�鶹ӳ��

Kyra Morgan is the �鶹ӳ�� State Biostatistician and a graduate of the Statistics Master’s program in the Department of Mathematics and Statistics—one of six Department alumni out of her team of seven who are actively working on the SARS-CoV-2/COVID-19 response. She oversees centralized analytics for the State of �鶹ӳ�� and is responsible for the State’s response as it relates to public health data and analytics, helping to inform both government leadership and �鶹ӳ�� citizens on pertinent data and public health information. She also oversees caseload projections for State programs like the Supplemental Nutrition Assistance Program (SNAP), Medicaid and Temporary Assistance for Needy Families (TANF), helping drive the State’s annual budget. It’s a big job, and in the Spring of 2020, her job got much bigger. As the COVID pandemic began to spread in �鶹ӳ��, Morgan’s workload doubled, at times tripled in response. She was immediately tasked with managing a public dashboard with basic data about the virus’s spread in the State. Her team created a “PPE calculator” to help with the allocation of personal protective equipment (PPE) to �鶹ӳ�� hospitals and medical professionals. She met with local health authorities, participated in active COVID investigations in rural areas, and made recommendations to the Governor’s office in collaboration with the Governor’s medical advisory team. All the while, public health still existed outside of COVID, and Morgan’s many other responsibilities still existed as well. Things haven’t slowed down, and Morgan continues to respond with math and science to this rapidly changing health crisis.

Paul Hurtado is a professor of Mathematics and Statistics and a mathematical modeler of biological systems. Hurtado is also part of the Evolution, Ecology, and Conservation Biology (EECB) graduate program faculty. He usually focuses on studying the ecology and evolution of infectious diseases affecting non-human hosts, but in early 2020 he turned his attention to the emerging pandemic. Since then, Hurtado has been working with Morgan and her team to develop COVID models for �鶹ӳ�� and help interpret the output of other publicly available models. In early July, Morgan and Hurtado (virtually) sat down together to discuss the pandemic response from their data-driven perspective.

Morgan: My job is exclusively COVID at this point, except that it’s not. We’re really, as a whole, getting hit from every direction.

Hurtado: I had a much easier transition at the start of all of this. The majority of my time was spent teaching and transitioning my classes to online learning. But COVID did throw a big twist into how I was thinking about research, and in both of my classes, we would spend at least the first 10 minutes talking about Coronavirus and data and models. Students could see some of the things we were talking about happening in real-time. Every week things were shifting.

Morgan:Are, I would say are shifting!

Hurtado: Very true! In terms of the early on March/April time period, how clearly did you have a sense of what data you needed and from whom?

Morgan: Early on, I was approached by my director and asked to get a dashboard up that same day and I had to think about what data we were currently collecting. All of the core data such as how many cases we had or how many people were getting tested came from mechanisms that have been in place for a long time. Mortality data we’ve already had access to, but we initially had a little bit of trouble getting in real-time. We have changed our process so that deaths from COVID are reported simultaneously to a local health authority and the State so we have as much real-time data as possible. 

"We’re back to seeing days where 15% of people tested are coming back positive. In the current atmosphere where testing isn’t limited, that’s a really high number. We know that our population is more saturated with the disease because of that percent positivity rate."

Hospitalization data is the biggest hole. In the past, that data was based on a hospital bill so you can imagine it was pretty delayed. But, at that time, there was no real reason we needed it sooner. Also, our disease investigation team typically has the capacity to call positive cases and request real-time updates on hospitalization status, among other things. With COVID, that has not been a stop-gap for us because we simply don’t have the people and resources to make enough of those calls. The way we’ve tried to fill that gap is by using data from the �鶹ӳ�� Hospital Association. A survey is put out daily to all of our hospitals to get some high-level information. Expanding on the data available is a priority for our leadership right now so we are actively working on getting more of that information. 

Hurtado: On the modeling end, things were similar. Early on, we thought, great, respiratory disease. We know how to model this. We have oodles of models and experience fitting them to data. It wasn’t like we were starting from scratch. What a lot of modelers did was start with flu models and go from there to see what needed to change. However, the pace of changing those models to keep up in real-time was difficult. There were a lot of unknowns early on.            

Morgan: We’re in a bit of a limbo where we’re seeing differences in the trend of the disease, but I’m not seeing that precipitate into different model outputs. For example, when I see every day we have a record high of new cases, and I go to these State-wide models that say we have a 99% chance of having reached the “peak”, obviously I know there is a disconnect between those two things. I’ve actually digressed from reporting models as much as monitoring and reporting on key trends in observed data.

Hurtado: When I started out modeling COVID, I collaborated with a friend of mine who does this stuff for a living and has an army of graduates and postdocs to help her play with different model assumptions. Early on, when looking at age-based models with less complexity built into them, they were all working pretty much the same in terms of forecasts. But as things started to progress and we had changes in who was getting testing and the age group demographics began to act very differently, all of those things started to break the models. As each of those things bent the rules of the models, we had to continually adapt.

