Using Synthetic Data to Improve Quality of Care

By Matt Phillion

If there’s one thing healthcare has plenty of, it’s data. Health systems are swimming in it. But clinicians sometimes struggle to access that data in a way that can improve care and performance in the moment. Traditionally, hospitals rely on teams of data analysts for insight into large data sets, but it can often take weeks or months for data to be ready to help influence change.

“Healthcare is a service industry, and it’s data driven. Care providers are all trained on how to assess data they’re picking up from patients at the individual level, and also at the population level,” says Josh Rubel, chief commercial officer with MDClone. “There’s a ton of data every time you see a patient, and the number of data elements are only growing. Often, the challenge providers run into is in decision-making.”

Providers work hard to define best practices, create context, and weed out the signal from the noise to make life better for their patients and communities. But still, they often struggle with understanding how to execute on that mission, Rubel explains. “Sometimes there’s just too much data at the population level, and they struggle with speed to value,” he says.

For example, he says, an anesthesiologist may want to know what happens to lengths of surgery on a population level when using one anesthesia medication versus another, or which medication creates the fewest complications. They may want answers about correlations between the patient’s age, social determinants of health, the procedure, and the choice of anesthesia.

“What we find is that providers aren’t able to [get answers to] those sorts of questions quickly,” says Rubel. “If you have any sort of correlative question, you have to engage with an analyst group or organizations, and the cycle time on data sets and analysis is very long. It’s not conducive with human curiosity.”

Simply put, the questions move faster than the data. “I [might] have a question at 2:36 on Wednesday, but have to wait three weeks for an answer … and I’ll be in that same OR tomorrow with a new question,” says Rubel.

What if providers had a better, faster way to make use of all this valuable data? This is where synthetic data comes in.

Defining synthetic data

Many of the elongated timelines involved with data analysis have to do with privacy standards around patient information, Rubel says. For example, what if you want to look at a patient’s data with a consultant who’s an expert on the topic but isn’t caring for the patient? “It can be difficult to go through the regulatory framework to look at this data.”

“What synthetic data does, if you do it right, is allow you to get a data set that is demonstrably similar to real patient data but has no real human subjects in that data set,” he explains. “It’s data assets on demand that tell you the same or very similar stories to those of real patients, which you can then access faster as you wouldn’t have any of those privacy regulations to navigate.”

Rubel differentiates synthetic data from de-identified or anonymous data by explaining it in physical terms. “What we and others like us do is look at the contours or shape of the data. Imagine patients are rows and elements are columns,” says Rubel. “We look at that shape against the real data and recreate that shape, that outcome, that same answer, but without any real human subject data.”

He notes that the use of synthetic data isn’t without risk, though. “if you don’t do synthetic data well, there’s a possibility that a patient could be identified,” says Rubel. That’s why organizations first perform an internal validation test to make sure they’re comfortable. After the first use case, peer-reviewed articles are available for further validation.

Institutions in the U.S., Israel, and Canada are currently using this type of technology, as is the U.S. Department of Veterans Affairs (VA). “The VA is making data available to third parties in projects like Mission Daybreak,” which works toward developing suicide prevention innovations for veterans, says Rubel. “They’re effectively, with synthetic data, creating this privacy-preserving data set and have made it available to organizations to try to predict patients who may be at risk. And this can be a very private subject, like all of healthcare.”

How does it all come together?

There are essentially three stages to making the best use of synthetic data, Rubel explains.

“First, enable these third-party relationships, whether it’s life sciences, academia, or peer institutions,” he says. “They’re excited to be able to let their data assets become more available out in the wild. There are many talented people on this Earth, and it’s useful to get more eyeballs on data sets to improve quality.”

Next is enabling wider distribution internally. “You’re able to push for more ask-and-answer dialogue with the data,” says Rubel. “If you combine synthetic data with data assets and without the three-week lag time, you’re able to engage in minutes and have an amazing impact.” This enables people out in the field to make use of data quickly, with enhanced patient privacy, while also providing clinicians what they need for speedy insights.

The third component, Rubel says, is a culture shift: pushing more data out to the departments themselves so they can collaborate and coordinate more. “Sometimes there are competitive dynamics, and there’s a limited footprint, a centralized footprint” for data, says Rubel. With the synthesized data model and easy-to-maneuver data, departments have visibility into each other’s operations, he notes. “If surgery is doing something innovative, maybe that can be applied in some way over in internal medicine, for example. This is where true data-driven, continuous-learning sorts of organizations start to take off.”

Access to data can also enable more practitioners to work at the top of their license. “It enables capacity planning, who sees who, what’s happening across the organization, how much time doctors are spending in the room, who performs which tasks. You cannot do studies of this kind using the EHR because most of the time the questions in play are novel,” says Rubel. “A nimble model like this can help you with the analysis you need to get to the top of licensure.”

All organizations are feeling the time and staffing crunch, so efficiency and throughput are top of mind. “We want to keep moving fast, answering questions like the efficacy difference between product A or product B, if there’s a difference in quality, if it impacts the time of surgery, resource allocation,” says Rubel. “All of these questions are difficult to answer without the right environment and framework but simple to answer with it.”

Healthcare organizations can understandably be hesitant to spend time and money on operational technology without knowing the payoff. “The best way to prove the economic value is customer examples,” says Rubel. “If you can provide this benefit for one customer, you can find others to make it real, installing the technology to allow for much more direct dialogue and exploration of data by physicians and nurses, folks in the departments rather than those in an analytics organization. It allows for more work at the node, and you don’t need to be a data scientist or coder or programmer. All you need is some limited familiarity with the data itself.”

With enough data literacy out in the field, synthetic data will become more than just a tool, but rather something users will demand to engage with. “We’ve seen stories from the field where we get feedback along the lines of: ‘I can’t believe this is possible. I’ve been asking for this forever,’ ” says Rubel. “It’s opened up a new world.”

A real-life use case for synthetic data has been in figuring out the best way to treat patients admitted in the ED with a blood clot and how to maximize care in a very scary situation. “It’s getting to one insight that can improve their 30- or 90-day mortality rate and make a difference in a big way,” says Rubel.

The end goal is greater access to accurate data to optimize care. “I want more data-driven everything,” says Rubel. “I want a world where we can reasonably shorten the time frame to engage with people and empower really talented, really smart people taking care of these populations to be a part of that conversation.”

Matt Phillion is a freelance writer covering healthcare, cybersecurity, and more. He can be reached at matthew.phillion@gmail.com.