Full Interview: Projecting the Future to Tackle COVID-19 Today With Jeffrey Shaman
By: Makena Binker Cosen (CC '21)
Tell me a little bit about your background as a researcher.
I study infectious diseases. I’m in some sense an infectious disease epidemiologist; although my training is in climate science, atmospheric science and the like--actually geophysics. I do a lot of work studying how the environment, whether hydrology and climate, affect infectious diseases. I look at some other elements of health as well, but my big focus has been infectious diseases. Over the years, I’ve worked on the two of those together, but I've also worked on each topic separately quite a lot. I’ve done some just straight research on climate; and I've done a lot of work where I’m just looking at infectious diseases. When I work on infectious diseases, I very often work on using mathematical modeling systems, in which I try to represent the virus as a series of deterministic equations that describe the propagation of the pathogen through a population, a community, some aspect of its transmission characteristics. We often use these models now, in the last decade in particular, in conjunction with what are called “Bayesian inference methods.” These are statistical methods that allow assimilation of observations and use these measures to arrive at a truer representation of what the model’s behavior should be. In other words, we inform the model by observations of what’s actually going on. This is something that is done in a lot of disciplines-- it’s done in engineering design, it’s done in numerical weather prediction-- where you take information about a system and use that information, which you know in and of itself is flawed, because observations have error and noise associated with them, and you’re trying to bring it into a system so that the system you’re mimicking, is better constrained. So, we’ve done a lot of work with those systems; and there are a lot of questions you can ask. You’re going to ask questions like, “Can I understand the fundamental properties of the virus, or pathogen, that I’m studying? What are its transmission characteristics? How long is a person infected on average? How long are they latent? How transmissible is this virus? How are environmental conditions modulating transmission? Are they important and to what degree?” You can also use it to study interventions. Once you have a model that you think is representative of the system it describes, you say, “Well, what if we did this or what if we did that? What if we had done these?” These are counterfactual simulations that allow examination of varied situations-- if we were to immunize this many people over time, how effective would that be, given this efficacy of the vaccine, this rate of compliance in uptake, this timing of deployment--what does that mean versus another strategy. And the last thing you can do with those types of systems, is you can also try to develop forecasts of what is going to happen. These methods, which combine mathematical models, data observations, and these Bayesian inference systems, are similarly used as the basis of numerical weather prediction. That’s how they make any and all the forecasts that we rely on all the time. We’ve taken that same suite of methods and we’ve transferred them into an infectious disease system, where we have a mathematical model of an infectious disease system. We have the same types of Bayesian inference methods, and we have observations of the disease system, like the number of cases of somebody with COVID-19 or Influenza, or a diarrheal disease. So, we work in sort of that milieu--where we’re trying to understand the properties of what’s going on, trying to use that understanding to inform what it means about what we should do and scope out types of interventions, controls, and consequences of our actions. We’re using it to make projections; and if the system is what is called stationary, when we’re not really impinging upon it, then we can make forecasts that have some sort of accuracy that we can quantify.
With this background, what features of the coronavirus do you think have made it particularly dangerous as a pandemic?
It has three features that are particularly problematic, that make it the most challenging pathogen we’ve faced since 1918, as a world population--humans that is. Firstly, it is a newly emergent respiratory virus, to which the majority of the world’s population has no immunity. We’re sort of like the dry kindling to it, and it can spread really quickly through our population. The second is that it behaves as a lot of common respiratory viruses do, in that the majority of infections are undocumented. People have little symptoms or mild symptoms; not enough to motivate them to go see a doctor, and not enough to often motivate them to stay at home. So, they’re out and about, spreading the virus broadly in the community, contributing to the silent or stealth transmission of it. The majority of time, you or I, get an infection from a cold in the winter time, we get a sniffle or sore throat or body aches, we don’t stay home. We still go out, go to school, go to work; do the things we need to, go out with friends. If we have a trip, we’ll still get on an airplane. What we’re doing when we do this is we’re spreading the virus broadly in the community. This is how the common cold gets around. This is how influenza gets around. Because there are enough of these infections that set up these silent, stealth trains of transmission that allow a virus to go along undetected by people unwittingly