Large Language Models for Accurate Mental Health Screening: Identifying Clinical Phenotypes in ED Healthcare Workers

Large Language Models for Accurate Mental Health Screening: Identifying Clinical Phenotypes in ED Healthcare Workers | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Large Language Models for Accurate Mental Health Screening: Identifying Clinical Phenotypes in ED Healthcare Workers Katharina Schultebraucks, Victoria Mueller, Sarah Birnbaum, Charlotte Hilberdink, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7087333/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The emergency department (ED) is crucial to the healthcare system. ED healthcare workers (HCWs) play a vital role in providing essential healthcare under demanding conditions. Traditional mental health assessment tools often rely on clinical interviews and psychometric assessments, which can be time-consuming, costly, and subject to biases. We aimed to identify heterogenous clinical profiles using a person-centered clustering approach based on burnout, depression, and PTSD symptomatology. We additionally investigated whether these phenotypes could be predicted by narratives about work-related future expectations using a Large Language Model (LLM). Based on n = 199 ED HCWs from an ongoing NIH-funded study (R01HL156134), k -means clustering revealed a High- and Low-Symptom Phenotype, significantly differentiating severity levels. Zero-shot LLM prompt engineering accurately predicted those clinical phenotypes from work-related narratives (accuracy = 70.9%; F1-score = 71.8%; sensitivity = 77.1%) based on key domain-specific indicators identified in the LLM’s reasoning. Our approach leverages LLMs based on unstructured narratives, offering an objective, time-efficient alternative that enhances early risk stratification and fosters a stigma-free environment for mental health assessment in high-stress healthcare settings, revealing subtle but meaningful variations in symptomatology. Future research should incorporate indicators such as risk factors and symptom dynamics to refine this tool for scalable, person-centered mental health monitoring across diverse healthcare settings. Health sciences/Diseases/Psychiatric disorders Biological sciences/Psychology/Human behaviour Health sciences/Biomarkers/Predictive markers Clinical Phenotypes Healthcare Workers (HCWs) Large Language Model (LLM) Emergency Department (ED) Burnout Figures Figure 1 Figure 2 Figure 3 1. Introduction The emergency department (ED) is the backbone of the healthcare system, with over 139.8 million visits in the US annually 1 . Frontline ED clinicians and staff serve a crucial societal role providing a vital health safety net. To provide the highest quality of care, ED healthcare workers (HCWs) must maintain a constant state of vigilance, focus, and resilience to operate effectively in a demanding and unpredictable environment on a daily basis 2 – 6 . The stressful environment of the ED poses a serious, ongoing risk to the psychological and physical well-being of HCWs. Alongside demanding work conditions in the ED, such as fast-paced work with high psychological demands 7 , 8 , excessive workload 7 , patient violence 9 , 10 , and shift work 11 , ED HCWs are regularly exposed to stressful and traumatic situations 2 , 3 . Accumulation of several long-term stressors can be extremely burdensome and leave many HCWs dissatisfied, leading to high attrition rates 10 , 12 , 13 . Studies have reported an increased prevalence of mental health issues, such as increased burnout rates in 26–82% of ED HCWs 8 , 14 , and around 20% of ED personnel met criteria for PTSD symptoms 15 – 19 . Up to 77.1% have depression 15 and HCWs in demanding roles, including ED providers, are at greater risk for suicide and suicidal ideation 16 , 20 – 22 . Burnout in HCWs can also be linked to physical consequences such as increased fatigue 11 , 23 , cardiovascular issues 24 , 25 , and disrupted sleeping patterns that can significantly affect emotional processing and other difficulties in cognitive domains, such as decision-making 23 , attention, and memory 26 – 28 . These consequences for ED HCWs can directly affect the delivery of high-quality healthcare. Poorer well-being can lead to increased medical errors 29 and negatively affect perceptions of patient safety 30 , ultimately compromising the quality of care 31 , 32 . Barriers to seeking psychological support, such as feelings of stigma, shame, fear of perceptions of weakness, and concerns about breaches of confidentiality that could affect career and licensing, hinder HCWs in general from seeking help 33 – 36 . The current traditional tools to assess well-being of HCWs mostly rely on clinical interviews and psychometric assessments that evaluate simplified dichotomous categorical