Self-triage, the critical process by which individuals evaluate the severity of their symptoms before seeking professional care, has traditionally been fraught with uncertainty and inconsistency. While online symptom-checkers and generic chatbots exist, they often suffer from a lack of medical validity, overwhelming users with conflicting information or impersonal interactions. The UC San Diego team’s new approach integrates trusted medical flowcharts, turning symptom assessment into a fluid, human-like conversation grounded in evidenced clinical pathways.

At the core of this AI system lies the incorporation of over one hundred detailed medical flowcharts developed by the American Medical Association. These stepwise decision trees provide the clinical backbone for the chatbot’s guidance, ensuring every recommendation is traceable to a validated medical standard. Unlike conventional large language models, which often operate as opaque “black boxes,” this multi-agent AI architecture offers unparalleled transparency and clinical reliability.

The chatbot works through a sophisticated triad of AI agents operating in concert. The first agent identifies the patient’s primary complaint by analyzing their natural language input, selecting the most appropriate medical flowchart while taking contextual factors such as age and sex into account. The second agent interprets nuanced patient responses—not just simple affirmations or negations—and dynamically determines subsequent questions, thus maintaining a logical and coherent diagnostic dialogue. The third agent translates technical clinical queries into patient-friendly language, enhancing comprehension and response accuracy.

Consider a prototypical interaction where a 35-year-old male reports abdominal pain. The system swiftly selects the abdominal pain flowchart, then poses questions akin to a clinical intake, such as pain intensity and associated symptoms, but framed in accessible terms. This iterative dialogue continues until the AI can confidently recommend whether symptom monitoring, primary care consultation, or emergency services are warranted. This patient-centered conversational design aligns with real-world clinical workflows, promoting user engagement and trust.

Extensive testing was conducted, involving more than 30,000 simulated patient conversations featuring diverse symptom descriptions and linguistic variations. The chatbot demonstrated remarkable efficacy, correctly selecting the appropriate medical flowchart approximately 84% of the time and adhering to the prescribed clinical decision-making process with over 99% accuracy. These robust performance metrics underscore the potential for reliable symptom evaluation in real-world settings.

Despite the impressive results, the developers emphasize that this AI system is not intended to replace clinicians but to serve as an adjunct resource capable of offloading routine triage tasks. By providing accessible, clinically sound guidance at home, the chatbot empowers patients with timely information while allowing healthcare professionals to focus on complex cases. Moreover, the design accommodates clinician oversight by enabling review of chatbot-patient interactions to ensure safety and quality.

Looking forward, the team envisions expanding this technology through integration with electronic health records, fostering seamless continuity of care. Plans include the development of a mobile application, incorporation of voice command capabilities, multilingual support, and the ability to process patient-shared images. These enhancements will broaden accessibility, addressing barriers faced by older adults and non-English speaking populations, and facilitating more comprehensive symptom assessment.

The innovation reflects a significant stride in combining the strengths of large language models with rule-based medical knowledge. By embedding trusted clinical algorithms within an AI conversational framework, the system merges accuracy with flexibility, navigating the complexities of human symptom description while adhering to the highest standards of medical ethics and practice.

Professors and researchers leading this initiative envision a future where AI-powered self-triage tools become integral components of healthcare delivery. Such systems could transform initial symptom evaluation, optimizing health system efficiency by guiding patients accurately and reducing the strain on emergency services caused by inappropriate visits. Ultimately, this technology aspires to bring high-quality medical triage guidance into the hands of everyday users, wherever they may be.

This pioneering chatbot exemplifies how artificial intelligence, when thoughtfully designed to respect clinical rigor and patient experience, can transcend current limitations of digital health tools. The elegant synergy between large language models and structured flowcharts exemplifies a new paradigm—one that prioritizes transparency, user-centeredness, and medical integrity in AI health applications. As real-world testing with hospital partners begins, this approach holds promise for transforming how people manage health concerns at home.

Full study: “A multi-agent framework combining large language models with medical flowcharts for self-triage,” published in the prestigious journal Nature Health, details the technical architecture, clinical validation, and extensive evaluation of this innovative system. The research represents a collaborative effort among experts in engineering, clinical care, and artificial intelligence, with affiliations spanning UC San Diego, Google Research, Kaiser Permanente, UC San Francisco, and Korea University Ansan Hospital.

Subject of Research:
Artificial intelligence in medical self-triage systems

Article Title:
A multi-agent framework combining large language models with medical flowcharts for self-triage

News Publication Date:
20-Apr-2026

Web References:
https://www.nature.com/articles/s44360-026-00112-2

References:
Liu, Y., Wang, E., Liu, X., et al. (2026). A multi-agent framework combining large language models with medical flowcharts for self-triage. Nature Health. https://doi.org/10.1038/s44360-026-00112-2

Image Credits:
Yujia Liu

Keywords:
AI chatbot, self-triage, medical flowcharts, large language models, artificial intelligence, digital health, symptom assessment, clinical decision support, conversational AI, patient-centered care

The challenge lies in ensuring that AI-driven tools like ChatGPT can mirror the intricate reasoning processes of healthcare professionals, especially when dealing with vulnerable populations such as the elderly. Geriatric medicine is particularly sensitive due to the complexity of conditions affecting older adults, which often involve multifaceted interactions between various health issues, medications, and social factors. The objective of the researchers was to examine whether different iterations of ChatGPT could provide responses consistent with the rigor expected from specialist experts in geriatric healthcare.

