Home  |   Subscribe  |   Resources  |   Reprints  |   Writers' Guidelines

March-April 2021

Locked & Loaded Against COVID-19
By Selena Chavis
For The Record
Vol. 33 No. 2 P. 24

How predictive analytics tools are arming health care organizations with insights on how to respond to COVID-19.

In recent years, the health care industry has made notable strides in its quest to improve data mining techniques. Technological advancement in the form of predictive analytics, artificial intelligence, and machine learning is ushering in a new day for extracting insights from the massive amounts of data that have been amassed by health care organizations.

The stakes for using these tools could not have been higher in 2020 when the COVID-19 pandemic made its way across the globe. Allison Viola, MBA, RHIA, director in the health care practice at Guidehouse, says that the pandemic required health care organizations across the country to report data on a scale unlike anything the industry has experienced in recent years.

As a result, this process shed light on the industry’s capabilities. It wasn’t necessarily a pretty picture, Viola says. “Unfortunately, this highlighted some vulnerabilities in approach, execution, and ability to scale at such a level,” she says. “However, the pandemic also demonstrated opportunities for the health care community and other stakeholders that support it to improve upon the way data are curated and reported. Data should not be viewed as an afterthought. Rather, careful planning and intentional steps toward a comprehensive approach through standardization and governance is essential to achieving quality data.”

Notably, the use cases of predictive analytics tools have been virtually limitless as health care executives have creatively harnessed data to become more proactive with everything from patient care and supply chain management to revenue integrity.

“Really, any way that you can think of a health system needing data to respond to COVID, there's a space for predictive algorithms to support that,” says Kyle McAllister, director of the data analytics practice with Pivot Point Consulting. “So, from a use case perspective right now, the industry is looking at things like vaccine administration, revenue recapture for elective surgeries, or even for the some of the self-pay things that happened throughout the last nine or so months. A very common one, obviously, is contraction rates of positive testing, utilization, and capacity management.”

In terms of managing bed capacity, McAllister points to a large multistate hospital system client in the Southwest. Spread across a wide geography, the organization needed a way to centralize management of data from local facilities. A predictive analytics model was developed to address this need, giving leadership visibility into such information as testing volumes and the number of positive cases.

Current state bed capacity is monitored, and future capacity needs can be forecasted in a matter of days based on what’s happening in their own environment. Data insights that just a decade ago would have been fraught with resource-intensive manual processes are now readily available for the entire network on a single dashboard, improving readiness for the unknown in terms of supply needs and staff resources.

“That has been huge for them,” McAllister notes. “Rather than having to react and potentially even miss the boat on sending those resources to the right place, they can at least be ahead of potential risk areas, even if the predictive model is not perfect.”

Due to the challenges of finding consistent methods to treat COVID-19, predictive analytics has become paramount in the health care system’s response to the pandemic, Viola says. Because patients have varying responses to the disease based on health factors and comorbidities, it’s difficult for providers to create an environment of readiness based on a standardized course of treatment.

“Hospitals are using data from EHRs to predict patient surges, emergency department overcrowding, and ventilator inventory,” Viola says, adding that various data points can help predict when a patient might require a ventilator or die. “EHR data are also being integrated with staffing data, which gives hospitals the ability to predict overcrowding statistics within just minutes of patient intakes and thus allows the facility to handle the surges created by COVID-19.”

How Many Data?
The question of how many data to use is difficult to answer, according to McAllister, who notes that often, the answer comes from understanding the specific use case.

“It’s going to vary widely across use cases,” he explains. “We spend a lot of time with the folks that we work with really trying to understand what use case they are trying to solve and what data really matter for that use case. Then, the volumes manage themselves to a degree because at some point, you're also restricted to what the health system has on hand.”

McAllister adds that health care organizations often are trying to accumulate as many data as possible in case they need it in the future. Referring to this information as “lazy data,” he warns that these assets often compound the challenge of analytics. “It only makes maintenance harder, and it takes up space,” McAllister says.

Viola emphasizes the importance of quality data over volumes of data. “We are experiencing a deluge of data from various sources from which to make decisions,” she says. “Ensuring data integrity is a critical step in making necessary decisions—whether clinical or nonclinical.”

Ensuring Data Reliability
“Ensuring accurate representation and reliability of the data in support of COVID-19 response efforts is imperative for national, state, and local planning, decision making, and thus, policy implementation,” Viola says. “These data also ensure an accurate picture of COVID response efforts.”

During the pandemic, many lessons have been learned about the critical nature of accurate data reporting. In August 2020, the Government Accountability Office released a report, “COVID-19: Brief Update on Initial Federal Response to the Pandemic,” which, among other evaluations, assessed the reliability of the number of higher than expected deaths from the Centers for Disease Control and Prevention’s (CDC) National Center for Health Statistics, which was intended for monitoring and reporting.

