OBJECTIVES: The RESULT study demonstrated that sulphonylurea (SU) plus rosiglitazone (RSG) provided a sustained and substantial increase in beta-cell function (BCF) from baseline (56%, p<0.0001) compared to SU alone (6%, p=0.41). This modelling study explores the impact on disease progression, health outcomes and healthcare expenditure in Portugal of different approaches to modelling RSG's effect on BCF. METHODS: DiDACT, a peer-reviewed published long-term model of T2DM, was used to replicate patient characteristics (73% Male, mean age 68.2 years, mean BMI 30 kg/m2) and the impact of SU+RSG on BCF observed in the RESULT study using an additive, a multiplicative or combined approach. Disease progression for 1000 hypothetical patients, projected total lifetime healthcare costs and health gains, measured in time to insulin and quality-adjusted life years (QALYs) were predicted. Following failure of intermediate SU dose to maintain glycaemic target, up-titrated SU therapy was compared to SU+RSG combination. The treatment change threshold was HbA1C≥7.5%. Resources were valued using national unit costs from a variety of sources. Costs and outcomes were discounted at 5% per year. RESULTS: Both revised calibrations yielded lower lifetime healthcare costs and additional QALYs, compared to the original calibration. Compared with SU alone both revised calibrations resulted in lower incremental cost-effectiveness ratios (ICERs). Each representation of RSG showed greater time to insulin than SU alone (13 years); the additive approach substantially extended viability of oral therapy to 27.5 years compared to 20 years for multiplicative and original calibrations. CONCLUSIONS: Each modelling approach resulted in reduced costs, increased QALYs and time to insulin when compared with the original calibration. The use of RSG in the management of T2DM appears to be cost-effective in all scenarios investigated. Forthcoming long-term studies of RSG may confirm the impact of RSG on BCF observed and enable determination of the most appropriate method for model calibration.