Performance of compliance and persistence measures in predicting clinical and economic outcomes using administrative data from German sickness funds.

Objective: To compare the performance of various compliance and persistence measures in predicting schizophrenia-related hospitalization rates and inpatient costs.

Methods: Retrospective, nonrandomized cohort study. Methods: Prescription claims databases from three German sickness funds (public health insurance programs). Methods: A total of 1484 patients who were hospitalized in 2003 for a schizophrenia-related episode and subsequently received long-term antipsychotic pharmacotherapy.

Results: Data on age, sex, schizophrenia, prescription drugs, hospitalizations, and inpatient expenditures were collected for each patient from the three German sickness funds. Refill compliance measures based on single-interval availability and multiple-interval availability, as well as refill persistence, were calculated for each patient over 1 year. Ten measurement variables were compared with respect to their performance in predicting disease-related hospitalization and inpatient expenditure, using multivariate logistic regression and gamma regression, respectively. C-statistics were calculated to determine each measure's predictive performance. Likelihood ratio tests showed that inclusion of compliance and persistence measures significantly improved (p<0.05) outcomes prediction in 6 of 10 hospitalization models and in 3 of 10 inpatient cost models compared with a baseline model that included only age, sex, and disease severity covariates. Refill compliance as a continuous variable of drug persistence, including transfer of oversupplies into subsequent periods, performed best in predicting hospitalization (C = 0.669). Availability ratios, capped at 100%, were superior to default availability ratios in predicting hospitalization. Allowing for cross-period carryover improved the discriminatory performance of the persistence models.

Conclusions: Refill persistence measures appear sufficiently flexible to account for hospitalizations common in schizophrenia and other psychiatric diseases. A continuous refill persistence measure should be used to assess compliance in psychiatric conditions when working with administrative data.