Vermont's COVID-19 Modeling And What It Means For the Weeks Ahead
Vermont's COVID-19 model was released Thursday and projects the "likeliest case" scenario of not having enough ICU beds, but having suffiecient stock of most PPE at the peak of the virus. This hour, the state's financial regulation commissioner, who's been overseeing modeling for the Scott Administration, breaks down how the model was created and what it mans for Vermont.
Our guests are:
- Michael Pieciak, Vermont Commissoner of Financial Regulation
- Maggie Koerth, senior science writer for FiveThirtyEight
Share your questions or personal experiences in the comments below.
Broadcast live on Monday, April 6, 2020 at 1 p.m. Rebroadcast at 8 p.m.
The following conversation has been edited for brevity and clarity.
A Q & A with Commissioner of Financial Regulation Michael Pieciak
Why is the Commissioner of Financial Regulation the lead on Vermont's COVID-19 modeling?
The Department of Health is very engaged in the work they are doing with regards to contact tracing. A lot of their epidemiologists are dedicating pretty much all of their time to contact tracing, so the governor asked us at the Department of Financial Regulation if we might be able to pitch in... and look at where we are headed with this information and data. I think we have a track record of getting difficult things done.
What did that process look like?
First, we had to make sure we had input from both internal and external stakeholders. We found some experts outside of our department who are leading experts in their fields to inform our decision-making and the work we were able to produce last week.
What does the modeling suggest for the number of cases in Vermont and the projected surge?
We are looking at the aggregate number of cases, but we are also very concerned about the day-to-day growth in cases. Those are some of the factors that are most important when considering the future and what it will mean for our hospital resources. In other words, what will be the demand for hospital beds, ICU facilities, for ventilators?
The best research out there right now contends that about 19% of those that fall victim to the virus will require some form of hospital care. About 14% will require care in a staffed hospital bed and about 5% will require ICU care.
By tracking the aggregate numbers of cases in Vermont and what the daily growth rate is, we are then able to estimate what a percentage or range would be, of hospital beds and ICU beds that might be needed.
What's critical is not just how many beds or ventilators will be needed between now and June, but at what point they will be needed. What will the peak look like? How much time do we have until the peak? As you mentioned, Jane, we laid out in some detail a worst-case scenario, a most-likely scenario and a best-case scenario.
What do those scenarios look like?
When it comes to hospital beds, ICU beds or ventilators, the worst-case scenario was certainly shocking in terms of the outcome.
In each of those instances — for the worst-case scenario — we were going to surpass our hospital resources by early April.
Fortunately, the information we provided last week tells us we are clearly not on that "worst-case" trajectory.
We are doing even better than the most-likely trajectory, and we attribute that to all of the sacrifices Vermonters are making, day in and day out.
When you look again at that most-likely scenario trajectory, the modeling is still saying we will need close to 600 hospital beds. When it comes to the ICU, we're going to need somewhere between 100 and 200 ICU beds at the peak.
We are thinking that the peak will come sometime in mid-to-late April or early May. That is when we anticipate, at this point, seeing the highest growth in COVID-19 cases and ultimately the greatest strain on our hospital resources.
What did you find in terms of what the state might need with regards to resources, given the expectation that the surge might peak at the end of April or in early May?
There are currently 622 hospital beds available in Vermont. In the worst-case scenario, the state could need at one time as many as nearly 2,600 beds.
Under the most-likely trajectory, the state has enough hospital beds to accommodate COVID-19 patients. You said we are actually doing better so far than what the most-likely scenario projected?
That is correct.
In that case, we would, in the most-likely scenario, get up to about 521 hospital beds being used at one time by COVID-19 patients. So we'd still have capacity there. But in terms of the ICU beds, even under the most-likely scenario the state is projecting, we would surpass the ICU bed capacity we currently have. We'd need about 211 ICU beds and we have 135 available.
So what can the state do about a shortage of hospital beds once you have identified that as a likely scenario?
