光华讲坛——社会名流与企业家论坛第 5855 期

（线上讲座）

主题： Improving suicide risk prediction through integrative statistical learning

主讲人：康涅狄格大学陈昆（CHEN KUN）副教授

主持人：统计学院林华珍教授

时间：2020年9月29日（周二）上午10:00-11:00

直播平台及会议ID：腾讯会议，289 217 199

主办单位：统计研究中心和统计学院 科研处

主讲人简介：

          Dr. Kun Chen is an Associate Professor in the Department of Statistics at the University of Connecticut (UConn) and a Research Fellow at the Center for Population Health at the UConn Health Center. Chen’s research focuses on multivariate statistical learning, machine learning, and healthcare analytics with large-scale heterogeneous data. He has extensive interdisciplinary research experience in a variety of fields including ecology, biology, agriculture, and public health. Currently Chen is funded by NSF for developing integrative statistical learning methods (DMS-1613295, IIS-1719798) and by NIH for improving suicide risk identification and prediction (R01-MH124740). He is an Elected Member of the International Statistical Institute since 2015 and serves as Secretary of the New England Statistical Society since 2017.

    陈昆博士是康涅狄格大学统计系副教授，康涅狄格大学健康中心研究员。他的研究集中在多元统计学习，机器学习，以及大规模医疗数据分析。他在生态学、生物学、农业和公共卫生等多个领域拥有广泛的跨学科研究经验。目前，他的在研项目包括整合学习的方法和理论（由NSF资助，DMS-1613295，IIS-1719798）和改善自杀风险识别和预测（由NIH资助，R01-MH124740）。他自2015年起当选为ISI的Elected Member，2017年起担任新英格兰统计学会秘书。详情请见其个人主页：https://kun-chen.uconn.edu

内容提要：

 Real-world integrative learning is a double-edged sword, in that while knowledge from different perspectives is integrated, various kinds of uncertainty, redundancy, and heterogeneity also come along. These notorious “side effects” together with curses of high dimensionality have to be properly treated in order to realize the power of integrative learning. In this talk, we discuss such challenges in the development of a data-driven suicide prevention framework by leveraging large-scale medical data scattered across different parts of the healthcare system. In particular, we present our work on (1) integrative survival analysis with uncertain event records for analyzing uncertain suicide outcomes due to either partial linkage of datasets or discrepancies in diagnostic coding, and (2) targeted learning with uncertain scope of data integration for improving condition-specific or provider-specific suicide risk prediction.

现实世界中的整合学习是一把双刃剑，在整合不同视角的知识的同时，也会产生各种各样的不确定性、冗余性和异质性。这些臭名昭著的“副作用”和高维度的诅咒必须得到适当的处理，以实现整合学习的力量。在本演讲中，我们将讨论通过利用分散在医疗系统不同部分的大规模医疗数据来开发数据驱动的自杀预防框架的挑战。特别地，我们提出了我们在以下方面的工作：（1）使用不确定事件记录的综合生存分析来分析由于数据集的部分关联或诊断编码的差异而导致的不确定自杀结果，（2）具有不确定数据整合范围的有针对性学习，以改善特定条件或特定提供者的自杀风险预测。