Viola points to a discovery in May 2020 that the CDC was conflating test results of viral and antibody tests, even though the two tests represent different information and are used for different purposes. The CDC has since taken steps to separate the test results to allow for a more accurate depiction of the pandemic, but this was certainly one area where important lessons were learned.

“It was critical to differentiate between the two as a positive viral test indicates a potential active case of the virus and a positive antibody test indicates a previous infection and the person has recovered,” Viola says. “By combining the two test results, it can possibly paint a very different picture of the pandemic’s severity and potentially lead to mandates and restrictions that do not adequately align with the true picture of the pandemic."

Even when presented with time-sensitive needs for data, McAllister emphasizes that optimal approaches to predictive analytics must always include a testing and validation period. And notably, in most data-related projects, these efforts tend to be the most time consuming, especially when it comes to developing machine learning or predictive algorithms.

“Essentially, the entire process of building a machine learning algorithm is starting with an algorithm that uses validation and training data to build itself,” McAllister explains. “It's all baked into the data gathering and wrangling portion of the work, as well as the actual training or development phase.”

McAllister goes on to note that the parameters for the testing and validation components of analytics, similar to data volume requirements, really depend on the intention and development of each algorithm in terms of determining “how reliable is reliable enough” and how deeply an organization will need to go from a validation perspective. Depending on the question being answered, he points out that sometimes the validation process can be a little looser.

One of the greatest challenges to creating algorithms is human and organization bias, McAllister says. For example, in a multihospital organization, patient populations can vary widely by facility. Consequently, an algorithm created at one institution may not work well for another. “Humans develop these algorithms, and they write the code and bring their bias to the table,” he says. “A major part of working on these [algorithms] is trying to understand and really test for those biases—trying to be really mindful about them.”

Health Care’s Data Scientist Conundrum
There are many names associated with the professionals who have expertise in data mining for predictive analytics—data scientist, data engineer, statistician, to name a few. Regardless of how they are identified, though, one truth is certain: They are in short supply.

“In recent years there has been an explosion of data science jobs and opportunities for companies and individuals to follow this path,” Viola says. “However, despite the growing demand, data scientists are in short supply.”

The result is that policy makers, public health professionals, researchers, health care systems, and amateur data enthusiasts are developing their own solutions and tools to address and support a variety of data needs such as tracking and visualization to help monitor the spread of COVID.

McAllister acknowledges that health care faces a significant disadvantage as top tech firms and other industries often get first dibs on the top talent. He notes that it’s difficult to compete with both the perceived and actual benefit of working with a large Silicon Valley firm.

One of Pivot Point’s sweet spots in terms of working with health care organizations is helping to recruit in this area, which is a struggle that often goes beyond data scientists. “We work across the full spectrum of the data analytics field, and there are a lot of roles in the data space, even outside of just developing algorithms,” McAllister says, pointing to roles such as data analysis and project management. “Really strong project managers can often be difficult to come by, especially for folks that are wanting people to be physically located wherever they are.”

The Future of Predictive Analytics
Viola believes the pandemic has shed light on the lack of infrastructure to support electronic case reporting and other electronic public health data needs. Therefore, moving forward there will be heightened focus on public health data infrastructure and its ability to exchange data.

The CDC recently launched the Data Modernization Initiative, which aims to “modernize data, technology, and workforce capabilities—together and at once.” The CDC claims the initiative will support public health surveillance, research, and, ultimately, decision making.
“In support of this effort and in light of recent regulatory changes issued by the Centers for Medicare & Medicaid Services and the Office of the National Coordinator for Health Information Technology, the CDC hosted a listening session and published a report in July 2020 that outlined what is coming in 2022 and what opportunities exist for public health,” Viola says.

McAllister notes that there is currently a lot of excitement about the promise of predictive analytics, which will advance these techniques in the coming years. “There's been rapid development in the space, and there are a number of organizations, including classic EHR companies, that are starting to build some of these things and incorporate them into their platform,” he says.

McAllister points to key initiatives related to interoperability and patient access on the national stage. “I actually feel that health care is on the precipice of significant positive change,” he says. “I expect health care IT is going to see pretty rapid development, disruption, and maturation in terms of processes and development, especially as large tech firms start to move into the space. We've seen Google and Amazon and others really jump in, and I think that's going to make a lot of change in the industry.”

Selena Chavis is a Florida-based freelance journalist whose writing appears regularly in various trade and consumer publications, covering everything from corporate and managerial topics to health care and travel.