RAPID: Data-driven Understanding of Imperfect Protection for Long-term COVID-19 Projections

该项目将利用COVID-19再感染和疫苗突破的数据，建立一个模型，研究不完善的免疫如何影响SARS-CoV-2病原体的传播动态，以及随后对病例数、死亡人数和住院人数的影响。决定COVID-19长期动态的一个关键因素是人群对COVID-19的免疫力如何随时间和暴露而变化。美国各州和世界各国关于疫苗接种突破和再感染的数据为研究人口水平上的免疫力下降动态提供了独特的机会。该项目将有助于了解COVID-19卷土重来的长期风险和严重程度，为美国COVID-19情景建模中心、美国COVID-19预测中心和欧洲COVID-19预测和情景建模中心做出贡献，从而为全球决策者提供信息。PI将整合本科编程课程和研究生健康机器学习课程的经验。该项目还将通过一个高级顶点项目和一个少数族裔服务项目提供研究和培训机会。在存在多种变体的情况下，该模型将代表一类不完善的保护。流行的模型，如全有或全无、泄漏和时间依赖性减弱，将与可解释的机器学习模型一起考虑。这些模型将通过它们在持有数据上的“普遍性”得到验证。将制定统一的免疫模型，使其能够与各种流行病学模型相结合。作为演示，它将与一个模型集成，该模型可以跟踪个人可能处于的各种状态，包括所有感染、再感染、一剂、两剂和增强剂的排列。在给定的不完全保护模型中，随着时间、年龄组和变体的不同，有了这些状态，就可以精确计算特定时间人群的免疫力。总体方法还将通过其产生的美国州级病例、死亡和住院预测的准确性来评估。该项目是与美国疾病控制与预防中心合作资助的，旨在支持快速反应研究项目，进一步提高联邦传染病建模能力。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。

This project will use data on COVID-19 reinfections and vaccine breakthroughs to build a model of how imperfect immunity affects SARS-CoV-2 pathogen transmission dynamics and subsequent effects on numbers of cases, deaths, and hospitalizations. A key factor dictating the long-term dynamics of COVID-19 is how population immunity against COVID-19 changes over time and exposure. Data on vaccination breakthroughs and reinfections in various states of the US and countries around the world create a unique opportunity to study immunity waning dynamics at the population level. The project will help understand the long-term risks of resurgence and severity of COVID-19, contribute to the US COVID-19 Scenario Modeling Hub, the US COVID-19 Forecast Hub, and the European COVID-19 Forecast and Scenario Modeling Hubs, and thus inform policymakers worldwide. The PI will integrate the lessons learned in an undergraduate on programming and a graduate-level class on Machine Learning for health. The project will also provide research and training opportunities through a senior capstone program and a minority-serving program. The model will represent a class of imperfect protection in the presence of multiple variants. Popular models such as all or nothing, leaky, and time-dependent waning will be considered along with interpretable machine learning models. The models will be validated by their “generalizability” on held-out data. The unified model of immunity will be developed in a way that it can be integrated with various epidemiological models. As a demonstration, it will be integrated with a model that tracks various states an individual can be in, including all permutations of infections, reinfections, one-dose, two-doses, and boosters. Having these states over time, age groups, and variants, for a given model of imperfect protection allows for precise computation of immunity in the population at a given time. The overall approach will also be evaluated by the accuracy of US state-level cases, deaths, and hospitalization forecasts it produces. This project was funded in collaboration with the CDC to support rapid-response research projects to further advance federal infectious disease modeling capabilities.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.