Of Epic Proportions

Rollins faculty pioneer new research using a groundbreaking database.

When you go to the doctor’s office, an assistant will wrap a blood pressure cuff around your arm. They’ll lead you to the scale to check your weight. When the physician comes in, they might type notes as you describe your symptoms. Then, they might order some tests.

What if researchers could access this kind of everyday data from billions of medical encounters? That’s exactly what’s happening at Rollins and institutions across the country. 

Most patient data is stored in an online system called an electronic health record (EHR). These systems organize your medical data so that your provider can easily access it. If a different provider uses the same system, they can see it, too. That’s how your endocrinologist might be able to review bloodwork results from your primary care provider.

A few years ago, the largest EHR provider, Epic, built a platform for sharing de-identified data with researchers. “Interest really started because of the COVID-19 pandemic,” says Howard Chang, PhD, professor of environmental health.

In the face of a desperate need for data, the EHR system was, according to Chang, “in a position to provide high-quality and real-time information on how many people are going to the hospital and their health outcomes." Epic saw this as an opportunity to expand from being just an EHR to being a steward of big data. They created a database called Cosmos, integrating data from several health systems into one large, de-identified database that researchers could use for different purposes.

At Rollins, faculty members like Chang; Jithin Sam Varghese, PhD, assistant professor of global health; Hui Shao, MD, associate professor of global health; and Ben Druss, MD, Rosalynn Carter Chair in Mental Health, are unlocking the secrets of this enormous database.

A Bigger Picture

The database offers many advantages, the greatest being its sheer size. Cosmos contains data on more than 300 million patients—over 85% of the U.S. population. In comparison, most health datasets are based on hundreds of thousands of individuals. Not only does this strengthen the power of statistical analyses, but it also enables researchers to collect sufficient sample sizes to study rare diseases.

For example, there are nearly 1 million people with multiple sclerosis in the United States, and Cosmos contains reliable records for over 440,000 of those patients. That’s a much larger sample than traditional hospital-based studies, which might only include a handful of patients. As a result, “diseases rare to study in one clinical setting now have large sample sizes for robust epidemiological analyses,” says Varghese.

But there’s an obvious question: If Cosmos only includes patients whose providers use Epic, can researchers really use it to say something about the entire country?

“Before Epic came along, we were limited to certain datasets that were large, but they had their own limitations.”

Howard Chang, PhD

Professor of Environmental Health

“Before Epic came along, we were limited to certain datasets that were large, but they had their own limitations,” says Chang. For example, Medicare claims data mainly covers the population of older adults, so you can’t use it to analyze children’s health. Meanwhile, private insurance claims might overlook the needs of people from low-socioeconomic backgrounds. To find out whether Cosmos is representative of the U.S. population, Shao recently mapped the penetration of Epic across every ZIP code in the country. The results revealed a high, consistent use of the platform, meaning the data should be generalizable.

Studying History in Real Time

But size is only part of what makes Cosmos different.

Historically, health policy and management research has relied on claims data and surveys, both of which are a few years old at best, says Druss. In contrast, Cosmos data has a lag time of only a few months. This near-real-time information allows researchers to study new trends, like the uses of GLP-1s. For example, Shao used the dataset to discover that even though GLP-1s are not indicated for people with type 1 diabetes, a large proportion are being prescribed them off-label.

Cosmos also offers researchers a more comprehensive, longitudinal view of how diseases progress over time. Varghese took advantage of this by studying variations in progression between individuals with type 2 diabetes to different complications like retinopathy, cardiovascular disease, cancers, and dementia. Prior to Epic, there were limited ways to study disease progression at a national level. Most large datasets are cross-sectional, so they don’t include long-term trends. Medicare claims data could have been helpful for studying the later stages of disease, but many people develop diabetes in their 30s or 40s, before they are eligible for this insurance program.

Druss is using Epic to study the evolution of a very different disease. “Psychosis is rare, it's hard to diagnose,” he explains. This makes it hard to study. But Druss is trying to understand how common first-episode psychosis is, and what happens before and after the diagnosis. He asks, “Did they have other mental diagnoses that are precursors to psychosis? Do they have more emergency room visits in the year afterwards?”

Together, these projects illustrate what Cosmos makes possible: Researchers can study not only how common a condition is, but also what happens before, during, and after it.

Getting Under the Hood

The enormous size and scope of Cosmos also creates new questions. The researchers are learning where the dataset falls short—and how to mitigate such limitations.

Cosmos “still suffers from a lot of issues with electronic health records,” says Chang. If a patient goes to a clinic that doesn’t use Epic, there will be a gap in the patient’s record. And even if a patient’s data does get entered into Epic, the detail and quality of that information will vary by provider.

It also doesn’t capture what happens outside the health care system. “It may be that your doctor prescribes a medicine for you, but you decide not to pick it up,” Varghese says, whether that’s because of cost, transportation, or other barriers.

Furthermore, Chang says, Epic “is not as established as other big health datasets.” That means researchers need to learn more about how it works and how to use it effectively.

“The only way to learn about this sort of dataset is to get in under the hood and start trying things out.”

Ben Druss, MD

Rosalynn Carter Chair in Mental Health

“We're using this as an opportunity to better understand Epic Cosmos. What are the strengths? What are some of the gaps? How can this be used more generally for medical health research moving forward?” says Druss. “The only way to learn about this sort of dataset is to get in under the hood and start trying things out.”

As some of the first researchers to use this dataset, Rollins faculty are helping pioneer new approaches to big-data science. For example, Shao is experimenting with a new method called trial emulation, which uses datasets like Cosmos to create an observational alternative to randomized controlled trials. It follows people who use—or don’t use—a specific intervention over time to see what the outcomes are. For instance, Shao is testing whether people who take semaglutide are more likely to develop thyroid cancer than people who don’t.

Meanwhile, Varghese hopes to integrate data from Epic with other sources. “One thing I really want to see is, can we use Epic data along with national surveys and other datasets to get better estimates? If so, what are the analytical methods required? What are the assumptions which must be considered when you're trying to combine data from Epic with non-Epic data systems?” This could potentially mitigate limitations from both datasets, one filling in where the other falls short.

For Druss, the possibilities are endless. “I can imagine many PhD students doing their dissertations using data from Epic Cosmos, faculty writing grants using it, using it for preliminary data, for other kinds of grants because the data is so up-to-date and high-quality...It's a wonderful resource for many different groups within the school.” To make this kind of resource more widely available, Druss is co-leading the Rollins Mental Health Research Accelerator program. Among other goals, this initiative is harnessing the power of shared infrastructure and datasets like Epic to support interdepartmental research.

He says, “I feel like it really is the future of health services research, and perhaps even more broadly, epidemiology and some other public health disciplines as well.”

A blood pressure reading. A weight. A doctor’s note. A lab order. On their own, each of these things might feel ordinary and inconsequential. But when combined with billions of other data points, they form a picture—one that Rollins researchers are just beginning to unveil.