Most data science in healthcare settings is separated into either clinical or business operations work. These projects aim to either improve the delivery of care or make the organization more efficient.
But that’s not the case at Symphony Post Acute Network, a Chicago-based healthcare provider, where Nathan Taylor, director of data science and analytics, has been working for the past year on data science projects touching both sides of the provider’s operations.
“We tend to be focused on one area and gain expertise in that area, but making that transition between the two is the challenge,” Taylor said.
Healthcare under pressure to improve
There’s never been more pressure for data scientists in healthcare to bridge this divide. More regulations are coming to healthcare that dictate how clinical care should be delivered, which can cause a ripple effect on how the business is managed.
For example, one rule created by the Affordable Care Act specifies penalties for hospitals through the Medicare program. Reimbursements for care are reduced when patient readmission rates exceed a certain threshold. The goal is to encourage providers to deliver effective, efficient care the first time rather than rushing patients out of the hospital only to see them return later with the same condition.
Reducing readmissions has been a primary target for data science initiatives at most hospitals since this rule went into effect, and Symphony is no different. Taylor is leading an initiative to reduce readmissions using a machine learning platform from DataRobot Inc. to develop a tool that assesses incoming patients and scores them on their readmission risk.
At first, Taylor and his team started developing the tool using R. They developed an algorithm that appeared to work, but translating the work into the health system’s production environment proved challenging. That was one of the main reasons for using the DataRobot tool, which comes with prebuilt APIs for connecting models to production environments.
Applying data science to the problem of readmissions is a good example of how clinical care delivery can affect the business operations of a provider. When the providers’ care is suboptimal, patients suffer, which also hits the health system’s financial results. Taylor said this shows why it’s so important for those doing data science in healthcare to not focus too closely on just a single application.
“Readmissions is a good example of how a clinical measure can impact operations,” he said.
Bridging clinical and business takes care
Still, building that link can be easier to conceptualize than to put into action. Most enterprises hiring data scientists today are looking for hybrids — people who have Ph.D.-level math skills, programming ability and business expertise. In effect, they are looking for data scientists who are paid to be experts.
But Taylor, who said he doesn’t consider himself this kind of unicorn of data science, said the ego that can come along with this kind of expertise can sometimes stand in the way of cross-department data science initiatives.
“I tell people to be humble about what you don’t know,” he said. “If you go into things trying to know everything yourself, it’s not going to go that well.”
Instead, Taylor recommends talking to people from all the departments that will be touched by a data science project or have ownership of a data source. They are the ones who know what they need from a tool and what certain data elements mean. Taylor said he regularly talks to doctors and nurses, as well as accountants when developing a new tool.
The business and clinical operations of healthcare providers may have traditionally been run as separate fiefdoms, but with evolving regulations and best practices in the delivery of care, there’s no reason they have to be separate from the perspective of data science in healthcare.
“Having bridged both of those, it makes the business stronger,” Taylor said.
Author: Ed Burns