A domain-adapted large language model to support clinicians in psychiatric clinical practice

2026-04-27 · Source: Nature Machine Intelligence · Field: Health & Wellbeing — Mental Health & Psychological Support, Clinical Care & Medical Practice, Medical Devices & Health Technology · Depth: Expert, long

Summary

PsychFound is a 7B-parameter, domain-adapted large language model (LLM) designed to support psychiatric clinical practice, developed by Shanghai Jiao Tong University and Beijing Anding Hospital. It was trained using a three-phase framework incorporating expert-curated psychiatric corpora and 64,588 Chinese real-world electronic health records. PsychFound integrates professional psychiatric knowledge and clinical reasoning across diagnosis, treatment planning, and longitudinal management in Chinese clinical settings. Retrospective evaluations across three knowledge assessments and five clinical task benchmarks showed PsychFound outperformed 22 other LLMs. A prospective study demonstrated that resident psychiatrists assisted by PsychFound achieved higher consultation quality, improved diagnostic accuracy, more appropriate medication selection, and reduced documentation time (all P < 0.01). A reader study with 60 psychiatrists found PsychFound's clinical reasoning matched that of attending psychiatrists.

Key takeaway

For AI Product Managers developing clinical decision support tools, PsychFound demonstrates that deep domain adaptation and integration into existing workflows are critical for real-world utility. You should prioritize developing LLMs with expert-curated data and validate their performance through both retrospective benchmarks and prospective clinical studies to ensure they meet or exceed human expert capabilities in specific tasks like diagnosis and treatment planning.

Key insights

Domain-adapted LLMs can significantly enhance psychiatric clinical workflows, improving accuracy and efficiency.

Principles

• Clinical workflow alignment is crucial for LLM adoption.
• Expert-curated data improves domain adaptation.
• LLM assistance can match expert human reasoning.

Method

PsychFound was developed via a three-phase framework, utilizing expert-curated psychiatric corpora and 64,588 Chinese real-world EHRs for domain adaptation and clinical reasoning integration.

In practice

• Use PsychFound for psychiatric diagnosis support.
• Apply PsychFound for treatment planning assistance.
• Integrate PsychFound to reduce documentation burden.

Topics

• PsychFound
• Large Language Models
• Psychiatric Clinical Practice
• Electronic Health Records
• Clinical Decision Support

Code references

• wrx33/PsychFound
• wrx33/PsychFound
• wrx33/PsychBench

Best for: Executive, AI Product Manager, AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• Towards AI-augmented decision making in psychiatry — AI-augmented decision-making can standardize care for heterogeneous psychiatric disorders.
• Active Evidence-Seeking and Diagnostic Reasoning in Large Language Models for Clinical Decision Support — LLMs exhibit a substantial performance gap in active, multi-turn clinical diagnosis compared to static, full-context evaluations.
• Ten Headache Specialists versus Artificial Intelligence for Clinical Literature Summarization: A Critical Evaluation and Comparison — Expert-written clinical summaries remain preferred over RAG-LLM outputs, despite AI's growing proficiency and occasional indistinguishability.
• Reading between the Lines: Leveraging Large Language Models for Global Dementia and Depression Assessment from Clinical Interviews — LLMs can effectively assess neuropsychiatric disorder severity from speech, enabling automated screening.
• Can AI Chatbots Reason Like Doctors? — Large language models demonstrate potential in clinical reasoning, but require careful integration and standardized evaluation in medical workflows.
• An Agentic LLM-Based Framework for Population-Scale Mental Health Screening — An agentic LLM framework enhances mental health screening by ensuring robust, adaptable, and trustworthy AI pipelines.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Nature Machine Intelligence.