Our Newest Metric: Contact Tracing

Today, we’re excited to announce a fourth metric added to our COVID warning system: contact tracing. We will lay out how we calculate whether states have sufficient contact tracing capacity, and why we think it is an important metric to assess reopening.

Why Does Contact Tracing Matter?

When people contract the virus, they do not show symptoms right away. Even as states begin to reopen, people will need to quarantine themselves if they have been silently exposed to someone with COVID.

How will they know? That’s where contact tracing comes in.

Because of this problem, it is critical that enough tracing capacity exists to rapidly trace the contacts of individuals who test positive for COVID. Those contacts can be tested, quarantined (if necessary), and asked about whom else they have come into contact with. Because exposed individuals begin infecting others, it is critical that this process be completed in less than 48 hours. If this routine of testing and tracing is done quickly and completely, it can contain COVID, as we have seen in South Korea and Taiwan, and without the need for costly lockdowns.

The White House Coronavirus Task Force’s Guidelines say that contact tracing is a “core responsibility” of states in order to be prepared to reopen. As of May 21, 27 states (CA, CT, DE, FL, HA, IL, IN, KS, KY, ME, MD, MI, MO, MT, NE, NV, NM, NY, NC, ND, OH, PA, RI, SD, WA, WV, and WI) call for contact tracing in their reopening plans. The American Enterprise Institute’s roadmap to reopening says states must massively scale up contact tracing and isolation/quarantine of traced contacts.

How Should We Measure Contact Tracing?

So how do we calculate contact tracing capacity? Experts recommend tracing contacts of someone who tests positive for COVID within 24 hours, to contain the potential of transmission. Based on conversations with practitioners and public health experts, our metric assumes that tracing all contacts for each new positive COVID case requires an average of five full-time contact tracers. 

Therefore, our contact tracing metric measures the percentage of new cases for which all contacts can be traced within 48 hours relative to available contact tracing staff in the state (assuming 1:5 new-positive-COVID-case:contact-tracing-staff ratio).

We use green if greater than 90% of the contacts can be traced within 48 hours, yellow if between 20% and 90% of the contacts can be traced within 48 hours, orange if between 7% and 20% of the contacts can be traced within 48 hours, and red if fewer than 7% of the contacts can be traced within 48 hours.

Here is an example from Wyoming:

As of June 13, Wyoming has an average of 12 new cases per day. If Wyoming needs 5 contact tracers per case, that would be 60 contact tracers necessary to trace all cases in 48 hours. Since Wyoming has 50 contact tracers, that is enough to trace 82% of cases.

What Should The Goals Be?

How did we choose our targets? Research estimates that the infection rate can be driven below 1.0 if 70-90% of cases are identified and 70-90% of those contacts are traced and isolated within 48 hours or less. Therefore, we chose 90% as the cut-off between green and yellow.

The boundaries between yellow, orange, and red are trickier. When less than 90% of positive cases have their contacts traced within 48 hours, contact tracing will likely be insufficient to contain COVID. We use 90% as the boundary between green and yellow. In the absence of expert consensus, we have set inclusive lower thresholds for yellow and orange. We peg the cut-off between yellow and orange at 20% — the number required for there to be one contact tracer per active case per 48 hours — and the cut-off between orange and red at 7%. Every state currently coded red is either currently experiencing a new outbreak or effectively has no tracing capacity.

A state can become green either by increasing the number of contact tracers, or by decreasing the number of new daily COVID infections. We hope that this new metric will help states factor contact tracing capacity into their reopening decisions.

Announcing the CovidActNow.org API

Covid Act Now (CAN) is a non-profit organization of technologists, epidemiologists, and medical professionals working to model how COVID-19 will spread in each U.S. state and county. 

We published the first version of our model on March 20. Since then, 10+ million Americans have viewed the model and we’ve engaged with dozens of government officials, including the U.S. Military and White House, to assist with response planning.

Today, we are launching an API to make our data programmatically available to everyone.

