Data, Analytics and Surveillance; UK Health Security Agency; Porton Down, United Kingdom.
PLoS Comput Biol. 2024 Sep 6;20(9):e1010817. doi: 10.1371/journal.pcbi.1010817. eCollection 2024 Sep.
Reverse epidemiology is a mathematical modelling tool used to ascertain information about the source of a pathogen, given the spatial and temporal distribution of cases, hospitalisations and deaths. In the context of a deliberately released pathogen, such as Bacillus anthracis (the disease-causing organism of anthrax), this can allow responders to quickly identify the location and timing of the release, as well as other factors such as the strength of the release, and the realized wind speed and direction at release. These estimates can then be used to parameterise a predictive mechanistic model, allowing for estimation of the potential scale of the release, and to optimise the distribution of prophylaxis. In this paper we present two novel approaches to reverse epidemiology, and demonstrate their utility in responding to a simulated deliberate release of B. anthracis in ten locations in the UK and compare these to the standard grid-search approach. The two methods-a modified MCMC and a Recurrent Convolutional Neural Network-are able to identify the source location and timing of the release with significantly better accuracy compared to the grid-search approach. Further, the neural network method is able to do inference on new data significantly quicker than either the grid-search or novel MCMC methods, allowing for rapid deployment in time-sensitive outbreaks.
反向流行病学是一种数学建模工具，用于确定病原体的来源信息，前提是病例、住院和死亡的空间和时间分布。在故意释放病原体的情况下，例如炭疽芽孢杆菌（炭疽病的病原体），这可以使响应者快速识别释放的位置和时间，以及其他因素，例如释放的强度，以及释放时的实际风速和风向。这些估计值可用于参数化预测性机械模型，从而估计释放的潜在规模，并优化预防措施的分配。在本文中，我们提出了两种新颖的反向流行病学方法，并演示了它们在应对英国十个地点模拟故意释放炭疽芽孢杆菌中的效用，并将其与标准网格搜索方法进行了比较。这两种方法——改进的 MCMC 和递归卷积神经网络——能够以比网格搜索方法更高的准确性识别释放的源位置和时间。此外，神经网络方法能够比网格搜索或新颖的 MCMC 方法更快地对新数据进行推断，从而可以在时间敏感的疫情中快速部署。
