OBJECTIVES: Among the challenges in assessing persistence with medication time to all-cause discontinuation with pharmacy claims data are the potential variations in persistence definitions and data-cutting criteria. This study compared persistence on atypical antipsychotics (olanzapine, risperidone, quetiapine, and ziprasidone) using different definitions of persistence and data-cutting criteria to assess their impact on the results and conclusions. METHODS: Using the Pennsylvania Medicaid database (January 1999 – June 2003), patients diagnosed with schizophrenia, aged 18-64 who initiated an antipsychotic of interest after a 3-month period without the index drug were identified. A treatment episode was defined as the period from the initiation date of index medication to the first medication gap. To assess the effect of methodological changes on study outcomes three methodologies were implemented: 1) 30- vs. 90-day gap to define treatment discontinuation; 2) multi-episode vs. first- or last-episod; and 3) 1-year fixed study duration. RESULTS: Using a 30-day gap to define treatment discontinuation, 43,491 treatment episodes were identified (olanzapine=16,709, risperidone=14,847, quetiapine=8,648, ziprasidone=3,287) for 24,365 patients. Average duration of these episodes was 211, 197, 180, 130 days, respectively for olanzapine, risperidone, quetiapine, and ziprasidone and increased to 253, 236, 201, and 144 days respectively using the last episode approach. Imposing a fixed 1-year study duration effectively truncated the longer treatment episodes and had a different impact on the persistence of olanzapine (190 days), risperidone (183 days), quetiapine (170 days), and ziprasidone (143 days). Similar patterns were observed using a 90-day gap criteria. CONCLUSION: In claims database studies, the approaches used to define persistence and treatment episodes affect the persistence of individual medications and may impact the outcomes and conclusions of a persistence study. It is critical, therefore, to carefully consider the analysis criteria and the use of sensitivity analysis with multiple data-cutting scenarios in order to provide a better understanding of the data.