Blog‎ > ‎

Bidirectional private automated contact tracing

posted Apr 8, 2020, 8:08 PM by Kevin Esvelt   [ updated Apr 8, 2020, 8:09 PM ]
Tracing and isolating the contacts of infected individuals to prevent downstream transmission has proven critical to effective pandemic control. Every country that has succeeded in limiting the spread of SARS-CoV-2 has relied heavily on contact tracing. Nations with little experience have struggled.

Private automated contact tracing (PACT), in which smartphones are used to notify people who may have been exposed by infected individuals while preserving the privacy of both parties, offers considerable advantages in scale and efficiency. It also protects everyone's anonymity better than the traditional form: not even health workers need to know who came in contact with whom. There's just one problem: everyone seems to assume we will use it to perform traditional contact tracing more effectively.

I suspect that we can do much better.

I'm a member of the MIT Safe Paths and PACT groups who have been developing relevant apps and protocols, and helped both of them connect with another group, Covid-Watch/CoEpi. To all of them I've advocated for one key feature: contact logs that are stored for many weeks. The reason is that contact tracing should be bidirectional.

Think of transmission as a family tree: each infected "generation" of people transmit the virus to the next generation of victims. During a pandemic, the family tree rapidly expands, as the average person infects more than one other individual. Some people might have dozens of "offspring"; others may have none. Traditional contact tracing only goes in the "forward" direction: it seeks to find the offspring of known infected individuals and isolate them so they don't spread the virus. But we could also use it in the other direction to find the "parent" of each infected individual.

By finding previously unidentified carriers from earlier generations, "reverse" contact tracing could discover entirely new branches of the transmission tree and isolate individuals known to be contagious by tracing each newly discovered past branchpoint out to the most recent generation.

That is not the case for SARS-CoV-2, for which somewhere between 1 in 6 and 1 in 2 carriers remain completely asymptomatic - yet still contagious. There are no warning signs indicating that these people should get tested, so they don't. These carriers are one major reason we observe so much community transmission with no obvious source of infection.

With bidirectional contact tracing, we have a chance to find asymptomatic carriers. Suppose Bob develops symptoms and tests positive for COVID-19 (or has a fever, which is highly indicative of infection under lockdown conditions when other pathogens are suppressed). With traditional forward contact tracing, we might identify Alice as someone who was exposed by Bob and advise her to self-isolate. With PACT in place, we could afford to trace in the other direction: we could notify Charlie, who was in contact with Bob around the time that he was probably infected, that he should get a serological test to see if he has antibodies to SARS-CoV-2.

If Charlie tests positive, he becomes our new "index case", and we repeat bidirectional contact tracing. "Forward" tracing might notify Beatriz and Brianna, who like Bob were infected by Charlie, as well as the people they infected: Aaron, Adam, Andrew, Alan, Alyssa, and Arianna, all of them "cousins" of Alice. We might even notify David, who infected Charlie, and through him Cameron and Cassie and Chelsea and their "descendants", ultimately limiting the spread of an even larger and previously undiscovered branch of the viral transmission tree.

The implementation of bidirectional contact tracing using a PACT protocol could be a powerful preventive tool in curtailing the spread of SARS-CoV-2. Just as importantly, it could defend against future pandemics that might be considerably more destructive. Indeed, coupled with a monitoring network that screens random people by next-gen sequencing to detect novel zoonoses or engineered agents, bidirectional PACT could become a powerful defense against almost any potential pandemic pathogen.