RAPID: COVID-19 comparative modeling and analyses of outbreaks using mechanistic and ensemble machine learning and the development of a platform for projection and management

点击翻译按钮获取中文摘要

Numerical projections of reliable infection and disease mortality rates within cities, counties, states, and countries, as well as identification of the factors most responsible for these rates are critical to rational management of the ongoing Covid-19 pandemic. In the absence of effective therapeutics and vaccines, a deeper understanding of the impact of societal measures (distancing, contact tracing, quarantining, etc.) on local Covid-19 outbreaks are needed by administrators and healthcare professionals in making decisions that affect the tradeoff between the physical health of individuals and the economic health of communities. The aims of the proposal are twofold. First, to use cutting-edge statistical models to uncover the factors that most affect SARS-CoV-2 transmission and mortality rates. Second, to provide decisions makers with a simple-to-use, extensive instruction supported, data and scenario analysis (DASA) platform for evaluating the implications of different policy measures, including the implementation and relaxation of social distancing behavior, surveillance, contact tracing, patient isolation, and vaccination (once suitable vaccines are available). Additionally, this DASA platform will be suitable for training students at the undergraduate and graduate levels in public health and allied programs, as well as providing an analytical tool for students carrying out epidemiological research.The epidemiological model that underpins the Numerus Model Builder DASA Covid-19 platform includes modifications of the standard SEIR (Susceptible, Exposed/Latent, Infectious, Recovered) formulation to incorporate an explicit contact (C) class, as well as dividing infectious individuals into pre/asymptomatic (A) and symptomatic infectious (I) disease states to yield a SCLAIV model (where V refers to naturally vaccinated/recovered class). Individuals in the C class can either thwart (return to the S class) or succumb to (move onto the L=E class) pathogen invasion after making contact with the SARS-CoV-2 pathogen. The formulation also includes a parallel series of Sr, Cr, Lr, Ar, Ir and Vr classes that correspond to individuals moving into these reduced-exposure SCLAIV-response classes at rates determined by the driving actions of Covid-19 policy measures that have been put in place. The values of the SCLAIV+reponse model parameters are influenced by various factors that will be identified using statistical machine learning methods. In particular, “superlearners” that are a mix of parametric and nonparametric ensemble machine learning methods will be used to identify the factors responsible for observed spatio-temporal patterns across different, regional Covid-19 outbreaks, using appropriate data scraped from the worldwide web. This RAPID award is made by the Ecology and Evolution of Infectious Disease Program in the Division of Environmental Bioloy, using funds from the Coronavirus Aid, Relief, and Economic Security (CARES) Act.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.