Big Data Approaches for Health Policy: Characterizing the Diffusion of New Medical Technologies
*Sharon-Lise Normand, Harvard Medical School, Health Care Policy
The last two decades have been characterized by dramatic changes in the use of new medical technologies. Atypical antipsychotic drugs, most of which are prescribed by psychiatrists and financed by public programs, and iterative changes to coronary drug covered stents are two examples. In this talk, approaches to summarizing antipsychotic drug-prescribing behaviors for three new therapeutically similar drugs using dispensing information for nearly 17,000 U.S. physicians between 1/1/1997 and 12/31/2007 are described. While logistic models are commonly used to study the diffusion path of a new technology, several features of prescription data complicate inferences. These include time-varying drug choice sets due to different launch dates, semicontinuous response data, and multivariate outcomes. We begin with examining time to first prescription of a new antipsychotic, estimating diffusion paths separately for each new drug using nonparametric approaches. Next, we estimate multivariate survival models to identify fixed physician characteristics related to adoption time. We then use all antipsychotic prescription data and estimate key parameters of the diffusion path for each physician individually. The physician-specific parameters are combined using Bayesian multivariate factor analysis to provide a parsimonious representation of drug-prescribing behaviors.