Natural language processing alerts and navigation tools have made it possible to improve detection of serious medical conditions, route patients to appropriate care and enhance the overall patient experience within an integrated, value-based care system.
The Southern California Permanente Medical Group in Pasadena has built such a tool into its clinical workflows. Recognized as a top scientific innovation of 2025 by the European Medical Journal, the Kaiser Permanente Intelligent Navigator (KPIN) uses advanced augmented intelligence (AI)—also known as artificial intelligence—to let patients describe their needs in their own words, then guides them to the next step, whether it is scheduling care, requesting a prescription refill or connecting with a physician.
"Online patient portals are now a mainstay for accessing services, communicating with providers and self-scheduling appointments. However, most health care portals provide patients with a list of options that may not adequately meet their specific care needs,” Khang Nguyen, MD, lead developer of the platform, told Healthcare IT News. He is medical director for Care Transformation at Southern California Permanente Medical Group, and chief medical officer for Care Navigation at The Permanente Federation.
Southern California Permanente Medical Group is part of the AMA Health System Member Program, which provides enterprise solutions to equip leadership, physicians and care teams with resources to help drive the future of medicine.
KPIN functions like a driver, taking patients where they need to go, Dr. Nguyen explained further in a Permanente Medicine op-ed.
“This tool was built into established workflows, providing 4.9 million patients in the region with a more seamless, timely and personalized experience through its patient portal,” he wrote.
A study published in Nature showed that KPIN identified high-risk symptoms with more than 97% accuracy, and performed well with its clinical models.
The custom-built digital customer service platform was designed in collaboration with physicians and clinical operational leaders, Dr. Nguyen explained in his op-ed.
“Safety was top of mind as we developed this tool. KPIN watches closely for any indication that a patient is facing a medical emergency,” he explained. If a patient writes that they’re having chest pain, KPIN will recognize the immediacy of the situation and connect them to a nurse to discuss symptoms and direct them to an emergency department, if needed.
KPIN builds on safety net
Patient portals are now a key part of digital health care self-service. However, ensuring that these systems consistently guide patients to the most appropriate clinical care remains a challenge.
To address concerns about self-scheduling risks, Southern California Permanente Medical Group initially developed the Virtual Safety Net, which used natural language processing to identify time-sensitive health risks. While effective, the system was difficult to scale, highlighting the need for a more flexible and comprehensive solution that could better integrate safety features, accurate care navigation and user-centered design into everyday digital interactions.
Responding to this challenge, the KPIN portal was developed to embed clinical alerts and navigation tools directly into the appointment booking process. The tool uses a multilingual natural language processing system that incorporates large language models and custom transformer technologies, which provide tailored electromagnetic solutions engineered for specific voltage, power, size and efficiency requirements.
The system analyzes patient-reported symptoms, detects high-risk conditions, incorporates demographic and clinical guideline data, and recommends appropriate care options. These options may include video or phone visits, asynchronous messaging, or other clinically appropriate services. An interaction is considered complete when a patient selects a care option.
KPIN accurately identifies high risk symptoms
In the Nature study of 3 million patient encounters, KPIN showed strong performance in identifying high-risk symptoms, with 96% accuracy, 97.5% precision and 96% recall. Its clinical navigation models also performed well, achieving 81.9% accuracy, 85.6% precision, 81.9% recall and an F1-score of 82.8%.
Additional results showed an adjusted successful booking rate of 53.68% and a low abandonment rate of 2.94%. “Abandonment” refers to an unfinished session. Patient satisfaction scores also increased by nearly nine percentage points.
The findings suggest that integrated digital navigation systems can improve clinical safety, simplify access to care, and support value-based care by guiding patients more accurately to appropriate services.
Future research should focus on establishing baseline measures before implementation, expanding conversational capabilities, and assessing real-world clinical outcomes to better understand how these tools can support clinical workflows and improve timely access to care, Dr. Nguyen and co-authors noted.
“Care is local, but at the same time it’s virtual and it’s become a global commodity,” Dr. Nguyen recently told Becker’s Health IT. “Patients are really expecting artificial intelligence to support health care in a way that it is supporting other industries, in the sense that people are able to describe what they want versus being given choices.”
“The whole point is care navigation,” Dr. Nguyen told Becker’s. “Where we’re going with this next is to add more specificity into different departments, different specialties, and then different use cases.”
From AI implementation to digital health adoption and EHR usability, the AMA is fighting to make technology work for physicians, ensuring that it is an asset to doctors. That includes recently launching the AMA Center for Digital Health and AI to give physicians a powerful voice in shaping how AI and other digital tools are harnessed to improve the patient and clinician experience.
