Cleveland Clinic's COVID-19 strategy driven by data modeling

Cleveland Clinic is using data models developed in concert with analytics software vendor SAS to prepare for a potential surge in patients due to the COVID-19 pandemic.

Cleveland Clinic is engaged firsthand in the fight against COVID-19.

Cleveland Clinic is an academic medical center in Cleveland that operates 11 hospitals and 19 health centers in three states, and like healthcare organizations worldwide, it's desperately working to treat people sickened by the new coronavirus. At the same time as it works to heal those already stricken, with the ultimate spread of the pandemic still an unknown, the healthcare provider is preparing itself to treat those who have yet to be infected, those now healthy who will eventually fall victim to COVID-19.

And to do that, and to try to be ready for whatever is yet to come, data modeling is driving Cleveland Clinic's preparations.

In mid-March, shortly after the first deaths from COVID-19 in Ohio were reported on March 9, the clinic began developing a series of data models to show potential levels of patients based on various scenarios -- one if no social distancing mandate was issued, another if distancing was ordered at a particular time, still another if the mandate came down at another time, and so on.

After beginning the work on its own during the week of March 16, by the following Monday Cleveland Clinic had contacted analytics and BI vendor SAS, a partner dating to 1982, to help develop the models. Within a week, Cleveland Clinic had predictive models in place to see how the spread of COVID-19 might progress given different scenarios, and had started to make data-driven decisions about how to prepare for the influx of patients to come.

The real power in these models is the ability to do the what-if analysis, to take a set of potential policies and test the impact.
Steve BennettDirector of global public sector and financial services practice, SAS

"The approach that we took, which I think is a little bit different than some other ones we've seen where people settle in on a curve or set of assumptions and then moderate, was that we wanted to run multiple scenarios and the show worst-case scenario and then iterations so that our leadership could see different possibilities," said Chris Donovan, executive director of enterprise analytics at the Cleveland Clinic.

With no stay-at-home mandate in Ohio until March 23, in the days after developing its data models Cleveland Clinic prepared for the worst-case scenario, the complete ineffectiveness of social distancing. As a result of the stay-at-home order, however, the reality in Ohio has followed the trajectory of a different, relatively less lethal model, allowing Cleveland Clinic to keep its supply of equipment and personnel ahead of demand.

Meanwhile, the healthcare provider remains prepared if the situation worsens.

"People look at these models and they want to see, 'What's the forecasted end point, how many sick, how many fatalities,' but the real power in these models is the ability to do the what-if analysis, to take a set of potential policies and test the impact," said Steve Bennett, director of global public sector and financial services practice at SAS. "That allows senior leadership, not only in a health system but in a government or in a public health agency, to make better informed decisions based on some sense of what the impacts of different policies are going to be."

The potential daily occupancy of Cleveland Clinic resulting from one scenario related to the COVID-19 pandemic is displayed in a data model developed by the hospital in partnership with SAS.
A data model developed by SAS and Cleveland Clinic shows the potential daily occupancy of the hospital depending on one scenario related to COVID-19.

The process

With the spread of COVID-19 on the rise and Ohio joining the list of states with multiple deaths resulting from the virus, during the week of March 16 the Cleveland Clinic began work on the data models it would need to prepare for a probable surge in patients.

At first it looked at available SIR (susceptible, infected, recovered) models -- as well as a model developed by the University of Pennsylvania to inform decisions in and around Philadelphia that other healthcare organizations could use by simply plugging in their own data.

One of the problems the Cleveland Clinic faced, however, was a lack of data. With so few COVID-19 tests available, there was no way to accurately know how many people in and around Cleveland had been infected with the novel coronavirus.

"As we were developing the models one of the challenges in this particular case was that there was so little-known data," Donovan said. "So, the approach we took here was that the one thing we knew for sure was how many patients at the clinic we were admitting to our hospitals."

After examining the code used in the Penn model as a way to get started coding its own models, the data scientists at Cleveland Clinic worked throughout the weekend of March 21-22.

