As the pandemic stretches into its fourth month, it may feel to many of us like we’re eating, sleeping, and living all things “coronavirus.” But in this respect, Rae Wannier outdoes even the most devoted COVID-19 newshound. A fourth-year doctoral candidate at the University of California, San Francisco, Wannier builds disease models using the computer programming language R at the university’s Proctor Foundation. That means she has spent the last few months living the pandemic reality so many of us share—long hours inside, childcare challenges, quarantine birthdays—while simultaneously immersing herself in models of what that reality might become.
Wannier, who earned a master’s in public health from Yale University’s Department of Epidemiology of Microbial Diseases before moving west to study disease modeling, worked mostly on modeling Ebola and measles before the COVID crisis. Now she’s helping produce some of the models that agencies like the San Francisco Department of Public Health rely on to make essential decisions about how to react as a community to a largely unprecedented disease threat. NOVA spoke with Wannier about what modeling looks like for her now, the lessons she’s learned from her work, and the importance of fondue in quarantine.
Alissa Greenberg: Your situation is interesting because you’re both a private citizen in a pandemic and a scientist studying what it’s like to be a private citizen in a pandemic. So many of us feel overwhelmed with information these days, and you’re probably inundated with information more than most people. But it also felt powerful to read the preprint [not yet peer-reviewed] paper you sent me, in which you state flat out that 44% of COVID-19 transmission happens before people are symptomatic—to see it quantified that way. Do you feel like the work you do makes you feel more or less in control?
Rae Wannier: Because of my work, I probably feel more in control, because I feel a greater degree of certainty in knowing which behaviors to engage in and which to avoid. For example, from this work, I know wearing a surgical mask is much better than cloth. So I make sure to only wear those masks. This work has also made me more willing to engage in outdoor activity, since I know the probability of transmission outdoors is relatively low. I’ve been quite focused instead on limiting any indoor exposure.
Having a greater appreciation for the uncertainties here gives me some degree of reassurance, in the sense that I have a better sense of what I can rely upon and what I can't rely upon. And that in of itself, I suppose, is helpful in terms of knowing how I can make plans—as opposed to other people, who don’t understand what's going on and don't know at all what they can rely upon.
AG: Is that why you got into disease modeling? What do you find compelling about it?
RW: I just am fascinated by infectious diseases. I appreciate the logic of trying to understand the transmission and how different behaviors increase the probability of transmission—just the spatial and interactive aspect of it all.
With infectious diseases, it all seems a little bit random. And there's this element of chance and probability that's inherent in the whole transmission chain. The interaction between individuals, and often animals, thinking about how people move and when they move and how that impacts this movement of these microbes…it’s just this whole complex network that I find really interesting to study. I also always wanted to study something that I felt was going to make a difference. But I didn't dream that it would become so immediately impactful at this early stage of my career.
AG: What were you doing before COVID hit? And how did you decide that it was time to change focus?
RW: I've been studying Ebola, and I have in the last two years personally seen the beginning of an outbreak then become a long, sustained, ongoing outbreak. In those situations, we were humming along, doing our work, and then all of a sudden the outbreak happened and we would just shift gears and start doing forecasting.
So looking at this in China, we looked at each other and said, “We don't think this is going away. It's spreading too rapidly.” It only takes a few COVID importations to escape before it ends up becoming community transmission.
Now, we've been asked to try and simulate the impact of different interventions and give some advice to the San Francisco Public Health Department on what we anticipate will be the impact of mask wearing and contact tracing and these shelter-in-place orders. Mine is not the only model they’re listening to, but still, this has been a wonderful experience and also an intimidating experience. I want to make sure I have a high level of confidence in what I'm doing, and I'm doing it on a very short timescale with a model that I'm fairly new to.
AG: Have you drawn on your previous Ebola work in this research at all?