spreading it to other people, not really being cognizant of just how ill they really are. And we’ve seen this ourselves because we’ve done testing of this in the broader community, where we can go to a major tourist attraction in NYC and we can solicit people for testing. They tell us their symptoms, we swab them, we take it back. In February, we found that one-in-nine persons walking around, ambulatory adults, was shedding a common respiratory virus. It’s out there. There’s a lot of it out there and there are many chances for an infected individual to potentially give it to somebody else unwittingly. This virus has that property. Not only do you have the majority of infections being undocumented, additionally for documented infections there is evidence that people shed the virus before they are actually symptomatic. That’s pre-symptomatic shedding. Even the people who will ultimately seek clinical care and be documented as having COVID-19, they were contagious before they were aware that they had symptoms. So, because of these properties, this virus gets around really fast. That’s why it spread so quickly in China, that’s why it started hopping on planes and going to Thailand, the US, Japan, and South Korea. That’s why it spread throughout the world in a matter of a month, really from mid-January to mid-February. The third ingredient unfortunately is that while the majority of infections are undocumented, there is still a fat tail. There is still a substantial number of people for whom this causes severe outcomes, critical complications, and puts them at risk for death. There is a substantial infection-fatality rate, that’s probably somewhere between 0.2 and 1.3% overall. That’s the infection-fatality rate, not the case-fatality rate-- which means it’s probably really between 0.5 and 1% I would guess. This means between one in one hundred, or one in two hundred persons, who get infected from this, including all the undocumented, will die from it. When you do the numbers, and you say, this is a novel, respiratory pathogen, to which the majority of the planet has no immunity-- everyone is susceptible to it--and given how aggressively it’s transmitted, you’re expecting 50-70% of the world to be infected with this before it really burns itself out or settles into a pattern of endemicity. That’s four billion people, let's say, who will be infected by this virus, if we do not check it, if we don’t come up with an effective therapeutic or vaccine, if we don’t slow its progress by all the non-pharmaceutical interventions that we’ve been utilizing in the intervening time while we wait for those pharmaceutical interventions. If four billion people were to get this, and one in two hundred of them were to die, that’s twenty million people. That’s the worst pandemic we have seen since 1918. The worst-case scenario is clearly very bad. The combination of characteristics this has: high susceptibility because its a novel pathogen, the fact that the majority of infections are silent, undocumented and they allow and support the transmission of this virus freely in the community, when people don’t know that they are actually contagious; and the fact that there is still a substantial number of people for whom this will have critical consequences. You put all that together, and that’s why we’re looking at such a really dire circumstance.
What is your team doing right now? What kind of simulations is your lab running? How do they work exactly?
The models can be more simple or they can be more complex. We’ve run models at NYC scale, in conjunction with the Department of Health and Mental Hygiene; this is work led by Dr. Wan Yang, who is an assistant professor in epidemiology, and tried to provide them projections of what is going on and understanding of where this virus is and where it may be going within the city. We’ve looked at it in China, and actually have tried to quantify those undocumented infections and how contagious they are. That is something that we did back in January and February. We’ve applied a similar model at a county scale to the entire United States, and used that to try to understand the epidemiological characteristics of the virus, whether particular locations were in a position they should be loosening restrictions. In other words, was the reproductive number is well below one. And one is that critical line between growth and diminution of cases as you go forward, and you’d like it to be below one, so the outbreak is dying out as you’re thinking about opening up things and allowing for more activity. We’ve used it to make projections as to what’s going to happen. I call these projections, and not predictions, because the system is not stationary. Not only do we have uncertainties as to the data quality, differences in testing rates, and things that make it hard to actually understand and identify the system well to make what would be a forecast, we also have a blindspot that we don’t know what happened over the last ten days, because there’s a lag between when someone acquires the infection and when they’re confirmed as a case. Even more importantly, we don’t know what people are going to do in the future. When you make a prediction, you’re assuming that behaviors are the same. When you make a weather forecast, your assumption is that you can’t change the weather, and nobody’s going to go out there and try to disrupt a hurricane if it’s coming your way. We have no means of actually doing that. When we make a flu forecast, there’s the assumption that we’re not going to do too much different than what