| This thesis deals with the subject of learning. In particular, I focus on physician learning about drugs in two key settings that clinicians typically work in: clinical trials (Chapters 1 and 2) and clinical practice (Chapter 3). In the first chapter, I propose a response-adaptive design for clinical trials, called Jointly Adaptive design, and show that it provides significant improvement over fixed randomized design (currently the gold standard for clinical trials) or existing adaptive designs. Further, I implement this design on a real clinical trial and demonstrate the extent of improvement that our proposed design provides. In the second chapter, I propose grid-based approximation as a way to reduce the problem dimensionality and the associated computational burden, thereby significantly widening the potential applicability of adaptive designs to large clinical trials. In the final chapter, I use data on adult patients with diabetes from the Department of Veterans Affairs to identify factors at individual and organizational level that lead to effective learning. Each of the three chapters is stand-alone and can be read on its own although connections exist between all three. |
