Analysis of Two-Stage Seamless Adaptive Design in Clinical Development and Its Applications

Abstract

In recent years, innovative two stage seamless adaptive designs are increasingly used in clinical development to accelerate decision making and improve efficiency while preserving confirmatory integrity. This thesis develops methods for two stage seamless adaptive trials in which interim and final stages may use different endpoints, and the enrolled population may shift over time. It introduces a design level mean framework that links early surrogate markers readouts to later primary clinical endpoints under a common objective, and derives a surrogate informed combined estimator for paired, partially overlapping and independent settings. Under certain conditions, the estimator is unbiased, uses variance minimizing weights, and applies correlation aware spending to preserve overall type I error.The second part of this thesis addresses efficient evidence for biosimilarity assessment. An innovative Biosimilarity Index (BI) is introduced to formalize decision making for biosimilarity when pharmacokinetic similarity provides the primary evidentiary basis, with performance compared against the conventional confidence interval approach. The idea of BI was then extended to an innovative Relative Biosimilarity Index (RBI) that explicitly incorporates inherent variability and provides a method to define a proposed regulatory similarity threshold for RBI testing that is interpretable across products. Collectively, these contributions expand novel statistical methods for two-stage seamless adaptive designs and biosimilarity evaluation, delivering procedures that are statistically accurate, efficient and aligned with practical development and review needs.