A novel framework for calibration and prediction of stochastic compartmental transmission dynamic models of novel pathogens

在新病原体最初出现的时期，准确估计关键的流行参数和预测未来的流行轨迹是具有挑战性的，因为观察只能部分反映疫情的真实状态。考虑到感染个体数量较少时偶发事件的重要性，随机传播动力学模型特别适用于新病原体出现期间的推断和预测。我的研究目标是开发和评估一个框架，用于使用观察到的但可能不完美的流行病数据对随机间隔模型进行实时校准。这一框架将使我能够估计关键的流行病参数，例如继发病例的数量或传染性的平均持续时间以及目前受感染的人数。此外，我们的框架将允许短期预测(例如，下周的预期诊断或未来三周的累积诊断)和长期预测(例如，总体发病率)。我的实时校准框架是基于我在系统生物学背景下开发的目标函数，并成功地应用于高度非线性的随机信号通路。在目前我和我的学术主持人一起做的工作中，我将这个目标函数纳入贝叶斯框架，并使用模拟研究来研究其估计关键流行病参数和预测流行病时间过程的能力。我已经提交了一份手稿，将我的方法与最先进的方法进行了比较。我能够证明，我的方法在推理和预测方面都优于当前的基准方法。如果获得资金，我将：1)将这一框架应用于最近的流感监测数据，并进行回溯性校准，并评估该方法的预测能力。这将使我们能够在真实数据上评估该方法的准确性；2)通过回溯预测，将我的框架在相同数据集上的性能与其他最先进的校准和预测方法进行比较；以及3)通过参加由美国疾病控制和预防中心(US CDC)主办的疫情预测倡议竞赛，使用该框架对下一季(即2016-2017)流感疫情进行实时预测。虽然我们写的这个项目是针对流感的特定应用，但我们的框架并不局限于特定的病原体。全球社会被迫面对传染病病原体出现或重新出现的频率表明，早期和准确预测流行病行为的改进方法对公共卫生和经济产生了巨大影响。这些方法将提高政策制定者更好地利用现有资源控制流行病和平衡重大疫情风险与社会和经济成本干预的能力。

During the period of initial emergence of novel pathogens, the accurate estimation of key epidemic parameters and prediction of future epidemic trajectory is challenging because observations only partially reflect the true state of the epidemic. Stochastic transmission dynamic models are especially useful for inference and projections during the emergence of novel pathogens given the importance of chance events when the number of infectious individuals is small. My research objective is to develop and evaluate a framework for real-time calibration of stochastic compartmental models using observed, but likely imperfect, epidemic data. This framework will allow me to estimate key epidemic parameters such as the number of secondary cases or the mean duration of infectiousness as well as the number of currently infected persons. Additionally, our framework will permit both short-term predictions (e.g the expected diagnoses in the next week or cumulative diagnoses over the next three weeks) and longer-term predictions (e.g the overall attack rate). My framework for real-time calibration is based on an objective function that I developed in a Systems Biology context and successfully applied to highly nonlinear stochastic signaling pathways. In current work I have been doing together with my academic host, I incorporated this objective function into a Bayesian framework and used simulation studies to investigate its capacity to estimate key epidemic parameters and to predict the epidemic time course. I have submitted a manuscript that compares my approach with state of the art methods. I was able to show that my method outperforms these current benchmark approaches for both inference and prediction. If funded, I will: 1) apply this framework to recent influenza surveillance data and perform a retrospective calibration and evaluate the predictive ability of this method. This will allow us to evaluate the accuracy of the method on real data; 2) compare my framework's performance to other state-of-the-art calibration and prediction methods on the same data sets by retrospective prediction; and 3) use the framework for real-time forecast on next season's (i.e. 2016-2017) influenza epidemic by participating in, the Epidemic Prediction Initiative Competition hosted by the US Center for Disease Control and Prevention (US CDC). While we have written this project with a specific application for influenza, our framework is not limited to a specific pathogen. The frequency at which the global community has been forced to confront the emergence or reemergence of infectious pathogens suggests a large public health and economic impact of improved approaches for early and accurate prediction of epidemic behavior. Such methods will improve the capacity of policy makers to better use existing resources to control epidemics and to balance the risk of major outbreaks with the social and economic costs interventions.