Artificial intelligence is only as strong as the data behind it. Healthcare leaders are working with their teams to build the infrastructure, governance, and culture necessary to support AI at scale.
Healthcare leaders gathered at the AI Summit during the Becker’s CEO+CFO Roundtable in early November to discuss the need for high-quality data, alignment between technical and clinical teams, and a cautious approach to security and interoperability.
Five emerging trends:
- Clean, well-structured data is the foundation for enterprise AI
Rajiv Kolagani, chief data and AI officer at Ann & Robert H. Lurie Children’s Hospital of Chicago, said hospitals can’t expect meaningful results from AI without first addressing the basics of data readiness. “We can’t really get any utility or value out of AI without data,” he said. “When you think about AI, you have to really think about preparing data for AI differently.”
Lurie has spent three years on a data modernization initiative, shifting to a modern cloud-based data platform that includes Snowflake and Fabric. “That became the foundation for us to do any kind of AI on top of it,” he said.
Lurie now uses a medallion architecture with bronze, silver, and gold layers to ensure trustworthy data sources. From there, the team builds knowledge graphs to “teach AI what the data means,” enabling semantic understanding of concepts like infections and diagnoses.
“You have to teach AI all of the context,” he said.
- Data quality isn’t an accident — it requires structure and accountability
Quality must be intentional and tracked through every stage of a health system’s data ecosystem. “It’s not an afterthought,” Mr. Kolagani said. “In order to make that happen, you have to have different tooling in your infrastructure.”
Lurie implemented data catalogs and lineage tools to trace where data originates, how it’s processed, and what errors occur along the way.
“There’s a ton of work that actually goes into making sure that you have high-quality data output,” he said. “Quality doesn’t happen by accident. You have to be super intentional about it, and you also have to have the right guy or gal running the data shop.”
- Healthcare’s data challenge is usability
Margaret Lozovatsky, MD, CMIO and vice president of digital health innovation at the American Medical Association, said that while healthcare is data rich, it remains information poor.
“Anybody that has ever used an EHR to deliver care to their patients knows that there’s a lot of stuff in there that is not accurate and shouldn’t be in there,” she said.
She described the problem as one of structure and interoperability.
“So much of our information that exists in the EHR today exists in a format that is a narrative,” she said. “We have struggled for years with interoperability, because every organization has this data in their own format.”
Large language models could help overcome those barriers, processing and translating disparate data into usable insights. “These tools can unlock the data and translate it into useful information that clinicians can use to make decisions,” Dr. Lozovatsky said.
The AMA focuses on AI with the conceptualization of augmented intelligence, emphasizing that while computers excel at processing, humans understand the context of that data.
- Governance and security must move as fast as innovation
Protecting patient data while encouraging innovation requires strong governance frameworks.
“We need to go slow to go fast,” Dr. Lozovatsky said. “Setting up these processes today to understand how your organizational data is being used in testing these tools and implementing these tools to ensure that at the end of the day, the patient information is not identifiable is really important.”
The AMA’s new AI Governance Toolkit, which offers an eight-step process for balancing privacy and progress. “It’s an eight-step process that is some of the best practices for organizations to think about as you’re setting up your governance within your institution,” she said.
- AI’s next leap will come from administrative efficiency and connected care
Early AI gains will likely come from automating administrative functions rather than direct clinical care. Dr. Lozovatsky predicted that “we’re going to start to tackle all of the administrative burdens first,” including patient access and scheduling. “That’s the next iteration of these tools.”
Anil Saldanha, chief innovation officer at Rush University System for Health in Chicago, added that AI represents the reset the industry needed. He envisions a future of connected, data-driven care across the continuum.
“We can’t just look at health systems in isolation. We live in a connected world,” he said. “I’m really bullish on diagnostics, machine learning, AI, cancer detection, stroke triage […] we’re building the plane as we fly, but I’m really bullish on the future of medicine.”
