We’re starting to get more data on how the coronavirus will impact the health of Americans and the U.S. economy. While the lockdowns and social distancing measures are far from over, many business leaders are looking at the costs of the pandemic and searching for ways to mitigate the impact.
Employee benefits leaders are no exception; many of you are now searching for ways to predict the impact of COVID-19 on employee health. Some of you may have started digging through health claims data, which are just starting reflect claims and episodes of care for coronavirus patients. We’re also hearing from Artemis Health customers that they’re searching for cost models to share with their C-Suite.
In today’s blog post, we’ll break down how benefits data and public health care can be used together to model risks and costs associated with COVID-19.
First, let’s review our story on risk assessment. We know there are three big factors that increase the likelihood of a severe case of COVID-19:
Each of these factors may help a benefits team focus their time and efforts when looking for ways to minimize risk to their organization.
As we’ve conducted weekly COVID-19 webinars for the past month, we’ve had a number of questions from current Artemis customers and webinar attendees on topics of interest to them during the pandemic. They’ve asked about the costs of chloroquine and hydroxychlorquine, the costs of pneumonia in-patient treatment, the effects of delaying elective procedures on benefits costs and employee health, drug shortages, and cost modeling. We wanted to tackle the question of cost modeling because it’s a concrete step that benefits professionals can take in uncertain times. So let’s dive in: how do you predict the costs of COVID-19 on your benefits costs?
First, benefits teams should keep in mind that they’ll need access to reliable data that serves this purpose. Artemis looked at a number of trustworthy sources when putting together our risk assessment analysis and cost modeling:
These sources are measuring a lot of the same things, but many are taking different approaches to how they’re measuring or what parameters they are setting. We’ll dive deeper on that in a moment. A lot of data models are selecting South Korean and Italian data because they are taking unique approaches to testing and controlling the spread of COVID-19, so you can see multiple possible outcomes by using data from these countries.
Secondly, predictive modeling is tough because the analyst is working in a “cone of uncertainty.” You have a maximum impact or a minimum impact, and you’re not sure which end of the scale will be the most accurate.
For example, think about how long before a domestic flight you would normally leave for the airport. There are a number of unknown factors that could affect your safe, minimum time and turn it into a missed flight: how long it takes to get an Uber, an accident blocking the freeway, a backed-up security line, holiday weekend travel, and more.
All of these factors increase the minimum or maximum travel time, and make it less certain that you can cut it close. The same is true for predictive modeling, especially around a disease as new and widespread as COVID-19. These functions will vary wildly, so you’ll see our cost predictions will also vary quite a bit. It’s up to benefits experts to present their variety of findings and make the best decisions they can.
Now, a note on making assumptions: you may need to assume some parameters when cost modeling, like the average length of a hospital stay or the percentage of people on your plan who will be infected. We have done so in our model as well, and we’ll be clear about our assumptions as we go through our cost models.
We started with a pretty simple question: how many people are on your health plan? This number should include anyone on the plan, not just the number of employees, but also dependents. Next, we used a few data sources to ask how many people on the plan will actually be infected with COVID-19. We looked to FAIR Health’s model for this number, which gives us low (20%), medium (40%), and high (60%) infection rate predictions.
The next step is to predict what percentage of the infected population will require hospitalization and Intensive Care treatment. As you can see, the CDC and Imperial College London have some very different numbers here:
There are a few reasons these predictions vary so much:
When you account for some of these factors, the CDC and Imperial College of London’s predictions for hospitalization and intensive care rates are actually pretty similar. Keep this in mind for your population as you prepare your cost model. For example, if your employees are largely located in dense urban centers with active outbreaks and travel to work via public transportation, your infection rates may be higher. If your employees are able to work remotely from home and follow social distancing guidelines, your infection rate (and subsequent hospitalization rates) will likely be lower.
So with this data, we can start to calculate the cost of hospitalization for any given member population. Again, we decided to use two trustworthy sources for costs for hospitalization episodes: The Kaiser Family Foundation and FAIR Health. Here’s what they predicted:
You can see the FAIR Health model assumes a 6-day hospital stay to come up with these numbers. They tried to normalize the data into a fixed time period. On the other hand, KFF looked at the average cost for pneumonia case hospitalization, and those stays are usually 2-3 days. So you can see that these models actually do agree with each other if you divide them into one-day stays. Artemis’ clinical experts are seeing mostly 6-7 day stays for COVID-19 right now, and while this could change, we think the FAIR Health model is the most reliable we have at this time. Actual claims data are just starting to roll in for COVID-19, so we may learn more in the next month from carriers about the actual length and cost of these hospital stays.
Kaiser and FAIR Health also modeled the cost of ICU care for coronavirus patients:
You can see that FAIR Health numbers don’t change, as they had already assumed ventilator/ICU care in their original hospitalization numbers. Kaiser’s numbers go way up as they account for a more severe illness.
Ok, so now we have a number of data points and measures, and we’re ready to model costs for a member population. Once we’ve gathered reliable numbers, we can do some relatively simple math.
Here’s what the equation looks like:
Benefits analysts will likely want to run this calculation several times to get a best case or worst case scenario. That will give you the widest “cone of uncertainty” so you can be more confident that you’re planning ahead.
We hope this cost modeling exercise is helpful to benefits leaders who are trying their best to mitigate risks, predict outcomes, and provide the best care for employees. If you’d like to learn more about how data can help you cope with the effects of COVID-19, we have a number of resources available.