OBJECTIVE: To present the design of a lifetime Markov model that compares the cost-effectiveness of primary prophylaxis versus on-demand treatment with recombinant factor VIII among children with severe hemophilia A. METHODS: Prophylactic infusions of rFVIII-FS have been shown to reduce the frequency of bleeding episodes and the risk of joint damage in children with hemophilia A with no pre-existing joint damage.  Clinical studies have shown significant improvement in outcomes with the use of prophylactic treatment, as well as apparent gains in health-related quality of life. However, recombinant clotting factors are also associated with relatively high cost.  Using a lifetime Markov model, the cost-effectiveness of primary prophylaxis treatment was compared to on-demand treatment.  This model is among the few that model long-term cost and effectiveness and is unique in that it takes into account the probability of inhibitor development, use of central venous access device (CVAD), and total bleeding risk including CNS and joint bleeds.  Prophylactic treatment is assumed to be from birth until 16 years of age.  Built in the model were also 5 health states: being alive, surgery, inhibitor development, disability and deceased.  SUMMARY:  From this model, cost-effectiveness estimations can be made for patients receiving on-demand treatment versus primary prophylaxis. Cost-effectiveness can vary by the frequencies of events between treatment arms, age where prophylaxis begins and ends, dose/frequency of factor VIII, cost of medications and key hospital-related events, and the probability of achieving specified clinical endpoints.  CONCLUSIONS:  The strengths and distinguishing characteristics of this model versus previously published hemophilia prophylaxis models include: long-term cost and effectiveness, probability of inhibitor development, use of CVAD, and CNS bleeds.  There are a few study limitations related to the lack of data for model assumptions.  Obtaining stronger evidence for these parameters may substantiate or potentially improve the model results.