OBJECTIVES: To compare refill compliance and refill persistence measures as to their accuracy in identifying patients with schizophrenia at risk of temporary discontinuance or complete cessation of antipsychotic pharmacotherapy. METHODS: Data was obtained from three German sickness funds with approximately 7 million insured (9.9% of SHI members). Information on age, sex, prescription related information and hospitalization were collected. A total of 1484 patients with schizophrenia (ICD-10 F20) who were treated in hospital and subsequently received antipsychotic long-term pharmacotherapy were evaluated. Refill compliance measures based on single-interval availability, multiple-interval availability, as well as refill persistence were calculated for each patient over one year. The resulting 10 derivative measures were compared with respect to their performance in predicting six-month rehospitalization using multivariate logistic regression. C-statistics were calculated to determine each model’s predictive performance. RESULTS: Likelihood ratio tests showed that the inclusion of a compliance variable significantly improved the predictive performance in six out of ten models over the baseline model with age, sex and severity (p<0.05). Refill compliance as a continuous variable of medication persistence including transfer of oversupplies into subsequent periods, performed best in predicting rehospitalization (C=0.669). Availability ratios capped at 100% were superior to default availability ratios in predicting rehospitalization. Allowing for cross-period carryover improved the discriminatory performance of our persistence models. CONCLUSIONS: Persistence measures appear sufficiently flexible to account for interruptive events, i.e. hospitalization, common in schizophrenia and other psychiatric diseases. It is recommended to use a continuous refill persistence measure to assess compliance in psychiatric conditions when working with administrative data.