Large language models show promise for clinical reasoning, yet psychiatric interviewing demands dynamic guidance within an evolving conversation, a facet that remains under‑explored. We introduce PsyCIDRA, a dual‑agent framework that links free‑form psychiatric interviewing with diagnostic reasoning for expert review.
The interviewer agent employs tools to keep working notes, load expert‑written interview skills, and retrieve ICD‑11 references to steer questioning. After the interview, the diagnostic reasoning agent receives the full transcript, reports diagnostic hypotheses together with supporting, conflicting, and missing evidence, and withholds a final hypothesis when none is sufficiently supported.
Evaluation used patient profiles generated by PsyCPG. Across four models on 53 evaluation cases, PsyCIDRA achieved higher diagnostic agreement than direct prompting. On 81 held‑out simulated cases, rank‑1 accuracy reached 60.5% versus 51.9% for direct prompting. In a blinded study with 101 human participants, PsyCIDRA agreed with psychologists on whether to propose a diagnostic hypothesis in 79.6% of cases, compared with 65.4% for direct prompting.
These findings support the potential of LLM agents to assist psychiatric assessment through free‑form dialogue. By jointly examining diagnostic reasoning, interview quality, and safety, the study contributes to a clearer understanding of the capabilities and limits of psychiatric interview agents.
Review