Lessons from a Physician-CIO on AI Governance with Dr. Stacey Johnston

Dr. Stacey Johnston, CIO and Digital Execution Officer at Beacon Health System, shares her journey from clinician to technology advocate, emphasizing the transformative potential of electronic health records (EHR) in enhancing workflow and patient care. She addresses the initial resistance to EHRs, where clinicians felt undervalued, and highlights the connection between efficient EHR use and improved patient outcomes, aligning with the quadruple aim of healthcare.

Dr. Johnston believes AI can significantly reduce physician burnout and streamline workflows, allowing physicians to focus more on patient care. She underscores the necessity of understanding workflows and aligning physician needs with hospital operations for effective AI implementation. Successful AI applications, such as agentic AI for managing appointment backlogs and improving colon cancer screening rates, demonstrate AI's impact on clinical processes.

She identifies underutilized AI opportunities in healthcare, such as referral management and medication refills, which could ease cognitive burdens on primary care physicians. Concerns about large language models (LLMs) include risks of inaccuracies and privacy issues. To address governance challenges, Dr. Johnston established a structured AI governance process at Beacon, including an executive steering committee and an AI council that developed policies to ensure responsible AI use.

AI policy mandates that all AI applications receive approval from an advisory council, with restrictions on inputting sensitive data into AI tools. Approved applications enhance workflows through solutions like ambient listening and clinical AI agents, improving efficiency in documentation and case prioritization. Governance is critical, requiring detailed submissions on data modeling and monitoring for bias, while an AI literacy program equips managers to oversee AI initiatives effectively.

Trust in AI is built through successful backend implementations, such as revenue cycle management, and the ambient listening solution has notably reduced documentation time. The gradual establishment of trust is essential, particularly in a healthcare environment with tight margins, where demonstrating ROI is crucial. The potential for automating decisions related to colon cancer screenings aligns with Medicare Advantage metrics, advocating for a centralized control plane for monitoring AI tools.

Dr. Johnston introduces the concept of an "AI citizen program," allowing users to create low-code or no-code AI tools under a centralized approval process. She emphasizes the need for runtime AI governance, particularly for redacting protected health information (PHI), balancing standardization with agility. Transitioning to a federated model encourages innovation, enabling teams to develop their own AI solutions.

A dedicated AI team is vital for effective implementation, separate from daily system maintenance, and should engage with business partners and stay updated on AI developments. The "Crawl, Walk, Run" approach suggests starting with high-potential use cases before broader expansion, with continuous evaluation necessary to demonstrate ROI. Building trust among clinical staff is crucial, requiring engagement with key physician partners to ensure AI tools are meaningful and integrated into workflows.

Concerns about AI agents interacting with patients are acknowledged, with studies suggesting AI can exhibit higher empathy than human callers. Future applications may include detecting emotional crises and predictive modeling for patient safety. The integration of lifestyle data from IoT devices into EHRs poses challenges, particularly in workflow management, while the importance of genomics in medication management is highlighted.

Dr. Johnston emphasizes the need for enhanced technical capabilities in hospitals to improve patient monitoring and the importance of upskilling healthcare workers, especially in light of current staffing shortages. Partnerships with community colleges are increasingly common to address workforce gaps, and IT teams also require training to manage AI models effectively.

In a lightning round, Dr. Johnston discusses AI's expected impact on healthcare over the next 5-10 years, noting it won't replace caregivers but will significantly change their roles. He admires Oracle's clinical AI agent as a leading example and envisions a future where interactions with technology are seamless and ambient, with AI serving as a co-pilot in clinical settings while managing administrative burdens.