Keywords: Simulated Patients, AI
Background:
There is a gap in virtual learning from clinical scenarios. Traditionally medical students have used “clinical cases”- presentations of symptoms with prompts as to clinical signs and diagnosis. However, these do not provide real-language interviewing of patients. Asking questions and processing the response to inform the most appropriate next question is a key part of clinical learning. It also helps students structure their interview, rather than have the structure provided as in a “clinical case” history
Methods:
To address this gap, I developed a prototype Virtual Patient (VP) web app. The VP can be accessed at https://patientinterview.pythonanywhere.com/. Students ask questions via text input, and the responses are presented in a chat format, displaying all questions and answers on the screen. The prototype is based on a case-of-the-week GP patient with depression, with 35 topic areas that users can explore
Results:
The key feature of the VP is its "training" mechanism, where the app learns the correct response to user input. If an appropriate response is not available a new response can be added to "train" the system. Users have been able to ask up to 20 consecutive questions with an appropriate response. The key is to have a strong narrative that draws students in. Additional cases are being developed and further analysis will be provided. The app uses free software.
Conclusion(s):
It is feasible to use free software to create a virtual patient. This provides a more interactive learning experience. Existing teaching resources such as OSCE simulated patient scripts could be added to such formats and in future validated VPs could be used for some assessments
Points for discussion:
Can students learn empathy remotely?
The current limitations of large language models (e.g ChatGPT) in patient simulation
How to convert existing teaching materials into virtual patients
#82
