Small Data, Big Implications

What we can learn from the latest indoor dining study

A striking, informative study was just released from South Korea, examining a transmission chain in a restaurant. It is perhaps one of the finest examples of shoe-leather epidemiology I’ve seen since the beginning of the pandemic, and it’s worth a deeper dive.

If you just want the results: one person (Case B) infected two other people (case A and C) from a distance away of 6.5 meters (~21 feet) and 4.8m (~15 feet). Case B and case A overlapped for just five minutes at quite a distance away. These people were well beyond the current 6 feet / 2 meter guidelines of CDC and much further than the current 3 feet / one meter distance advocated by the WHO. And they still transmitted the virus.

That’s the quick and dirty of it. But there’s a lot more detail here, and like many stories, it is best told through a picture:

First, just reading the study is an exercise in what it means to do a study really, really well, with the resources of a government that’s committed to generating useful information. They first found an index case (A) in Jeonju, a person with no history of overseas travel or travel outside Jeonju. They then used location data and found a single overlap with another known infection, case B. (Yes, using such data does raise privacy considerations, but the reality is that in countries like the United States, such location data is routinely collected by all sorts of apps on your phone as well as by telecom companies. Our data is then harvested, accessed by law enforcement  and others, and heck, freely sold to headhunters, with little to no oversight. At least this is something beneficial). Once they identified case B and the location of the overlapping encounter, the restaurant, they went back and tested everyone who worked there, as well as the people who also overlapped with case B, the infector. They found one more infected person, case C. They went back to the restaurant and recreated the airflow and the conditions of the 30 minutes during which the infections happened. 

The first field investigation was started from June 19 based on assessment of CCTV, table locations, timeline, and movement route of case A and other people in the restaurant were verified. Also, the internal structure, distance between visitors, and exact locations of the ceiling type air conditioners were investigated. Air speed and direction at several specified positions were precisely measured using a portable anemometer (Kestrel 2500; Nielsen-Kellerman Co. Boothwyn, PA, USA) on June 24 and July 2. To measure air flow, we set the air conditioner at the same fan speed and direction of June 12. The chairs of cases and visitors were also occupied by people to simulate the same situation. A total of 39 environmental samples of inlets and outlets of air conditioners, table seat of case A, and nearby tables and chairs in consideration of air flow direction were collected on June 23 for testing of SARS-CoV-2 in the environment and were analyzed by rRT-PCR test.9

They watched the CCTV footage to make sure there was no other interaction. They then undertook genome sequencing of the three patients to confirm that they were not likely to have different sources of infection. In other words, the testers determined that these three people weren’t just three infected people who  just happened to be at the restaurant at the same time, but that this was a transmission chain. Since they tested everyone else, too, they were able to confirm that nobody else was infected. They also knew about symptom onset days, and like much of the literature suggests, the initial case infected others the day before his symptom began—the fact that the window of highest infectivity includes the day before symptoms start has made this a very challenging pandemic. 

This study confirms the results of an earlier fascinating study highlighting the risk of indoor dining. There, too, on January 24th 2020, in a restaurant in Guangzhou, one person (case A1) on the cusp of becoming symptomatic infected 10 other people (out of a total 91, including the initial case). There, too, CCTV was used to confirm that it was the airflow which was investigated carefully. (See the graph!). There, too, people were infected at far greater distances than 6 feet, two meters.

On the one hand, these studies confirm how hard it is to make indoor dining safe,since masks cannot be worn while eating. Airborne transmission is not only possible while dining, it is possible beyond our current distance guidelines in time periods as short as five minutes! What we consider close contact for purposes of tracing and testing in this country (generally an unmasked interaction within less than six feet for more than fifteen minutes) is inadequate.

On the other hand, notice the striking number of people not infected in the study, despite sometimes being seated next to the initial infection case. People sitting in the same table as A1 were not infected. People sitting the next table over who had their backs to the airflow were not infected, even though two others in the direction of the airflow were. In other words, not facing the infected person mattered greatly. People sitting the same distance away from the index case but on the other side of the airflow, tables E and F, were completely spared. They were spared despite three out of four people in table B, where the air flew back again at them, getting infected, even though, like the unaffected table, were just a single table over.

Looking back at the Korean study of the Jeonju restaurant that was just released, we see something similarly striking. Even though someone more than 20 feet away was infected after just five minutes of exposure, many other people in the restaurant not in the line of the airflow were not infected, despite being closer. They were also not infected if they were in the line of the airflow but had their backs turned against case B, the initial case. As the paper explains, which the first graph above demonstrates:

Only the visitors (cases A and C) sitting in the air flow path of case B were infected with COVID-19, while other visitors (V2, V3) closer to the infector for a longer period of time but in the absence of direct air flow did not become infected. In addition, the visitors sitting at tables with cases A and C (V1, V6, and V7) were not infected with COVID-19 because they faced away from the infector’s face. 

To complete our understanding of indoor transmission here, we need to bring in a third example, the Skagit County choir chase. The initial report and a follow-up examination document an unusually high “secondary attack rate”—the number of people infected by what was almost certainly a single case. (Supplemental information is available here).

In this particular case, a stunning 52 people (out of 61 total, including the initial case) were infected in one choir practice in March, in Skagit County, Washington. It was after the outbreak had become public, so there were some precautions taken, like putting hand-sanitizer around and restricting hugging and handshakes. Still, a massive outbreak occurred, crucially, almost certainly because the people were singing! It was a choir practice! We know from a lot of research that singing generates more aerosols (tiny droplets that can float around) than talking, which in turn generates more than just breathing. 

While some of this may look discouraging, I find these “small data” studies greatly empowering. They don’t just tell us what happened, they tell us what didn’t happen. 

These small studies cannot tell us the proportion of transmission that occurs indoors, but they highlight how it occurs: droplets and aerosols being carried through the air. Clearly, the closer you are to the person, the more likely you are to get hit But equally clearly, air flow and the positioning of people are huge variables, too. Just talking doesn’t seem to generate aerosols to the point that everyone within the room is infected (though this is possible), but singing does—at least in these well-studied cases.

Sometimes, though, the most important information is that the dog did not bark.

As Sherlock Holmes explains to the Scotland Yard detective in The Adventures of Silver Blaze:

Gregory (Scotland Yard detective): Is there any other point to which you would wish to draw my attention?

Holmes: To the curious incident of the dog in the night-time.

Gregory: The dog did nothing in the night-time.

Holmes: That was the curious incident.

For example, like these authors, I have not found a single outbreak traced to a movie theater. Surely, some must have happened, I used to think, since the virus was circulating in so many countries in February and March, before the distancing and shutdowns occurred? But I don’t know of any, and if I missed any, it’s still likely not a large number. The absence of evidence is probably not because of lack of tracing; everyone going to see the same movie is a memorable event, easier to trace than many other kinds of gatherings. Meanwhile, we have multiple choir outbreaks all over the world. But people don’t face each other or talk much at a movie theater! We can’t rule out future transmissions, and I’m not rushing back into any theater anytime soon, but I think we can use this as useful information. 

What can we conclude from all this? I think there are three broad lessons here. One, small data can be extremely illuminating. Two, air flow and talking seem to matter a great deal. Three, sadly,  indoor dining and any activity where people are either singing or huffing and puffing (like a gym) indoors, especially with poor ventilation, clearly remains high risk. However, given these pieces of information, it also seems that both masks (which dampen the emission of droplets/aerosols from the infected person and which can also lessen the amount one breathes in) and ventilation remain crucial tools.,We should follow Japan’s lead and discourage talking, especially loud talking indoors or on public transportation. Knowledge is power, and sometimes small data is more power.