It’s 2 am and you’re still awake. Your surgery is coming up soon, and you have questions. What if there are complications? How long will recovery take? And more.
More than 70 per cent of patients experience anxiety before surgery, studies show, and they seek reliable information on what lies ahead. But anecdotal experiences and online information can be confusing.
Expert-in-the-loop (EITL) chatbots offer a potential solution. They use large language models (LLM) and curated knowledge to respond to queries. They also have an additional step of verification and correction by human experts. The bot can selectively update its knowledge base using expert insights.
In the healthcare industry, EITL chatbots can provide patients on-demand access to doctor-verified medical information, thereby easing the load on physicians.
A team at Microsoft Research Lab India, led by Mohit Jain, partnered with doctors and patient coordinators at Sankara Eye Hospital, Bengaluru, to create CataractBot, an EITL chatbot that can answer queries about cataract surgery.
The team conducted several interviews to understand common questions and misconceptions about cataract surgery. The bot’s knowledge base drew from resources such as hospital procedures, treatment guidelines, and pre- and post-surgery guidelines. However, it did not include patient data to ensure privacy.
CataractBot was developed in nine months, after an iterative process of feedback and refinement.
Design decisions
A major challenge for an EITL chatbot is in catering to users from diverse linguistic, educational, and technical backgrounds.
For instance, cataract surgery patients are usually above 60 years old. The bot must, therefore, take into account their comfort level with technology usage. So it uses a chat window on WhatsApp. Patients and their attendants can use text, speech, or tap-based interactions with the bot in five available languages. Responses are in text and audio formats to ensure literacy levels and language proficiency are no bar.
EITL chatbots need expert input for verification. This can be challenging given time constraints. Intelligent design can mitigate this challenge. For instance, doctors on the CataractBot expert panel receive a one-click verification prompt to confirm the accuracy of an answer. Where needed, doctors can provide informal text feedback without editing the original response.
Streamlined workflows
EITL chatbots can streamline workflows for experts, who merely need to verify LLM-generated responses instead of answering common inquiries directly. Over time, the bot’s ability to provide accurate answers is expected to improve.
Dr Kaushik Murali, a paediatric ophthalmologist at Sankara Eye Hospital who was involved in developing CataractBot, says patients often have trouble remembering what the doctor says. They can unhesitatingly repeat their queries to the chatbot multiple times, freeing up doctors to focus on more complex issues during in-person visits.
Dr R Sowmya, another paediatric ophthalmologist, says, “It offers privacy to me and the patient… and encourages them to ask ‘silly’ questions that they may hesitate to ask otherwise.”