AI in action: How 4 health systems are deploying clinical AI tools

Artificial intelligence has captivated the healthcare industry like no other technology has managed to do before. And it’s no wonder: even though we are just at the beginning of exploring what AI can really do, some organizations are already seeing extraordinary results as they work to optimize and enhance the experiences of patients, care providers, and entire communities. 

Leaders at these health systems have good reason to be proud of these early wins, especially since a successful deployment of an AI idea is by no means guaranteed. huge number of pilots die on the vine before they can reach maturity due to inadequate governance, lackluster buy-in, and even fundamental flaws in the technology itself. 

So it makes sense not only to celebrate the ones that have managed to make it through, but for executive leaders to share how they are approaching their AI strategy to up the odds that a greater proportion of future initiatives can move forward in a seamless and successful manner. 

At the Trace3 Evolve Technology Conference, panel participants,, and keynote speakers from leading healthcare organizations across the country outlined their current and upcoming AI activities, with a strong focus on bringing clarity and insight to the clinical environment.   

Here are four stories of how health systems are putting AI into action with an eye toward enhancing proactive, data-driven clinical decision making. 

Maximizing productivity and precision with ambient listening and radiology tools

Smoother workflows and better support for clinicians are among the top goals for James Kluttz, VP and CTO at Sutter Health. 

“We have a laundry list of very exciting projects in play,  We’re bringing in ambient listening, which we’ve now deployed across our entire ambulatory division. nd we’re moving it into the inpatient space to further reduce the pajama time,” he said. “We’re leveraging assistive AI technologies for our radiologists to make quicker diagnoses, and even help them identify anomalies that they might not be directly looking for.”  

“We’re also using computer vision for risk fall detection, where we can scale our team’s visibility from one patient at a time to 20 patients or 100 patients. That’s an enormous boost in capacity to provide care without needing to ramp up our staffing.” 

Working methodically to identify promising tools, as well as collaborating directly with clinicians to inform next steps and iron out any wrinkles are critical for success, he added. 

“There is no substitute for collaborating hand-in-hand with our clinicians to make sure that what we’re doing in the AI world actually is delivering the results that we expect,” he stated. “Some of our best ideas come from our frontline teams, and we absolutely prioritize harnessing that input to generate better outcomes for these projects.”  

Surrounding the smallest patients with AI-enabled situational awareness

At Children’s Hospital Colorado, the littlest things can make a big impact for tiny patients, said Patrick Guffey, MD, MHA, Chief Medical Information Officer. 

“Deterioration detection is so important,” he stressed. It’s nearly impossible to tell if a patient is getting better or worse just from EHR notes, which only provide static data from a snapshot in time. What if we had the ability to create a sort of ‘film strip’ in the EHR that showed their trajectory? Or visually see on a color-coded chart of the patient’s organ systems which ones are doing well, and which ones might need attention immediately or in the near future? We’re building this with a leading EHR vendor and working through that concept.” 

Similar technology could make it easier to monitor patients from home instead of keeping them in the hospital, which may reduce emotional, financial, and logistical burdens on the patient and their families. We’re going to have to use technology to zero in on the patients that really need to stay in the hospital, those that may need more frequent follow-up, and those that can go back to living their lives as normal with only periodic follow-up once or twice a year,” said Guffey. 

For those that do need to stay in the inpatient setting, however, leaders need to make sure that their environment doesn’t cause more problems than it solves.  

“Smart cameras for room monitoring are so important in pediatric settings,” he said. I know that my toddler could escape their crib with ease, despite my best efforts.. t when a toddler crawls out of bed in the hospital and bonks their head on the ground, that could mean another surgery that didn’t have to happen. . ing able to detect unsafe conditions in a room, like a bedrail being down when it shouldn’t be, could be very powerful, especially if the care team gets an alert when the system identifies a potential problem.” 

Leveraging AI to detect neuromuscular conditions in premature babies

Sticking with pediatrics, Jason Peoples, Director of Technology and Innovation at Mary Free Bed Rehabilitation Hospital, shared how AI can help identify neuromuscular conditions, like cerebral palsy, in premature infants. 

“Even with the right workflows established, it’s not easy to be extremely accurate about assessing the movements of the tiniest babies without some help from technology,” he said, describing a project that started in the pre-COVID era. We thought about how we could capture, retrieve, and analyze video to make these assessments, which would be especially useful in care settings that don’t have easy access to specialists specifically trained in this area.”  

Using LIDAR cameras to capture the movement of upper and lower extremities, the team developed a method to incorporate augmented reality into an overlay of the images to give clinicians an enhanced visual understanding of the patient’s situation.  

“Our goal was to develop a machine learning algorithm that’s as accurate, if not more so, than what our human eyes are seeing,” he said. We’ve developed the model and refined it for better precision, more accuracy, and a more expedited output. Our model is testing at about 80% accuracy. Right now, an MRI today is about 85% accurate, so that’s pretty good.. ’re continuing to work on this project to assist with earlier diagnoses, which can positively change the trajectory of the patient’s lifetime experiences as well as reduce the overall costs associated with their condition.” 

Aggregating and summarizing data for dialysis patients

In dialysis care, clinical data comes in from all directions, making it difficult for providers to easily extract key features in the record, synthesize the information, and generate an accurate portrait of the patient’s status on their own, said Adam Weinstein, MD, CMIO at DaVita Kidney Care.   

The challenge gets even more complicated when using that data for billing, reimbursement, and care coordination with other facilities – especially as best practices change and medicine continues to advance. 

“We’re finding emerging trends in the terminology used in documentation, which can be confusing for natural language processing (NLP) algorithms that aren’t familiar with how those terms or definitions of terms are changing,” he said in a keynote address at the conference. For example, the acronym ‘HFPEF,’ which has to do with heart failure, only started to show up in the last five or six years, but now it’s the most common diagnosis that dialysis patients have.” 

“AI algorithms have to be built with these types of changes in mind, because they need to be flexible enough to adapt to new terms and concepts in the record. That involves building in an AI governance process that has appropriate testing of our NLP, appropriate sampling with perhaps deeper reasoning models so that we can check our original model, and building a single source of truth that we can add to over time as things change.” 

Success with these tools to surface key data to providers relies on following structured processes and educating users about how to best work with AI-enabled functions.  

“Not only do we have to make sure the AI technology is working as expected, but that users are engaging with the tools in an appropriate manner so that we stay ahead of the drift that is clinically validated and avoid drift that isn’t,” he said. “There is so much we can do with AI to change the experience of caregivers, and therefore the experience of patients.  It’s hugely promising for us and for the rest of the healthcare ecosystem.” 

Jennifer Bresnick is a journalist and freelance content creator with a decade of experience in the health IT industry.  Her work has focused on leveraging innovative technology tools to create value, improve health equity, and achieve the promises of the learning health system.  She can be reached at [email protected].