As the newly appointed chief AI officer of Phoenix-based Mayo Clinic Arizona, Bhavik Patel, MD, described his new role as the opportunity to be at the forefront of long-overdue disruptive innovation.
While Mayo Clinic Arizona has various deep learning, natural language processing, and generative artificial intelligence initiatives underway, Dr. Patel discussed three specific areas of focus with Becker's: predictive models for quality measures, opportunistic screening and cancer detection and treatment response.
The health system has successfully developed and is now in the process of implementing AI models that can predict patients' likelihood of a hospital-acquired infection and 30 day readmission. The tools' prediction abilities allow medical providers to triage resources more efficiently, such as obtaining additional tests and equipment to have readily accessible.
Mayo Clinic Arizona has implemented an AI model that identifies patients opportunistically to determine if they are at high risk of experiencing a stroke or heart attack, which Dr. Patel pinpointed as the number one preventable death in the United States. The screening does not require patients to undergo additional processes during their visit, and it provides physicians with essential information to know if they must intervene early and begin preventive measures.
The health system has several initiatives in place that aim to identify cancers at an earlier, more treatable stage, then predict the treatment response for each patient. Dr. Patel told Becker's that the tool's ability to expedite the diagnosis process grants physicians "an upfront view of what that patient's treatment course or therapy is going to be." Providers can then begin to tailor the treatment process that day, opposed to having to play the "waiting game" to see if the patient is doing better or worse.
As there is no specific framework for AI in healthcare, Dr. Patel said the most important way to ensure these models are successful is by measuring impact to patient care.
"If we've developed models that align with solving clinically significant challenges that directly impact patient outcomes, then there's a tangible way to go back and do an impact analysis," Dr. Patel told Becker's.
While these impact analyses are beneficial to demonstrating success and building clinician trust and adoption, they have a dual responsibility to expose the technologies' flaws. Dr. Patel emphasized the importance of identifying and correcting instances when a model makes an incorrect prediction, but acknowledged the way this clashes with the risk-averse nature of healthcare providers
"If you have an AI model in a manufacturing business … and you have a way to predict which product is defective and you're wrong 1% of the time, it's probably not as big of a deal," Dr. Patel said. "There might be a revenue cycle or financial sort of consequences, but, in healthcare, obviously 1% means patient lives."
Dr. Patel agreed with the cruciality of healthcare providers being risk averse, but not to an extent that prevents the progress of adopting solutions and technologies that could improve the patient and provider experience.
"When we look at technology and AI models, there's a tendency to want them to be perfect … I understand why we have that expectation. … The issue is no human being is perfect," Dr. Patel told Becker's. "Sometimes we have to understand that … even if it helps a little, even if it incrementally adds value, we need to embrace that."
Still, he said embracing these technologies must be done in a responsible way by prioritizing robust studies, scientific rigor and clinical trials in the preboarding and implementation process. Along with conducting these forms of research, Dr. Patel said that mindfully framing the advantages of new AI models, incorporating change management and reiterating the "shared goal to preserve the humanity of medicine and to do what's right" will deter risk averseness from preventing progress.
"I'm looking forward to the day when AI is seamlessly woven into the fabric of our healthcare operations," he said. "We can't continue to live with some of the healthcare statistics that we have. That's why I'm so passionate about AI in healthcare because I truly believe it is the technology, the tool and the innovation that will finally help us get to where we need to get."