Special Seminar: "Statistical Methods for Quantifying Predictive Accuracy of Absolute Risk Prediction Models"
Jinbo Chen, PhD
Associate Professor, Department of Biostatistics and Epidemiology, Perelman School of Medicine, University of Pennsylvania
Statistical Methods for Quantifying Predictive Accuracy of Absolute Risk Prediction Models
ALL ARE WELCOME
ABSTRACT:
Statistical models for predicting cancer absolute risk are useful tools for stratifying patients into different risk groups. A popular method to develop these models is through integration of data from multiple sources that provide an odds ratio function, composite cancer hazard rates, and competing risk of mortality hazard rates. Emerging risk predictors can be incorporated into the odds ratio function to improve predictive accuracy. Age-specific area under the receiver operating characteristic curve (AUC) has been the most commonly used statistic to evaluate the performance of absolute risk prediction models. Alternative statistical measures for evaluating the accuracy of prediction models with binary or time-dependent outcomes have been well developed. But their application to absolute risk prediction requires extensions to accommodate multiple data sources as well as the time-dependent nature. Here, we develop appropriate statistical measures for quantifying predictive accuracy of absolute risk prediction models, considering a general scenario where the odds ratio function is developed from a two-phase stratified case-control study. We demonstrate the performance of our methods through extensive simulation studies and application to a breast cancer risk prediction model. In the recent literature, to evaluate the predictiveness of risk-associated single neucleotide polymorphisms, the odds ratio function was approximated by the product of marginal odds ratio functions. We evaluate the applicability of our method to this approximated method for cancer risk prediction.
BIO:
Dr. Jinbo Chen is currently a tenured associate professor at the University of Pennsylvania Perelman School of Medicine, Department of Biostatistics and Epidemiology. She graduated from the University of Washington, Department of Biostatistics, in 2002, and subsequently worked at the NCI, Division of Cancer Epidemiology and Genetics for 3 years before moving to Penn in 2006. Jinbo's area of research interest includes design and analysis of two-phase outcome-dependent studies, the development and assessment of risk prediction models, and statistics genetics.
Ìý