diagnoses without comprehensive consideration of multiple symptoms and potential comorbidities. They can be cost-intensive and time-consuming for large-scale use. They are often burdensome and have low response rates 37 , potentially excluding those who experience more stress and burnout as they may be less likely to participate 38 . Furthermore, self-reported symptoms are susceptible to recall bias 39 and social desirability 40 . Qualitative methods may effectively address these biases by capturing valuable individual and complex experiences and emotions through techniques such as semi-structured interviews, providing a holistic, in-depth, and contextual view of these experiences, but are labor- and time-intensive. Studies have shown that mental health symptomatology may be underreported in many high-performing professions, such as healthcare workers 36 , as was found in police officers 41 and military personnel 42 . Therefore, it is crucial to shift away from traditional psychometric assessments that focus on classical diagnoses. Instead, there is an urgent need to adopt more low-burden, objective, and person-centered measures to identify clinical phenotypes that account for heterogeneous, interdependent symptoms and their severity. Going forward, this will facilitate the early detection of stress-related symptoms and the targeted implementation of interventions to strengthen the mental well-being and resilience of HCWs. This study aims to identify relevant profiles of clinical phenotypes based on burnout, depression, and PTSD symptomology in ED HCWs by analyzing narratives of work-related stressful situations using large language models (LLMs). We hypothesize that LLMs can effectively classify these heterogeneous clinical phenotypes, thereby providing a more nuanced understanding of mental health risks. This approach can potentially replace traditional assessments, offering a scalable, time-efficient, and unbiased alternative that healthcare providers are more likely to engage with. 2. Methods An overview of the methodological framework employed in this study, including participant assessment, clinical phenotyping, the LLM approach, and evaluation, is illustrated in Fig. 1 . 2.1 Participants Participants were enrolled as part of an ongoing NIH-funded observational longitudinal study (Early Signs, R01HL156134) with the primary aim of investigating burnout and well-being in N = 350 ED and Trauma Unit HCWs (clinicians and staff) in the greater New York City (NYC) area. Participants were eligible if they were ≥18 years old, fluent in English, had direct patient contact, and typically worked full-time clinical hours (including those who worked temporarily part-time due to burnout or temporary leave). Medical students and temporary HCWs were excluded. Participants were recruited through online advertising (Facebook, iConnect (research trial database)) and recruitment visits by research staff to EDs in the NYC area. All participants signed the informed consent and were informed that participation was voluntary and would not affect their employment status or performance reviews. This study was approved by the Institutional Review Board of NYU Grossman School of Medicine. For this sub-study, 259 assessments (199 assessments at baseline; 66 assessments at 6-month follow-up) of N = 199 participants, who were enrolled from April 2022 through August 2024, were included. 2.2 Procedures Data collected during the baseline and 6-month follow-up assessments were included in this sub-study. Participants provided demographic and work-related information, and completed self-report questionnaires to measure current psychological symptom levels for burnout, depression, and PTSD at home prior to or during the assessment. For weekly alcohol intake and daily smoking behavior, the Smoking Alcohol Intake Questionnaire was used. During the assessment, a semi-structured interview was conducted to capture emotion-driven narratives on stressful work experiences through a HIPAA-compliant video platform (WEBEX). 2.3 Measures 2.3.1 Psychological Symptomatology Burnout symptoms were assessed using the 9 items of the Maslach Burnout Inventory (MBI-9; range 0–6, Maslach, et al. 43 across the domains Personal Achievement (PA; items 1, 6, 9), Emotional Exhaustion (EE; items 3, 4, 7), and Depersonalization (DEP; items 2, 5, 8). Depressive symptoms during the past 2 weeks were assessed using the 8-item Patient Health Questionnaire (PHQ-8; range 0-3 44 ). PTSD symptomatology in the past month, with regards to COVID-19 or another traumatic experience from working in the ED, was measured using the PTSD Checklist for DSM-5 (PCL-5; 20 items, range 0-4 45 ), for which subdomain scores for PTSD’s DSM-5 diagnostic symptom clusters were calculated; Cluster B Intrusions, Cluster C Avoidance, Cluster D Negative Cognitions and Mood, and Cluster E Arousal and Reactivity Alterations. Higher scores on all items and subdomains indicated higher self-reported symptom levels, except for PA, with lower scores indicating higher symptom levels. 