All Times EDT
Keywords: adaptive designs, evidence thresholds, interim monitoring
Mirroring dramatic advances in medical therapies for many disease conditions, statistical approaches relevant for treatment development and regulatory approval have also shown major advancements during recent decades. Current practices for design, analysis, and interpretation of clinical trial data have evolved to a point where some practices used in the not-so-distant past would seem outdated or unsound today. As is the case with medical and other technologies, it is natural to expect this beneficial evolution in statistical practice to continue into the future. We’ll describe some outdated past practices, and the motivations and impetuses that led to improvements. Some improvements were driven in part by advances in related technologies, for example, methodologies that built on improved computational capabilities; some carried over to operational aspects of clinical trial conduct, such as interim monitoring and decision making; some allowed added flexibility within trials and clinical programs, as well as possible redefinition of the traditional clinical development paradigm, such as adaptive designs and master protocols. We’ll discuss challenges and change-resistance that were initially encountered for some novel approaches that we now recognize as highly beneficial, and how these were overcome. We cite some landmark literature contributions and regulatory guidances that facilitated and expedited improvements in our practices. We’ll mention recent controversies involving the relevance and interpretation of traditional statistical measures such as p-values. We can anticipate that many future advances in our practices will relate directly to the potential to collect huge amounts of data, quickly, and globally, from new sources not traditionally included in clinical trials, as well as real-world and real-time evidence; this clearly promises huge potential benefits, but challenges as well, in terms of most correctly drawing conclusions from this data.