Wednesday, March 6, 2019 - 12:20pm - 1:10pm
Pardee 217
Dr. Sandy Balkin
Free & there will be pizza!

The PCSK-9 Inhibitor class of cholesterol-reducing drugs has faced unprecedented payer scrutiny resulting in only about one in five prescriptions approved within 24 hours and about one-third of prescriptions that were ultimately approved, but never picked up. Our hypothesis is that the information contained in the characteristics and complete medical histories of those patients that did successfully obtain payer reimbursement for a PCSK9 therapy can be identified using statistical feature identification methods and leveraged to size the addressable or “reimbursable” market.

The ability to perform the computationally sophisticated analytics necessary to identify patients that look just like those who are on a PCSK-9 Inhibitor, but have not yet received a prescription, is only recently possible due to the general availability of open and closed patient claims data, which include robust lab results, and the computation power afforded by cloud based analytic platforms like AWS Redshift, Google BigQuery, Apache Spark and machine learning libraries, such as H2O, that can be efficiently deployed across multiple CPUs and computational clusters.

This presentation will describe one such approach, called Look-Alike Analysis, and how we developed a predictive model that could be then used for many traditional marketing activities including market sizing, forecasting, segmentation and targeting. 



Sponsored by: 
Department of Mathematics

Contact information

C. Jayne Trent