Let's talk about hospital beds first. The 622 beds I referred to cover what is available in hospitals at this time. There are also opportunities to create additional surge facilities outside of the hospital system. That is what the state leaders at the Vermont Office of Emergency Management and National Guard are working on and preparing for.
The other thing the governor has done and hospitals have done, is to cut back on elective surgeries — procedures that could wait until after the pandemic is over. That is really what has driven Vermont's hospital bed availability up.
For example, we were at around 300 available hospital beds just a couple of weeks ago. Now, we are trending between 600 and 700 beds. That's because there has been less routine work that has happened at the hospitals; it's been put off so that any surge that's seen over the next few weeks can be handled.
What can be done to increase the number of ICU beds available during a surge?
When it comes to the ICU, that becomes a little trickier because there are still individuals that need ICU care that is unrelated to COVID-19. You can't eliminate that demand. You have to be thinking about ways to build out ICU capacity inside the hospital, or potentially near the hospital as well.
What is the main difference between an ICU bed and a hospital bed when it comes to planning for the surge?
The main differences are in the level of staffing and some of the equipment. The equipment can be dealt with, but the staffing is really the key here.
The governor has called on those with a medical background to step up and volunteer here, and that's really the reason why.
We can build out the facility, purchase the equipment we need, but it really comes down to staffing these facilities to make sure that Vermonters are getting the care and the treatment that they require.
Even though we are not on the worst-case trajectory, you want to plan against that worst-case trajectory. This is a very fluid and volatile virus, and it's equally fluid and volatile when you are modeling it. Things are trending in the right direction right now, but that can change very quickly.
We need to prepare for the worst while we do everything we can to create the best outcome.
What will be the biggest deciding factor in whether we stay on our current course?
Even under the likely scenario you pointed out, we are going to go past our hospital capacity. We are working hard to make sure that doesn't happen on the resource side, but we truly have the power in our own hands and in our own decisionmaking as Vermonters to make sure that we don't come close to that, by following through with social distancing measures as Vermonters have been doing.
It's good to see that those sacrifices are having an impact and slowing the spread of COVID-19 in Vermont. That's why we say that we see some glimmers of hope, but that the worst is still ahead of us and we need to double down.
Vermonters and their individual choices have much more control over this than anyone else does. They've changed the trajectory in Vermont, they've clearly helped us with our hospital resources, and they have saved lives.
How dramatically have Vermonters changed their behavior?
When you look at mobility data from cell phone use — how often people are going on their daily commutes, moving around their towns — the information for Vermont shows a considerable decrease in commuting and in how frequently people are running those errands, by about 50%.
to be clear, the data I'm talking about is the sort of data collected when you download an app and say, "Yes, you can use my location." It's collected in aggregate and it's anonymous and the state of Vermont isn't doing the tracking.
How can you expect the number of cases to peak if there is no vaccine and we aren't sure that those who recover will have immunity in the future? What would cases not just continue to increase with more exposure?
Our modeling is assuming that social distancing measures stay in place until at least mid-April and show that the total number of daily cases would start to decline sometime around the middle to late part of the month. That's because people are in less contact with each other, so you're seeing reductions in transmission within the population.
However, it's true that one of the consequences of having success at social distancing is that a greater percentage of your population is still susceptible to getting the virus when those social distancing measures are released.
What is the purpose then of slowing the growth rate down?
It lets scientists and medical professionals get a better understanding of the virus and more experience treating it. It gives hospitals more time to build up their resources, so that if we do see a spike further down the road, we have a greater capacity to respond to those increases than we would today or a month ago.
It also allows, across the country, for expanded testing. If we get more testing in New England and Vermont, we increase our ability to jump on any outbreaks. That changes the trajectory and type of social distancing required.
And, this gives doctors and medical professionals more time to think about treatment for the disease. Even without a vaccine, if we had the opportunity to deploy a treatment that might work for a more severe case or even a mild case in early May, that is a very different scenario from where we are even today.
Does your modeling rely only on data from those who test positive? What informed your assumptions?
The assumptions being used in our model and those around the country draw on the experiences of researchers in China, in Italy and increasingly in Spain.
We recognize that there is under-testing and underreporting going on.
