As doctors and nurses treat patients during the COVD-19 pandemic, Casey Kuhn and his team of seven people at Metro Health-University of Michigan Health have been busy working to identify what they face in the days and weeks ahead. The manager of business intelligence and analytics at Metro Health, Kuhn leads ongoing efforts to use modeling to project the pandemic’s trajectory so the hospital can plan, prepare and respond accordingly. Kuhn, whose staff normally works on operational reporting and analytics, spoke with MiBiz about the modeling.
What are some of the key factors and variables that go into the modeling equation to come up with projections?
Doubling rates is the one key variable, and the doubling rate of basically the initial rate of infection, and that’s where we’re using these SIR models (an epidemiological model for projecting infection rates). Those numbers apply to the infected population and because of the lack of testing, we can’t really know what that is, and so we have to make some approximations based on what we are actually observing.
How large of a sample does that take into account?
For Metro, for example, a difficulty is if we are only looking at our patients, how many patients we have and how many are hospitalized. With all statistical modeling, the larger it is the better. That’s why we are not just looking at our own patients. We’re looking at Wuhan (China) and South Korea and Italy and whatnot in order to look for common trends.
Are you partnering on this with other organizations across the market?
What we are doing is sharing data among the three largest health systems (in Kent County) so we all have more complete data sets. The best part of this is that all of our colleagues — from our sister organizations across Michigan (that are part of University of Michigan Health) to the other health care organizations in West Michigan — are all working together openly, sharing and helping each other. That open cooperativeness transcends through the ranks as a testimony of how real this is.
What’s the biggest challenge to doing modeling?
The biggest X factor is in fact the human (element) and what are the effects of social distancing. How far have we gone? What we know is that in order to make a change to that curve of infected (people), it’s a one- to two-week lag from the time that we do something. And then, because of the 14-day infection period of COVID-19, it’s another week to two weeks before you actually see a difference in the health system. You look at an executive order going into effect, we can’t see that for three to four weeks later to know how effective this really is.
How far ahead can you confidently project?
With some confidence, we can make projections for the next seven to 10 days. When we get to that 30-day mark, (the factors are) not knowing what’s going to happen with that social distancing and even weather — although there are a myriad of factors. Any model that shows what we’re going to look like over a month from now is like looking at a Farmer’s Almanac.
How does the health system use this data to prepare?
We really need to know: Are we going to hit capacity for our general beds, for our ICU beds, and for our ventilators? When is that going to be and how many might we need totally? Like every other health system, we need to be ready for when we do peak. By developing these bell curves and demonstrating what they’re like, we can plan for reasonable scenarios so that we’re proactive, rather than reactive, and then we’re using these in order to combine our models and our understanding as a community for overflow planning.
What have you learned throughout this pandemic that you can use in your everyday work once this crisis passes?
The need for more collaboration before we get to crisis mode, and to have a lot of these processes built in advance to be ready for the next crisis. As an example, as we’re working on these models — and my number one struggle is access to complete and accurate data — it occurred to me weeks ago that what we really need is a national registry and a process for health care institutions to contribute on a national level and to be able to localize that data to their particular locale. If we had that earlier, it would have helped us create better models faster, instead of digging around looking for good data.
Bottom line: Have something in place before a crisis hits?
Our number one mission is to treat the patients. That’s our priority. Now isn’t maybe the time as a nation to have compulsory reporting, because that’s an extra burden for the health care institutions that are trying to treat patients at a time when there has been scarce resources. If we had defined this ahead of time — ‘for this crisis, this is what’s going to happen’ — and those data feeds were already there … maintaining it is not as difficult. The hard part of setting up a data feed is just the initial analysis and understanding what you need to do and building it out.
We need to have an efficient flow of data so scientists have access to the data when they need it.
What kind of pressure do you feel as you go about modeling and preparing projections?
I do feel a constant weight to go beyond our best because there’s clearly not only lives at stake, but then there’s the financial implications as well and folks’ livelihoods. We could certainly take a worst-case scenario and say, ‘Here, let’s make sure we do this,’ but that wouldn’t be in our best interests, so we’re developing the most reasonable model that we can plan toward in hopes that we save as many lives as possible.