OBJECTIVES: Cost-effectiveness assessments in T2DM are commonly based on models that predict only primary cardiovascular events. This study aimed to assess the implications of incorporating secondary cardiovascular events on predicted cost-effectiveness.  METHODS: Routine UK hospital data, between 2000 and 2005, were analysed to quantify the cumulative incidence of first, second and third myocardial infarction (MI) or stroke events in T2DM subjects. Adjustments were made for out of hospital mortality and under-diagnosis of T2DM. Cardiovascular risk equations, used in a previously published cost-utility model, were re-calibrated, using the ratio of primary plus subsequent event to primary event, to predict subsequent MIs and strokes consistent with the observed UK data. The cost-effectiveness analysis compared two treatment strategies: A: 1st line metformin; 2nd line DPP-4 inhibitor add-on; 3rd line sulphonylurea add-on. B: 1st line metformin; 2nd line sulphonylurea add-on; 3rd line thiazolidinedione add-on. RESULTS: Of the 1,124,846 T2DM patients identified, 55,868 and 65,436 experienced primary MI and stroke events, respectively. There were 2159 (3.86%) and 185 (0.003%) second and third MI admissions, and 5808 (8.88%) and 755 (0.012%) second and third stroke admissions, respectively. Modelled risk multipliers of 1.04 for MI and 1.1 for stroke were required to predict cumulative incidence consistent with the UK data. Incorporating subsequent events had little impact on the cost-utility analysis with the ICER decreasing from £3129 to £3105 per quality adjusted life year. More noteworthy, was the impact on cost per life year gained, which decreased from £257,902 to £90,055, with subsequent events included.   CONCLUSIONS: The inclusion of subsequent cardiovascular events in models of T2DM provides greater face validity but has little impact upon cost-effectiveness. Thus, economic assessments of therapies that modify glycaemic control, using models that do not incorporate subsequent MI and stroke events, are not significantly biased.