Bayesian nonparametric common atoms approach for creating synthetic controls in early-phase glioblastoma drug trials

Abstract  We introduce a model-based system to complement treatment-only glioblastoma (GBM) trials with a synthetic treatment arm based on external data, using control arms from earlier studies. The main innovations of the proposed system are a model-based approach that accounts for all relevant uncertainties, implied inference on homogeneous patient subpopulations and the use of possibly varying subsets of covariates. As a case study we consider a hypothetical repeat of the InSIGHT study, augmenting the common control arm with a synthetic control cohort. We develop an R shiny app to implement the proposed approach, using inference that avoids sharing of actual patient records.

He will also share his experience on step by step through each of my PhD years and the lessons learned, give an overview of his advisor’s research, and talk about Graduate Student Assembly.

Slides

Participants: 22