By Monday, Cleveland Clinic engaged SAS to help develop its models, and in the ensuing days the data analysts were ready to share them internally with management to plan for the impact of the pandemic in Northeast Ohio.

"The power of a collaboration like this ... is getting two groups of people who understand the science and the medical implications of these things sitting down together and figuring out what these models mean and how to best apply them," Bennett said. "That's when you can figure out that maybe a model isn't telling us something that's completely accurate and can make a hybrid model with insights from not just one model."

The result, he added, is better preparation for healthcare providers and ultimately better outcomes for patients.

SAS, however, did more than just help the Cleveland Clinic write the code for its data models. Because the vendor works with healthcare organizations worldwide, it was able to provide data from countries and regions worldwide where COVID-19 hit before it started to spread in the United States. Cleveland Clinic then combined that data with its own information to create better forecasting models.

Data-driven decisions

Ultimately, the Cleveland Clinic and SAS teamed up to develop about a dozen different data models projecting the number of potential COVID-19 patients based on different levels of effectiveness of social distancing.

In the first days after developing the models, at a time when the stay-at-home order had just been issued and its effect was unknown, the health system decided to prepare for the worst-case scenario.

With the worst-case scenario being nearly 10,000 patients in June and July, Cleveland Clinic officials decided to open a 1,000-bed temporary hospital on the clinic's medical education campus, and to convert many of the beds in a hotel it owns to hospital beds. In addition, that worst-case scenario informed the health system's decision to order the extra medical equipment that would allow the clinic to deal with the potential of having to accommodate nearly 10,000 patients at once.

"Those assumptions were fed into our [personal protective equipment] planning, our supply planning, our bed utilization planning, all that kind of stuff, and that was the plan for the worst-case scenario," Donovan said.

The reality, however, has been better than that worst-case scenario. Social distancing is having a limiting effect on the spread of COVID-19, and as the stay-at-home mandate has lowered the number of potential patients the Cleveland Clinic has adjusted its preparations.

Instead of opening the 1,000-bed hospital and converting hotel beds to hospital beds, for example, the clinic is prepared to take those measures if needed but hasn't had to, including using the temporary hospital.

"As we got further into it, we started to see things mitigate, and leadership started to make some changes," Donovan said. "When we felt we were comfortable to handle a surge if we needed it, they started to roll back some of that and put it in a standby, be ready to execute in a certain number of days kind of mode."

The present situation

Six weeks after starting work on its data models, looking at a dozen different 'what-if' predictive models developed in partnership with SAS, the Cleveland Clinic has been able to remain prepared for the actual number of COVID-19 patients needing treatment and hospitalization.

And data modeling has been a critical factor in that preparedness, according to Donovan.

"I definitely think it helped the organization be prepped for this and to make informed decisions," he said." They executed quickly, they didn't wait to see if maybe we're not going to be on that [worst-case scenario] curve."

With Cleveland Clinic able to meet demand, and with stay-at-home measures so far keeping the number of COVID-19 patients in Northeast Ohio lower than in many other regions, the clinic has been able to help areas hit harder by the pandemic.

The predictive models are behind those decisions as well as the clinic sent doctors and nurses to New York City and personal protective equipment to Detroit.

"We felt like we had enough and we're looking at the models right now. We have enough to get us through where we expect to be at this point. [Data modeling] really helped the organization be prepared."

SAS, meanwhile, is working with many organizations in addition to the Cleveland Clinic engaged in the fight against COVID-19.

The vendor is working with ministries of health in Europe and Asia, state governments in the United States, and a variety of hospitals and healthcare providers worldwide, Bennett said. He declined to identify these other customers noting that they're not yet ready to publicize their work, though he specified the German Ministry of Health as one.

"We're seeing a great need everywhere for this sort of insight to be able to do the sorts of things the Cleveland Clinic is doing," Bennett said. "They're asking, 'How can we get ahead of planning so we have what we need?'"

Dig Deeper on Data science and analytics

Data Management
Content Management