Online Program

Return to main conference page
Friday, September 14
Fri, Sep 14, 8:00 AM - 9:15 AM
Thurgood Marshall East
Incorporating Historical Data in Prospectively Specified Analyzes

Discount Function Approaches for Incorporating External Sources of Data in Clinical Trials (300682)

*Donald Musgrove, Medtronic 

Keywords: Bayesian, dynamic borrowing, discount function

Medical devices present a key opportunity to address clinical trial design concerns arising from both feasibility of clinical studies and ethical constraints on sample size. Importantly, it is desirable to reduce the number of subjects required for enrollment to efficiently investigate the safety and effectiveness of the therapy being studied. To address both feasibility and ethical considerations, borrowing of suitable data external to the trial has been proposed as a solution to reduce the sample size of clinical trials. External sources of data may include historical data from similar device trials, computational modeling results, or electronic health record data. Statistical concerns of external data borrowing include increased type 1 error, decreased power, and biased effect estimates. Dynamic borrowing of historical data has been viewed as a key method for incorporating historical data in the analysis based on consistency with the current data under consideration, and thus at least partially alleviating statistical concerns. The idea behind dynamic borrowing is quite simple: if the current data under consideration are very similar to the available historical data, include 100 percent of the historical data in the analysis, otherwise weight the historical data toward zero. In this presentation, I will introduce and discuss several dynamic borrowing methods related to recent developments around the so-called discount function approach. The talk will focus on the discount function concept, ease and/or difficulty of comprehension by laypersons, implementation, computational considerations, and statistical concerns and controversies. The goal of the talk is to provide a clearer understanding of discount function approaches and key issues involved in external data borrowing.