2.3.2 Work-related Narratives During the semi-structured interview, HCWs were asked the following emotion-provoking question verbatim to capture an unstructured narrative on work-related future expectations: “What are your expectations about your role and your work as an ED clinician in the future?”. Participants had a predetermined time of three minutes to answer the open-ended question. 2.4 Statistical Approach 2.4.1 Clustering analysis To identify heterogeneous profiles of clinical phenotypes, we performed clustering analysis using the k-Means algorithm 46 , 47 in Python (v3.9, ‘scikit-learn’ library v1.5.0). Item scores of the MBI-9 and PHQ-8, and PTSD symptom subdomain scores for PCL-5 Cluster B, C, D, and E were included in the clustering analysis. Data were preprocessed including checking for missingness (≥30% missing values, applied when calculating subdomain scores) and near-zero variance (threshold = 0.1) for each feature. We excluded n = 1 participant due to missing questionnaire data and removed PHQ-8 item 8 from the analysis due to near-zero variance. Multivariate outliers of 9 assessments (3.4%; 8 at baseline and 1 at 6-month follow-up) from n = 7 participants were identified and removed based on the Mahalanobis Distance (threshold = 0.001), resulting in a total of N = 254 assessments (190 at baseline; 64 at 6-month follow-up). In total, N = 191 participants were included in the final clustering analysis. For clustering, the k -means algorithm partitions data into k clusters by minimizing the sum of squared distances between data points and their assigned cluster centroids. The variables were standardized by centering them around the mean and scaling to unit variance. Model evaluation was based on pre-identified indicators of the best-fitting model by testing the fit of a 1-profile model and subsequently increasing the profile number by 1 until the addition of a profile was no longer optimal or improved. Indicators were the Silhouette score (higher value: better fit), Davies-Bouldin score (lower value: better fit), Calinski-Harabasz score (higher value: better fit), Dunn Index (higher value: better fit) and the Elbow Method, which involved plotting the within-cluster sum of squares (WCSS) against the number of clusters to identify the ‘elbow point’, where the slope rate of decrease is reduced and is sharpest (see Supplementary Material S1 on best-fitting model evaluation). 2.4.2 LLM zero-shot prompts To determine whether LLMs can effectively predict and identify clinical phenotypes, we employed a pre-trained LLM with zero-shot prompt engineering. In a zero-shot setting, the model generates predictions without having seen any labeled examples related to the specific task beforehand. Instead, it relies entirely on its pre-existing knowledge from training and the information provided in the prompt. The LLM analyzed narratives about work-related future expectations from the participants whose video recordings of the interview were available to distinguish and classify clinical phenotypes. In total, 126 narratives were utilized from a subgroup of N = 90 participants with available video recordings. Using a three-step approach, narratives were transcribed by 1) manually segmenting the video into separate clips per question, 2) applying speaker diarization (partitioning speech recordings according to the identity of each speaker) using PyAnnote (v3.1.1), and 3) transcribing audio using Whisper (v20231117) and checking manually for quality. To evaluate the capability of pre-trained LLM in predicting clinical phenotypes, we developed zero-shot prompt templates composed of four components: $$\:\text{P}\text{r}\text{o}\text{m}\text{p}\text{t}=\:Task\:+\:Clinical\:Phenotypes\:+\:Further\:Information\:+\:Narrative$$ The ‘ Task ’ component specifies the primary assignment for the LLM to perform. The ‘ Clinical Phenotypes’ component describes the characteristics of the distinct clinical phenotypes derived from the cluster analysis. The ‘ Additional Information’ component provides supplementary context (e.g., details about symptom comorbidities). Lastly, the ‘ Narrative’ component contains participants' narratives about their work-related future expectations. Example prompts are provided in Supplementary Material S2. We locally employed Meta’s open-source, HIPAA-compliant, Llama3-8B-Instruct model 48 as LMM on an NVIDIA GPU using the “transformers” Python package (v4.44.0). Additional implementation details are provided in Supplementary Material S3. It is important to note that we disabled Llama-3 sampling in order for the model to consistently generate the same response for a given input. Instead of randomly sampling from the model's probability distribution during generation, the model always selects the highest probability token at each step, resulting in deterministic predictions. Predictive model performance was measured by precision, recall, F1-score and ROC-AUC, which are established performance measures used for machine-learning classification tasks 49 . 