Doug Fullam is a credentialed actuary with experience in the life, health, and property/casualty industry. He’s helped firms and organizations across insurance, financial services, and the public sector provide detailed expert risk assessment for their portfolios and risk transfer needs. His technical background includes actuarial science, stochastic modeling, longevity analysis, pandemic risk assessment, data science, and financial risk assessment. At AIR, Doug leads the Life and Health Modeling group, helping customers probabilistically model pandemic, earthquake, and longevity risk to their life and health portfolios.
Doug Topken: I'm Doug Topken host of a new Verisk podcast series on business issues related to the COVID-19 pandemic. As the pandemic continues governments, businesses, and people are figuring out how to manage its impacts on human time, the economy and business around the world. To do that, access to reliable data is critical. In today's podcast we'll provide some perspective on the numbers and projections we're seeing in the news every day, and different factors that come into play regarding your accuracy. With me today is Doug Fullam, director of life and health modeling at AIR. Doug is closely involved with the AIR pandemic model, which is providing forward looking simulations of potential COVID-19 impact. Thanks for being with us today, Doug. To start, let's talk about the pandemic models that are out there and the different projections and numbers being generated. How can we know which models to trust?
Doug Fullam: Well thank you for having me. And it's definitely tough, different models have different levels of detail, and they are and sometimes even for the ones that are dealing with this every single day, you know, reading through all the little nuances about it is very challenging. What are they trying to estimate where are they pulling their data from it all gets muddier and muddier to distinguish between the two and we fully understand that. We should note that when you're trying to build these models, there's kind of three major areas that you want to make sure that you take consideration. You don't have models and assumptions. You want to have it validate well, and you want to have a good data to trust that you're pulling from. So if you don't have good data you know the old adage garbage in equals garbage out - isn't reasonable that we're getting is coming from a reputable source, all that kind of stuff is super important to make sure you understand. When you talk about these methods and assumptions and things that go into that aspect of it, you know, we really want to make sure that your methods and the practices that the model is incorporating into the system, really adhere to what you'd expect in reality as best as possible, so all models are simplifications but they really should try to make sense when you're looking at the parts and the moving aspects of it. And then you need to make sure it validates well, you know, does it make sense when you look at it this event relative to historical events or is the results that are coming out of that seem implausible or just unrealistic. All that kind of stuff is moving within the system and so understanding these models becomes very challenging, but this is why we know. We do have experts in the field and people that are at this level do try to make sure that you provide valid information to people. And you should really get your information from multiple sources to help kind of bring an aggregate sense of the results together.
Doug T: There must be plenty of factors that these models take into account. Can you give us a few examples of some of the tangible factors as well as intangible factors that are looked at.
Doug F: Sure, a few different things that are going on there. So you would really want to think about your transmission rate. So how quickly is this thing spreading from one person to the next – it's super important to understand, these things do grow exponentially. And so thinking about how it spreads from one person to two people to four people is super important to understand but you also need to understand, you know how it's connected from different places. How are we connected from you know let's say New York City to London, or how is London connected to Milan, or how is Milan connected to pretty much any other place in the world. So you really need to understand the transmission rate and the global scale of connectivity that we have. We also need to understand the response metrics that we're doing so how are we responding to it in the US versus how is Italy responding or China or Egypt any of these kind of countries that will affect the continuous spreading of the, of the impact. And then we also need to think about it from a historical context, how are we thinking about this event, and how does it spread relative to what we know today and maybe the technology that we have today versus maybe technology of the past. And so all of those parts are challenging in their own right, but they really allow for a detailed assessment of what's going on. And so if you kind of could put them all together you can really understand where the event might be heading the next week, two weeks, four weeks, etc.
Doug T: Well, we certainly been hearing a lot about transmission rates. Can you explain to me what that really means?
