Using AI to Train Future Clinicians in Depression Assessment: Feasibility Study

Comprehensive Summary

This study evaluates how well ChatGPT can be used to train clinicians in identifying stages of depression and suicidal ideation. Three patient role scripts were developed with varying degrees of depression and suicidal ideation. These scripts were then put into ChatGPT-4 to aid medical students in clinical history-taking. 148 medical students with an average age of 22.71 years were randomly assigned to one of the three scripts. The chats were then evaluated for diagnostic accuracy, suicidal assessments, and how well the AI stuck to the script. In 90% of the answers, ChatGPT stuck to the script. On average, students identified the correct severity of depression 60% of the time and the phase of suicidality 67% of the time. Of note, the majority of students failed to sufficiently assess for suicidality. In the discussion, the authors noted that some of the script errors from the AI were due to deficiencies in the prompt.

Outcomes and Implications

Continuing to develop AI training tools in all aspects of medical training can help reduce the costs of training new professionals and can also better prepare future clinicians due to AI’s repeatability. Absolute accuracy is also not required from the AI as patients can provide false information in real life, thus mistakes made by the large language model can train students for these situations. Due the online setting, the AI could easily be adjusted for a variety of different settings and languages.