ChatGPT-4 in Psychiatric Discharge Summaries: Toward a More Comprehensible and Emotionally Sensitive Clinical...

Area of Science:

• Psychiatry
• Medical Informatics
• Artificial Intelligence

Background:

• Psychiatric discharge summaries often use technical language, hindering patient comprehension and potentially causing distress.
• Increasing patient access to medical records necessitates patient-centered communication tools.
• Developing comprehensible and emotionally safe communication is crucial for psychiatric care continuity.

Purpose of the Study:

• To evaluate ChatGPT-4's effectiveness in rewriting psychiatric discharge summaries.
• Assessing diagnostic fidelity, linguistic clarity, emotional sensitivity, treatment comprehension, and readability of AI-generated summaries.
• This study is the first in South America to examine generative language models for this purpose.

Main Methods:

• A mixed-methods, observational cross-sectional design was employed.
• Twenty-five anonymized psychiatric cases were rewritten using ChatGPT-4.
• Psychiatrists evaluated summaries for diagnostic fidelity, clarity, emotional risk, and treatment understanding; readability was assessed using standard indices.

Main Results:

• ChatGPT-4 summaries showed significant improvements in readability and clarity, exceeding general comprehension thresholds.
• AI-generated summaries were particularly effective in explaining treatment.
• Minor diagnostic inaccuracies and instances of emotionally charged language were noted, requiring oversight.

Conclusions:

• ChatGPT-4 shows promise in making psychiatric discharge communication more accessible and emotionally appropriate.
• Professional oversight is essential to maintain clinical accuracy and contextual sensitivity.
• Future research should incorporate patient feedback and explore human-AI collaboration models.