RAPID: Computational Modeling of Contact Density and Outbreak Estimation for COVID-19 Using Large-scale Geolocation Data from Mobile Devices

2019冠状病毒病的爆发突出表明，新出现的大流行病的全球风险日益增加，迫切需要加强数据驱动的工具，以识别、控制和减轻其影响，特别是在人口稠密的城市地区。人们越来越关注智能手机的位置数据，将其作为加强流行病学建模和预测疫情进展、传播和暴露风险的一种方式。当与人工智能或机器学习算法相结合时，这些高分辨率数据有可能大大提高感染和住院估计的粒度和精度。然而，位置数据的使用引发了严重的社会、伦理和技术挑战。在当前大流行的背景下，已经开始认真讨论潜在的公共卫生利益与对隐私和公民自由的影响之间的权衡，特别是考虑到私营公司越来越多地使用这些数据来宣传有针对性的广告、评估零售消费者行为和模拟旅行需求以及其他应用。此外，由于对这些数据的代表性和偏见的不完全理解，特别是对代表性不足和弱势社区的数据，这些数据在公共利益方面的使用受到了破坏。目前尚不清楚的是位置数据的这种偏差程度，以及使用这些数据的公共卫生效益如何随着空间和时间的聚集而减少，这可能有助于最大限度地减少收集和使用这些数据时的隐私问题。为了解决这些问题，该项目将开发基于大规模位置数据的计算模型，以(1)估计时间范围（小时、每日等）和空间范围（人口普查街区、社区等）的暴露密度，这将使官员和研究人员能够评估和预测特定地区的传播率；(2)衡量和评估社会（物理）距离努力的程度和有效性，并在县市内部和城市之间进行比较，并了解对脆弱社区和人群的不同影响；(3)基于邻里和社区之间的运动和旅行模式来衡量疾病传播的程度，这将支持疾病爆发的时空模式预测，并根据已确定或怀疑感染的地区的聚集流动轨迹确定“危险”地点。项目团队特别关注就地安置令和暴露风险如何不成比例地影响低收入和少数族裔社区，以及在评估社会经济差异时地点数据可能存在的偏见。该项目将评估这些模型对流行病学家和公共卫生官员的有用性如何随空间聚合（例如，社区级数据是否优于县级数据）和时间聚合（例如，近实时模型是否优于每日或每周时间尺度）而变化，并提供定量绩效评估，可用于在健康效益和隐私风险之间权衡的集体决策。项目产出将酌情开放资源并向公众提供。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The outbreak of COVID-19 has highlighted both the growing global risk of emerging pandemics and the urgent need for enhanced data-driven tools to identify, contain, and mitigate their effects, particularly in dense urban areas. There has been increasing attention given to locational data from smartphones as a way to enhance epidemiological modeling and predict outbreak progression, transmission, and exposure risk. When combined with artificial intelligence or machine learning algorithms, these high resolution data have the potential to vastly improve the granularity and precision of infection and hospitalization estimates. However, the use of locational data raises serious social, ethical, and technical challenges. Trade-offs between the potential public health benefits and the impacts for privacy and civil liberties have started to be debated in earnest within the context of the current pandemic, especially in light of increasing use of these data by private companies to promote targeted advertisements, evaluate retail consumer behavior, and model travel demand, among other applications. Furthermore, the use of these data in the public interest is undermined by an incomplete understanding of the representativeness and bias embedded in these data, particularly in relation to under-represented and vulnerable communities. What is not yet known is the extent of this bias in locational data and how the public health benefits of using these data diminish with spatial and temporal aggregation, which could help to minimize privacy concerns in the collection and use of these data. To address these questions, this project will develop computational models derived from large-scale locational data to (1) estimate the exposure density across a range of temporal (hourly, daily, etc.) and spatial (census block, neighborhood, etc.) scales, which will enable officials and researchers to evaluate and predict transmission rates in a particular area; (2) measure and evaluate the extent and effectiveness of social (physical) distancing efforts over time and comparatively within and across counties and cities, as well as understand the disparate impacts on vulnerable communities and populations; and (3) measure the extent of disease spread based on movement and travel patterns between neighborhoods and communities, which will support predictions of the spatial-temporal patterns of disease outbreak and identify “at-risk” locations based on the aggregated mobility trajectories for areas were infections have been identified or suspected. The project team is particularly concerned with how shelter-in-place orders and exposure risk disproportionately impact low-income and minority communities, and the implications of potential bias in locational data in assessing socioeconomic variations. The project will assess how the usefulness of these models for epidemiologists and public health officials varies with spatial aggregation (e.g. is neighborhood level data superior to county level data) and temporal aggregation (e.g. is a near-real-time model superior to daily or weekly timescales) and provide quantitative performance assessments that can be used for collective decision-making on the trade-offs between health benefits and privacy risk. Project outputs will be made open-source and publicly available as appropriate.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.