RAPID: Retrospective COVID-19 Scenario Projections Accounting for Population Heterogeneities

COVID-19的长期负担可能因种族和族裔而异。为了解决这一变化，该项目将扩展当前的模型，以考虑种族和民族。在美国，按种族/民族划分的结果和疫苗摄取数据的可用性创造了一个机会，可以明确地对这些变量进行跨组建模，并评估来自现实世界数据的结果。该项目将帮助我们了解COVID-19结果和疫苗接种的不公平现象，并帮助美国为未来的COVID-19和其他疫情做好准备。只要有有关族裔和种族的数据，该项目就有可能适用，并可扩展到其他类型的群体。该项目将整合本科编程课程和研究生健康机器学习课程的经验教训。该项目还将通过一个高级顶点项目和少数族裔服务项目（如南加州大学JumpStart项目和维特比暑期学院）提供研究机会。拟议的项目将整合关于种族和族裔的数据以及各种其他数据集，以说明人口健康。集成的关键创新是能够从数据中学习接触矩阵。该项目将使用一种新颖的方法，其中n×n接触矩阵由n个隐藏参数生成，这些参数表示一个群体与随机选择的个体接触的可能性。所学接触矩阵将与PI目前在美国情景建模中心使用的流行病学模型相结合，以生成病例、死亡和住院的长期预测。该方法将学习接触矩阵与从调查数据和高分辨率移动数据中导出这些矩阵的其他方法进行比较。新方法将使亚种群的相互作用建模时，这种流动性数据是不可用的。该模型将与COVID-19情景建模中心合作，利用过去三年观测到的真实数据进行评估。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 long-term burden of COVID-19 may vary across races and ethnicities. To address this variaiton this project will extend a current model to account for race and ethnicity. The availability of outcomes and vaccine uptake data by race/ethnicity in the US creates an opportunity to explicitly model these variables across the groups and evaluate the results from real-world data. The project will help us understand the inequities of COVID-19 outcomes and vaccination uptake and prepare the US for the future of COVID-19 and other outbreaks. The project has the potential to be applicable wherever relevant data on ethnicity and race is available, and can be extended to other types of groups. The project will integrate the lessons learned in an undergraduate course on programming and a graduate-level class on Machine Learning for health. The project will also provide research opportunities through a senior capstone program and minority-serving programs such as the USC JumpStart program and the Viterbi Summer Institute.The proposed project will integrate data on race and ethnicity along with various other datasets to account for population health. The key innovation in the integration is the ability to learn contact matrices from data. The project will use a novel approach, where the n×n contact matrix is generated by n hidden parameters that indicate the likelihood of contact of a group with a randomly selected individual. The learned contact matrix will be integrated with an epidemiological model currently being used by the PI in the US Scenario Modeling Hub to generate long-term projections of cases, deaths, and hospitalization. The appoach will compare learning contact matrices with other approaches that derive those matrices from survey data and high-resolution mobility data. The new approach will enable the modeling of sub-population interactions when such mobility data is not available. The model will be evaluated with ground truth data observed over the last three years in collaboration with the COVID-19 Scenario Modeling Hub.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.