Abstract:
|
A prediction model commonly used in clinical research involves selecting features such as demographic characteristics, genomic biomarkers, and health history that could be signi?cantly associated with the response. Bayesian shrinkage prior models have emerged as a popular and ?exible method of variable selection. A uni?ed Bayesian hierarchical framework that implements and compares global-local shrinkage priors in binary and multinomial logistic regression models is presented here. The key feature is the representation of the likelihood by a Polya-Gamma data augmentation, which admits a natural integration with a family of shrinkage priors. Measures of accuracy, AUC, Brier score are used as evaluation criteria in the extensive simulation studies conducted under di?erent data dimensions and parameter settings. The simulation study achieved excellent predictive performance. The model identified signi?cant variables for disease risk prediction in a broad range of applications such as Pima Indians Diabetes, Colon cancer, and ADNI Alzheimer’s data sets, and is computationally e?cient.
|