Strategies for using prior information in assessing safety
David Ohlssen, Novartis Pharmaceuticals Corporation  *Jerry J Weaver, Novartis 

Keywords: Bayesian methods, single conjugate prior, meta-analytic, robust prior

Assessing safety in drug development naturally incorporates an accumulation of knowledge as we progress from pre-clinical studies through to the various clinical trial phases. While pre-clinical animal studies are limited in their direct applicability to human patients, they do provide for a reasonable qualitative assessment with respect to identifying organ specific toxicity. However, it is the early clinical trial data that gives us the greatest opportunity to leverage historical information with key phase II data for making important phase III development decisions. In this presentation we will discuss a number of approaches in utilizing Bayesian methodology for the assessment of safety, namely adverse events. Specifically, we will introduce and evaluate the following three different priors:

1) A single conjugate prior, which potentially could be based on discounting prior information from previous studies;

2)A meta-analytic method, which automatically discounts prior information;

3) A robust prior that involves adding an additional mixture component to either a simple conjugate prior or meta-analytic prior. The additional component would be weakly informative, and therefore act as a corrective feature in the case of prior-data conflict. Details regarding the construct of the priors will be provided, along with the frequentist operating characteristics associated with estimation and testing properties for all three priors. The methods will be exemplified in a case-study involving the development of a compound for schizophrenia. In conclusion, we will suggest some areas of application on where these three priors may have the greatest impact for assessing safety in late phase trials.