Statistical estimation and decision-making for the COVID-19 pandemic

There are multiple sources of data giving information about the number of
SARS-CoV-2 infections in the population, but all have major drawbacks, including
biases and delayed reporting. Representative random prevalence surveys, the only
putatively unbiased source, are sparse in time and space, and the results can come
with big delays. Reliable estimates of population prevalence are necessary for
understanding the spread of the virus and the effectiveness of mitigation strategies. We
develop a Bayesian framework to estimate viral prevalence by combining several
of the main available data sources. It is based on a discrete-time
Susceptible–Infected–Removed (SIR) model with time-varying reproductive parameter.
Our model includes likelihood components that incorporate data on deaths due to the
virus, confirmed cases, and the number of tests administered on each day. We anchor
our inference with data from random-sample testing surveys in Indiana and Ohio. We
use the results from these two states to calibrate a model on positive test counts and
proceed to estimate the infection fatality rate and the number of new infections on each
day in each state in the United States between March 2020 and March 2021. We
estimate the extent to which reported COVID cases have underestimated true infection
counts, which was large, especially in the first months of the pandemic. 

Building on this work, we consider decision-making vis-à-vis non-pharmaceutical
interventions (NPIs) – which include mask mandates, social distancing, and workplace
and school closures, among other policies. NPIs have proven to be effective tools for
mitigating the spread of COVID-19 over the course of the pandemic. However, the
individual effects of NPIs on viral transmission remain uncertain, which complicates
decision-making concerning which policies to implement and when to loosen or tighten
restrictions. Indeed, initial attempts to quantify the effects of NPIs were hampered by
high correlation in the implementation of interventions early in the pandemic. With more
data, subsequent studies have been able to disentangle the effects of NPIs at various
geographic scales, although with substantial uncertainty remaining. Furthermore, the
economic and social costs incurred by these interventions can be significant, including
disrupted economic output, job losses, and learning loss.

We take a statistical data-driven approach to decision-making during the pandemic that
weighs the effects of NPIs against their costs, in combination with a mechanistic
epidemiological model, to navigate the economic, social, and public health trade-offs of
COVID restrictions. We develop a Bayesian hierarchical epidemiological model to
estimate the impacts of NPIs on SARS-CoV-2 transmission in the United States. Our
model combines state-level data on cases and deaths due to COVID and intervention
policies implemented over time in all 50 states and D.C. We link these data – which
track SARS-CoV-2 transmission and government intervention policies, respectively –
via a regression function nested within a mechanistic epidemiological model simulating
viral spread. We take a Bayesian approach to account for uncertainty in
epidemiological parameters. Our model is able to capture the complex temporal
dynamics of SARS-CoV-2 transmission, providing validation for its use in NPI policy
planning. Going further, we combine this model with estimates of the costs of
interventions from the literature to quantitatively evaluate those policies that have been implemented during the pandemic and to formulate more effective and less costly NPI
strategies.