“Is Attention All We Need?” - A Systematic Literature Review of LLMs in Mental Healthcare (Preprint) DOI
Andreas Bucher, Stephan Egger, Inna Vashkite

et al.

Published: June 2, 2025

BACKGROUND Mental healthcare systems worldwide face critical challenges, including limited access, shortages of clinicians, and stigma-related barriers. In parallel, Large Language Models (LLMs) have emerged as powerful tools capable supporting therapeutic processes through natural language understanding generation. While prior research has explored their potential, a comprehensive review assessing how LLMs are integrated into mental healthcare, particularly beyond technical feasibility, is still lacking. OBJECTIVE This systematic literature investigates conceptualizes the application in by examining implementation, design characteristics, situational use across different touchpoints along patient journey. It introduces three-layer morphological framework to structure analyze applied, with goal informing METHODS Following methodology vom Brocke et al. [1], was conducted PubMed, IEEE Xplore, JMIR, ACM, AIS databases, yielding 807 studies. After multiple evaluation steps, 55 studies were included. These categorized analyzed based on journey, elements, underlying model characteristics. RESULTS Most assessed whereas only few examined impact outcomes. used primarily for classification text generation tasks, safety, hallucination risks, or reasoning capabilities. Design aspects such user roles, interaction modalities, interface elements often underexplored, despite significant influence experience. Furthermore, most applications focused single-user contexts, overlooking opportunities care environments, AI-blended therapy. The proposed framework, which consists L1: Situation-layer, L2: Interface-layer, L3: LLM-layer, highlights trade-offs unmet needs current research. CONCLUSIONS hold promise enhancing accessibility, personalization, efficiency healthcare. However, implementations overlook essential contextual factors that real-world adoption underscores “self-attention” mechanism, key component LLMs, alone not sufficient. Future must go feasibility explore models, experience, longitudinal treatment outcomes responsibly embed ecosystems.

Language: Английский

Can Artificial Intelligence be used to teach Psychiatry and Psychology?: A Scoping Review (Preprint) DOI Creative Commons

Julien Prégent,

V. V. CHUNG,

Inès El Adib

et al.

Published: March 30, 2025

BACKGROUND Artificial Intelligence (AI) is increasingly integrated into healthcare, including psychiatry and psychology. In educational contexts, AI offers new possibilities for enhancing clinical reasoning, personalizing content delivery, supporting professional development. Despite this emerging interest, a comprehensive understanding of how currently used in mental health education, the challenges associated with its adoption, remains limited. OBJECTIVE This scoping review aims to identify characterize current applications teaching learning It also seeks document reported facilitators barriers integration within contexts. METHODS A systematic search was conducted across six electronic databases (MEDLINE, PubMed, Embase, PsycINFO, EBM Reviews, Google Scholar) from inception October 2024. The followed PRISMA-ScR guidelines. Studies were included if they focused on or psychology, described use an tool, discussed at least one facilitator barrier education. Data extracted study characteristics, population, application, outcomes, facilitators, barriers. Study quality appraised using several design-appropriate tools. RESULTS From 6219 records, 10 studies met inclusion criteria. Eight categories identified: decision support, creation, therapeutic tools monitoring, administrative research assistance, natural language processing, program/policy development, student/applicant Key availability tools, positive learner attitudes, digital infrastructure, time-saving features. Barriers limited training, ethical concerns, lack literacy, algorithmic opacity, insufficient curricular integration. overall methodological moderate high. CONCLUSIONS being range functions training assessment support. While potential outcomes clear, successful requires addressing ethical, technical, pedagogical Future efforts should focus faculty institutional policies guide responsible effective use. underscores importance interdisciplinary collaboration ensure safe, equitable, meaningful adoption

Language: Английский

Citations

0
“Is Attention All We Need?” - A Systematic Literature Review of LLMs in Mental Healthcare (Preprint) DOI
Andreas Bucher, Stephan Egger, Inna Vashkite

et al.

Published: June 2, 2025

BACKGROUND Mental healthcare systems worldwide face critical challenges, including limited access, shortages of clinicians, and stigma-related barriers. In parallel, Large Language Models (LLMs) have emerged as powerful tools capable supporting therapeutic processes through natural language understanding generation. While prior research has explored their potential, a comprehensive review assessing how LLMs are integrated into mental healthcare, particularly beyond technical feasibility, is still lacking. OBJECTIVE This systematic literature investigates conceptualizes the application in by examining implementation, design characteristics, situational use across different touchpoints along patient journey. It introduces three-layer morphological framework to structure analyze applied, with goal informing METHODS Following methodology vom Brocke et al. [1], was conducted PubMed, IEEE Xplore, JMIR, ACM, AIS databases, yielding 807 studies. After multiple evaluation steps, 55 studies were included. These categorized analyzed based on journey, elements, underlying model characteristics. RESULTS Most assessed whereas only few examined impact outcomes. used primarily for classification text generation tasks, safety, hallucination risks, or reasoning capabilities. Design aspects such user roles, interaction modalities, interface elements often underexplored, despite significant influence experience. Furthermore, most applications focused single-user contexts, overlooking opportunities care environments, AI-blended therapy. The proposed framework, which consists L1: Situation-layer, L2: Interface-layer, L3: LLM-layer, highlights trade-offs unmet needs current research. CONCLUSIONS hold promise enhancing accessibility, personalization, efficiency healthcare. However, implementations overlook essential contextual factors that real-world adoption underscores “self-attention” mechanism, key component LLMs, alone not sufficient. Future must go feasibility explore models, experience, longitudinal treatment outcomes responsibly embed ecosystems.

Language: Английский

Citations

0