时 间：2026年6月17日 10:00-11:00
地 点：普陀区理科大楼A1414
报告人：李鹏飞 加拿大滑铁卢大学教授
主持人：刘玉坤 永利集团教授
摘 要:
In diagnostic studies, researchers frequently encounter imperfect reference standards that may produce misclassified disease labels. Treating such standards as gold standards can bias receiver operating characteristic (ROC) curve analysis. To address this issue, we propose a likelihood-based method under a nonparametric density ratio model. The proposed approach enables reliable estimation of the ROC curve, area under the curve (AUC), and Youden's index. An efficient expectation-maximization algorithm is developed for implementation. Extensive simulations show that the proposed method has smaller mean squared errors in estimating the ROC curve and Youden's index than existing methods. We apply the proposed approach to a malaria study.
报告人简介:
Dr. Pengfei Li received his Ph.D. in Statistics from the University of Waterloo in December 2007 and completed postdoctoral training at the University of British Columbia in 2008. He has been a Professor at the University of Waterloo since 2019. His research interests include finite mixture models, empirical likelihood, missing data analysis, ROC curve analysis, and non-probability survey sampling. Dr. Li has published approximately 85 papers in refereed journals and book chapters, including sixteen articles in leading statistical journals such as The Annals of Statistics, Biometrika, Journal of the American Statistical Association, and Journal of the Royal Statistical Society: Series B. He was elected a Fellow of the Institute of Mathematical Statistics (IMS) and received the CRM-SSC Prize in Statistics in 2022.