RW: One model I’ve been working on, I’ve used very similar methods from my Ebola work to try and analyze the impact of these shelter-in-place orders on transmission. The way that we've chosen to do it is different than you'll see in most papers. We've broken it down into individual policies. When you think about all of the different travel restrictions that are put on incoming travelers, quarantining them, and putting out rules for increased contact tracing, all of that, it’s a lot of policies. We estimated the impact in aggregate of all these policies is probably to reduce transmission by about 60%. It's quite impactful—like, 60% is a huge reduction. This first paper we’ll publish is a proof-of-concept paper, and then we’re going to extend it to more countries.
AG: What can you tell me about that project with the Public Health Department in San Francisco? What’s your modeling work like, and what kind of results are you finding?
RW: The model that I've been primarily driving here in San Francisco is using an “agent-based” model to estimate the impact of mass quarantine and contact tracing on continued transmission. With an agent-based model, you actually simulate 10,000 specific people (or “agents”), and you know their gender and their age and where they live and what household they belong to. It’s a toy image of the San Francisco Bay Area.
You use census data, what we call “synthetic population,” where within each census tract there are approximately the right number of households and then approximately the right number of people in each household. And then you “infect” people in this population, and what that means is that you have something resembling an accurate network, where we know roughly which people are interacting and how far they're commuting. You can give people very specific characteristics that inform how likely they are to transmit or be symptomatic or die.
The impact of masks is likely to be felt more as the community reopens more.
What was interesting, actually—but it makes sense—is that the impact of masks is greatest when community openness is greatest. We think cloth masks probably reduce the transmission potential by about 30% per contact. So it's not actually an overwhelming amount, but it helps. When people are still sheltering in place and there's not a lot of community contact, and workplace contacts are greatly reduced, the opportunity for masks to be impactful is just lessened. The impact of masks is likely to be felt more as the community reopens more. With shelter-in-place, we think cloth masks will only reduce transmission by 8% or 9%. But as the community reopens, that’s likely to increase to 13%. Thirteen percent doesn’t sound like a lot, and it certainly won’t control transmission on its own. But it does help—because it means that you don't have to find 13% somewhere else, from some social distancing measure. And also, if there was the willpower for us to continue to shelter in place, it could drastically speed up the decay of cases and mean that it would shorten the length of time that we would have to continue our current interventions to achieve a halt of transmission.
We also found contact tracing is not as impactful as you would hope. By the time you’re identified as a contact and have a test and have a positive result—which is about when contact tracing starts—that's normally four or five days after symptom onset. And most transmission has already occurred before that. Also, more importantly, most of your contacts have already likely progressed through at least half of their transmission periods. Then, when you combine that with the fact that we're only capturing maybe 10% to 20% of cases, you begin to feel less optimistic.
The thing that we certainly find with contact tracing is that when you do it badly, it doesn't have much impact. But if you can do it well, like you actually put resources in to do it well—which are a lot of resources—it can have a greatly increased impact. But it's never going to do everything.
AG: What do you hope the general public will learn from your models? And what do you hope other modelers will find exciting about your work?
RW: First of all, I very much hope that they model the same things and with different assumptions and different types of models. What would be most exciting to me is if, with their different models and different assumptions, they come to similar conclusions—because that would actually be the strongest thing that could happen to reinforce our conclusions.
We're doing our job right if every model is different. It's very difficult to write a model that considers every single aspect of this ongoing outbreak. Some people focus on underreporting and some people focus on the minutiae of being asymptomatic to symptomatic to maybe no longer going into work, then going into the hospital and every single step of that path. Some people focus on transmission from travel. You can't focus on the minute details of all of these aspects of transmission simultaneously. But each one of those aspects helps inform and guide the response, in terms of helping people understand what parts of this matter. And if you start getting a lot of disagreement, then that's also really interesting because then you can ask: “Well, what are the different assumptions that people are making to get these different answers? Should we be more concerned about this particular part of the disease transmission when we're making these estimates?”