Doug F: Sure. So, let's just think of it as a nice little city that we're all in and, and you come in contact you know with 50 people or 100 people over a given time period. And the question is how quickly do you spread it from yourself to the other person. And so normally there's a couple of different metrics that epidemiologists use one of the main ones that you might hear is the transmission rate, sometimes it's just denoted as, "R", R standing for the word reproduction, and when we talk about that you have the R naught which is the transmission rate at the beginning of the outbreak. When pretty much everybody in society susceptible to the pathogen, it can be spread from one person to the next, without any follow up from maybe herd immunity and past outbreaks things of that nature. And so in essence it, let's say the transmission rates two. So that means, the first person that's sick will spread it on to two other people, those two people will spread it to two more people each, so you go from one person to two people to four people to eight people, etc., and it grows in this exponential fashion. And as it's growing it's spreading to more and more people. Over time though as we start implementing different things maybe a vaccine comes online and so we effectively can reduce the transmission rate because we can inoculate some parts portion of the population, maybe people who got sick and are now better, maybe have their own immunity to it and so it kind of reduces the transmission because everybody in the population is no longer becoming infectious and you get to the concept of herd immunity. And so this is an essence how it spreads and at the beginning of the outbreak is generally the worst time, and then it slows up over time so it starts out with very high transmissions, and then comes down. The objective is to really get less than one. When that rate is lower than one, the outbreak will slow down so if we think of 1000 people are infected, and that rate is less than one let's say it's 0.8, then 1000 people were only spreaded on to 800 more people and those 800 people might only spread it on 640 more people, and it'll kind of come to an end on its own. This kind of a natural progression that outbreaks have over time.
Doug T: Interesting. What are your thoughts on underreporting, can that account for unexplained variations in the numbers of different countries or states?
Doug F: Yeah, so we should be mindful that pretty much any historical event, any material size has underreporting. It's just kind of a nature of the beast that we have to deal with this has been borne out study after study looking at historical outbreaks. So the same is true for this event. Underreporting does vary between countries, but what we should know is that what you hear in the news or generally speaking, what you get is laboratory confirmed tests. But if you really understand the bigger aggregate picture you should think of that as the tip of the iceberg, and a lot of epidemiologists are really trying to infer what the actual case size is. So one way to do this is we can think about it from an antibody test and they're starting to do more and more of those. And they can tell us how many people have antibodies for the given pathogen. We can also look at other aspects – we can look at how quickly it spreads from country to country and that can kind of give us a sense of how many actual cases might be out there. And then we also think about it that generally speaking, there's differences between well resource countries, and poor resource countries, countries with greater resources and better healthcare infrastructure will likely have lower underreporting, but countries with poor healthcare infrastructure, have higher underreporting. Each of them will have underreporting but the scale will be different.
Doug T: What about variations some countries have as many reported cases as others but far fewer deaths?
Doug F: Yeah, there's no real easy answer to this, unfortunately. There can be differences in reporting, so, in certain countries, maybe doctors are really focused on making sure they list everything, and add the COVID-19 as one of the causes of death. Other countries might not list that and they just maybe list the primary cause. We should note that if you do have COVID-19 disease, a lot of times I think come with other things and the person might die of something specific such as a heart attack where they might die of acute respiratory distress syndrome or something like that. So maybe COVID-19 led to a heart attack or whatever the case may be, but sometimes the doctors might not record COVID-19 as the cause of death. And so there's problems with that there's also as you know, going back to the underreporting certain countries are going to be having a harder time reporting than other countries. And then there's also issues about how far along in the outbreak. So generally speaking, in the very early stages, there's going to be more difference between the deaths relative to cases, then you are maybe later in the outbreak. And this is just has to do with the fact that generally speaking you need to get sick and then it might take a few weeks to die from the event and so usually you see the cases before you see the deaths, so you can see a wider range at the start of the outbreak.
Doug T: Sure. Got it. Tell me about population density, how does it play a role in the spread of the virus?
Doug F: There are a lot of things that play a role but population density is, is one of them. It stands to reason that if you live in a more densely populated area you just naturally come in contact with more people per day, you know, wherever you're walking on the street there's probably more people that are passing. Maybe you live in an apartment building where you all ride in the same elevator, all that kind of stuff. So a person that maybe lives in a more rural countryside may only come in contact with a dozen people in a given day on average, but a person who might be living in an urban environment maybe come in contact with 40 people 50 people 60 people. And so, just by that sheer contact that you have, and potentially even being in more confined spaces because generally speaking, you know, denser cities have smaller size homes and smaller apartment buildings and the rest. There's more likely chance that you're going to be in close contact with somebody, and when somebody coughs, they could spread it on to you or vice versa. And so naturally there's just more transmission within that.
