Development of an Open-Source and Data-Driven Modeling Platform to Monitor and Forecast Disease Activity

项目总结 对传染病活动进行可靠和实时的市级预测建模和预测 改变公共卫生决策者设计信息等干预措施的潜力 在世界各地存在健康威胁的情况下，开展运动、预防性/反应性疫苗接种和病媒控制 世界。虽然疾病活动与以下因素之间的联系：人类流动性、气候和 环境因素、社会经济决定因素和社交媒体活动在中国早已为人所知 在流行文献中，很少有人关注开发开源平台的明显需要 能够利用多个数据源、因素和不同的建模方法 和异质国家，以监测和预测疾病传播，在四个地理范围(国家， 州、市和市)。这个项目的总体目标就是开发这样一个平台。 我们的长期目标是探索有效的方法来整合来自多个不同研究的结果 关于全球疾病动态与当地和全球因素，如天气条件，社会和 经济状况、卫星图像和在线人类行为，以开发可操作的、稳健的和真实的- 时间数据驱动的疾病预测平台。 这笔赠款的目的是利用三个互补的科研团队的专业知识。 以及来自各种数据源的丰富信息，以构建一个能够 结合信息，在地方一级产生实时的短期疾病预测。作为这项工作的一部分，我们 将评估用于监控和预测的不同数据流和建模方法的预测能力 疾病在多个地理范围--国家、州、市和直辖市--以巴西为测试案例。 此外，我们将使用机器学习和机械模型来了解疾病动力学 多个空间尺度，横跨巴西这样的异质国家。 我们的具体目标是：(1)评估单个数据流和疾病建模技术的实用性 预测；(2)融合建模技术和数据流，以提高四个方面的准确性和稳健性 空间尺度；(3)描述构建运营性网络所需的基本计算基础设施 疾病预测平台；以及(4)在真实环境中验证我们的方法。 这一贡献意义重大，因为它将促进我们对科学知识的准确性和 用于疾病预测时不同数据流和多种建模方法的局限性 变速箱。我们的努力将有助于在地方一级(城市-- 和市级)。此外，我们的目标是为 流行病学社区将其用作知识发现工具。最后，我们的目标是开发这个平台 在由世卫组织、疾控中心、学术界和地方政府组成的主题专家小组的指导下 以及巴西境内的联邦利益相关者。 提议的方法是创新的，因为很少有人致力于开发开放源码的 能够跨异类数据源和驱动程序组合不同数据源的计算平台 和大国，进入监测和预测疾病传播的多种建模方法，结束 多个地理比例..此外，我们还建议研究如何最好地组合建模方法 到目前为止，已经被独立地发展和解释，即传统的流行病学 机械模型和新的机器学习预测模型，以产生准确和健壮的 实时疾病活动估计和预测。

PROJECT SUMMARY Reliable and real-time municipality-level predictive modeling and forecasts of infectious disease activity have the potential to transform the way public health decision-makers design interventions such as information campaigns, preemptive/reactive vaccinations, and vector control, in the presence of health threats across the world. While the links between disease activity and factors such as: human mobility, climate and environmental factors, socio-economic determinants, and social media activity have long been known in the epidemic literature, few efforts have focused on the evident need of developing an open-source platform capable of leveraging multiple data sources, factors, and disparate modeling methodologies, across a large and heterogeneous nation to monitor and forecast disease transmission, over four geographic scales (nation, state, city, and municipal). The overall goal of this project is to develop such a platform. Our long-term goal is to investigate effective ways to incorporate the findings from multiple disparate studies on disease dynamics around the globe with local and global factors such as weather conditions, socio- economic status, satellite imagery and online human behavior, to develop an operational, robust, and real- time data-driven disease forecasting platform. The objective of this grant is to leverage the expertise of three complementary scientific research teams and a wealth of information from a diverse array of data sources to build a modeling platform capable of combining information to produce real-time short term disease forecasts at the local level. As part of this, we will evaluate the predictive power of disparate data streams and modeling approaches to monitor and forecast disease at multiple geographic scales--nation, state, city, and municipality--using Brazil as a test case. Additionally, we will use machine learning and mechanistic models to understand disease dynamics at multiple spatial scales, across a heterogeneous country such as Brazil. Our specific aims will (1) Assess the utility of individual data streams and modeling techniques for disease forecasting; (2) Fuse modeling techniques and data streams to improve accuracy and robustness at the four spatial scales; (3) Characterize the basic computational infrastructure necessary to build an operational disease forecasting platform; and (4) Validate our approach in a real-world setting. This contribution is significant because It will advance our scientific knowledge on the accuracy and limitations of disparate data streams and multiple modeling approaches when used to forecast disease transmission. Our efforts will help produce operational and systematic disease forecasts at a local level (city- and municipality-level). Moreover, we aim at building a new open-source computational platform for the epidemiological community to use as a knowledge discovery tool. Finally, we aim at developing this platform under the guidance of a Subject Matter Expert (SME) panel comprising of WHO, CDC, academics, and local and federal stakeholders within Brazil. The proposed approach is innovative because few efforts have focused on developing an open-source computational platform capable of combining disparate data sources and drivers, across a heterogeneous and large nation, into multiple modeling approaches to monitor and forecast disease transmission, over multiple geographic scales.. In addition, we propose to investigate how to best combine modeling approaches that have, to this date, been developed and interpreted independently, namely, traditional epidemiological mechanistic models and novel machine-learning predictive models, in order to produce accurate and robust real-time disease activity estimates and forecasts.