Steffen Ventz, of the Department of Data Sciences, Dana-Farber Cancer Institute and Department of Biostatistics, Harvard T.H. Chan School of Public Health, will present:
“Adaptive Statistical Designs for Subpopulation-Driven Biomedical Experiments”
Abstract: The development of anti-cancer treatments has focused increasingly on therapies that target genetic alterations common to multiple cancer types. Consequently, clinical investigations have shifted towards subgroup-driven (Basket) studies that seek to match treatments to subpopulations that benefit from them. The fragmentation of the overall study sample size into smaller subpopulations stimulates the development of statistical designs that maintain adequate power at the subpopulation level.
In the first part of the talk, I introduce Bayesian Uncertainty Directed (BUD) study designs, a class of adaptive statistical designs in which the investigator specifies an information measure tailored to the experiment. All decisions during the study are selected to optimize the available information at the end of the study. The approach can be applied to several designs, ranging from early-stage multi-arm trials to biomarker-driven Basket studies. I illustrate operating characteristics of proposed BUDs through several examples, including biomarker-stratified Basket trials.
In the second part of the talk, I introduce a statistical procedure that integrates datasets from multiple biomedical studies to predict patients’ survival and define risk-stratified subgroups based on individual clinical and genomic profiles. The proposed procedure accounts for potential differences in the relation between predictors and outcomes across studies due to distinct patient populations, technologies to measure outcomes and biomarkers. These differences are modeled explicitly with study-specific parameters. Hierarchical regularization is used to shrink the study-specific model parameters towards each other and to borrow information across studies. Using a collection of gene expression datasets in ovarian cancer, I show that the proposed model increases the accuracy of survival predictions compared to alternative meta-analytic methods.
All are Welcome.