The Patient Journey Now Includes a Large Language Model
A Large Language Model (LLM) like GPT-4 is trained on all of the world’s knowledge it can access. Not just text in one domain like medicine, but law and engineering, voice and photos, German and Chinese. Cross-domain knowledge improves performance over narrow specialization. Studies show that patients prefer chatting with LLMs over physicians because the advice is the same but the LLM is less rushed and can appear more empathetic. Top experts are saying that patients benefit from chatting with an LLM before and after talking with a physician.
The patient journey will no longer be centered on the physician-patient dyad. It’s becoming a triangle with the patient and the physician consulting a LLM before, during and after an encounter.
A patient with a new symptom or concern can benefit from chatting with the LLM rather than Google.
The physician will use an LLM to listen in on the patient encounter and draft a note for the health record before they leave the exam room.
The patient may do the same with their smartphone and save the LLM description of the encounter in a health diary they keep.
If the physician prescribes a test or a treatment, the patient will discuss it with the LLM as they might with a second physician opinion.
As the LLM becomes more accessible, this cyclical patient journey will repeat more frequently and become more routine.
The context for the patient-LLM-physician conversation includes the health records of the patient as well as the technology to access the LLM. The hospital electronic health record (EHR) system is not guaranteed to offer a complete and accessible LLM context, even if it includes the physician, nurses, and allied health professionals. Patient records may be spread across multiple health systems and sophisticated wearables, and may include a health diary where the patient keeps track of symptoms, concerns, and photos as if they were messages to self. A physician could be running an LLM app on their phone instead of using the one provided by the hospital. Unlike the hospital EHR, a physician’s LLM with access to the hospital EHR would be accessible by voice and camera, able to cut through the maze of generic EHR screens, and readily customizable to the individual physician’s preferences.
The inclusion of a patient’s health history as context in a chat comes in two forms. One is formal, standardized, and regulated under the 21st Century Cures Act (2016) as an EHR. The other is a free-form patient diary that looks like a familiar thread of messages or emails. Either way, the context as well as the chat are private. Access to a regulated EHR requires a username and password. Access to a patient diary might be through an installed app or a separate username and password.
Physician access to the patient’s EHR and diary as context for an LLM chat requires patient authorization. Unlike the patient’s own access, privacy concerns typically require the patient to limit the scope of physician access to their diary and even their EHR. For example, a patient seeking a second opinion would not want to have previous diagnoses or treatment recommendations accessible to the second opinion physician.
From the physician’s perspective, a chat with the LLM is a risk. It needs to be documented in case of dispute and may need to be included in whatever EHR the hospital is mandating for care coordination purposes. EHR integration also drives billing and can play a role in reducing the more than $100 billion of insurance fraud in the US each year. In some cases, the physician, rather than the hospital, controls the EHR and the physician can choose the LLM, but the need for patient authorization and legal documentation remains.
The patient journey now includes chatting with an LLM in the context of regulated physician EHRs and patient-friendly health diaries. Fortunately, there are now standards that can meet the challenge of bridging between the patients, physicians, and providers of large language models. Trustee® by HIE of One provides the essential authorization mechanism for inclusion of LLMs into the patient journey.