Some percentage of your population is either asymptomatic and so have COVID-19 but are not displaying symptoms, or it's mild and they may think they have some other illness. We assume in our modeling that that is going on.
Do Vermont's models include the return of snowbirds to their second homes and how would that affect the projected peak?
Yes. That is why the governor issued the order asking people to self-quarantine for 14 days. It's to help them protect their communities in Vermont and to protect themselves.
On the hospital side, if you have 30,000 to 40,000 additional people in the state [that is the current estimate], that is something we need to take into account when considering whether we have enough hospital resources to accommodate these people. We've taken that into account as well, to make sure our resources count not just for Vermonters but for others who may be in Vermont.
How granular can you get in Vermont? Is the state modeling spread of disease county by county?
There is an element of predictive power that you gain when you pass the 100-case threshold. This is something our experts have been telling us and we saw it play out in Vermont, that our growth rate did stabilize when we crossed 100 cases, and then start to slowly decrease.
So when the number of cases is below 100, which is currently the case for every county except Chittenden, it does become less predictable to try to model. At the state level, we wanted to wait until we crossed that threshold to make our modeling public.
Could the state take a county-by-county approach to lifting social distancing orders?
Vermont is a small state geographically. All the states in the Northeast had this concern, that if you don't act in lockstep, you might have people moving across borders, exposing themselves, and bringing COVID-19 back to their community. I think this has to be a statewide effort at this time. But we are looking at it.
Are there pieces of Vermont's unique character that affected this modeling?
In terms of density of population, we compare very differently to some of the urban metropolitan areas. We also tend to be a pretty healthy population, more so, for example, than the data drawn upon in China. We also happen to be an older population, so there are a few things that may help us and some that hold us back.
How do you hope Vermonters will use these models?
Vermonters should understand that the sacrifices they are being asked to make are not done in vain. They appear to be having an impact on Vermont as a whole and on their communities. Now is a good time to double down.
Q & A with FiveThirtyEight Senior Science Writer Maggie Koerth
How should people know what kind of trust to put in the modeling they are seeing and how should they interpret it?
Models are really useful for people in a public health decision-making context. They allow them to get a general idea of: If we do this type of intervention, is it likely to work out for the better or for the worse?
They are used for relative comparison. But they are also useful for those people to get an idea of how bad this is likely to get.
However, none of these models is "the right model." As an average person, you shouldn't be going around trying to find "the right model" that will tell you exactly what is going to happen, because none of them will. They are going to be wrong not because they are bad, but because that is just how this works.
The goal is not to give you a perfectly accurate image of the future; the goal is to help us make decisions about what to do next.
Usually we hear that more data is better when it comes to building a model. Why have national models been so wrong about COVID-19 in Vermont?
The expertise ends up mattering a lot. It's easy for people to end up thinking that because they know the math, they could make a model.
You have this wide range of variables, and you don't know the answer necessarily. On paper, that could lead you to an outcome of anywhere from zero to infinity. Where the expertise becomes really important, is in helping to determine which variables are likely and which variables are connected to each other.
Knowing how a state works, probably ends up mattering a lot for how you come up with answers to these things.
Subjectivity flows up through the data collection process. There is a tendency to look at a number and think, 'Somebody counted this, and so we know for a fact that that's what it is.' But that's not often the case when you are talking about complicated public health data collection.
What would you advise people about the value of a model and how they should apply it to their own behavior and perhaps their ideas about what could happen in the future?
Models are interesting and supply information, but you can't really think about them as one thing at a time. As a science journalist, a lot of what I do with evidence and data is to look not at one science paper or one set of findings, but at what those findings across multiple papers are telling me in aggregate. Any one finding could be way off, but the more you see them coalescing, the more you know something closer to the truth.
That's kind of the same for models. The more you see, the better idea you have of what is likely. You won't know exactly what is going to happen, but once you start to see those models coalescing, that's where you start to understand what the most likely scenario is.
And it's really important to know that that are constantly changing. So everything we do right now, is going to be affecting what the models say is most likely, and how accurate the model is.