Still, the most satisfying part of all this is when you spend a huge amount of work creating this model and coding it and figuring it out, piecing it all together—and then you run it, and it works. Not only that, but sometimes it works and it gives you a result that you didn't expect. And sometimes you think about it, and you're like, “Ohhh, I understand why doing it’s doing that. I hadn't thought about that before.”
What I have gained the most appreciation for since I started this job is that models are really best designed for relative answers rather than absolute answers. Trying to ask how will this change, not saying, “It will become exactly this.”
AG: Does it drive you crazy to see all these people on the internet playing around with modeling? What kind of misconceptions does it breed to have all these models floating around?
RW: Many people point to models that have gotten things wrong. And part of that is the modelers’ fault. They don’t make a lot of effort to explain what their predictions mean. But I wish that people would maybe have a better understanding that these predictions are not made in a vacuum. These models that we make are only as good as our assumptions, and no model is perfect.
If a model is good, its actual goal should be to investigate and question how we expect the dynamics of the disease to change based upon our actions and our choices and the environment that we're in. And that means when they make these predictions, they're going to have certain assumptions about either continuing not to have interventions, or continuing to shelter in place. But hopefully, if we do it right and we pay attention to the model, the bad things that we're predicting will never happen. That doesn't mean that the models were wrong.
What I have gained the most appreciation for since I started this job is that models are really best designed for relative answers rather than absolute answers. Trying to ask how will this change, not saying, “It will become exactly this.” For example, people seem not to appreciate that the main expectation is only the mean of a distribution. By which I mean, if we think the mean outcome is 1,000 cases, and the distribution is from 300 to 2,000, we really mean that the distribution is from 300 to 2,000. You should not expect it to be 1,000 cases—because the probability of it being exactly 1,000 is actually fairly small.
People do the same thing with weather. They say, “Oh it’s 50%, 60% chance of rain.” Then it doesn't rain and they say, “Oh, they got it wrong.” But the weatherman only said there was a 60% chance; he didn't say 100%. You should actually understand that that “40% no rain” prediction is actually a large probability.
We don't pretend that we can say the precise number of cases, especially with infectious disease. Every single person who gets the disease is going to transmit to anywhere from zero to 40 people. Trying to predict whether any individual is a “40 person” or a “zero person,” is nearly impossible. And that is incredibly impactful in terms of the growth of the outbreak—if you get a handful of super-spreaders and suddenly it takes off, or you don't get any super-spreaders for a while and then it grows at a more sedate pace.
I don't think that it's impossible for people to understand the concept of uncertainty. But I think that it's been underemphasized in these predictions. Some dedicated scientific journalists actually do a very nice job, but for the most part the lay journalist doesn’t get it right. I sort of wish they didn’t present the mean at all; the mean presents a false sense of certainty. I think if they just said, “They predict between 300 to 2,000 cases,” and just left it at that, then maybe the reader would understand the inherent level of uncertainty here.
AG: Has the hardest part of your COVID experience been professional or personal?
RW: The most challenging part has been that I have my son Leo at home with me 100% of the time—while I'm trying to work more than I normally do, and my husband also is still trying to pretend to work full time. Leo turned four recently, our first pandemic birthday. We had a little party; I made fondue and brownies. Tomorrow, I turn 32, and we're going to have fondue. You may have noticed a trend at this point. [laughs] I can’t remember the last birthday I didn’t have fondue.
I’ll admit, it’s very hard doing work with Leo at home. I love him, but I also worry a lot about the schools getting opened and closed and what that means to him in terms of not having a routine. The school is still doing a half-hour meeting each day, but the hour changes every week, and my meeting schedule also tends to change.
When his school shut down, and they said, “We're going to close for two weeks intially,” I said, “Well, that's a laugh. It's gonna be at least three months.” I knew that going in.
But also it’s meant that I have a greater appreciation for just how difficult it is to predict what's going to happen, even once you have very few cases—if we ever get to that point. Or, I should say, we'll get to that point eventually. One way or another, we'll get there.
This interview has been edited for length and clarity.