Algorithms for Threat Detection: Detection and management of emerging diseases

该提案概述了流行病学监测和暴发管理的新方法，充分结合了传染病数学建模和贝叶斯统计的思想。对于监测问题，我们建立了分层混合模型来识别异常值，该模型考虑了疫情的演变及其空间分布。在这些模型中，混合物中的成分对应于与共享一组共同症状的不同疾病一致的替代动力学。对于疫情管理，我们开发了基于模拟的算法，用于不确定情况下的序贯最优设计，可用于流行病干预措施的最优前瞻性设计。该算法描述了部分观察到的马尔可夫决策问题的更一般的背景，并可能应用于其他领域，如临床试验设计、控制和经济。新疾病的出现，无论是自然的还是由于生物攻击的结果，都是对国家安全的最重要威胁之一；及早发现和充分干预是拯救生命和将此类爆发造成的损害降至最低的关键。然而，当症状与人群中已经流行的疾病的症状相似时(例如，发烧和喉咙痛是感冒和流感的常见症状，也是呼吸道炭疽病的症状)，早期发现可能非常困难，有必要对时空模式进行详细分析，以便能够将新疾病的存在与先前存在的疾病动态所固有的随机波动区分开来。同样，在这种情况下，由于缺乏对新疾病的传染性和现有疫苗效力的了解，干预措施(疫苗接种、检疫和扑杀，仅举三种可能性)的设计也受到了损害。这项研究为不确定情况下的疾病监测和暴发管理开发了新的算法，有可能极大地提高政府和国际机构识别和干预新出现的传染病爆发的能力。

This proposal outlines new approaches to epidemiological surveillance and outbreak management that fully integrate ideas from mathematical modeling for infectious diseases and Bayesian statistics. For surveillance problems, we develop hierarchical mixture models for outlier identification that take into account the evolution of the epidemic and its spatial distribution. In these models, the components in the mixture correspond to alternative dynamics consistent with different diseases sharing a common set of symptoms. For outbreak management, we develop simulation-based algorithms for sequential optimal design under uncertainty that can be used for optimal prospective design of interventions in epidemics. The algorithm is described in the more general setting of partially observed Markov decision problems, and can potentially be applied in other areas such as clinical trial design, control and economics.The emergence of new diseases, either naturally or as the result of a biological attack, is one of the most important threats to national security; early detection and adequate intervention are key to saving lives and minimizing the damage caused by such an outbreak. However, early detection can be extremely difficult when symptoms are similar to those of diseases already endemic in the population (for example, fever and sore throats which are common symptoms of colds and influenza, are also symptoms of respiratory Anthrax), and a detailed analysis of the space-time patterns is necessary in order to be able to separate the presence of a new disease from the random fluctuations inherent to the dynamics of the preexisting disease. Similarly, the design of interventions (vaccination, quarantine and culling, just to mention three possibilities) is impaired in this setting by the lack of knowledge about the infectiousness of the new disease and the efficacy of available vaccines. This research develops new algorithms for disease monitoring and outbreak management under uncertainty that have the potential to greatly improve the ability of government and international agencies to identify and intervene in outbreaks of emerging infectious diseases.