Doug T: There's talk the spread of the virus may slow as temperatures get warmer will seasonality affect the virus Doug?
Doug F: Seasonality effects virus transmissions, Yes. So you do see this especially – let's take the normal flu that happens every year or general flu, seasonal flu. Generally speaking, we do see the more intentional winter less intense in the summer. This is really not anything surprising to most people. That being said, when we think about general seasonal flu transmission rates aren't as high. So, this event as transmission rates and we, you know, going back to what we talked about before, R not factor somewhere in the mid twos, but usually seasonal flu might be a little bit in the low ones meaning 1.1, 1.2, 1.3, etc. And so if that golden objective is to get less than one, it's a lot easier to get from maybe 1.2 down to less than one... there's less social distancing that needs to happen there's less isolation and quarantine that needs to happen to kind of reduce that transmission and then getting something from two and a half, all the way down to one. And so, just the seasonality of maybe we spend less time indoors with each other we have less contact with each other or at least when we do it's further apart from each other, will naturally help bring that that transmission rate down. And so, you know, we do see that and we kind of expect at least some level of seasonality to play a role here, but we shouldn't think of it as being the silver bullet that's going to reduce that transmission rate down from two and a half, all the way below one. What we really need to think about is a multitude of factors that are interacting to help reduce the transmission.
Doug T: I didn't want to end our podcast today before you could tell us about the Verisk COVID projection tool.
Doug F: Yeah, of course. So it's Verisk COVID-19 dashboard. And this tool effectively is designed to give information to all seven plus billion people around the world. We want people to take the information and use it to whatever benefit themselves or their community. So if we're talking about the insurance community, obviously we want whether it's the P&C industry or the life and health industries to take that information and try to understand what that means to impact to their particular business, whether it be life, health, workers comp, liability, whatever the case may be. And they can kind of get a sense of what different scenarios would produce severity of low, moderate and high, but obviously if we extend beyond that you know the public sector of trying to make plans for how the event might happen and unfold in the next month or so, use that information to better make decisions and hopefully make and help reduce the impact of this event has, or even just for the average person who's trying to learn more information about this, about what might be coming just for their own edification, they can log in, get information from that and they can even ask us questions. So there's an email that they can click and shoot notes out and we do try to be responsive to that as much as possible and provide more detailed information if they want. So it's really just kind of our own way of helping the global community in some kind, and get additional feedback, get information and and you know if we can have a have a benefit to society, that's all the better.
Doug T: Right, I'd like to understand more about this model and your approach and building it. Can you tell me something about you know what goes into building the model the approach that you take?
Doug F: So, the baseline of the model is called an SEIR model. Sometimes that's abbreviated as SIR model, but what it stands for is this Susceptible Exposed Infectious and Removed. And if you think about how a person will progress from being potentially infected to potentially getting better or unfortunately dying, they're going to go down this path of being first susceptible, then they'll become exposed, and infectious and then removed, out of the population and as I mentioned removed can be death or get better. And so in essence the modeling is taking that information, and when people become infectious it then allows it to spread on to susceptible people, and they will become an exposed, infectious people will move to the R category removed, the E people will become I, and they'll kind of cycle through this over and over again. This model allows us to implement the impact within cities in different parts of the world, because each city is going to be in their own unique place during the outbreak, some are going to be earlier on the outbreak, some are going to be later in the outbreak. It also allows us to model the implementation of effective isolation quarantining social distancing. You can do that region by region or even city by city, and we can even implement if and when a vaccine comes available, after that vaccine has by vaccinating part of the susceptible population and effectively removing them from the overall system. And this model is is a pretty common framework that people use to really understand how disease spreads from person to person and it's effectively we're taking this model, which has been developed many decades ago, but allowing computer simulations to really run through many different iterations that could happen and unfold and provide more detailed analytics to help them whatever the planning maybe like I said insurance, public sector, or just general edification.
Doug T: Well thanks for joining us today, Doug, and thanks for sharing your insights with us we certainly appreciate you taking the time to speak with us today. To learn more on this and other COVID-19 related topics, be sure to follow us and visit Verisk.com. We hope you enjoy this podcast and invite you to join us again. Until next time, stay healthy everyone.