By Justin Gnau, MHSA, RHIA, Region CIO of Information Technology Services at St. Luke’s Health

 

Artificial intelligence (AI) is changing health care, and we’ve only begun to understand its potential. Regardless of the hype—or alarm—surrounding AI, health leaders should view it as another advanced technology among many, a tool with a purpose. AI developments should be accepted only following intentional study to ensure they help achieve the ultimate goals of the health organization.

Health leaders bear the responsibility of ensuring any technology, whether AI or a new imaging machine, promotes the betterment of patients served. This responsibility is why St. Luke’s Health leaders are cautious when investigating and implementing new AI applications. Before implementation, AI advances must prove useful in positioning our providers to do one or more of the following: 

• Improve patient turnaround time

• Reduce administrative burden

• Treat patients in a way that improves health outcomes

• Improve the physician/patient relationship

 

Understanding when AI meets these benchmarks opens the door to using AI in ways that help us remain a national leader in value-based care.

First Steps: Adopting AI for Patient Safety

In 2018, SLH took the first big step into AI. At the time, we were investigating technologies and solutions to improve code blue and sepsis care. We approved and adopted an application using algorithms, AI and care modeling to notify staff when patients are at risk for coding or experiencing critical issues with sepsis.

The results speak for themselves. Since implementing this solution, the number of patients who experienced code blues at Baylor St. Luke’s Medical Center has dropped by nearly 50 percent. This success led us to research and ultimately adopt an AI toolset to aid in stroke intervention. 

Previously, emergency staff waited for a radiologist to review CT scans before moving toward treatment. Now, a new AI device scans CT images, discerns occlusions and stroke potential, and texts the results and images to members of the stroke team. Stroke team members review the AI data, confirm the findings and immediately refer for appropriate treatment. As the team prepares to treat the patient, radiology validates the findings.

Thanks to this radiology assistance, our stroke intervention time has dropped from approximately two hours to nearly 25 minutes, yielding faster care and improved health outcomes.

More, Improved Face-to-Face Experiences

Outside of its radiology and patient monitoring capabilities, AI is giving clinicians more time for bedside care. Generative AI allows clinicians to engage more patients and be less concerned with transcribing, as AI speeds the process along by doing the following:

• Listening to patient-provider conversations while maintaining patient privacy

• Transcribing conversations

• Creating more accurate notes based on conversations

 

When rolling out such AI at SLH, we’ve experienced nearly wholesale acceptance. Clinicians advocate for this technology, as it relieves them of an administrative burden, improves their work-life balance and empowers them to provide more in-depth care for more patients. Clinicians who are nervous about the transition often buy in once they understand our intent is not to replace clinicians but to augment their work.

Once clinicians see the potential of AI—that it can transcribe patient conversations securely, produce more accurate and complete documentation of interactions and even enter notes into a patient’s chart—they’re on board. Though such recording technology is not yet a common solution, our clinicians are eager for its arrival, and the Association of American Medical Colleges suspect such technology could reduce physician burnout.

Proceed With Caution

Most data-driven health organizations sit on mountains of data collected through years of careful documentation. AI uses our stored data in ways that allow us to take meaningful action with immediate effect on patient outcomes. Through AI, we can hone our stored data with precision and speed and derive outcomes that drive population health. There is, however, one potential pitfall: bias.

All AI is only as helpful as the information fed to it. Present biased care models to AI, and your AI solution will be biased as well. The risk encouraged a panel to publish principles to address algorithm bias in JAMA Network Open.

When developing or modifying AI, you can reduce bias by doing the following:

• Be aware of bias potential. You cannot avoid bias if you don’t anticipate its presence.

• Look for bias. As you feed data to an AI program, ask whether you recognize bias of any sort in the data. If so, remedy accordingly.

• Seek outside perspective. Pull in experts beyond your core problem-solvers (clinicians, researchers, etc.). Ask ethicists and others to determine whether your models segment data in an improper manner.

• Review data at regular intervals. AI is not a “set-it-and-forget-it” technology. Best results occur when leadership remains engaged in its ongoing maintenance and development. As you become aware of additional potential biases, ensure such bias does not taint your work and your AI serves to close care gaps.

 

Successful Rollout Starts With Team Involvement

When implementing AI, I encourage health care information technology and physician leaders to consider the wider team who will use the technology. Of utmost importance is identifying the end users—whether physicians, nurses, researchers or other members of the care team—and involving them in every step of the process. Early inclusion makes the transition more natural and prevents end users from feeling forced to learn a new technology on the job. Involving end users on the front end also gives them ownership, which increases the likelihood team members will use the technology and provide actionable feedback on its efficiency and effectiveness.

Used properly, AI may bring about the same level of change brought on by the calculator. Calculators don’t do away with math. They allow us to perform more complex calculations, and AI will reap similar results in the medical field, allowing us to push care to the next level to the benefit of patients and clinicians alike.