OBJECTIVE: To demonstrate the feasibility of using Net Benefit Regression Model (NBRM) methods to measure cost-effectiveness (CE) of new drugs that delay disability progression in multiple sclerosis (MS). Using the net-benefit statistic to estimate cost-effectiveness within a regression framework, the NBRM simplifies statistical work involved in economic evaluation (e.g., avoiding problems associated with ratio statistics) and also offers useful insights (e.g., exploring the importance of covariates on the marginal cost-effectiveness of a treatment). METHODS: A single-period-single-observation-per-person ordinary least squares (OLS) NBRM was generalized to a multi-period-multi-observation-per-person mixed-effects (ME) NBRM. Phase I regressions estimate marginal treatment effects, costs and net benefits. Phase II methods use marginal results to estimate area-under-the-curve total health outcome, cost, incremental cost effectiveness ratio (ICER) and net benefit (NB) for each person-with-MS (PwMS) treated and for the treatment program. DATA: Feasibility was tested using MS Research Unit person-level clinic-visit observational data, 1980 - October 2003: 2296 unique PwMS; 15,589 clinic-visit records; 646 PwMS treated with new drugs from 1998 - 2003. Disability progression is measured by Extended Disability Status Scale (EDSS). Health outcomes are measured by EDSS-weighted Disability Adjusted Life Years (DALY) avoided. Clinic data were linked to health services utilization and cost data, 1989 - 2003. RESULTS: Feasibility of NBRM methods was demonstrated using Nova Scotia data. The CE for a treatment program and for selected subgroups of PwMS treated can be estimated using NBRM methods. Results of a study on the Effectiveness and Cost-Effectiveness of New MS Drugs in the 'Real World' are expected in late 2004. CONCLUSIONS: It is feasible to apply NBRM methods to multi-period-multi-observation-per-person data using mixed effects models. CE estimates derived from person-level observational data, using NBRM methods, complement CE estimates derived from group-level observational data, using simulation model methods.