Self-triage chatbots were adopted more rapidly during the COVID-19 pandemic, when they were trained to evaluate COVID symptoms and connect high-risk users with details for telehealth providers or local testing centers. Buoy Health, GYANT and Conversa Health all launched COVID-19 symptom screeners to support health systems during the pandemic. Providence Health has a bot called Grace that helps with a variety of patient issues and directs them to the appropriate venue. The quoted accuracy rates are impressive.1 These virtual assistants are also helping with the management of chronic conditions, and there are a number of them for conditions like diabetes, hypertension and weight loss. There are also solutions for helping seniors to manage their health and daily lives.
LLM-powered chatbots could also help care teams to more quickly and easily access training and education resources. Frontline care team members frequently reference these materials as protocols and have to review them to find answers to their questions. LLM-trained chatbot-style search could offer a far more efficient method for users to quickly find the specific information they need to best navigate the technology and shift focus back to their patients.
The Mayo Clinic is exploring the use of search powered by generative AI to make finding information easier for team members. Traditional enterprise search, with queries producing a list of links based on pattern matching and significant manual investigation required to find the more relevant answers, can potentially be improved using generative AI. With generative AI, there’s an opportunity to leverage enterprise data from EHRs, practice management software and other systems to more effectively find answers to questions. This could include applying generative AI to conversational apps that can answer complex questions, produce accurate summaries that synthesize many sources, and help people get the information they need.
Virtual assistants can also be used to respond to patient messages to their providers. Using large language models, responses can be drafted and after the provider reviews and approves them, those messages can be sent to patients. This would save time for the clinicians but give them full oversight over the responses that are eventually sent to the patients. Another option is for patients to use a voice-based system to populate medical questionnaires before they visit their healthcare practitioner, saving them time and frustration and cutting down on wasted resources. Voice assistants could therefore be a useful way to educate patients and to engage them at the clinic and continue that engagement at home.
