California rolled out a COVID-19 contact tracing app this week, and officials — including Apple CEO Tim Cook — touted it as an advancement that would help slow their ongoing surge in cases. Using the app will be easy. Measuring whether the app can deliver will be harder.
Nine months after Apple and Google first announced their partnership, contact tracing apps’ role in reducing viral transmission is still difficult to measure. Relatively few people have downloaded the apps, and because of the apps’ focus on privacy, it may be nearly impossible to quantify how well they’re actually able to help prevent disease.
“It seems to me that it’s incumbent upon those who are behind these efforts to show evidence that they’re having some effect,” says Ryan Calo, law professor and a director of the Tech Policy Lab at the University of Washington. “So far, I’ve personally been unconvinced that there’s been any kind of significant showing of efficacy.”
“I’ve personally been unconvinced that there’s been any kind of significant showing of efficacy”
States each spent hundreds of thousands of dollars to develop these contact tracing apps, rebranded over the summer as exposure notification apps. New York’s cost $700,000, for example, while Virginia’s had a $229,000 price tag. The apps were designed to supplement state efforts to manually track and stop the spread of COVID-19. Anyone who tested positive for the coronavirus would still, ideally, get a phone call from a contact tracer. Contact tracers would ask who they’d interacted with during the time they may have been contagious, and ask those people to quarantine or be tested for the coronavirus.
Manual contact tracing isn’t perfect — contact tracers can only find the people that a sick person knows they were in touch with, and the process can be slow. That’s the gap automated contact tracing apps aim to fill. In theory, they would let someone who tested positive alert strangers they stood by on a long subway ride or grocery store employees they interacted with that they’d been exposed to the virus. And it would happen fast: telling the app you’d tested positive would automatically notify those contacts, without waiting for a tracer to call.
There’s one major catch: that entire scenario relies on lots of people downloading the app. Those strangers on the subway have to use it in order for them to get any benefit from an alert. If only a handful of people turn on exposure notifications, the chances that the random people who they interact with in their daily lives will also have the system running is fairly low.
In many states, uptake has been slow. In New York, for example, only around 5 percent of people have downloaded COVID Alert NY, which launched at the start of October. Fewer than 3,000 of the 180,000 people who tested positive since the app launched had it installed, and only around 800 people were notified of an exposure. Nevada’s app was installed by around 4 percent of the population, a spokesperson told The Verge. Around the same percentage of Michigan’s population uses MI COVID Alert, and only 142 people have reported positive test results, according to a spokesperson.
Virginia, the first state to launch an exposure notification app, has seen more success: around 10 percent of the population, and around 20 percent of the population between the ages of 18 to 65 who own smartphones, uses the app, a health department spokesperson told The Verge. Colorado’s app is also more popular — 20 percent of the population uses it, according to a statement from the state’s Emergency Operations Center.
The numbers reported by states are still far below the levels needed to make a dent in the pandemic, analysis suggests. It would take around 60 percent of a population using a digital contact tracing system to stop transmission of COVID-19, according to a study by a team at the University of Oxford. Other studies found that it would take that level or higher of adoption, along with other public health measures, to control outbreaks.
“We wouldn’t be expecting a huge effect”
“We wouldn’t be expecting a huge effect based on the kind of 20 or 30 percent uptake. It would be a small effect,” says Isobel Braithwaite, a clinical fellow in public health at University College London and author on a review of automatic contact tracing. It’s not all or nothing, she says — a small effect could still nudge transmission down a few percentage points and keep people from getting sick. Another still-unpublished study from Google and the University of Oxford found that if 15 percent of people in a Washington state county used an exposure notification app, infections could go down by 8 percent.
“I’ve come to think about it in terms of a spectrum, where you increase the benefit with the higher level of uptake. But even at low levels, there’s still some benefit,” Braithwaite says.
The University of Oxford study does show that even at low levels, there’s some utility to the apps, says Mike Reid, an assistant professor in infectious disease at the University of California, San Francisco, who works with the San Francisco Department of Public Health on contact tracing. “My feeling is that anything that can help us to reduce transmission has got to be leveraged,” he says. COVID-19 spreads exponentially, so cutting off even one potential infection matters. “Every time you reach a case and help them understand their need to isolate, you know you prevent thousands of subsequent infections,” Reid says.
The challenge is, researchers aren’t able to pick apart whether the apps are actually preventing those infections. Most of the research on the way exposure notification technology changes the trajectory of an outbreak is based on modeling, including the University of Oxford study. Experts use data and estimates to map out what could happen if a certain number of people download an app, based on what they know about the way COVID-19 spreads and how contact tracing typically works. But they don’t track the progress of apps to find out what actually ended up happening.
Google and Apple made a big point of prioritizing privacy and security when developing the apps. They wanted to minimize concern that the companies were tracking users’ whereabouts, so none of the information collected by the app is identifiable. But by reducing the amount of data they collect, there isn’t an easy way to evaluate how these exposure notification programs actually work once they’re implemented.
“By the very nature of how they’re done in terms of prioritizing privacy, it’s very challenging to truly assess their impact,” Reid says.
The United Kingdom initially attempted to develop its own contact tracing app outside of the Google and Apple system, which would have collected more information on usage. “They moved to the more privacy preserving Apple and Google approach for the wider national rollout, because of political and public perception considerations,” Braithwaite says. “That makes studying the effect much, much harder.”
“That makes studying the effect much, much harder”
Researchers don’t know, for example, how many of the people who receive notifications on the apps follow isolation guidelines or get tested for COVID-19. They also don’t know how many of the people receiving notifications are people who wouldn’t have been flagged by a manual contact tracer. “You have to see if there are situations where they never would have known,” Calo says. A robust evaluation would figure out how many times someone was in a situation where they got a notification from a stranger, actually isolated, and ended up developing symptoms themselves.
That’s key to evaluating whether apps were worth the investment, Calo argues. “They never would have quarantined or gotten tested, and we saved a lot of people that way,” he says. “And then you have to put that number against the numbers of how much was paid for the app.”
For Braithwaite, even a slight dip in virus transmission would make up for the millions of dollars it took to develop and roll out exposure notification apps. “We’re buying a bit more time until the vaccine by suppressing transmission slightly,” she says. Not having clear-cut evidence isn’t a reason to disregard the apps, she says. It took time for researchers to study how well masks helped slow the spread of the coronavirus, she notes, but public health experts still asked everyone to wear them while that process was ongoing. “We don’t have any evidence that they don’t work, either.”
There’s no precedent for using this type of technology
There’s no precedent for using this type of technology as part of a public health response to any disease, Reid says. It’s never been used before and was developed in the middle of a pandemic. Its use against COVID-19 is, in some ways, just a trial run. Despite the intense data difficulties, researchers will likely still try to sort through how effectively they drove down viral transmission — maybe by comparing coronavirus spread in areas with an app to those without, Braithwaite says. Having at least a hint toward answers will help public health decision-making going forward.
“The lessons we will learn will impact how we respond to future pandemics,” Reid says. “I think evaluating its impact right now, and determining whether this is going to be part of our plan for responding to future pandemics, is critical.”
Correction December 14th, 10:46AM ET: An earlier version of this article attributed a research study to the University of Washington rather than the University of Oxford. We regret the error.