The research team conducted comprehensive evaluations of ChatGPT’s responses to geriatric script concordance tests, a format designed to assess clinical reasoning in a standardized way. By juxtaposing AI-generated outputs with assessments made by healthcare experts, the study sought to discern the level of agreement and discrepancies between artificial intelligence and human judgment. This approach serves not only to validate the capabilities of the AI but also to highlight the potential limitations which must be acknowledged and addressed.

One of the fundamental observations made during the research was the variability inherent in the responses produced by different versions of ChatGPT. Each model demonstrated unique strengths and weaknesses across various scenarios tested, shedding light on the continuous development process required for such technologies. It became evident that while these models could occasionally produce expert-like responses, inconsistencies often arose, particularly in complex case scenarios where medical nuances are plentiful. This observation is crucial, as it underscores the importance of refining AI tools for specific medical applications, where a one-size-fits-all approach is unlikely to suffice.

Moreover, the researchers took a careful look at the context in which these AI models operate. By evaluating responses through a lens of expert opinion, the study aimed to articulate a more contextual understanding of how AI can be effectively integrated into clinical settings. Geriatrics demands a nuanced comprehension of patient history, socio-economic factors, and individual patient needs, aspects that require more than just regurgitated medical data. This research emphasizes that successful AI implementation hinges on the ability to account for these contextual factors effectively.

Equally noteworthy was the consideration of how AI can serve as a tool for augmenting rather than replacing human expertise. The notion that AI could supplement clinical judgment and enhance the decision-making process emerged prominently throughout the study. By leveraging the analytical capabilities of models like ChatGPT, healthcare professionals may be able to better prepare for consultations, providing deeper insights and richer dialogues with patients, particularly elderly ones who often encounter systemic barriers to quality care.

Furthermore, the ethical implications surrounding the use of AI in healthcare cannot be underestimated. The researchers acknowledged the potential risks associated with relying too heavily on AI-generated information which may misalign with the actual needs of geriatric patients. Turning a blind eye to the ethical facets of AI application could ultimately harm the very population that these technologies aim to assist. The study therefore advocates for an ongoing conversation that incorporates diverse perspectives across medicine, technology, and ethics to create a framework that safeguards patient welfare while embracing innovation.

Expertise in geriatric medicine is built on years of education, clinical experience, and emotional intelligence—traits that are challenging to replicate in artificial intelligence. The findings of this research accentuate the need for vigilance in the integration of such tools into practice, advocating for a model where AI operates in tandem with seasoned professionals. Technology should be a bridge, not a barrier, fostering enhanced communication between caregiver and patient while ensuring that human insight remains at the forefront.

As the healthcare landscape continues to evolve with the infusion of AI, the responsibilities of both technologists and healthcare professionals will be paramount. Collaboration across domains will not only push the boundaries of what is technologically possible but also ensure that patient-centered care remains robust during this transformation. Future research endeavors should continue to assess and refine these AI models, ensuring they align with the evolving standards of care while maintaining the humanity that underpins effective healthcare delivery.

The findings from this study are poised to inform not only academic discourse but also practical applications of AI in the geriatric field. As these conversations advance, it will be essential to cultivate an environment where ongoing feedback between AI developers and healthcare practitioners becomes standard practice. This can help create a more sophisticated understanding of how these technologies can truly add value in the complex domain of geriatric health.

In conclusion, the research led by Lilamand and colleagues brings to light the crucial dialogue surrounding the integration of AI in medical fields, particularly geriatrics. It highlights opportunities for improvement, while recognizing the challenges that must be addressed. The journey towards creating an AI framework that aligns with expert opinion is ongoing, but studies like this pave the way for a future where technology and medicine collaborate harmoniously. As the dialogue evolves, the hope is that these advancements will yield a healthcare model that is not only efficient but also empathetic and tailored to the individual needs of every patient.

With the rapid pace of AI development, there is a pressing need to stay abreast of technological advancements while ensuring that human welfare remains at the heart of all endeavors in healthcare. The outcomes of this research serve as a call to action for all stakeholders in the field to engage thoughtfully and collaboratively in shaping the future of geriatric medicine in the age of AI.

Subject of Research: Evaluation of ChatGPT alignment with expert opinions on geriatric script concordance tests.

Article Title: Evaluating how different versions of ChatGPT align with expert opinions on geriatric script concordance tests.

Article References: Lilamand, M., Decaix, T., Gourraud, PA. et al. Evaluating how different versions of ChatGPT align with expert opinions on geriatric script concordance tests. Eur Geriatr Med (2025). https://doi.org/10.1007/s41999-025-01334-5

Image Credits: AI Generated

DOI: https://doi.org/10.1007/s41999-025-01334-5

Keywords: geriatric medicine, artificial intelligence, ChatGPT, clinical reasoning, ethical implications, healthcare technology.