A state-space SIR model with time-varying quarantine protocols

This talk discussed a data analytic toolbox developed by Song Lab at UM, which enables public health workers to analyze and evaluate time-course infection dynamics of the novel coronavirus disease (COVID-19) using the public available data from the China CDC. This toolbox provides forecast of some key turning points. The data analytics are built upon a state-space SIR model, a hierarchical modeling framework in which the two-dimensional observed time series of daily incidences of infected and removed cases are emitted from the underlying infection system governed by the classical SIR infectious disease mechanism. We extend the SIR model to incorporate different types of time-varying quarantine protocols, including government-level macro isolation policies and community-level micro inspection measures. Part of the output includes forecast of two key turning points: the time of daily infected proportions smaller than the previous ones and the time of daily infected proportions smaller than that of daily removed proportion. An R software package is made available for the public, and some examples on the use of this software package are illustrated. Some possible extensions of our toolbox are also discussed. This talk is in Chinese.

本次报告介绍了密歇根大学（University of Michigan, UM）宋实验室开发的一套数据分析工具箱，该工具箱可依托中国疾病预防控制中心（China CDC）公开的公共数据，协助公共卫生工作人员分析与评估新型冠状病毒肺炎（Corona Virus Disease 2019，COVID-19）的时序感染动态。该工具箱可对若干关键转折点进行预测。本数据分析方法基于状态空间SIR模型（state-space SIR model）构建，该模型属于层级建模框架，其中每日新增感染与移除病例的二维观测时序数据，由遵循经典SIR传染病传播机制的底层感染系统生成。我们对SIR模型进行了扩展，使其能够纳入多种时变防疫管控策略，包括政府层面的宏观隔离政策与社区层面的微观排查措施。部分输出结果包含两类关键转折点的预测：一是每日新增感染占比低于前一日的时间点，二是每日新增感染占比低于每日新增移除病例占比的时间点。团队已面向公众开源一款R软件包，并对该软件包的若干使用示例进行了演示说明。报告同时探讨了该工具箱的若干潜在优化方向。本次报告为中文报告。