报告时间：2020年9月15 日（周二）10:00-11:00
报告形式：腾讯会议
报告人：孙韬
报告题目：New Statistical Methods for Complex Survival Data with High-Dimensional Covariates
报告摘要：
Complex survival outcomes such as multivariate and/or interval-censored endpoints become more commonly used in clinical trials, for example, in bilateral diseases or diseases with multiple comorbidities. The revolutionary development of genetics technology allows the generation of large-scale genetic data in modern clinical trials. In this talk, I will present two new statistical methods for modeling and predicting complex survival outcomes with high-dimensional covariates, motivated by two large clinical trials for studying a bilateral eye disease, the Age-related Macular Degeneration (AMD).
In the first part of my talk, I will briefly discuss a flexible copula-based semiparametric regression model for bivariate interval-censored data. The model parameters are estimated by the sieve approach and the asymptotic properties of the sieve estimators are rigorously proved. With simulation studies, I will demonstrate that the proposed method achieves satisfactory estimation and inference performances. Then I will present the novel discoveries of genetic risk variants associated with progression to late-AMD by applying the proposed method to a large-scale clinical trial study: Age-Related Macular Degeneration Study (AREDS) in a genome-wide association study (GWAS). The second part of the talk is inspired by the extraordinary achievements of deep learning in establishing powerful prediction models in the biomedical field. I will introduce a multiple-hidden-layer deep neural network (DNN) survival model to effectively extract features and make accurate and interpretable predictions. Through simulation studies and GWAS data from two large-scale clinical trial studies of AMD, I will demonstrate that the DNN model improves predictive accuracy as compared to existing methods, and provides valuable insights into early prevention and tailored intervention for AMD.
个人简介：
孙韬，中国人民大学统计学院，生物统计与流行病学讲师，匹兹堡大学生物统计学博士，获得2019年ENAR和ICSA优秀论文奖。主要研究领域为复杂生存数据模型，半参数统计模型，深度学习疾病预测模型，copula模型及其诊断，论文发表在Biostatistics, Statistics in Medicine, Lifetime Data Analysis, R Journal。医学统计方向包括流行病学调查和生物信息学，成果发表于Science, Nature Immunology等。
主持人简介：
林存洁，中国人民大学统计学院副教授，中国科学院数学与系统科学研究院博士，中国青年统计学家协会理事，全国工业统计学教学研究会理事，中国现场统计研究会资源与环境统计分会理事，北京生物医学统计与数据管理研究会会员。主要研究方向包括：生存分析、有偏抽样、缺失数据分析、复杂数据分析、模型平均等。承担国家自然科学基金青年项目、全国统计科学研究重点项目等多项科研课题。研究论文发表于Statistics in Medicine、Statistica Sinica、Journal of Multivariate Analysis、统计研究等国内外权威期刊上。