Until effective COVID-19 treatments and/or vaccines are available, contact tracing will play a major role in many U.S. cities’ and states’ plans to resume some semblance of normal life. With so much public attention focusing on contact tracing as a path forward, department faculty Aaron Scheffler, PhD, MS, John Kornak, PhD, and Travis Porco, PhD, MPH, of the Proctor Foundation have received funding through the UCSF COVID-19 Rapid Response Pilot Grant Initiative to start modeling its effectiveness to help guide local decision-making.
Contact tracing consists of identifying people who have come into contact with someone confirmed to have COVID-19, and then testing them to see if they test positive and need treatment or if they test negative, in which case they may be asked to quarantine. The practice, a basic tool of public health, aims to replace community-wide quarantine with more targeted forms.
A number of factors can make contact tracing more or less effective, Scheffler notes, and those are what he plans to model. Using an agent-based computer model, the researchers plug in parameters that reflect the factors that shape behavior. These might include how densely populated households are, what percent of residents have reduced their everyday travel, what percent comply with imposed quarantine periods, and so on.
The computer program then creates a specified number of hypothetical individuals and assigns them behaviors according to the parameters the researchers have set. Just as in real life, there is a degree of randomness in who bumps into whom and who successfully fights off an infection.
Agent-based models, along with compartment-based and statistical models, form the basis of most disease-modeling. We’ve seen compartment-based models, like the now famous Washington Post example, visually communicate the effects of social distancing measures, as dots changed from blue – susceptible, to brown – infected, and finally to pink – recovered. We described Paul Wesson’s statistical modeling of local COVID-19 infections a couple of months ago.
“To make these models realistic, you really need to tailor them to the area under study,” Scheffler said. “Running the programs is not hard, but we need to dig into the results to be sure we’re finding something useful.”
The early work suggests that contact tracing can help delay the growth of the ongoing pandemic but alone may not fully curb it. One hypothesis is that contact tracing works best during periods of low transmission because, in an outbreak, contact tracers would likely fall behind, leaving infected people to expose others before they’re notified. Shorter turnaround times for test results may also boost the effectiveness of contact tracing by enabling tracers to call contacts closer to the time they were exposed. How many citizens will cooperate with their quarantine will prove to be a vital variable, as well.
Each parameter is set in keeping with the best available data both in terms of infection dynamics and local conditions. After parameters are set and the models run, Scheffler and his collaborators will validate their modeling against actual trends, such as reported deaths.
Watching how changing each of these variables impacts the success of contact tracing and the spread of the virus will help local governments set policy and allocate resources in the coming months.
To help disseminate their findings, Scheffler and his colleagues will create an online dashboard displaying key results from these models where users can manipulate a few basic variables – not unlike the Washington Post COVID-19 infographics that allowed users to select no social distancing, light social distancing and aggressive social distancing measures.