Artificial intelligence (AI) has already changed how health care organizations collect and analyze data. Now, it’s changing how they build new solutions for challenges such as automating administrative work and supporting clinical decision-making.
A Harvard T.H. Chan School of Public Health webinar on June 9 explored how health care organizations are using AI tools and best practices for ensuring the safety of patients and the security of health data. It was part of the Pressure Points series, co-hosted by The Studio and the Advanced Learning Academy (ALA).
Panelists included Trishan Panch, CEO of Lunar Studio and executive chair and chief strategy officer of Lumin Health, and Heather Mattie, lecturer in biostatistics, both of whom are co-directors of several ALA courses on AI in health care; John Brownstein, senior vice president and chief innovation officer at Boston Children’s Hospital; and Aman Bhandari, chief AI and analytics officer at SCAN Group. Rifat Atun, vice dean for non-degree education and innovation, moderated.
Panch summarized various types of AI tools and explained some of the differences between them: Chatbots such as ChatGPT pull together information in response to users’ questions, and generative AI performs actions such as creating an image or lines of computer code in response to prompts. Agentic AI differs in that it’s proactive. The user enters a goal, such as planning a multi-step workflow and the tool figures out the steps to achieve it.
Boston Children’s Hospital has been using AI for years, Brownstein said—for example, to record conversations between providers and patients (with consent) and transcribe them into clinical notes. They’ve also partnered with OpenAI on a solution for improving diagnoses of rare diseases, he said.
“[AI] is really changing the game in terms of delivering care, but it’s also changing our ability to support operations of the hospital,” he said. For example, he said clinicians are working with technology partners to develop solutions that can be used in the ICU to create summaries of patients’ health information from across disparate sources.
Keeping humans in the loop
Bhandari noted that there are two different categories of AI solutions being developed in health care settings: operations (dealing with high-volume administrative tasks) and clinical (which can directly affect patient outcomes). He and other panelists stressed the importance of creating safeguards, particularly around clinical solutions.
It’s important to have a human in the loop at every step of the process in developing an agentic AI solution, including performing regular checks and evaluations, Mattie said.
She also said that clinicians and health care leaders need to ensure that solutions are built on data reflecting the diversity of people served by the health care organization and don’t compound existing health inequities. “They may not realize they’re encoding assumptions about which patients are included, which data fields matter, which outcomes they’re optimizing for,” she said. “What population does this perform well on? Where was it tested? What happens in subgroups? Those questions have to become reflexes and not afterthoughts.”
An AI governance group at Boston Children’s Hospital includes both clinicians and representatives from a range of the hospital’s administrative departments, Brownstein said. The group reviews efforts to develop AI solutions across the hospital, assessing risk, prioritizing projects based on the hospital’s overall needs, and fostering collaboration across clinical departments. They try to support innovative solutions coming from researchers and clinicians, he said, but added that the organization is very careful about which solutions go beyond the prototype phase to being actually deployed.
The panelists agreed that health care leaders need to understand how to use AI tools. Brownstein said, “Many people think every CEO basically needs to be the chief AI officer of an organization at this point, because they need to live and breathe this,” he said. “[AI] has to be part of the fiber of every health system going forward.”
Panch said that having access to AI tools for developing software is “training people to think about the problems that they’re facing and to have belief and, dare I say, agency that they can solve those problems themselves. And I think that’s really important, especially in health care.”
