All Times EDT
Keywords: statistical innovation, Bayesian methods, adaptive design, prior elicitation, historical data, assurance, probability of success, quantitative decision-making framework
The problem of high attrition rates in pharmaceutical industry has led to many statistical innovations over the past decade, including increased utilization of adaptive design and Bayesian methods as well as incorporating Probability of Success (PoS) into development decision-making. The high attrition rate is often attributed to lack of rigor in design of early face studies. We build upon the framework of Lalonde et al 2007 to create a fully Bayesian version of the quantitative decision-making framework (QDM) which has won Royal Statistical Society PSI Statistical Excellence award and now is embedded at GSK. The emphasis is on fully documenting success criteria upfront (as well as set of assumptions) and then quantifying the likelihood of transitioning into the next stage of development via calculating the assurance PoS. While some authors employed similar approach focusing on PoS of a single study (and optimize the design accordingly), we extend that approach by optimizing the current trial design taking into account the next development stage. The goal is to design a study delivering relevant data and enough of it to allow substantial reduction in risk of failure of the next trial. In that sense, our approach is similar in-spirit to optimizing study design at the portfolio/program level. A corner-stone foundation of this approach is prior distribution: it reflects what we DO know and what we DO NOT know, which feeds into study design and decision-making, helping to quantify the uncertainty/risks.