PS29: Borrowing External Data in Rare Disease Trials: Methods, Applications, and Beyond

Developing therapies for rare diseases often presents unique and complex challenges. One of the challenges is the limited availability of patient samples for clinical trials. The limited patient population makes it nearly impossible to implement conventional study design approaches. Statisticians are increasingly turning to the use of external data to overcome the limitation of the rare disease trial. Bayesian approach is one of the most powerful tools for leveraging external data. It integrates prior information from external data into current analyses, thereby enhancing the statistical power, precision and robustness of inferences. The FDA recognizes the potential of Bayesian methods as part of complex and innovative designs. It has provided guidance on the use of Bayesian methods, emphasizing the importance of incorporating prior information with current data to make more informed decisions. Although Bayesian methods have not been widely used to provide substantial evidence of effectiveness in new drug developments, their adoption is growing due to the need for more flexible and efficient trial designs. Despite offering several advantages, Bayesian external data borrowing methods inherently present some challenges and potential issues. First, it can be difficult to choose appropriate external data to be used as prior information. The external data being chosen must be in good quality and relevant and comparable to the current study to avoid introducing bias or misleading results. Second, there is no established guideline for determining the appropriate amount of information to borrow from the external data. Due to the inherent heterogeneity between the external data and the current data, it is crucial to quantify this borrowing to avoid making false positive decisions. Lastly, the complexity of Bayesian methods can make their implementation and interpretation challenging. Compared to the traditional frequentist methods, the concepts of posterior distributions and credible intervals in Bayesian analyses require more time and efforts to be understood and accepted, particularly by stakeholders who may not be familiar with Bayesian concepts. In this session, speakers from the FDA, industry, and academia will come together to share their insights and experiences in developing and implementing various external data borrowing approaches. Their presentations will explore the technical and regulatory benefits as well as the challenges associated with these methods.