3. Results 3.1 Clinical phenotypes A descriptive overview of demographic information, psychological symptomatology, and behavioral characteristics is presented in Table 1 . The best-fitting clustering model consisted of two clinical phenotypes (Fig. 2 ). Their descriptive profile labels were based on means of total and subdomain scores (Table 2 ). We interpreted the profiles as follows: “High-Symptom Phenotype” (39.0%, 99 assessments from n = 79 distinct participants) with higher symptom levels on items for burnout with regards to emotional exhaustion and depersonalization, as well as depressive symptoms and PTSD symptoms across all clusters, re-experiencing, avoidance, negative alterations in cognition and mood, and hyperarousal; and “Low-Symptom Phenotype” (61.0%, 155 assessments from n = 120 participants) with lower symptom levels on items for burnout with regards to emotional exhaustion and depersonalization, as well as depressive symptoms and PTSD symptom clusters for re-experiencing, avoidance, negative alterations in cognition and mood, and hyperarousal. No significant difference was observed in burnout-related personal accomplishment between the two clinical phenotypes. 3.2 LLM zero-shot prompting The mean word count of the narratives was 360.90 (SD = 124.31, for a descriptive overview of both profiles in the LLM participant sample, see Supplementary Material S4). The Llama3-8B-Instruct model predicted both clinical phenotype profiles using work-related narratives with a sensitivity of 77.1%, accuracy of 70.9%, and F1-score of 71.8% (Fig. 3 ). Based on the zero-shot prompt, the LLM identified several key domain-specific indicators extracted from the narratives that served as reasoning for classifying participants into clinical phenotypes (Table 3 ). Classification into the High-Symptom Phenotype was most frequently predicted based on the following five domain-specific indicators: emotional exhaustion (68.1%), feeling overwhelmed (62.3%), emotional numbness and avoidance (33.3%), hopelessness/helplessness (30.4%), and stress/anxiety (27.5%); and for the Low-Symptom Phenotype: positive outlook (94.6%), clear future vision (65.5%), sense of control (58.2%), positive language (56.4%), and coping mechanisms (27.3%). See Supplementary Material S5 for an example of such domain-specific indicators. Table 3 Key domain-specific indicators extracted from the narratives for reasoning of the LMM to predict the High-Symptom or Low-Symptom Phenotype. High-Symptom Phenotype Low-Symptom Phenotype Categories n (%) Categories n (%) Emotional exhaustion 48 (67.6%) Positive outlook 52 (94.6%) Feeling overwhelmed 45 (63.4%) Clear future vision 36 (65.5%) Emotional numbing and avoidance 25 (35.2%) Sense of control 32 (58.2%) Hopelessness/helpless 21 (29.6%) Positive language 31 (56.4%) Stress/anxious 19 (26.8%) Coping mechanisms 15 (27.3%) Cognitive issues 13 (18.3%) Self-awareness 12 (21.8%) Frustration 13 (18.3%) Work-life balance 11 (20.0%) Future uncertainty 11 (15.5%) Acknowledgement of the problem 10 (14.1%) Fatigue/drained 9 (12.7%) Disappointment 8 (11.3%) Feeling guilt or self-doubt 8 (11.3%) Seeking a change 7 (9.9%) Dissatisfaction 7 (9.9%) Physical exhaustion 5 (7.0%) Distress 4 (5.6%) Impact personal life or relationships 3 (4.2%) Lack of sustainability 2 (2.8%) Sleep problems 1 (1.4%) n (%): number (percentage) of narratives in which the key domain-specific indicator was identified as predictive of either the High- or Low-Symptom Phenotype. The percentage was calculated within the two phenotypes separately. 4. Discussion This study is the first to utilize LLMs to predict heterogeneous clinical phenotypes in ED HCWs. The clinical phenotypes were identified using a person-centered clustering approach based on self-reported burnout, depression, and PTSD symptomatology. These mental health issues are highly prevalent in this profession, given the high-stress and trauma-intensive nature of the ED environment. Using a zero-shot LLM approach with a single narrative response to an open-ended question about work-related future expectations, we accurately classified ED HCWs into these two phenotypes. In addition, we identified several key domain-specific indicators used by the LLM for clinical phenotype classification. Using a person-centered clustering approach, we identified a High- and Low-Symptomatology Phenotype in ED