One of the more important early lessons learned from the pandemic has been the significant role that data and science plays, particularly in the context of medical professionals and public health leaders sharing critical information for the benefit of proactively identifying surges and ultimately improving patient care.
At the Des Moines, Iowa-based UnityPoint Health, a team of data scientists—hired years ago—was able to quickly react and use data analytics to help the health system measure and predict key COVID-19 trends. For instance, UnityPoint’s analytics team, spearheaded by data scientist Ben Cleveland, created a dashboard that allowed health system leaders to manipulate variables to predict the impact their decisions would have throughout the organization. What’s more, it enabled UnityPoint to make plans for the ongoing spread of COVID-19 and forecast how expanding testing would impact the health system’s resources, care models and service reactivation plans in the communities where it provides care, according to health system officials.
In a recent interview with Healthcare Innovation detailing the specifics of these efforts, Cleveland says that from the UnityPoint perspective, the pandemic unfolded in three distinct phases. The first was back in mid-March with the initial outbreak, “when uncertainty and fear were at their highest, nationally. And that was our experience too,” Cleveland recounts, “At the time, New York City was really struggling and everyone was basically wondering if we were a few weeks behind them, so the questions we were asking were with New York City as a backdrop early on.” Cleveland recalls one of the health system’s senior leaders asking him if he was able to tell if COVID-19, at that time, was like standing in front of a tsunami or something far more manageable. “That sharpened our focus early on,” Cleveland acknowledges.
UnityPoint’s analytics team had to make an early call on which specific key indicators would be used to make organizational-wide decisions. Given all the limitations of testing data, the prevalence of testing, and how it was constantly changing at the time, the data team decided to mostly focus on COVID-specific hospitalizations as the key indicators early on. “So we built algorithms to monitor and essentially tell us if our own individual ‘curves’ were heating up, cooling down, or plateauing, and then we took those algorithms and combined them with some of the national models being produced by major universities and public health institutions,” says Cleveland. He notes, “It was a time of great national collaboration; everyone was open-sourcing all of their code and being very transparent. It was very much a [feeling] that we’re all in this together, so let’s share ideas as rapidly as we could. So we imported those models into our environment and set them up for our own local markets.”
The goal here, explains Cleveland, was to look at the trajectories UnityPoint’s models were showing, consider the different scenarios that could take place, and then combine those with internal algorithms. That way, the data science team could essentially map what its models were showing to the scenarios the national models were predicting. “Based on the characteristics of those curves, we were able to show leadership early on that these were the key milestones where we’ll know if we’re standing in front of that tsunami, if things will get bad, or if they will stay manageable. So early on in that initial outbreak stage, it was very much about setting up the capabilities to understand what the virus is doing locally, and then map those to the predictive models that our national peer groups were developing at the time,” he says.
Later in April, efforts moved to the next phase, which focused on reactivating so that the health system could safely reopen for patients and providers. “It was clear to everyone at that time the virus wouldn’t be going anyway anytime soon, so we needed to figure out how to safely reopen,” says Cleveland. He continues, “At that time we were [focused] on trying to understand all the interconnected links in the chain to providing safe care, and where our areas of emphasis and risk needed to be. As our markets ramped up their volumes, we needed to forecast PPE usage, replenishment rates, as well as understanding the [volume] of testing that would be necessary, and compare that with what we were doing that day. We also needed to keep a handle on the virus’s behavior from the perspective of bed capacity and staffing. So that was really where the broad analytics capabilities were put on display. Our supply chain team has developed a ton of great dashboards to indicate how many days on hand we have for all kinds of PPE and what their projected usage rates are,” says Cleveland.
All of this data enabled UnityPoint leaders to gain strong visibility into very detailed specifics. For example, they could tell if one of the health system’s facilities in rural Iowa was likely to run out of masks, when that exact day would come, and then bolster their supply accordingly, Cleveland explains. The testing and lab teams also came together to study the data on testing usage and growth, as well as relationships with suppliers and other national demand that could affect how many tests could be conducted, acquired and sustained on a consistent basis, he adds. Meanwhile, UnityPoint’s clinical analytics team was studying various different treatment patters and protocols, and giving clinical leadership lots of insight into what the experience has been with its COVID-19 patients—particularly what was working well and what wasn’t.
That second phase was a four-to-five-week process, and once UnityPoint was able to safely open up, it has since moved into the “monitor and manage” stage in which patients are being cared for, though Cleveland admits “there probably will be times where we need to pump the breaks based on individual disease incidence, supply chain constraints, and testing demands as things flare up [around] the country.”
Nonetheless, his data science team has committed to updating each of the aforementioned analytics on a weekly basis, as well as distributing them to essentially the health system’s entire leadership team. “We’re also simulating our census levels now that we have both COVID patients and non-COVID patients in the hospital, evaluating their different needs, and [trying] to understand how those will change over time,” he says.
Developing analytical models that health system decision makers will ultimately rely on can prove quite challenging since “perfection” is sometimes expected but rarely possible. Cleveland, when asked about the many COVID-19 national predictive models that were criticized for constantly changing, says that while he understands the frustration from leadership and the public when models vary so greatly, it really highlights how unprecedented a time this has been. “A lot of those model methodologies that were developed have never actually been validated against a real-world global pandemic before,” he points out. “Like many others early on, we didn’t know how things would play out, and many models were showing more incidence that ended up happening, so our message early was an [admission] that we don’t really know, and that these models have huge confidence intervals.”
An important lesson learned in this area, Cleveland says, was that health system leadership needed to be educated on the constraints and limits of models, while instead focusing on what they could tell you. For example, he offers, “If we’re really headed toward the edge of a cliff, those earliest indicators [could] tell us when that might happen, and how bad could things get in those cases. Tempering expectations was quite important, as well as being able to distill some of the noise out there,” Cleveland says.
Ultimately, he adds, “The pandemic has showed us that the appetite for analytics and data science is much bigger than previously thought, and that leadership has started to understand that data science won’t be perfect, and won’t solve all our problems, but it gives us an objective way to make decisions that we haven’t been able to make before.”
