Bayesian Inference of Treatment Response Rates in a Dose-Response Study with In-Treatment Dose Modification According to Early Response: Application to a Phase II Study

• Activity Number: 658
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Dominique Williams*+ and Ming-Dauh Wang and Scott Beattie and Damon Disch
• Companies: Eli Lilly and Company and Eli Lilly and Company and Eli Lilly and Company and Eli Lilly and Company

Abstract

Bayesian dose-response modeling using the Normal Dynamic Linear Model is being used more frequently in drug development. The Normal Dynamic Linear Model allows flexibility in the shape of dose-response relationship and meanwhile renders inferential efficiency by recursive borrowing of information from the previous dosage level. In a recent Phase II study, patients were allocated to new or continued dosing regimens at a mid-trial time point based on the outcome of the primary endpoint. We wanted to estimate the probabilities of response at later time points via a dose-response model as if patients would have remained on their original dose assignment. A modified two-step NDLM model was proposed and applied under both continuous and categorical frameworks. The final results explained the data reasonably. We will discuss what we attempted - what worked, what did not - with key take-always that we learned in applying this methodology.