Systematic review details LLMs for depression detection, therapy support, and ethics

A new systematic review from Yisong Chen, Yifan Gao, Sijing Yu, Chuqing Zhao, and Yang Lu, published in the Journal of Industrial Integration and Management (2025), provides a comprehensive map of how large language models (LLMs) are being applied in mental health care. Drawing on interdisciplinary studies, the review integrates data from social media posts, electronic medical records, and multimodal inputs to assess current capabilities. The authors identify five core application areas: social media analysis for early screening, clinical conversational agents, therapy support tools, prompt engineering for domain adaptation, and emerging multimodal learning that fuses text, speech, and sensor data.

The review finds that LLMs already enable early detection of depression, suicide risk assessment, personalized therapy support, and generation of psychoeducational content. Techniques like prompt engineering and annotation strategies are highlighted as critical for improving interpretability and clinical relevance. However, the authors stress ongoing ethical, sociotechnical, and regulatory challenges, including bias, privacy, and accountability gaps. They advocate for frameworks that ensure safe, equitable, and accountable deployment of LLMs in real-world mental health settings. The paper doubles as a practical roadmap for clinicians and AI researchers, while warning that clinical adoption remains constrained by validation standards and governance infrastructure.