we typically do to disrupt the movement of the flu and its propagation through the population. That’s not the case here. We’ve completely disrupted our economy, we’ve shut businesses and schools. We’re leading very different lives right now than we were formerly. That has had a marked impact on disease transmission. We don’t know how much we’re going to stick with that going forward. We don’t know what kind of loosening will take place. We don’t know, even if we have that loosening, which we know when they’re going to do it, they’ve already done it for a lot of these...we don’t know what the outcome will be. We don’t know what the consequences of those loosening restrictions will be on opportunities for the virus to be transmitted from person-to-person, which is really the fundamental outcome we’d like to know. The reality is that if a particular location opens their businesses, or allows their businesses to open, we don’t know how many businesses will open. Some will choose not to do so, because they consider it unsafe for their employees or their customers; or it’s not economically viable for them to work in a partially open economy. Then, we don’t know how many customers will frequent those businesses. Are people going to restaurants? Are they actually going to use them, or are they going to be shy about doing it, and how that might vary from place to place? We don’t know what kind of social distancing will be enforced--if at all. Will they enforce it or will they not? Will people wear masks? Will it be outdoors or indoors? What are the effects of changing weather conditions and humidity on this, is this virus seasonal? There are so many things unknown, that making a prediction is not possible. So we instead make projections. And you might say, “so, why make them at all?” The reason is to scope out what are the possible outcomes. When we make a projection, we assume certain things will happen. We assume that maybe they’ll be a five percent increase on a weekly basis, in the transmissibility as more and more people get relaxed, and there’s less and less compliance with control measures, more businesses opening, and less compliance with social distancing and face-mask wearing. Maybe we’re getting more cavalier. That might be one scenario- when you say that the ultimate outcome of all those effects is this gradual increase in the force of transmission. What’s the consequence of that? Or maybe there will be no effect at all, and you can just say that it’s all going to stay the same. Business as usual. But you have to lay out these scenarios. By doing that, you can see what the sensitivity of outcomes are to those future paths. No matter what we do looking at growth, or no matter what we do looking at the virus declining in locality, or is it dependent on what we do? And if so, what does that mean about tracking and understanding what needs to be done, and the responsiveness that we need to have. It’s really important in this situation that we’re very responsive. When there are outbreaks taking place, we would like to see the states and communities responding aggressively. Unfortunately, that has not been the case.
How easy has it been to get reliable data? Do you expect there to be any limitations or biases based on this?
Certainly there are biases in the data. The deaths are undercounted. Not everybody is tested, not everybody is known to be having COVID. If you look at some of the preliminary studies, the NY Times even did one for some European countries showing excess mortality versus COVID deaths; and they found that there was like a 20-30% gap between them, indicating that excess deaths were higher than they would be otherwise. Right now, you might expect, maybe there are a few more deaths because they can’t do heart surgery the same way in places where hospitals are overrun with COVID, or they can’t treat accidents the same way; but you could also expect there to be fewer traffic accidents because there are fewer people on the road. So how that all comes out, I don’t know. But what it really indicates is that there is a sizable undercount. I’ve also had someone reach out to me from a hospice place in rural Georgia saying that, “We used to get 1-2 people coming in with sepsis each year. Now, everybody has sepsis, whose coming into die, and none of them have been tested. They all are COVID patients, but they are not known to be.” This can be very bad in some places. Brazil, news reports suggest, is grossly undercounting cases in deaths due to COVID. I mean, substantially, maybe by a factor of two or three or more. So, there are real problems, and it varies considerably from country to country based on what their practices and policies are for it, and what their capacity is to test; and even serve people in their health needs, particularly when you get into the much more marginalized portions of the developing world. In the US, there are biases in the death data. There are shifting biases in the case data that depend on testing and what not. Our job on using these models in inference systems is to allow for and account for biases in the noise in the data. Remember, in the very beginning I said, “We know that data is not perfect. They are noisy in and of themselves.” That has to be explicitly accounted for, because you can’t trust the observations. They are not the Gospel. There is some unknown truth out there that we’re trying to estimate better, by using both the model and the observations in conjunction with one another.
With that in mind, what is the importance of testing and contact tracing right now?
