Abstract:
|
An understanding of prescription drug utilization patterns can provide insight into disease prevalence, drug expenditures, and prescribing patterns across time and space. Additionally, the ability to forecast future utilization can aid the decision-making of insurers, pharmacies, manufacturers, and policymakers. However, seasonal and spatial utilization patterns may differ greatly between between different drugs and therapeutic classes, and may depend on various external covariates. In this work, we employ multivariate versions of Bayesian structural time series models (Scott and Varian, 2013) to forecast future prescription drug utilization across multiple classes simultaneously. We discuss trade-offs between model flexibility and computational tractability involved in inference procedures, and demonstrate the performance on data provided by an insurance company.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.