Collaborative Research: IHBEM: Data-driven multimodal methods for behavior-based epidemiological modeling

在这个项目中，通过开发一个统一的建模框架来解决基于行为的流行病学建模的挑战，该框架包含了整合新数据源、扩展流行病学模型和政策干预评估的新方法。捕捉人类行为既复杂又具有挑战性，因为新的和意想不到的行为模式不断出现。这在流行病期间尤其明显，因为流行病是由各种各样的行为形成的，反过来又加速了行为变化的速度。例如，COVID-19大流行的轨迹在很大程度上受到社交距离、佩戴口罩和接种疫苗等行为的影响，这些行为在大流行期间也随着疾病风险、社会规范、政府决策和激励措施的变化而出现并发生了巨大变化。该项目的目的是开发流行病学模型，根据现实世界的行为进行预测，并捕捉行为、流行病和政府决策之间的反馈循环，从而实现更有效的公共卫生决策。该项目的目标是通过改进处理现实世界流行病的数学和机器学习方法，并引入捕捉现实世界人类行为和将行为反应整合到流行病学模型中的新方法来实现的。首先，提出了从多模态、真实世界的传感器（如手机和搜索引擎日志）中去噪和提取意义的新方法，这些传感器的数据通常非常不完美，但却提供了捕捉人类行为的独特机会。这使得实时捕捉复杂的人类行为成为可能。其次，为了连接流行病学模型和现实世界行为，基于主体的疾病动力学模型与人类行为模型相结合，这些模型捕捉个人如何根据感知的成本和收益选择行为。这种模型在计算上很复杂，需要新的方法来校准和验证实际数据，以便能够对流行病进行现实的预测。最后，为了评估公共卫生决策对行为和流行病结果的复杂影响，开发了新的情景建模工具和因果推理方法，以估计存在混杂因素和干扰的情况下此类决策的影响。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。

In this project, challenges of behavior-based epidemiological modeling are addressed by developing a unified modeling framework that incorporates new methods for incorporating novel data sources, extended epidemiological models, and evaluations of policy interventions. Capturing human behavior is complex and challenging, as new and unexpected behavioral patterns emerge constantly. This is especially evident during epidemics, which are shaped by a wide variety of behaviors and, in turn, accelerate the speed of behavioral changes. For example, the trajectory of the COVID-19 pandemic was greatly shaped by behaviors such as social distancing, mask-wearing, and vaccination, and these behaviors also emerged and changed dramatically over the course of the pandemic in response to changing disease risks, social norms, government decisions, and incentives. The aim of this project is to develop epidemiological models that can make predictions based on real-world behaviors and capture feedback loops between behaviors, epidemics, and government decisions, thus enabling more effective public health decisions.The aims of this project are accomplished by improving mathematical and machine learning methods for dealing with real-world epidemics and introducing novel approaches to the capture of real-world human behavior and integration of behavioral responses into epidemiological models. First, novel methods are proposed to denoise and derive meaning from multimodal, real-world sensors, such as mobile phones and search engine logs, the data from which is often highly imperfect but which provide unique opportunities to capture human behavior. This allows the capture of complex human behaviors in real time. Second, to bridge epidemiological models and real-world behaviors, agent-based models of disease dynamics are coupled with models of human behavior that capture how individuals choose behaviors based on perceived costs and benefits. Such models are computationally complex and require new methods to calibrate and validate on real data to enable realistic forecasting of epidemics. Finally, to evaluate the complex effects of public health decisions on behavior and epidemic outcomes, new scenario modeling tools and causal inference methods are developed to estimate effects of such decisions in the presence of confounders and interference.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.