Making AI Guidance a Part of Digital Health Governance

At Yale New Haven Health, AI governance has been incorporated into the organization's existing approach to digital health governance.

Yale New Haven Health has been fine-tuning its approach to AI governance.

The health system, which operates hospitals in Connecticut and Rhode Island, was among more than a dozen health systems and hospitals that participated in the HealthLeaders AI in Clinical Care Mastermind program this year. In addition to attending several virtual discussions, participants met in person in September for roundtable discussions as part of the HealthLeaders CMO Exchange in Utah.

Like other leading academic health systems across the country, Yale New Haven Health has been working through finding the right model for AI governance, according to Lee Schwamm, MD, chief digital health officer.

"AI governance is a different twist on digital health governance in the sense that AI tools are not static and they are not predictable," Schwamm says. "AI tools are not like an electrocardiogram machine, where you connect it to the patient, then you get a consistent and predictable output."

Rather than setting up a separate AI governance structure, Yale New Haven Health has determined that AI governance needs to be an enhancement of the organization's existing approach to digital health governance.

"Today, about 15% of our digital health applications are overtly enabled with AI, but a year from now, that figure could be 60%," Schwamm says. "So rather than setting up two parallel governance structures, we have committed to creating a thoughtful model for how we govern tools that have AI."

Part of the enhanced approach to digital health governance is the creation of an AI implementation advisory committee.

"The committee is a multidisciplinary group, with representation from both the health system and the school of medicine faculty," Schwamm says. "The committee's recommendations get advanced into the deployment process, or the recommendations are sent back for refinement or are declined."

Schwamm serves as the chair of the committee, and members include the health system's chief information officer, chief health information officer, chief research information officer, chief data analytics officer, chief quality officer, vice president of clinical applications, and subject matter experts who develop and implement AI tools.

Lee Schwamm, MD, is chief digital health officer at Yale New Haven Health. Photo courtesy of Yale New Haven Health.

Adoption of New AI Tools in Clinical Care

Yale New Haven Health is implementing or evaluating several AI tools that impact clinical care, according to Schwamm.

The health system is assessing an Epic clinical toolkit that includes automatic creation of discharge summaries and summarizing clinical notes from previous patient visits. The organization is also looking at AI tools in the pharmacy space, including tools that would create prior authorization documentation and match the right patients to the right medication opportunities.

The health system is also planning to deploy an algorithm that supports a clinician in making the best assessment for the cause of a stroke. And it’s implementing several billing, coding, and revenue cycle AI tools, which help code care more effectively, help providers pick the right codes, and draft appeal letters when services are denied payment.

Yale New Haven Health also has several homegrown AI applications.

"These include looking at electrocardiograms and deriving findings that usually would require an echocardiogram or cardiac ultrasound to detect," Schwamm says.

Measuring Outcomes and Value for AI Tools in Clinical Care

Outcome metrics should be determined before the adoption of an AI tool in clinical care, according to Schwamm.

"This is similar to a clinical trial, where we would define the primary outcome of interest in advance," Schwamm says. "This enables you to be more honest in your assessment of whether AI tools are adding value to your care."

These measures can go beyond clinical outcomes, Schwamm explains.

"Sometimes, there are AI tools in clinical care that are also associated with important measures of efficiency, such as reducing the rate of patient no-shows in a clinic or [increasing] the rate of patients getting appropriate follow-up care," Schwamm says.

Assessing the value and ROI for AI tools can be challenging.

"It is hard to measure the value of AI tools in clinical care, particularly over short periods of time," Schwamm says. "Some of our AI programs have not been in place long enough for us to see measurable results."

There are early indications that AI tools in clinical care are generating value for Yale New Haven Health. For example, the health system has an AI tool that was adopted last year to monitor hospitalized patients and identify those at risk of significant clinical deterioration, including death, cardiac arrest, and going to the ICU.

"By focusing on patients who are at high risk of deterioration, we can intervene sooner to decrease negative outcomes such as mortality," Schwamm says. "Although it is difficult to determine whether this AI tool is responsible, we have seen a significant reduction in mortality over the past year in our hospitals."

AI tools could also add value to remote patient monitoring programs by continuously connecting patients to healthcare in a way that is comforting, according to Schwamm.

"The point is not surveillance like Big Brother," Schwamm says. "Instead, we can ensure that a patient's condition is on the right trajectory, which is a huge opportunity."

So far, the biggest gains from AI tools in clinical care have been better and earlier diagnoses, which are not primarily associated with a financial return on investment, Schwamm explains.

"Ultimately, the reasons you should want better diagnoses and earlier diagnoses are to improve the quality of the patient experience, improve clinical outcomes, and improve the experience of clinicians so you can retain as many of them as possible," Schwamm says.

For AI tools in clinical care, there needs to be a broad definition of ROI, according to Schwamm.

"The value that we achieve is more of a blend of financial ROI and improved health that does not necessarily translate into financial performance in the fee-for-service payment model," Schwamm says.

The HealthLeaders Exchange is an exclusive, executive community for sharing ideas, solutions, and insights.

To find out more about the HealthLeaders Exchange program, visit the program’s webpage or the program’s LinkedIn page. To inquire about attending a HealthLeaders Exchange event and becoming a member, email us at exchange@healthleadersmedia.com.