The challenges of modeling and forecasting the spread of COVID-19.

Bertozzi AL, Franco E, Mohler G, Short MB, Sledge D. · 2020

Read The challenges of modeling and forecasting the spread of COVID-19. by Bertozzi AL, Franco E, Mohler G, Short MB, Sledge D. (2020) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc7382213.txt.

Introduction

Bertozzi and co-authors lay out three regional models for forecasting COVID-19 in the Proceedings of the National Academy of Sciences. Pandemic modeling sat at the center of policy in 2020, yet forecasting a new virus stayed hard. The authors favor deliberately simple models over elaborate ones.

Their three models connect to each other and to a region's case data, letting officials test what distancing does to the curve. The results warn against relaxing non-drug measures while no vaccine or antiviral exists, since spread rebounds. The paper doubles as an accessible framework for generating policy-relevant insight. Readers get a clear look at how modest mathematics can guide big decisions under deep uncertainty.

Catalog record

CREATORBertozzi AL, Franco E, Mohler G, Short MB, Sledge D.
YEAR2020
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC7382213

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The challenges of modeling and forecasting the spread of COVID-19 By Andrea L. Bertozzi, Elisa Franco, George Mohler, Martin B. Short, Daniel Sledge Published in: Proceedings of the National Academy of Sciences of the United States of America DOI: 10.1073/pnas.2006520117 ABSTRACT Significance The coronavirus disease 2019 (COVID-19) pandemic has placed epidemic modeling at the forefront of worldwide public policy making. Nonetheless, modeling and forecasting the spread of COVID-19 remain a challenge. Here, we present and detail three regional-scale models for forecasting and assessing the course of the pandemic. This work is intended to demonstrate the utility of parsimonious models for understanding the pandemic and to provide an accessible framework for generating policy-relevant insights into its course. We show how these models can be connected to each other and to time series data for a particular region. Capable of measuring and forecasting the impacts of social distancing, these models highlight the dangers of relaxing nonpharmaceutical public health interventions in the absence of a vaccine or antiviral therapies. The world is in the midst of an ongoing pandemic, caused by the emergence of a novel coronavirus. Pharmaceutical interventions such as vaccination and antiviral drugs are not currently available. Over the next year, addressing the coronavirus disease 2019 (COVID-19) outbreak will depend critically on the successful implementation of public health measures including social distancing, shelter in place orders, disease surveillance, contact tracing, isolation, and quarantine (1, 2). On 16 March, Imperial College London released a report (3) predicting dire consequences if the United States and the United Kingdom did not swiftly take action against the pandemic. In both nations, governments responded by implementing more stringent social distancing regulations (4). We now have substantially more case data from the United States, as well as the benefit of analyses performed by scientists and researchers across the world (5–12). Nonetheless, modeling and forecasting the spread of COVID-19 remain a challenge. Here, we present three basic models of disease transmission that can be fit to data emerging from local and national governments. While the Imperial College study employed an agent-based method (one that simulates individuals getting sick and recovering through contacts with other individuals in the population), we present three macroscopic models: 1) exponential growth, 2) self-exciting branching process, and 3) the susceptible–infected–resistant (SIR) compartment model. These models have been chosen for their simplicity, minimal number of parameters, and for their ability to describe regional-scale aspects of the pandemic. In presenting these models, we demonstrate how they are connected and note that in different cases one model may fit better than another. Because these models are parsimonious, they are particularly well suited to isolating key features of the pandemic and to developing policy-relevant insights. We order them according to their usefulness at different stages of the pandemic—exponential growth for the initial stage, a self-exciting branching process when one is still analyzing individual count data going into the development of the pandemic, and a macroscopic mean-field model going into the peak of the disease. The branching process can also track changes over time in the dynamic reproductive number. These models highlight the significance of fully implemented and sustained social distancing measures. Put in place at an early stage, distancing measures that reduce the virus’s reproduction number—the expected number of individuals who an infected person will spread the disease to—may allow much-needed time for the development of pharmaceutical interventions. By slowing the speed of transmission, such measures may also reduce the strain on health care systems and allow for higher-quality treatment for those who become infected. Importantly, the economic consequences of such measures may lead political leaders to consider relaxing them. The models presented here, however, demonstrate that relaxing these measures in the absence of pharmaceutical interventions may allow the pandemic to reemerge. Where this takes place, social distancing efforts that appear to have succeeded in the short term will have little impact on the total number of infections expected over the course of the pandemic. The epidemiological perspective on modeling infectious disease spread involves consideration of a larger number of modeling parameters detailing the spread of and recovery from the disease, additional compartments corresponding to age categories, and other related choices (e.g., refs. 3 and 13). A data-driven approach to modeling COVID-19 has also emerged, in which statistical and machine learning models are used for forecasting cases, hospitalizations, deaths, and impacts of social distancing (14, 15). Our work demonstrates the utility of parsimonious epidemic models for understanding the pandemic and provides an accessible framework for a larger group of quantitative scientists to follow and forecast the COVID-19 pandemic. It includes explanations that will help allow scientific researchers to develop insights that may contribute to public health policy making, including contributing to public health forecasting teams. Importantly, the branching process model that we detail is relatively new and underutilized in epidemiology. It provides a method for quantitatively estimating dynamic reproduction numbers, which can be critical in assessing the impact of distancing measures over time. The results from the parsimonious models presented here are consistent with recent analyses from public health officials in California (16) and with the original Imperial College model (3). We present examples of forecasts for viral transmission in the United States. Where other studies have typically developed and presented one model (choosing to fit parameters within the chosen model), our analysis compares three different forecasting models using a fitting criterion. The results of these models differ depending on whether the data employed cover confirmed cases or mortality. In addition, many aspects of disease spread, such as incubation periods, fraction of asymptomatic but contagious individuals, seasonal effects, and the time between severe illness and death, are not considered here. In some cases (e.g., seasonal), relevant data do not exist, and in other cases (e.g., age of patients), we choose not to include additional parameters in favor of parsimony. RESULTS A. Exponential Growth. Epidemics naturally exhibit exponential behavior in the early stages of an outbreak, when the number of infections is already substantial but recoveries and deaths are still negligible. If at a given time t there are I(t) infected individuals and α is the rate constant at which they infect others, then at early times (neglecting recovered individuals), I(t)=I0eαt. The time it takes to double the number of cumulative infections (doubling time) is a common measure of how fast the contagion spreads: if we start with Ī infections, then at time Td=ln2/α we achieve 2Ī infections. For the COVID-19 outbreak, exponential growth is seen in data from multiple countries (Fig. 1), with remarkably similar doubling times in the early stages of the epidemic. For COVID-19, we expect an exponential growth phase during the first 15 to 20 d of the outbreak in the absence of public health interventions such as social distancing, isolation, or quarantine. This estimate is based on patient data from the Wuhan outbreak, which indicate that the average time from illness onset to death or discharge is between 17 and 21 d for hospitalized patients (20, 21). Because deaths are a fraction of infections, they initially increase at a similar exponential pace, with some delay relative to the beginning of the outbreak. These observed doubling time estimates (2 to 4 d) are significantly smaller

Read the complete text