Morgan: It’s great to have a model and forecast and a beautiful curve. That’s really valuable, but in practice, you don’t always need a fancy forecast to see what has happened or what is happening. Some of the things we monitor internally are not forecasts, but we can see a certain trend in measures. We’ve said early on as a State we can’t just look at the number of positives because we know testing is going to ramp up. But, alternatively, we know that some of the ramping up of testing isn’t necessarily sustainable. When we first saw an increase in cases while we were increasing testing, that made a lot of sense. But now, testing has actually flattened out. We’ve reached a testing capacity and we’re still seeing an increase in cases. Because of this, we look at a number of things beyond just the number of cases because that is obviously heavily subject to testing. We look at things like percent positivity—how many of the people we are testing come back positive. Early on in the pandemic when we were mostly testing people with severe symptoms, a high of 21% of the people tested came back with a positive result. When we had a really high testing activity, that number dipped down really low. We saw days where less than 2% came back positive. Now, we’re back to seeing days where 15% of people tested are coming back positive. In the current atmosphere where testing isn’t limited, that’s a really high number. We know that our population is more saturated with the disease because of that percent positivity rate.

It’s also really important to look at hospitalization data. Theoretically, there shouldn’t be a lot of bias in the number of people hospitalized for COVID, and we have at this time record-breaking numbers of people hospitalized on a daily basis with confirmed COVID. We are back to a peak that is worse than our first peak, and not just from a perspective of new cases because that we understand can be subjective. Our percent positive is peaking and continuing to go up and hospitalizations have surpassed the previous peaks in hospitalizations. We’ve been able to tell we’re on this upward trajectory for about a month now. I don’t feel as tied to needing a fancy statistical model as I did early on because we have enough data to see what is happening.

Hurtado: I would agree with that. There are so many different factors that go into transmission going up and down. It’s so complicated that if you were to take the time to write down the models that would do a good job of predicting what your intuition is already telling you, first, you wouldn’t have enough of the data needed to feed into those models and second, you would have to make some interesting assumptions and contortions about the mathematical terms. However, once you’ve got a good understanding of how to carefully think about data, expert opinion is a real thing. I’ve seen a lot of my modeling friends make this same transition that you’re talking about. They went to grad school and studied math and stats, and now they’re talking with policymakers about how hair salons are different than bars. They’re able to take a lot of that ability to think critically and carefully about quantitative stuff and do better off the cuff than they could make a model do in a moment like this where things are rapidly changing.

"Public perception is driving the response to COVID. Public perception has a lot to do with what policymakers are deciding and the push back they are getting to open or close, wear masks or not."

Morgan: When you have a math degree, you don’t think you’re going to be the public face of Health and Human Services. That’s been a difficult transition for many of us in the quasi-science/quasi-leadership roles. As statisticians, we’re not taught a lot about the real-world part of things. Things I thought were common knowledge are not common knowledge to other people. For example, we know modeling is based on real data. When I was watching my press conference back, a lot of the comments referenced a belief that we weren’t using real data to create our models. I didn’t realize how many people in the general population misunderstood how we were creating these models. I could have done a better job of explaining that. Public perception is driving the response to COVID. Public perception has a lot to do with what policymakers are deciding and the push back they are getting to open or close, wear masks or not. I could have done a better job of explaining how the models actually work and how I addressed the public early on.

Hurtado: Kyra, that happens to everybody. You always find an audience where you will struggle to figure out where their head is at. That’s a really important lesson to learn and one that’s really hard to teach. Especially for folks that get mathematical and statistical training, there’s not a lot of opportunity to practice. You are not the first person with a nice quantitative background who’s had to get up in front of a bunch of people and explain math to them.

Morgan: And that dialogue is so important! The social interpretation of where we are as it relates to COVID is so important to the direction the numbers are going in. I know that at least from my anecdotal world that I function in, which obviously is made up of a group of people of a certain demographic and societal background, COVID was this huge emergency and everyone was doing a great job of social distancing initially. Now there’s this idea I see reflected in the news and the media and in people’s behavior that we’re somewhere near the downhill and we’re so not. It’s important to recognize that people are getting COVID-fatigue. It is easy to think “this has been around for a while now, things are opening, it must be getting better, it must be safe.” The data do not suggest that for �鶹ӳ��. The data do not suggest it’s a safer time to go out and do things. The data do not suggest you should be more lenient on your mask-wearing habits. The data really suggest we’re in a period of growth and transmission is increasing. We are in full swing resurgence right now.

Hurtado: The number one most important information that drives the modeling forecasts that I make which I don’t have good data on and I don’t expect to ever have good data on is human behavior. The ups and downs that we see in transmission right now are purely driven by whether or not people are doing what they can to use PPE, stay at home, not go out and mingle. Those sorts of ups and downs in human behavior are not easy to predict. We need to be able to predict what people are going to do in three weeks to understand what is going to happen in two months, and we’re just never going to know that. People are all over the map in terms of whether it’s a thing to take seriously or not. And even those who do take it seriously, it’s tiring! Even they eventually crack and need to go visit another person and socialize which is totally understandable. Hopefully, we are able to get some good vaccines and more treatment options. But like you said, this is not the end or even the middle of COVID. 

One thing that’s been great to see is the collaboration within the scientific community. The pace of science has picked up quite a bit. When there are these rapidly evolving situations, it helps that you can just call someone on the phone and have a conversation. Conversations just like this!