Coronavirus in Context: Expert Unpacks Data Behind COVID-19 Model, And Why Projections Differ

[MUSIC PLAYING]

Welcome to "Coronavirus in Context". I'm Dr. John Whyte, Chief Medical Officer at WebMD. We hear a lot about these different models. How long will the virus last? When's the peak? How many will die? I'm delighted to be joined today by Dr. Teo Vos. He is professor of Health Metrics and Sciences at the University of Washington. Thanks for joining me, Dr. Vos.

Pleasure.

Let's start off with, can you explain to us at a high level how these models are created? What type of data do you use, and where do you get it?

We, early on, made a choice to focus on daily counts of death as the starting point to understand the shape of the epidemic. The reason for that is that the numbers of positive tests is much influenced by testing practices. Not only differences between countries, states in the US, in how plentiful these tests were available and were being used, but also over time, there are big shifts in the positive numbers. And so we said, there's less ambiguity in the counts of death, although there are still problems, with late reporting and deaths being missed. But by and large, it is a more solid number to tell us about the shape of the epidemic.

But where are you getting the data? Is it from public sources, or are people sending you the information about testing and deaths and resources?

THEO VOS: It's other institutions who are collating the information, that we rely partially on what other people do, we look at websites of countries, of states, and we identify the most reliable source for the daily updates of the counts of death.

Now, what is your model showing right now, in terms of where we are in the peak of cases?

It depends on where you are, and we clearly have seen the peak and, even the end of the first wave in Wuhan, in China. We are now increasingly observing, in parts of Italy, parts of Spain, some of the US states like Washington state, that have already reached that peak in deaths, and are on a trajectory of decline in the number of daily deaths.

From the model of deaths, the key things that we project are the peak and then the time when we think the numbers are going to go down to very low levels. What we then do is we back calculate from the curve of the deaths, of what the curve of hospitalized cases is. And from that, we can then make statements about the need for health services to cater for symptomatic patients, and then what the likelihood is, among those symptomatic patients, how many are so seriously ill that they would need intensive care treatment, and/or ventilation.

Now, your colleague, Dr. Christopher Murray, has predicted that coronavirus deaths will come to a halt this summer, with zero deaths after June 21. Yet other models from Harvard, published in science earlier this week, says that we'll have to have some intermittent social distancing till perhaps 2022. How do we reconcile these different models?

I don't think you're quoting Chris Murray very well, because the going back to zero cases and maintaining a level of zero cases would require a large effort, and ongoing social distancing measures, even if they are not at the level of the restrictions that are currently being mandated.

But it's a key thing that we need to start looking at now, when is there a safe level of numbers of new cases, and we can back calculate that from the shape of when the deaths occur, at which the public health agencies and the laboratories and the health system would be able to move to a phase of active surveillance, where anyone with even the mildest symptoms that could indicate a COVID infection are getting tested, get a very quick return on that test, and then public health folk do aggressive contact tracing of every positive case, and yet again, contact tracing of contacts that are found to be positive.

If that is put in place, that would be safe guarding us from seeing a second wave, or possibly a third wave, or a fourth wave. And it's not unlikely that the second wave, or a third wave, is going to occur, because we estimate that the current epidemic has impacted less than 5% of the population, even in places with very large numbers of cases and deaths. That means that the vast majority of the population remains susceptible.

Yeah.

And if we don't take these measures, they'll be exposed to the virus, and we can see this explosive growth coming back.

Now, I wanted to ask you, are policymakers using the models correctly?

I don't think I can give 100% yes or no answer to that. I must say that said that both on the side of policymakers that we interact with, and the media, that there is a surprising degree of savviness-- that people ask very good questions. And so I think with some help that we can give, there is a reasonable understanding of what the models mean. And we are the only modeling group that has come close to predicting what the shape of the curve was going to be.

You probably remember a number of these more traditional infectious disease models that go from susceptibles to exposed to infected to recovered, and they have this assumption in it that there's an ongoing explosive growth, and many of those models end up with 70% to 80% of the population infected, and extremely large numbers of deaths.

I think we've come much closer to the mark. I think we're actually adjusting, in some of our locations, the peak of the curve to an earlier date, and the total number of deaths set to lower numbers than we originally had projected.

But then it's the underlying assumptions that are so important, that you just described. Are people not taking time to either look at them, to understand them, to report on them?

Well, I think the best way to deal with that is to calibrate your model frequently, with newly observed data. Because the further we get into the experience of various locations with the epidemic, the better we get at predicting what the shape of the curve is in places that are still earlier on.

When we started this several weeks ago, all we had was the Wuhan experience, and so we were making everyone follow Wuhan. We now have a whole lot more locations that have gone over the top of the hill of the curve. And so we're now feeling much more comfortable of what that shape is, and it actually hasn't influenced the shape of the curve that much since the early days, when we only had Wuhan to work with.

That's a good distinction. So you're fairly confident in your predictions and your estimates?

I think for the trajectory, to the peak, from the peak, to where the cases will drop to significantly lower numbers, yes. I think the big sticking point there is what is going to happen with the social distancing measures, and more importantly, apart from the actual measures that are in place, how is that shaping behavior?

Are you trying to predict the future with these models, or are you using these models to influence the future?

I think it's a combination of the two. We wouldn't be doing this if we didn't want to influence policy, and therefore by changing policy to influence what the predictions might be. But mind you, the predictions that we are making have an inbuilt assumption that social distancing measures continue until the end of this first wave of the epidemic. And even in places where currently, there are inadequate social distancing measures, or less strict, we make this assumption that they will inevitably move to a complete set of measures within a week. So it's already-- what we are showing is an optimistic scenario, but taking into account that all these measures are actually in place.

Those are good points. And in our last minute or two, when people come to your site at IHMS or HealthData.org, what do they need to know when they come and look at some of these data points and some of these graphs?

The first thing I would suggest is that people click on the frequently asked question button, because there's a very good section that answers a lot of the first type of questions that users may have. We also regularly produce updates, where we say, OK, this is what has changed, this is why we are seeing differences in the numbers that are being produced.

We're working extremely hard today to build in, rather than a concept that is more coarse on individual types of social distancing measures, what the actual effect is as measured by cell phone data. And so we're hoping that that will give a more granular sort of measurement of what these measures have brought about in terms of behavior change, including that places that may not have fully mandated certain measures, how have people reacted to it?

So those are some of the things that you need in the future, to continue to refine the models?

Yep. And we are also planning to put out a set of dates by state in the US, at which point each state is reaching such a low level of estimated new infections in the community, that it would be a point where we can start thinking about lifting some of the most strict social distancing mandates, as long as it's, as we will advise, is accompanied by this very aggressive surveillance that will be needed to prevent a second wave.

And I think it's an important point worth repeating, Dr. Vos, how you pointed out that even though we like to talk about the peak and when it's reached, it really depends on where one lives. And we want to put it in that framework, correct?

There may be partial opening up in the country, where some of the states that have gone further ahead in the epidemic may open up earlier than others, where we may then still have measures that travel between states is being discouraged.

Well, Dr. Vos, I want to thank you for taking the time and helping to explain these models, and I hope we can have you on again.

Pleasure.

And I want to thank you for watching "Coronavirus in Context". I'm Dr. John Whyte.