OBJECTIVES: To demonstrate a unique approach to modeling long-term time-on-therapy and cardiovascular disease (CVD) outcomes of patients with schizophrenia treated with atypical antipsychotics (AAPs). METHODS: A 5-year Markov cohort analysis among adult patients with schizophrenia was undertaken to compare time-on-therapy and CVD outcome differences lurasidone, generic-olanzapine, aripiprazole, quetiapine, and ziprasidone. Modeled health states were: patients on initial AAP; patients switched to a second composite-AAP; and patients on clozapine after failing a second composite-AAP. The composite-AAP health state simulated frequent treatment switching and was operationalized by averaging outcomes, costs, and discontinuation rates (for transition probabilities) of the AAPs. Patients discontinuing composite-AAP due to lack of efficacy were switched to clozapine. Time-on-therapy was modeled using sub-states based on time of switching. Baseline characteristics of the modeled cohort, data for discontinuation rates, and average weight change were obtained from CATIE, a comparative clinical trial of lurasidone vs quetiapine XR, and an open-label study comparing aripiprazole and olanzapine. Relative risk of diabetes obtained from a retrospective analysis predicted CVD events using Framingham BMI risk equations. RESULTS: Over 5 years, patient time-on-therapy for the initial-AAP, composite-AAP, and clozapine, respectively, was 0.85, 3.13, 1.01 years (lurasidone); 1.00, 2.98, 1.02 (generic-olanzapine); 0.51, 3.41, 1.08 (aripiprazole); 0.47, 3.45, 1.08 (quetiapine); and 0.54, 3.37, 1.09 (ziprasidone). In a 10,000 patient cohort, there were 407, 434, 415, 416, and 412 CVD events, respectively, in the lurasidone, generic-olanzapine, aripiprazole, quetiapine, and ziprasidone arms. CONCLUSIONS: This long-term Markov cohort model simulates multiple treatment switches by using a composite health state from sub-states and also enabled outcome assessment of time-dependent patient characteristics, such as CVD events. The results were consistent with published Markov micro-simulation models showing that lurasidone and generic-olanzapine had favorable discontinuation rates and that lurasidone and ziprasidone had fewer CVD events. This analysis represents an effective alternative for modeling cohort-level outcomes.