One of the key issues that has always plagued the healthcare system has been the disparity of information between providers and patients. A provider knows where to send a patient if they have a cough with a mild fever (as well as where not to send them), but the patient doesn’t know if they should go to the ER at 2 am, call their doctor the next day or just wait it out. The answer depends on factors like their history, general health and co-morbidities. Similarly, many questions and issues that come up in the course of people’s daily lives could be answered easily if they had a physician nearby that they could consult.
This seems simple enough, but the reality is that few people have access to an on-demand physician. As such, we have people making decisions about how to use the healthcare system without knowledge or guidance. This leads to over or under-use of the system, depending on the situation. Also, when patients interact with the healthcare system, it’s often a manual and inefficient process. It involves waiting, being re-routed, not being able to reach anyone, being asked to fill out forms, and more.
It’s been a dream of many of us to have systems that can guide patients remotely and automatically on the basics of how serious their issues are and where they can seek care. This dream is now closer to reality thanks to AI-based virtual assistants. We discussed these in the chapter on decision support, but they can be considered as both a patient decision support system and a clinical workflow tool because with good guidance and by removing a lot of contacts for mundane and basic issues, we can improve the workflow of providers and free them up to focus on their patients. Also, by ensuring that only patients who need to show up to medical facilities are directed there, we can cut down on unnecessary visits and lighten the load of burned out providers.
Soon, medical bots trained using large language models and clinical data can start to impress everyone with their patient triage and support capabilities, just as ChatGPT is impressing everyone now. Given that these systems can be trained on large amounts of historical medical literature, they’ll be much better than current chatbots at handling patient questions or issues. If they have access to patient data, they can personalize their interactions with patients and provide responses that are more appropriate given the patients’ conditions and medications. Regardless of what system you use, including generative AI systems, the most effective virtual assistants need access to patient data to provide personalized advice.
Many of the solutions here use AI to create voice and text-based virtual assistants that can talk to patients, analyze the responses they provide and then offer up guidance. The most well-funded healthcare AI startup is called Babylon Health, and after using its bots to triage patients to the right setting based on their issues, it then allows them to talk to doctors within a couple of minutes and to receive medical advice from phone calls, text messages and video consultations. AI chatbot uses NLP to understand symptoms and to check that against the data stored in the patient’s medical history. Working within the UK’s NHS, Babylon Health trained its chatbot to act as an on-demand symptom checker to evaluate patient symptoms. A validation study of the technology found that the bot could safely triage 94% of test patients and that it could match expert decisions for 85% of them. New York-based K Health uses AI to navigate a large dataset of medical conditions and respond to patient inquiries about their medical conditions.
