An AI application that succeeds as a pilot can still fail across a health system, because scaling is less a technology problem than a change management one, Cleveland Clinic executives said at the institution's A.I. Summit for Healthcare Professionals.

Deploying AI at enterprise scale means improving the habits and workflows of tens of thousands of caregivers, said Rohit Chandra, PhD, Executive Vice President and Chief Digital Officer of Cleveland Clinic, and it depends on three layers of readiness that Tomislav Mihaljevic, MD, CEO and President and the Morton L. Mandel CEO Chair, defined as data, computing infrastructure, and applications.

Moderated by Scott R. Steele, MD, MBA, FACS, FASCRS, President of the Cleveland Clinic Main Campus Submarket, these leaders and others discussed how Cleveland Clinic approaches such challenges with scaling AI in a large health-care system. 

From Pilot to Enterprise

One of the central questions asked of the panel was why an AI application that works successfully as a pilot may prove more difficult to implement across a health system.

Dr. Chandra explained that health care’s complexity makes scaling fundamentally different from testing a technology in a limited setting. He said it’s essential that technology introduced into clinical or near-clinical environments make every caregiver’s work easier.

“It’s easy to implement an idea, try it out, and do something interesting,” Dr. Chandra said. “But when you’re driving this flavor of technology adoption at [Cleveland Clinic's] size and scale, we have to be appreciative of the fact that we’re trying to improve the habits, the behaviors, the lives, the culture, and the processes of tens of thousands of clinical caregivers.”

Scaling therefore requires understanding existing work, determining how technology can improve it, training caregivers, and managing the transition toward new standards of care.

“Just the change management involved at our size and scale is essential if you want to accomplish improving the care that we provide, but it doesn’t come easy,” Dr. Chandra said.

Choosing What to Scale

The discussion then turned to how health systems decide which projects are worth scaling in the first place. Asked how Cleveland Clinic chooses among competing pilots and proposals, Dr. Chandra said the organization begins with its enterprise strategic objectives, including making health care safer, more affordable, and more scalable. Potential applications are then evaluated based on their potential impact on patients, caregivers, or the organization, he said.

Jorge A. Guzman, MD, Executive Vice President of U.S. Markets, described a similar framework when explaining how Cleveland Clinic determines which opportunities offer the greatest value. He said decisions are based on the magnitude of challenges involving quality improvement, patient and caregiver experience, access, and affordability. “We try to identify what are the biggest challenges we’re trying to overcome, the impact that we’re going to achieve by doing that and then figure out how we fund it and prioritize [implementation].”

Three Layers of Enterprise Readiness

Beyond selecting which applications to pursue, health systems must determine whether they have the infrastructure needed to deploy AI more broadly.

Dr. Mihaljevic pointed out that implementing a defined AI application is considerably different from integrating AI throughout a health system.

“The implementation of AI for discrete tasks in health care is relatively straightforward,” he said. “What is not so easy is to integrate AI at the enterprise level on multiple different parts of the enterprise and make it all make sense.”

Dr. Mihaljevic identified three layers of enterprise readiness: data, computing infrastructure, and applications. He also emphasized the growing importance of cybersecurity as health systems become more reliant on technology. “We know that the [cyber] attacks are going to happen. There is absolutely no doubt about it.” He said Cleveland Clinic is considering how to segment its technology ecosystems so that [a cyber] attack would 'limit the blast radius' and affected areas could be identified and fixed quickly.

Dr. Mihaljevic said broader AI implementation could also require reimagining the health-care delivery ecosystem. Rather than adding AI on top of existing processes, he suggested that health systems begin with an “AI-first” approach. However, he added that doing so raises questions about which tasks AI cannot perform, which require human involvement, and what work environment would best support a redesigned model of care.

AI-Ready Data and Rapid Experimentation 

When the panel was asked what technological capabilities would be most important to future AI implementation, Dr. Chandra identified two priorities: AI-ready data and the ability to experiment rapidly. “Data is actually the oil that powers AI,” he said.

Dr. Chandra explained that Cleveland Clinic has petabytes of data, a volume that continues to grow daily, but making that information usable by AI requires bringing it together in an AI-ready form. He described that process as “easier said than done” but said AI-ready data are a “critical enabler” of health care and administrative transformation. He cited patient charts, supply-chain information, and administrative contracts as examples of clinical and nonclinical data that need to be made accessible to AI tools.

Dr. Chandra also said health systems need the ability to test multiple approaches and conduct experiments more rapidly. “Technology is evolving at a rapid pace,” he said. “Health care needs to find a way to accelerate the exploration, the experimentation, and the learning.”

Realistic Expectations

In a closing question, the panelists were asked to identify the biggest misconception about AI in health care. Their responses emphasized the importance of realistic expectations about what technology can solve.

Kelly Hancock, DNP, RN, NE-BC, FAAN, Executive Vice President, Chief Caregiver Officer, and Chief Administrative Officer, said one misconception is that AI will provide a solution to all of health care’s problems, describing the idea as “magical dust.”

Miguel Regueiro, MD, Executive Vice President and Enterprise Chief of Staff, emphasized the need for health systems to prepare for AI implementation. “We need to be ready for what’s coming and not stick our head in the sand, the ostrich mentality,” he said.

Dr. Guzman pointed to patient access to care as an example of a challenge that technology alone cannot address. “Access is not a technology problem,” he said. “It’s a workflow problem. It’s a culture problem.”

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.