Discussing my exposure notifications reporting at the webinar!
This week, I had the opportunity to participate in a webinar about the future of exposure notifications, the digital contact tracing systems used in about half of U.S. states. The webinar was hosted by PathCheck Foundation, a global nonprofit that works on public health technology—including exposure notification apps.
I talked about my recent feature in MIT Technology Review, which investigated usage rates and public opinion around exposure notification technology. Other panelists included Jeremy Hall, project manager of Hawaii’s exposure notification system, Sam Zimmerman, director of exposure notification programs at PathCheck, and Ramesh Raskar, technology professor at MIT and PathCheck founder.
It was a great session, with discussion ranging from the challenges of implementing exposure notification technology in the U.S. to the ways this technology may be used for future infectious disease outbreaks. With a year of work under their belts, Zimmerman and Raskar brought insider perspectives to the challenges that I had seen from the outside in my reporting. For example, Raskar discussed how Massachusetts’ own exposure notification app is still in a trial run even though PathCheck approached the state public health agency offering to provide that technology in summer 2020.
I was also excited to hear from Hall on how Hawaii’s public health agency promoted exposure notification technology in their state. At the time I collected data for my Technology Review piece, Hawaii had about 650,000 people in the state’s exposure notification system, including those who downloaded the app and those who turned on the EN Express option in their iPhone settings. That represented 46% of the state’s population—a larger share than any other state.
Since I did my data collection, Hawaii has added an additional 250,000 users, I learned from Hall. This includes both Hawaii residents and tourists; tourists with iPhones get push notifications encouraging them to opt into EN Express when they enter the state. Hawaii has also worked with county public health departments and local organizations to publicize its exposure notification system. I think the state could be a model for other public health institutions working to implement exposure notification technology.
If you’d like to watch the webinar, it was recorded and is available at this link—you’ll just need to put in a name and email. The conversation starts about one minute in.
The majority of states do not collect or report detailed information on how their residents became infected with COVID-19. This type of information would come from contact tracing, in which public health workers call up COVID-19 patients to ask about their activities and close contacts. Contact tracing has been notoriously lacking in the U.S. due to limited resources and cultural pushback.
Exposure notification systems are availalbe in 25 states and D.C.
This week, I have a new feature out in MIT Technology Review. It’s an investigation into the usage rates and public opinion of exposure notification apps—those Bluetooth-enabled systems that promised to function as a method of digital contact tracing. You can read the story here; and for the CDD this week, I wanted to provide kind-of an extended methodology behind the piece.
The inspiration for this feature came from my conversation with Jenny Wanger, which was published in the CDD back in March. Wanger is the Director of Programs at Linux Foundation of Public Health, a software development network that has worked on exposure notification systems. We discussed the privacy-first design of this technology, as well as how difficult it has been to evaluate how well the apps are working.
That conversation got me thinking: someone should actually try to collect comprehensive data on how many Americans are using exposure notifications. The federal government doesn’t provide any data on this topic, and most of the states that have exposure notification systems available don’t publicly report data, either. But, I thought, there might be other ways to gather some estimates.
When I talked to Lindsay Muscato, an editor at Technology Review’s Pandemic Technology Project (and a longtime CDD reader!), a few weeks later, she agreed that such an investigation would be valuable. The Pandemic Technology Project has done a lot of reporting on exposure notification apps, but hadn’t yet pursued the kind of novel data collection project I was envisioning.
The project started with a hypothesis: that exposure notification systems are underutilized in the U.S. due to a lack of trust in governments and in new technology.
Initially, I planned to use app reviews from the Google Play and Apple stores as the main data source for the story. The two online stores provide download estimates, which I intended to use as a proxy for app usage rates—along with ratings and reviews that I could use as a proxy for public opinion. (Shout-out to machine learning engineer Venelin Valkov, who has put together a great tutorial on scraping and analyzing app store reviews with Python.)
But an interview early in the reporting process caused me to change my data collection plans. I talked to two exposure notification experts at the New Jersey public health agency, who told me that the app download count I saw on the state’s COVID-19 dashboard was actually a significant underrepresentation of the state residents who had exposure notifications enabled on their smartphones.
This data disconnect was due to something called Exposure Notification Express, or ENX. ENX is an upgrade to the digital contact tracing system, released by Apple and Google last September, that made it easier for states to develop new apps. The upgrade also built exposure notifications directly into the iPhone operating system, allowing millions of people to opt into the notifications without downloading a new app.
