Recap of Charting the Course: Artificial Intelligence at Kaiser Permanente
On March 6, 2024, KP International hosted a conversation on artificial intelligence at Kaiser Permanente’s Center for Total Health in Washington, DC, with leaders discussing how we are approaching governing, designing, and implementing artificial intelligence in our healthcare delivery system. Four key themes emerged.
Rapid growth, with regulation trying to catch up
First examples of AI go back to the 1950s and 1960s, but it is in the last 18 months, since the launch of ChatGPT that we’ve seen exponential growth and evolution in AI capabilities.
Walter Suarez, MD, Executive Director, Health IT Strategy and Policy, noted that there are 31 countries that have AI laws. The United States has only one federal law to date, and that simply enabled the work to figure out regulation of AI to start start. Complicating the regulatory picture in the United States is that there are more than 100 state bills designed to put safeguards around AI use, but those are not yet official laws. Two bills of note in California and Virginia, both states where Kaiser Permanente operates, are trying to address algorithmic discrimination.
Since the event, Kaiser Permanente has announced our participation in the Coalition for Health AI, a coalition of health care systems, patient advocate communities, technology companies, and the government, working to ensure the safe, equitable, and trustworthy use of AI.
AI’s history: augmenting human intervention
Daniel Yang, MD, Chief AI Officer for Kaiser, provided a brief and insightful historical view on AI.
AI existed for years, functioning as expert systems, codifying knowledge into rules. The challenge with this capability is that it did not scale.
In the last 10-15 years, the shift focused to classification via deep learning. One notable example was Google’s work training AI to distinguish cats from dogs. In the healthcare realm, the greatest advancements have been with imaging and radiology.
More recently, the focus of AI has been on predictive capabilities. One Kaiser Permanente example Daniel highlighted is the Advanced Alert Monitor, an AI tool that predicts patient decline in the hospital. While the model is clinically validated, what is most exciting is that this has been deployed in 21 hospitals, with workflows and operational models to support the alerts. And, it is saving 500 lives per year. This is AI supporting human intervention.
Generative AI is what has sparked the rapid evolution of AI in the past 18 months. The best impact in health care is likely on how AI can reduce cognitive burden for care teams.
Potential within integrated care models
Vivian Tan, Vice President of Strategic Information Management and Global Relationships, noted that AI is about platforms and capabilities. Kaiser Permanente has more than 100 petabytes of data, with the largest growth coming from unstructured data. In her role, Vivian is looking at how to match high value AI capabilities with the best platform, and look to the scale of our full enterprise, not only at our regional level.
Kaiser Permanente has an incredibly diverse member population, and when we think about implementing AI, we need to look at who is creating the models, and look at diverse hiring practices, continued maintenance of models, and protocols to manage model drift.
Due to our focus on value-based care, Kaiser Permanente is motivated to look at potentially different problems that can be solved with AI. We aim to keep people healthy, and focus on empowering patients to engage with their healthcare. We have the opportunity to leverage AI to support the “health” within “health care.”
In Kaiser Permanente Mid-Atlantic States region, Ainsley Maclean, MD, Chief Medical Information Officer & Chief AI Officer, Mid-Atlantic Permanente Medical Group, and Shirley Moncada, Director of Digital, Mobility and Virtual Care Team, shared how we are implementing AI within our integrated model. Dr. MacLean shared examples in imaging and clinical documentation where early tests have shown that AI can serve as an assistant for physicians, helping to make data more accessible and supportive of clinical diagnosis and charting. Shirley highlighted how AI chatbots are helping members navigate to the right care more easily, and connecting members with people when needed. This makes accessing care simpler for members, and enables Kaiser Permanente’s teams to focus on the complex cases that need more care and guidance.
Implementing AI responsibly
A theme woven through all the examples and strategies is how Kaiser Permanente is most focused on the discipline needed to deploy AI responsibly. AI is complicated and risky to implement. Kaiser Permanente is focused on how do we apply AI to the right problems, the ones that deliver the most value. Guardrails are essential and will facilitate implementation. The goal is to keep bad, biased, and potentially harmful AI out of the system while reducing time to deploy and scale good AI. Daniel Yang has recently published a more detailed look at Kaiser Permanente’s approach to responsible AI.
Fundamentally, Kaiser Permanente believes that AI tools alone cannot improve healthcare, but they can, when deployed responsibly, support physicians and care teams in providing high-quality and equitable care.
We look forward to continuing conversations about this rapidly evolving capability. Stay tuned for more updates. A KP International blog highlights recent news and intiatives related to AI.