BACKGROUND Large language models are increasingly deployed in mental health applications, yet growing evidence suggests they encode algorithmic biases that influence clinical outputs. Because these models now mediate patient-facing decisions, such biases carry the potential for direct harm. Whether they systematically affect psychiatric diagnosis across demographic groups remains underexplored. OBJECTIVE To examine whether large language models (LLMs) exhibit implicit demographic biases when generating psychiatric diagnoses. METHODS We developed 1,152 synthetic clinical vignettes using a matched-pair design that manipulated gender, race/ethnicity, age, socioeconomic status, English proficiency, and urbanicity while holding clinical content constant. Vignettes were divided into control (unambiguous anorexia nervosa) and ambiguous conditions designed to permit differential diagnosis. Ten LLM configurations across five model families were tested. RESULTS Control vignettes produced near-unanimous anorexia nervosa diagnoses (M = 100.0%), while ambiguous vignettes elicited greater variability (M = 23.6%). Inter-model agreement was moderate for ambiguous vignettes (Fleiss' κ = 0.410, 95% CI: 0.397–0.422). Mixed-effects logistic regression with LLM as a random intercept revealed significant demographic biases: Black patients were over six times more likely to receive a major depressive disorder diagnosis than White patients with identical presentations (OR = 6.09, 95% CI: 5.13–7.24), Latine patients were over nine times more likely (OR = 9.57, 95% CI: 8.00–11.45), and Asian patients were nearly three times more likely to receive an anorexia nervosa diagnosis (OR = 2.88, 95% CI: 2.44–3.42). Female patients were less likely than males to be diagnosed with anorexia nervosa (OR = 0.43, 95% CI: 0.37–0.49). CONCLUSIONS These findings demonstrate that LLMs exhibit systematic demographic biases in psychiatric diagnosis even when clinical content is held constant, revealing measurable patterns that can inform improvements to training data, model architecture, and clinical deployment frameworks.

Background/Objectives: This study examines the academic discourse surrounding Critical Incident Stress Debriefing (CISD) and Critical Incident Stress Management (CISM) for first responders using Latent Dirichlet Allocation (LDA) topic modeling. It aims to uncover latent topical structures in the literature and critically evaluate assumptions to identify gaps and limitations. Methods: A corpus of 214 research article abstracts related to CISD/M was gathered from the Web of Science Core Collection. After preprocessing, we used Orange Data Mining software’s LDA tool to analyze the corpus. We tested models ranging from 2 to 10 topics. To guide interpretation and labeling, we evaluated them using log perplexity, topic coherence, and LDAvis visualizations. A four-topic model offered the best balance of detail and interpretability. Results: Four topics emerged: (1) Critical Incident Stress Management in medical and emergency settings, (2) psychological and group-based interventions for PTSD and trauma, (3) peer support and experiences of emergency and military personnel, and (4) mental health interventions for first responders. Key gaps included limited focus on cumulative trauma, insufficient longitudinal research, and variability in procedural adherence affecting outcomes. Conclusions: The findings highlight the need for CISD/M protocols to move beyond event-specific interventions and address cumulative stressors. Recommendations include incorporating holistic, proactive mental health strategies and conducting longitudinal studies to evaluate long-term effectiveness. These insights can help refine CISD/M approaches and enhance their impact on first responders working in high-stress environments.

This study investigates the criterion concordance of the Bolante Threat Assessment Guide (B-TAG), a structured professional judgment (SPJ) tool designed to assess the risk of targeted violence. We compared ratings from seven trained human raters against ratings from three large language model (LLM)-based raters, each simulating a distinct professional perspective relevant to threat assessment. Both human and AI raters evaluated 12 anonymized case summaries against a criterion standard established by a majority-rule panel of three expert reviewers with more than 20 years of threat assessment experience each. Concordance with the expert panel was assessed at the case level using Gwet's AC1, selected over Cohen's kappa and Krippendorff's alpha for its robustness to prevalencerelated paradoxes under skewed base rates. Across the 12 cases, mean concordance with the expert panel was similar for human raters (M = 0.647) and AI raters (M = 0.604), though human raters agreed with the panel slightly more often. AI raters identified the risk indicators the panel judged present about as often as human raters did, but flagged more indicators the panel judged absent, producing more false positives. These findings suggest that AI may serve as a supplementary reviewer but is not a substitute for trained threat assessors, and offer preliminary evidence regarding the potential for AI-based tools to approximate expert human judgment in structured threat assessment coding, while underscoring specific limitations in the current evidence base.

