Generating synthetic patient vignettes from real medical texts for the teaching of clinical reasoning

Experience with simulated clinical cases is a relevant component in the development of clinical reasoning (CR). Generating and vetting cases that are locally relevant is, however, a complex and time-consuming process.

We propose the use of generative artificial intelligence (AI) to create synthetic patients (SyP), in the form of narratives, based on real-world data describing patients' symptoms. We pilot tested this solution with self-reported questionnaires of patients with chest discomfort using a chatbot.

Automatically creating vetted clinical narratives that are locally relevant would amplify the teaching of CR, allowing for a larger exposure of students to clinical cases. We synthesized SyP from narrative data that retained the initial diagnostic hypothesis of the original patients as defined by a general practitioner. Our results indicate that a more efficient process of generating cases for educational purposes mediated by AI is feasible.

We plan to fine-tune the process to improve the narratives while preserving confidentiality. In the future, the process could be used on a large scale for the development of diagnostic abilities and communication skills.