Incorporation of Supplemental Data Into the Design and Analysis of Randomized Trials

Abstract

Randomized trials are the gold standard for assessing causal relationships between an intervention and a clinical outcome. The challenge with randomized trials is that they are very expensive to perform, often take place at different times and locations, take many years to complete and often struggle with enrolling an adequate number of patients. This leads to several pressing issues in today's health research: first, the high cost of clinical trials directly translates into high cost of medications for patients. Next, clinical trials can be outdated by the time they are actually completed, especially in a rapidly evolving field. Also, trials of the same interventions performed in different populations can be difficult to aggregate. Lastly, trials investigating uncommon conditions may face great challenges or be abandoned altogether if enrolling enough patients is not realistic. In this dissertation we develop methods that can help to alleviate the aforementioned issues. In Chapter 2 we propose a method for borrowing of historical data when the marginal treatment effects observed are considerably different between studies, but this difference can be explained by differences in study populations. Chapter 3 proposes a group-sequential trial method which incorporates supplemental data at interim analyses and allows for early stopping of the primary trial. Chapter 4 proposes a SMART trial analysis method that uses within-trial borrowing of patients to increase precision and aid in identifying the ideal dynamic treatment regimes, as well as in clustering of dynamic treatment regimes that lead to the same expected outcomes. Overall, data borrowing can dramatically improve efficiency of clinical trials in the aforementioned settings.