It’s vital. If you look at the countries that have been very successful with dealing with this. It’s a hard problem. Basically, we are looking at something where we’ve been tasked with a very hard problem. We need to suppress the virus using non-pharmaceutical interventions, and not substantially disrupt economies in the process of doing so. If you look around the world, every country in the world is tasked with this problem, and almost all of them have failed. They are ignoring the problem, or they are outright not dealing with it well. There are a handful of countries that have done it well. Vietnam, Taiwan, Thailand, South Korea did well after dealing with a large outbreak. China, in a very different, authoritarian way has been able to handle it. It’s interesting that these are countries, near the epicenter, have also had to deal with SARS and MERS, have had experience recently in dealing with infectious diseases, which I would argue has primed them for it. There are other places as well. Germany has done better. Israel, New Zealand, Iceland. They have done well as well. South Korea, because it is a densely populated, highly populated modern country, that is well-connected with the world economy, and people fly through it all the time, is probably the best example to look at. They aggressively developed tests when this virus emerged, the same day by the way that it emerged in the US. They had their pharma companies develop a lot of diagnostic tests, that were then verified by government labs, and then flooded the market and made ample opportunity for anybody who really wanted to be tested. They had laws in the books already that allowed them to use people’s credit cards, video surveillance, and GPS records from phones to contact trace. They could actually go into people’s information to actually support this effort to control the virus. They also had a quarantine act that allows them to enforce quarantine for individuals who have been in contact with someone who is infectious. They used this, they shut their schools. High compliance in use of face masks there. They didn’t shut their businesses. That’s not to say that their businesses didn’t take a big hit initially-- no one was going to them comparatively. But, they managed to really squash the virus. Since then, they’ve further reopened things. More people are using the businesses, high use of face masks. They’ve opened houses of worship, baseball season going at stadiums. They’ve opened night clubs. They’ve driven it down to single digit numbers of cases per day. Since opening the night clubs, they’ve had clusters of outbreaks. They had up to eighty cases in a day, usually around forty. Now, forty cases a day, if you were to just go with that number, for a thousand days, which is almost three years, that would be forty thousand cases, right? If we were to say that maybe there are ten times as many infections overall as there are cases, which is probably a high number for Korea, because of how well that they test. Let’s just go with it. That would be 400,000 people infected in the next three years roughly. That’s less than one percent of their total population. What it means is that South Korea, in their new normal, with reasonable economy, maintaining their unemployment rate at four percent, could hold on for three years while the world tries to develop an effective vaccine or therapeutic. And they could deploy it before 99% of their population is ever infected with this virus. Now, Korea’s population is 51 million people. It’s about one-seventh the size of the US, actually a little bit more than that. Forty cases a day would be 280 cases a day in the US. In the US right now, we are at 20,000. We’re nowhere near that; and we’ve disrupted our economy and skyrocketed unemployment. So we’ve unfortunately failed at both. We’re not alone. Lots of countries have. Right now, we seem unfortunately to be ignoring the problem too much, and ignoring the fact that this can take off exponentially again, and that we may be catching a break in summer, because it may be somewhat less transmissible in summer. But if we’re really complacent about this, and we’re not willing to reimpose the kinds of isolation and shelter-in-place methods--this virus will run over us. And accepting that in the US, accepting that a million people will die from it, seems really, really like a failure of leadership.
That’s a lot to take in. With how quickly everything is evolving, how often do you find yourself needing to update your projections, or how often should people that are following up on projections check again to say that they still hold over.
We update our projections twice a week actually. So, we bring in new information from the last three or four days, and we update the projections again. We also change our scenarios. When we were first doing this, we were just imposing scenarios where there were further restrictions. In March and April, we were looking at people using shelter-in-place, restrictions on mass gatherings, closing of businesses and schools, face mask use, and social distancing, to control it. We had scenarios where there was increasing control of the virus invariably so. Then, once they started to loosen restrictions, we had to change our scenarios and say, “Well there are possibilities of growth because they are loosening them now.” We don’t know how much. We’re again having to reconsider them based on what is going on.
Has your team modeled how protests and the response to protests may influence COVID-19 outcomes?
It’s too early to tell, because they’ve only been going on for twelve days now. That’s about the length of time between somebody acquiring an infection and being confirmed as a case. Certainly, you look at the protest and demonstrations; and in some places you see a lot of face mask wearing, and it can be during the day time and it’s outdoors, and those all may help preclude the transmission. But you bring people together and you would figure that you are providing more opportunities for transmission. This is not to diminish the need for the protests, because that is a choice, and these issues are very important, equally important, if not more important, and they are public health issues as well. There are other locations where people are not wearing face masks at all. You just don’t see that level of compliance there, and that’s more concerning. But the truth is that we don’t know. We don’t know what this all comes together to mean. If being outdoors in the open air, where the wind whisks things away, and if people are wearing face masks, and if the majority of protestors are in the sunshine, and UV radiation breaks down the virus when it’s exposed to it, then the risk may be slight, may be nominal. On the other hand, it may be large. We don’t know. As of yet, we don’t know because we haven’t also had enough time to see whether or not this is resulted. The same thing goes for all the other loosening restrictions. So we’ve had people getting together and going to the beach on Memorial Day. We’ve had them walking on boardwalks and going in stores. You see shots of people going into nightclubs, bars, and restaurants; and not wearing masks as well. Those environments may be more problematic because they are indoors, at night, and people are pretty packed together as well. They look like opportunities for super spreading. It’s really hard to pin down why we’re seeing growth in some locations in the US and not in others, why some places are still declining, and why some places like California, Arizona, Texas, Florida, the Carolinas, Alabama, are seeing growth in cases. It’s difficult to know what’s going on. There are changes in reporting and testing practices over time as well. We don’t have access to information on the microscale processes that might be supporting or not supporting transmission locally.
Your comment will be posted after it is approved.
Leave a Reply.