Our API exposes:

Reported Data: State and county level data for confirmed cases, deaths, and hospital bed capacity. The data is collected from a number of sources, including The New York Times, and is updated daily.Forward Projections: State and county level projections for hospitalizations and deaths based on several possible interventions. This data is generated from our model.

By launching a public API, we are making the data that powers CovidActNow.org available to anyone, free of charge, under the Creative Commons 4.0 license. The data is updated daily around midnight U.S. Pacific Time, and is available in both JSON and CSV format.

You can view the documentation on GitHub.

By making this data available, we are hoping it will be a helpful input into COVID efforts such as response planning, reopening initiatives, data visualization, and the creation of new tools.


Since Covid Act Now launched on March 20, our team has spent significant time refining our model. It is originally based on a traditional SEIR model by Dr. Alison Hill at Harvard, and now Dr. Rebecca Katz and her team at the Georgetown Center for Global Health Science and Security audits our work. In the past few weeks we’ve made several improvements, including adding hospitalization and severity rates, data sources, and inference projections.

Going forward, our organization will be focusing on providing data to leaders to inform their decisions around reopening safely. 

Our work is being done in the open, and you can find our model open-sourced on GitHub.

Using the API

To get data from our API, you’ll need to construct a URL with the state or county, the intervention type, and append .json or .csv. For example:

+ [ STATE CODE (CA, PA, NV) ] + [ .INTERVENTION_OPTION ] + [ .timeseries.json ]

JSON objects can be easily manipulated with code, while CSV (comma separated values) can be easily imported into Google Sheets or Excel.

The intervention choices are:

API ParameterIntervention TypeR AssumptionsNO_INTERVENTIONNone3.7WEAK_INTERVENTIONSocial Distancing1.7STRONG_INTERVENTIONShelter in Place1.1 for 4 weeks, 1.0 for 4 weeks, 0.8 for 4 weeks.OBSERVED_INTERVENTION–A dynamic forecast based on the observed data in a given U.S. state

Two notes on interventions:

OBERSERVED_INTERVENTION infers an R(t) value from recent cases, hospitalizations, and deaths in each state.To see more details on each intervention and sources for our R assumptions, please see our model references and assumptions.

For example, this link returns an aggregated list of how COVID will spread in every U.S. State based on observed data, in CSV format:

And this is Dakota County, Minnesota’s data for how COVID would spread if it reopened everything tomorrow, in JSON format. (Note that we’re constructing the URL with the FIPS code for the county.)

Open this link in your browser now to see the data:

Hopefully these examples give you a taste of how to use the API. For all the details, see the documentation on GitHub.

In the future, we intend to make more data available, including:

Additional file formats like shapefiles for GIS systems.Integrations with data visualization products, like Tableau.

Please Play Around!

In order to accelerate decision making, safely re-open the country, and, ultimately, save lives, a massive, unprecedented collaboration across government, business, and individual citizens is needed. 

We hope this data plays a small part in that collaboration, and we’re excited to see how it is used, visualized, integrated, and transformed to make a difference.

If you have any questions, or want to share with our team what you’ve done with the API, please drop us an email at info@covidactnow.org.

Happy coding!

Thanks to Addy Osmani, Chad Arimura, Ilya Voldarsky, Josh Dzielak, Max Lynch, Paul Irish, and Steve Wilmott for reading drafts of this post and giving feedback on the API.

Inference Projections for States

Today, Covid Act Now’s U.S. Interventions Model launched inference projections at the state level. You can see inference projections on state graphs at Covid Act Now under the purple label, “projected based on current trends.”

Covid Act Now & IHME: Why Two Models Are Better Than One

IHME and Covid Act Now's U.S. Interventions Model are two common forecasting tools used by decision-makers in the United States. We often get asked how these two models compare. This blog post lays out the answer to that question, to help bring clarity to the conversation around COVID-19 modeling.

Return to covidactnow.org


Recent Posts