In 2010, Seattle Children’s created a clinical effectiveness program under its center of quality to look up all medical literature available for tough questions. The program includes millions of pieces of literature, and is growing; how could they stay updated with the best guidelines for the most common conditions?
Seattle Children’s and Google decided to apply generative AI to the process. They created Pathway Assist, an AI-powered agent designed to improve clinician access to complex information and evidence-based best practices. Healthcare providers often need information about the latest medical research and personalized outcomes quickly to provide the best care for patients. Seattle Children’s identified the issue and decided to leverage the expertise of dozens of providers and comprehensive medical literature to create clinical standard pathways for 70-plus diagnoses.
The project was released in early April to great fanfare, a nod to the hospital’s proactive approach to AI.
“We’re not really going out there to look for technology and then trying to retrofit it into our environment. We’re starting with what the frontline users are asking for and where their needs are,” said Clara Lin, MD, CMIO of digital health and informatics at Seattle Children’s, during an episode of the “Becker’s Healthcare Podcast.” “We came up with a list of use cases that we thought would be very beneficial to the organization from an operational and efficiency standpoint; from an improving quality of care perspective. This one in particular that we developed in partnership with Google ended up being proof of concept for us, an early use case that we wanted to see if it’s possible in a ‘fail fast’ model.”
Those pathways are now available to all healthcare providers with Pathway Assist, which streamlines the information for caregivers. It was built with Google’s Gemini models to share information within seconds instead of the 15 minutes it typically takes manually. For each of the 70 conditions, they developed a pathway for clinicians to standardize care as much as possible. The large language model also takes flow charts, spreadsheets and images that are part of academic articles into consideration, not just the written words.
“We took thousands of pages of really great content developed by the clinical effectiveness team under our center of quality and we layered basically Google Gemini on top of it. It’s not just the straight Gemini that we can see online, but it’s really a Gemini that’s been fine tuned specifically for these pathways that we developed,” she said. “It’s only referring to these pathways. We did it purely through prompt and fine-tuning of Gemini, and right out of the box, it had decent accuracy, but it wasn’t good enough for clinical use. We kept fine-tuning and prompting and changing the prompts and making sure that Gemini understands what we were trying to get out of it.”
All the work paid off. The end product is a conversational chatbot that can ask the clinician questions if the prompt isn’t specific enough and then provide an answer back. The clinician can ask, for example, “I have a four year old in front of me in the ER with an asthma exacerbation. What should I do next?”
“The chatbot will basically ask you questions about the patient, and then in the end it doesn’t tell you what you should do next, though that was the question I asked as a physician. But it will collate all of that evidence-based information from the pathways that we’ve created over the last 15 years by these dozens of experts at Seattle Children’s,” said Dr. Lin. “Then it presents that information to the request or to the person they’re chatting with.”
The platform also provides the reference article and page number for all information and suggested guidelines for the next steps.
“The human ultimately is still the one that’s making the decision, and ultimately deciding if this is the information that is most appropriate for the patient in front of them,” said Dr. Lin. “But using Pathway Assist effectively shrunk the information searching time to just seconds for clinicians.”
Dr. Lin and her team are continuing to develop in the AI space as technology rapidly evolves. The big challenge is supporting the new technology with the right infrastructure and governance, and then scaling throughout the organization, whether it’s purchased, licensed off the shelf or developed internally.
“I’ve been thinking a lot and trying to figure that out because that, ultimately, will be a challenge not just for Seattle Children’s, but everybody out there in healthcare IT that is seeing a rapid expansion of their AI portfolio right now,” she said.
