The Role of Personal AI Assistants in the Patient Journey

The most advanced technology tools should be controlled by patients and physicians, not hospital business.

Generative artificial intelligence, often called Large Language Models (LLM), represent a qualitative departure from previous automation and information technologies. The combination of multi-modal (text, speech, image, code) integration and vastly democratized access to information promises solutions to problems previously thought unsolvable. LLMs, however, are as impersonal and generic as medical science itself and their application needs to be contextualized by inclusion of individual health records and the supervision and accountability of a clinician. This process of LLM contextualization can itself be based on local, private, and personalized AI assistants. Adding context to a LLM consultation is sometimes called Retrieval Augmented Generation or RAG. 

An AI Assistant adds key information in the patient’s health record to the question being asked of the LLM by a physician or the patient. The assistant can present relevant parts of the health record automatically and withhold others to protect privacy. An assistant capable of machine learning can adapt to the concerns and learn from the experiences of the patient or the physician. The assistant can make more efficient use of costly LLM resources.

Better patient care is a combination of more education and more attention to details. For physicians, who are legally accountable for medical errors and are compensated based on their reputation, the AI assistant is an essential partner in harnessing the same undifferentiated LLM. Electronic Health Records (EHR) and related information technologies are imposed on physicians and often contribute to physician burnout. Importantly, if the AI Assistant is also controlled by the employer (such as a hospital or hedge fund group), the individual physician is commoditized and cannot take their assistant’s experience with them to another job. Linking the AI Assistant to the physician encourages investment in the assistant technology and elevates the level of patient care across the board.

Portability of the AI Assistant empowers the patient as well as the physician. There are thousands of hospitals and medical practices, each with their own EHRs. There are also thousands of medical devices, wearables, apps and other sources of patient information that an AI Assistant could use. Some of this personal data, especially social determinants of health, is too sensitive to trust to hospital or government databases. Any practical solution to connecting the AI Assistant to these thousands of private data sources requires standards for access authorization and communication.   

For over a decade, the founders of HIE of One have led and promoted standards efforts designed to give patients and their physicians authorized access to health records held by hospitals and medical devices. These standards, combined with Patient Right of Access regulations, have come a long way. But until now, the benefit for patients and physicians to take control of the patient medical record has not been worth the trouble. Leveraging LLMs and personal AI Assistants changes the value equation for both principals in the patient journey. 

Trustee® by HIE of One will soon include a demonstration of a personal AI Assistant with connections to both the OpenAI GPT-4o LLM as well as to the open-source NOSH patient-centered health record, using the new IETF GNAP standard. If the physician is to be responsible and totally in control of their AI, then it should not be subject to censorship by the hospital or the patient. An IETF Personal Digital Agent Protocol mail list, pdap@ietf.org discusses other standards beyond GNAP to enable this highly efficient and privacy-preserving option.