715 Broadway, 12th Floor, Room 1221
Please join the Department of Biostatistics for a Seminar Series featuring Dr. Yongzhao Shao, who will deliver a presentation on integrated regression analysis.
Multi-group studies are increasingly used to borrow strength from multiple centers to obtain broadly generalizable findings including regression-based predictive models. However, among the often numerous predictors, some of them are nonlinear with differential distributions among centers leading to group composition heterogeneities. Also, novel biomarkers are measured using different emerging technical platforms producing not-directly-comparable data between centers. Data incoherence and heterogeneous group composition can lead to biases in covariates and misleading or contradictory findings for the conventional pooled data analysis. In this talk, we introduce a new regression analysis for multi-group studies via systems of partially linear models (SPLMs) with integrated data harmonization using latent construct to make measurements comparable across centers and eliminate potential biases in covariates. The SPLMs allow the nonparametric component to be infinite dimensional in some Reproducing Kernel Hilbert Space to account for potentially complicated nonlinear predictors. A constrained regularization method is introduced with variable selection capacity to filter out non-informative predictors, and consistently estimate effects of selected predictors. In particular, when the effects of predictors are all homogeneous, the SPLMs can automatically reduce to one single predictive model which facilitates unified predictions for all groups. The proposed method is easy to implement with a fast computing algorithm as illustrated using a study on Alzheimer's disease. This talk is based on joint research with Lei Yang at NYU School of Medicine.
Dr. Yongzhao Shao is a professor in the Division of Biostatistics, Department of Population Health and Department of Environmental Medicine at NYU School of Medicine. He is an elected fellow of the American Statistical Association. His current research interest includes developing and evaluating risk prediction models in precision medicine for cancers and Alzheimer’s disease and developing methods for studying effects of environmental exposures on diseases.
Registration required to attend.