In short, I couldn’t use app downloads as a proxy for usage rates. I also couldn’t use Apple app store reviews, because the majority of iPhone users were using ENX rather than downloading a new app. Many state apps are listed on the Google Play store but not on the Apple store, for this reason.
I still used Google Play reviews for the public opinion piece of my story. But to determine usage rates, I developed a new plan: reach out to every state public health agency with an exposure notification system and ask for their opt-in numbers. This involved a lot of calling and emailing, including multiple rounds of follow-up for some states.
The vast majority of state public health agencies to whom I reached out did actually get back to me. (Which I appreciate, considering how busy these agencies are!) The only one that didn’t respond was South Dakota; I assumed this state likely had a low number of residents opted into exposure notifications because South Dakota shares an app with two other low-activation states, North Dakota and Wyoming.
Based on my compilation of state data, 13 states have over 15% of their populations opted into exposure notifications as of early May—passing a benchmark that modeling studies suggest can have an impact on a community’s case numbers.
(I used the U.S. Census 2019 Population Estimates to calculate these opt-in rates. I chose to base these rates on overall population numbers, not numbers of adults or smartphone users, because that 15% benchmark I mentioned refers to the overall population.)
This is a smaller number than the engineers who developed this technology may have hoped for. But it does mean these 13 states—representing about one-third of the U.S. population in total—are seeing some degree of case mitigation thanks to exposure notifications. Not bad, for an all-new technology.
I was also impressed by the five states that reported over 30% of their populations had opted into the notifications: Hawaii, Connecticut, Maryland, Colorado, and Nevada. Hawaii had the highest rate by far at about 46%.
For anyone who would like to build on my work, I’m happy to share the underlying data that I collected from state public health agencies. It’s important to note, however, that the comparisons I’m making here are imperfect. Here’s a paragraph from the story that I’d like to highlight:
Comparing states isn’t perfect, though, because there are no federal standards guiding how states collect or report the data—and some may make very different choices to others. For example, while DC reports an “exposure notification opt-in” number on its Reopening Metrics page, this number is actually higher than its residential population. A representative of DC Health explained that the opt-in number includes tourists and people who work in DC, even if they reside elsewhere. For our purposes, we looked at DC’s activation rate as a share of the surrounding metropolitan area’s population (including parts of nearby Maryland, Virginia, and West Virginia).
So, did my research support my hypothesis, that exposure notification systems are underutilized in the U.S. due to a lack of trust? Somewhat. I definitely found that the technology has failed to live up to its potential, and the app reviews that I read indicated that many Americans did not trust the technology—or simply failed to understand its role in COVID-19 prevention.
At the same time, however, I found that some states have seen significant shares of their populations opting into the new technology. Plus, the app reviews showed that a lot of people not only trusted the technology—they bought in enough to advocate for it. The majority of states actually had more five-star ratings than any other category, and a lot of those reviewers tried to combat the mistrust they saw elsewhere in the comment section with explanations and endorsements. This is a job that should’ve been done by public health agencies themselves, of course, but the positive reviews may indicate a promising future for this technology.
In the story’s conclusion, I argue that exposure notification technology is still in its trial run here in the U.S. State public health agencies had limited budgets, limited resources, and limited capacity for trust-building. As a result, they focused simply on getting as many people to opt into the technology as possible—rather than any kind of comprehensive data collection or analysis.
“The ultimate goal [of exposure notifications] is for more folks to know they’ve been exposed,” says Hanna Sherrill, an Eagleton Science and Politics Fellow at Rutgers University who worked with the New Jersey public health agency on its exposure notifications system. “Hopefully some of them will take the advice to quarantine, and then they will stop the spread from there. Even if there’s one or two people who do that, that’s a good thing from our perspective.”
Other state public health staffers who responded to Technology Review’s data requests echoed her sentiment—and their attitudes suggest that digital contact tracing in the US may still be in its trial run. We have 26 different prototypes, tested in 26 different communities, and we’re still trying to understand the results.
“In the US, the existing apps and tools have never hit the level of adoption necessary for them to be useful,” Sabeti says. But such success may not be out of reach for future public health crises.
I’m hopeful that, with more investment into this technology, public health agencies can build on the prototypes and develop community trust—before we see another pandemic.
I plan to keep reporting on this topic (including investigation into Google and Apple’s role in the technology, which a couple of readers have pointed out was lacking in the Technology Review piece). If you have further questions or story ideas, don’t hesitate to reach out.