Background: Large language models (LLMs) are increasingly deployed in mental health applications, yet growing evidence suggests they encode algorithmic biases that influence clinical outputs. Because these models now mediate patient-facing decisions, such biases carry the potential for direct harm. Whether they systematically affect psychiatric diagnosis across demographic groups remains underexplored. Objective: This study aims to examine whether LLMs exhibit implicit demographic biases when generating psychiatric diagnoses. Methods: We developed 1152 synthetic clinical vignettes using a matched-pair design that manipulated gender, race and ethnicity, age, socioeconomic status, English proficiency, and urbanicity while holding clinical content constant. Vignettes were divided into control (unambiguous anorexia nervosa [AN]) and ambiguous conditions designed to permit differential diagnosis. Ten LLM configurations across 5 model families were tested. Results: Control vignettes produced near-unanimous AN diagnoses (mean 100%, SD 0.1%), while ambiguous vignettes elicited greater variability (mean 23.6%, SD 10.1%). Intermodel agreement was moderate for ambiguous vignettes (Fleiss κ=0.410, 95% CI 0.397‐0.422). Mixed-effects logistic regression with LLM as a random intercept revealed significant demographic biases: Black patients were over 6 times more likely to receive a major depressive disorder (MDD) diagnosis than White patients with identical presentations (odds ratio [OR] 6.09, 95% CI 5.13‐7.24), Latine patients were over 9 times more likely (OR 9.57, 95% CI 8.00‐11.45), and Asian patients were nearly 3 times more likely to receive an AN diagnosis (OR 2.88, 95% CI 2.44‐3.42). Female patients were less likely than males to be diagnosed with AN (OR 0.43, 95% CI 0.37‐0.49). Conclusions: These findings demonstrate that LLMs exhibit systematic demographic biases in psychiatric diagnosis even when clinical content is held constant, revealing measurable patterns that can inform improvements to training data, model architecture, and clinical deployment frameworks.

Background/Objectives: This study examines the academic discourse surrounding Critical Incident Stress Debriefing (CISD) and Critical Incident Stress Management (CISM) for first responders using Latent Dirichlet Allocation (LDA) topic modeling. Its aim is to uncover latent topical structures within the literature and critically evaluate underlying assumptions to identify gaps and limitations. Method: A corpus of 214 research article abstracts related to CISD/M was gathered from the Web of Science Core Collection. After preprocessing, we used Orange Data Mining software’s LDA tool to analyze the corpus. Models ranging from 2 to 10 topics were tested for log perplexity and topic coherence, with LDAvis visualizations guiding interpretation and labeling. A four-topic model offered the best balance of detail and interpretability. Results: Four topics emerged: (1) Critical Incident Stress Management in medical and emergency settings, (2) Psychological and group-based interventions for PTSD and trauma, (3) Peer support and experiences of emergency and military personnel, and (4) Mental health interventions for first responders. Key gaps included limited focus on cumulative trauma, insufficient longitudinal research, and variability in procedural adherence affecting outcomes. Conclusions: The findings highlight the need for CISD/M protocols to move beyond event-specific interventions and address cumulative stressors. Recommendations include incorporating holistic, proactive mental health strategies and conducting longitudinal studies to evaluate long-term effectiveness. These insights can help refine CISD/M approaches and enhance their impact on first responders working in high-stress environments.