied to PML and post-PML states, as well as disease states defined by EDSS scores. RESULTS: Over a 20-year time horizon, treating anti-JCV antibody negative patients with natalizumab resulted in a total of 6.49 quality-adjusted-life-years (QALYs) per patient. Treatment with natalizumab in the anti-JCV antibody positive sub-groups, with and without previous IS history resulted in 6.27 and 6.43 QALYs per patient, respectively. Interferon-β treatment resulted in 6.05 QALYs per patient across all PML risk groups, whereas no treatment led to 5.82 QALYs per patient. In the highest risk group (anti-JCV antibody positive with prior IS use), utility decrement due to PML was small compared to the lowest risk anti-JCV antibody negative group (0.22), and overall QALYs were higher than interferon-β and no treatment. When considering the overall RRMS patient population, the risk of PML associated with natalizumab had to be more than 10.43 times higher in order to result in equivalent QALYs with interferon-β. CONCLUSIONS: The risk of PML associated with natalizumab treatment is offset by the benefit of natalizumab across all patient groups compared to interferon-β and no treatment.