More details about the TRACE model

This page provides additional technical details about the simulations generated by the TRACE model, including the policy dimensions and underlying epidemiological assumptions explored. For context and more general information on the model and our analysis, please visit the Main Results page and the About TRACE page.

Policy Dimensions

Note: The number of tests reported above are for a population the size of the entire United State; in the simulations we use a reduced population of 50,000 agents and scale these numbers appropriately (see About TRACE).

Scenarios

Policies Robust to Uncertainty

The model identifies several robust policy options that are effective across the full range of epidemiological uncertainty that we explored. This means they produce epidemic outcomes that are statistically indistinguishable from a policy of mass shelter-in-place under even the most pessimistic assumptions about the epidemic that we tested (Most Unfavorable). These “highly robust” policies generally still retain some amount of social distancing coupled with intensive test and trace policies. They include:

1. Self-quarantine of all individuals with symptoms (High adherence) + Low capacity to trace contacts of those reporting symptoms + Low PCR testing capacity (with at least Low accuracy) administered to these contacts (with those testing positive quarantined) + continued school closure
2. High PCR testing capacity (with High accuracy) administered first to those who are newly symptomatic and then used to randomly sample, with only those who test positive quarantined (High adherence) + Moderate capacity to trace contacts generated from those who tested positive; contacts self-quarantine (High adherence) + continued school closure
3. High PCR testing capacity (with High accuracy) administered first to those who are newly symptomatic and then used to randomly sample, with only those who test positive quarantined (High adherence) + High capacity to trace contacts generated from those who tested positive; contacts self-quarantine (High adherence) + continued remote work for 33% of the workforce

If the underlying epidemiological conditions are less pessimistic (Slightly Unfavorable), another set of policies become equally effective. Sufficient testing could yield greater certainty about the underlying parameters of the epidemic and make one of these policy options a robust choice:

1. Relax all social distancing measures + High PCR testing capacity (with High accuracy) administered first to those who are newly symptomatic and then to a random sample, with only those who test positive quarantined (High adherence) + Moderate capacity to trace contacts generated from those who tested positive; contacts self-quarantine (High adherence)
2. Self-quarantine of all individuals with symptoms (High adherence) + Low capacity to trace contacts of those reporting symptoms + Low PCR testing capacity (with at least Low accuracy) administered to these contacts (with those testing positive quarantined) + remote work for 33% of workforce
3. Only Moderate adherence to quarantine, with either High PCR testing capacity (Low accuracy) & Low contact tracing OR Low PCR testing  capacity (Low accuracy) & High Contact tracing + schools closed + remote work for 33% of workforce

Policies effective only under Somewhat Favorable conditions or better include:

1. Relax all social distancing + self-quarantine of all individuals with symptoms (High adherence) + Low capacity to trace contacts of those reporting symptoms + Low PCR testing capacity (with at least Low accuracy) administered to these contacts (with those testing positive quarantined)
2. Relax all social distancing measures + Low PCR testing capacity (with High accuracy) administered first to those who are newly symptomatic and then used to randomly sample, with only those who test positive quarantined (High adherence) + Low capacity to trace contacts generated from those who tested positive; contacts self-quarantine (High adherence)

Policies effective only under the Most Favorable conditions include:

1. Relax all social distancing measures + Low PCR testing capacity (with Low accuracy) administered first to those who are newly symptomatic and then used to randomly sample, with only those who test positive quarantined (only Moderate adherence) + Low capacity to trace contacts generated from those who tested positive; contacts self-quarantine (only Moderate adherence)
2. Low PCR testing capacity (with High accuracy) administered first to those who are newly symptomatic and then used to randomly sample, with only those who test positive quarantined (only Low adherence) + Moderate capacity to trace contacts generated from those who tested positive; contacts self-quarantine (only Low adherence) + continued school closure

The Importance of Adherence

• With low adherence (only 50%), we found no policy that safely allows full release of social distancing under any condition simulated (to achieve suppression, one must either keep schools closed or partial remote work)
• With moderate adherence (75%), full release of social distancing is sometimes possible, but only with either (a) the Most Favorable epidemiological assumptions or (b) “go slow” (wait until initial infections are low to start releasing). There is no policy that allows you to release social distancing quickly with less favorable epidemiological assumptions in our simulations.
• With high adherence (90%), it is possible to release social distancing measures more quickly, even under fairly unfavorable conditions, with only a moderate test-and-trace capacity in place.

Why Gradually Relaxing Social Distancing is Effective

Waiting to release social distancing until the background rate of infections is low enough allows some much less-demanding policies to work well even under Very Unfavorable epidemiological assumptions. These include:

• Release social distancing, without asking self-quarantine, as long as adherence is high, with a moderate PCR testing regime
• Withstand lower adherence by keeping schools closed and partial remote work in place, under various testing conditions

If you assume more favorable epidemiological scenarios (either no asymptomatic transmission OR higher background immunity), you can:

• Release social distancing with a minimal PCR testing regime as long as adherence is high
• Withstand lower adherence while opening schools, without self-quarantine, by using partial remote-work (as long as you have a moderate PCR testing regime)
• Withstand lower adherence by keeping schools closed, even with minimal test-and-trace-capacity

Prioritizing Tests for Non-Symptomatics

One problem created by prioritizing PCR tests towards the symptomatic is that evidence suggests that some individuals who present with COVID-like symptoms may be sick with respiratory illnesses other than COVID-19 (for instance, the flu). To incorporate this possibility into the TRACE model, 4 out of every 1000 agents begin each simulation with an influenza-like illness (ILI) and attempt to receive testing where available. This strains PCR test capacity when the number of daily tests is limited, and diminishes the effectiveness of a policy that first targets tests towards those with symptoms.

By contrast, a policy that asks agents presenting with symptoms of COVID-19 to self-isolate even in the absence of a positive test performs much better, by reserving limited testing to detect asymptomatic cases or protect priority populations (e.g. agents over 65 years of age). In addition, we find that this testing strategy results in many fewer actual numbers of agents in quarantine across many scenarios, limiting disruption to economic and other activity.