Mayo Clinic has launched a multitude of AI efforts, from its ongoing alliance with Google Cloud to a joint venture to develop ECG-based algorithms. Now, the hospital system is striking a two-year partnership with Verily to build decision support tools, starting with cardiology and diabetes management.
Although Verily and Google may work under the same parent company, the project’s leaders emphasized that this is a separate effort. The goal, Mayo Clinic Cardiologist Dr. Rick Nishimura said, was to build a suite of tools that are better than the options physicians currently have at their disposal.
Most tools today are “one size fits all,” leaving physicians to figure out how to make the software’s recommendations relevant to the patient sitting in front of them. They also don’t solve the problem of surfacing all of the information that’s needed to make a decision. And with little regulatory oversight of most clinical decision support systems, it’s difficult to know if they are truly accurate, or sounding too many false alarms.
“The current clinical decision support systems now, physicians hate… The computer says you have to do this or else we’re going to keep on bugging you with these red lights in your EHR,” Nishimura said in an interview with MedCity News.
So, how do you build something physicians would actually want to use?
Mayo Clinic and Verily plan to design it to surface records on a patient that would be relevant to that case, and pull in recommendations from Mayo Clinic’s subspecialists on medication, dosing or tests to order, for example.
“Instead of the physician being told what to do, the physician or caregiver is able to take these inputs and make their own decision, which is a lot more palatable,” he said.
To start, the companies will work with subject matter experts to determine what will be the most important information to help them make a decision, said Mayo Clinic’s IT section head Jane Shellum. Then, they have to find where that information is actually stored in the patient’s record — is it in an echocardiogram report, or a clinical note?
After testing to ensure they got the data they expected, the next step is to bring it in front of clinicians to see if there’s anything else that would be critical to their decision.
“I just had a conversation with a cardiologist today, who said if you’re going to show me this piece of data and this piece of data… there’s one more piece of data that tells the whole story about what I’m thinking about,” she said. “There’s the technical validation to make sure the data is correct and there’s the clinical validation to say, is it meaningful data?”
Later on, they plan to run a study to determine if the tool was successful in reducing the cognitive burden for clinicians.
While they may be starting with cardiometabolic conditions, Shellum said they are designing the tool in such a way that it could be used in different settings. It’s a good place to start because of the potential impact while still being “the right level of hard,” said Verily’s head of health informatics, Paul Varghese.
“It requires the synthesis of a range of information that might be captured in different formats and different forms within an EHR. It might be encoded in structured data. It might be fully documented in a clinical note,” he said. “When we looked at a solution for this, we realized it required a reasonable level of sophistication, knowledge and experience.”
They plan to design it in such a way that it will be able to integrate with most commercial electronic health records using Smart on FHIR standards. To start, it will be deployed internally at Mayo Clinic, but the companies plan to extend it to Verily’s health system partners and customers.
Photo credit: ipopba, Getty Images