Tools & Strategies News

How Real-Time Data Informs Decision-Making During COVID-19

Providence’s DetectionMap uses real-time data extracted from the EHR to show COVID-19 prevalence at the neighborhood level.

How real-time data informs decision-making during COVID-19

Source: Getty Images

By Jessica Kent

- Throughout the era of COVID-19, it’s become apparent that the only constant is change. The crisis continues to rapidly evolve, with updated guidelines and fluctuating case numbers, highlighting the need for up-to-date, real-time data.

At Providence, leaders recognized this need early on. As the first health system to treat a confirmed COVID-19 patient in the US, Providence’s analytics and insights team quickly got to work on keeping their 120,000 caregivers informed about the impact of the pandemic.

This effort led to the development of the DetectionMap, a publicly available online tool that shows where COVID-19 is prevalent in geographical areas where Providence cares for patients.

“The idea for the DetectionMap came about early in the pandemic. It was a period where there was a lot of uncertainty about what was going on inside our communities,” Ari Robicsek, MD, chief medical analytics officer at Providence, told HealthITAnalytics. “We hypothesized that we could use data from our EHR to paint a picture of the pandemic in different areas.”

Ari Robicsek, MD

Providence cares for thousands of patients each day, enabling the health system to capture a large volume of information about the illnesses patients are experiencing. This was especially helpful when it came to developing the DetectionMap, Robicsek noted.

READ MORE: Working Toward Seamless Care Delivery with Real-Time Data

“For every single patient we see in the outpatient setting, we take two pieces of information from the EHR. We apply natural language processing tools to those data points to determine whether or not they described symptoms that could be COVID-19. In a particular region, we might have seen 2000 patients over a period of time, and 20 of them may have had symptoms of COVID. That gives us critical information about a point in time in a particular location,” he said.

“We also analyze the percentage of all people who test positive for COVID in a particular area. We use an algorithm that combines neighborhood-level data on patients who have COVID-like symptoms with data on people who have tested positive and live in the same neighborhood.”

The team then takes that data and projects it onto a map in a way that protects the privacy of individuals’ information. The map is accessible to anyone, and Providence expects that it to benefit the communities it serves.

Fundamentally, the tool relies on natural language processing-based symptom data and test data. However, this kind of analytics technology can sometimes come with challenges in the medical field, Robicsek said.

“When you’re dealing with the complexity of natural language processing and the context of medical language, you're looking for specific items. For example, one of the clinical features we’re interested in is whether or not a patient had a fever. Doctors will sometimes say a patient has a fever, or doctors will sometimes say a patient doesn't have a fever, and a lot of natural language processing tools are good at differentiating assertions from negations,” he explained.

READ MORE: COVID-19 Claims Enhance Real-Time Data at the County Level

“But there are a lot of hypotheticals or irrelevant mentions in medical notes. A provider might mention that she told patient to go to the emergency department if they develop a fever. That's a hypothetical. We needed to find a way to deal with that. We had to develop data science tools to overcome that issue.”

The organization also had to keep patient privacy top of mind when developing the DetectionMap. While the timeliness of the data used to build the tool is one its main advantages, it also creates potential security and privacy risks.

“Our data is extremely up-to-date because it’s based on fresh electronic medical record data. It’s also extremely geographically granular, so it can show you what's going around in your specific neighborhood,” Robicsek said.

“We needed to find a method of processing our data so that no patient's data could be re-identified from the map. After we developed a data science pipeline for that de-identification, we hired a third-party privacy analytics company. They went through our code line-by-line to confirm that the risk of de-identification is extremely low.”

With the real-time data provided by the DetectionMap, users can make more informed choices as the pandemic wears on, which will hopefully reduce virus transmission in communities.

READ MORE: Real-Time Data Analytics Critical for Improving Heart Health

“We really are in a crisis, and we're trying to keep safe. There are so many day-to-day decisions that we need to make. When we're practically in lockdown, some of those decisions are easier. But in more ambiguous times – for example, this past September – if you had to decide whether to hold church services in person or send your child to an in-person birthday, it’s helpful to know how much COVID-19 activity there is in your neighborhood,” said Robicsek.

“As an epidemiologist, I would argue that all of those decisions could be informed by your knowledge of how much COVID is in your community right now. There are many different tools available online that give you information about COVID-19 activity. But the most granular level that they present is the county level, and the data lags.”

In the future, Providence plans to expand the DetectionMap tool to show the prevalence of seasonal flu. Leaders also believe that their de-identified information can help hospitals and health systems in the US and around the world make more informed decisions during the pandemic.

“Our health system tries to sit at the intersection between compassion and novel technology, so this is a way for us to use technology to give back to our communities. It feels very aligned with our system's mission,” Robicsek concluded.