The majority of states do not collect or report detailed information on how their residents became infected with COVID-19. This type of information would come from contact tracing, in which public health workers call up COVID-19 patients to ask about their activities and close contacts. Contact tracing has been notoriously lacking in the U.S. due to limited resources and cultural pushback.
This week, I wrote a story for Popular Science that goes over what we know (and don’t know) about the most common settings for COVID-19 infection.
Most of the main points will probably be familiar to CDD readers, but it’s still useful to compile this info in one concise article. Here are the main points: Outside events are always safer. Surfaces are not a common transmission source. Communal living facilities and factories tend to be hotspots. Indoor dining and similar settings carry a lot of risk. Essential workers are called essential for a reason. And don’t rule out small gatherings, even though such events are safer for those of us who’ve been vaccinated.
This story gave me an excuse to revisit one of my favorite COVID-19 datasets: the Superspreading Events Database, a project that compiles superspreading events from media reports, scientific papers, and public health dashboards. I interviewed Koen Swinkels, the project’s lead, for the CDD back in November.
At that time, the database had about 1,600 events; now, it includes over 2,000. All of the patterns I wrote about in November still hold true now, though. Notably, no event in the database took place solely outside (though Swinkels told me he’s seen some events with both an indoor and outdoor component). And the vast majority of events in the database took place in the U.S.
For those U.S. events, most common superspreading settings are prisons (166,000 cases), nursing homes (30,000 cases), rehabilitation/medical centers (24,000 cases), and meat processing plants (13,000 cases). By this database’s definition, a superspreading event may comprise a sustained outbreak at one location over a long period of time—and prisons have been continuous hotspots since last spring.
You can check out the U.S. superspreading events in the database below. I made this visualization in November and updated it this past week.
One of the reasons why I like the Superspreading Events Database is that Swinkels and his collaborators are extremely clear on the project’s limitations. If you load the database’s public Google sheet, you’ll see a prominent note at the top reading, “Note that the database is NOT a representative sample of superspreading events. Please read this article for more information about the limitations of the database.” The article, a post on Swinkels’ Medium blog, goes in-depth on the biases associated with the database. It’s easier to identify superspreading events in institutional settings, for example, since many of them employ frequent testing. Still, I think that—when carefully caveated—this database is an incredibly useful resource for identifying patterns in COVID-19 spread.
Swinkels additionally pointed me to another great source for exposure data: the state of Colorado publishes outbreak data in weekly reports. A few other states publish similar info, but Colorado’s data are highly detailed and complete. In this past week’s report, released on March 10, the state says that 6,900 out of a total 28,000 cases in active outbreaks are linked to state prisons. 3,900 more cases are linked to jails.
I’ve visualized the March 10 Colorado outbreak data below. As you may notice, the next-biggest outbreak setting after prisons and jails is higher education—colleges and universities represent 6,700 active outbreak cases. Colorado’s dataset does not specify how many of those cases are linked to the mask-less University of Colorado party that drew wide criticism last weekend… but we can assume that party was no small player.
Finally, this PopSci story also gave me an excuse to revisit one of my favorite COVID-19 data gripes: the lack of contact tracing info we have in the U.S. I’ve written about this issue in the CDD before; I surveyed state dashboards in October, and drew connections from the Capitol invasion in January. But it was still disheartening to find that now, in March, we continue to be largely in the dark about how many contact tracers are actively employed in most states and how many people they’re reaching.
Here’s a clip from the story:
In the US, though, the practice is done unevenly, if at all. Most states and local jurisdictions, struggling from years of underfunded public health departments leading up to the pandemic, have not been able to hire and train the contact tracers needed to keep tabs on every case.
Many states have attempted to supplement their limited contact tracing workforces with exposure notification apps, which are theoretically able to notify users when they’ve come into contact with someone who tested positive. Though these apps became more widespread in the US this past winter, they’re still not used widely enough to provide useful information. New Jersey, one state that provides data on its app use, reports that about 574,000 state residents have downloaded the app as of March 6—out of a population of 8.9 million.
This situation is not likely to improve much in the coming months as Americans aren’t about to change their perspectives on privacy any time soon. But if you have the opportunity to download an exposure notification app for your state, do it! The more data we have on where people are getting exposed to COVID-19, the better we can understand this virus.
The majority of states do not collect or report detailed information on how their residents became infected with COVID-19. This type of information would come from contact tracing, in which public health workers call up COVID-19 patients to ask about their activities and close contacts. Contact tracing has been notoriously lacking in the U.S. due to limited resources and cultural pushback.
COVID-19 Data Dispatch 