Stat Colloquium [In-Person]: Dr. Ming-Hui Chen
Chair, Department of Statistics, University of Connecticut
Title: An Informative Prior for Borrowing Historical Data under the Mixture Cure Rate Model
Abstract: Motivated by Eastern Cooperative Oncology Group's (ECOG) melanoma cancer trials E1684 and E1690, a new informative prior called ``borrowing-by-parts power prior'' is proposed under the mixture cure model. The proposed prior not only allows leveraging different amounts of information from the historical data for different sets of parameters under the mixture cure rate model but also yields the desirable prior propriety. We also propose a new approach to determine the power parameters based on the volumes of the overlapping regions. Some attractive theoretical properties on the overlapping regions are also derived. In addition, an efficient algorithm for computing the volume of the overlapping region is developed via the Importance-Weighted Marginal Posterior Density Estimation (IWMDE) method. We carry out an in-depth analysis of the ECOG’s melanoma cancer data E1690 by borrowing the historical data E1684 using the proposed borrowing-by-parts power prior to further demonstrate the usefulness of the proposed methodology. This is a joint work with Hongfei Li.