HCWs, capturing the entire range, severity, and interdependence of symptoms rather than focusing on dichotomous categorical diagnoses. These two heterogeneous phenotypes, symptomatic and resilient, are consistent with previous studies 50 – 54 . The prevalence of resilient individuals was higher, aligning with other studies that report high resilience in HCWs despite their high workloads and high-performing nature 50 , 53 . Yet, a significant portion still reported mental issues, which is also aligned with previous findings 50 , 52 . Our results extend previous findings 50 – 54 by including individual symptom levels across various mental health domains, revealing that our High-Symptom Phenotype encompasses complex comorbidities across burnout, depression, and PTSD, which could influence and exacerbate symptom development, outcomes, and diagnosis 15 . Pei, et al. 53 conducted one of a few studies in COVID-19-affected Chinese HCWs that also included multiple symptom types (depression, anxiety, insomnia, and resilience) and identified four distinct profiles: low, mild, moderate, and high severity. Importantly, they did not include measures of occupational burnout in their phenotyping approach, neglecting a top-of-mind concern that may lead to attrition from Emergency Medicine and even poorer patient care 12 , 29 . Leiter and Maslach 50 identified five profiles in HCWs based on MBI subdomain scores, which were later replicated in an online survey study of Canadian nurses 51 . The profiles include: burnout (high symptomatology across Emotional Exhaustion (EE), Depersonalization (DEP) and Personal Accomplishment (PA)); engagement (low symptomatology across EE, DEP and PA); and three intermediate profiles with varying patterns of low, medium, and high symptomatology across subdomains. In our study, we found two phenotypes that differed significantly in EE and DEP severity but not in PA. The consistent PA levels across phenotypes suggest that ED HCWs maintain work-related self-efficacy despite their exposure to trauma and demanding work conditions, including high work pace and shift work, that may lead to experiencing burnout in other domains. This seems to align with Önder and Basim 52 , who highlighted that PA in nurses had a non-linear relationship with DEP and EE and may exist independently of the other domains. The role of PA remains poorly understood and should be further investigated 50 . Regarding comorbid symptoms of burnout, mild depressive symptoms were observed in the High-Symptom Phenotype, and roughly one-third of HCWs in this phenotype reported PTSD symptom levels above the probable diagnostic threshold. These high severity levels may reflect the mental health risks of vicarious trauma exposure in the ED, and possibly the impact of working during the COVID-19 pandemic. For example, two latent profiles of post-traumatic stress symptoms (High- and Low-PTSS) were identified from an online survey of Brazilian HCWs affected by COVID-19 54 , underscoring the necessity to include PTSD symptomatology in analyses of HCW well-being. Using zero-shot prompting in an LLM, we accurately predicted these complex phenotypes from a single unstructured narrative about work-related future expectations. Our results show promising potential in predicting mental health outcomes using zero-shot prompting and are comparable to other studies. For instance, Ohse, et al. 55 used GPT-4 to predict self-reported PHQ-9 scores from 84 transcribed clinical interviews in a general population with an F1-score of 76.0%. In addition, Danner, et al. 56 aimed to identify risk for PTSD and depression using GPT-3.5 on semi-structured clinical interviews from N = 193 veterans and general population individuals, predicting probable depression with an F1-score of 78.0%. Remarkably, our study uniquely classified mental health symptomatology based on a three-minute free-speech text about work-related future expectations rather than structured or semi-structured clinical interviews. We also predicted complex, heterogeneous symptomatology with reliable accuracy, adding complexity to our model. In a recent study, Lho et al. 57 applied multiple LLMs and embedding-based models to analyze self-concept narratives derived from sentence completion tests of psychiatric patients, achieving ROC-AUCs of 0.72–0.75 for depression and 0.72–0.73 for suicide risk using zero-shot LLMs. While this demonstrates the effectiveness of LLMs on clinically enriched data from individuals already engaged in psychiatric evaluation, our study extends this potential into a more subtle and challenging context. Specifically, we focused on high-functioning ED professionals who are trained to suppress or mask emotional distress 58 . The ability to identify latent psychopathology from unstructured narratives in response to one open-ended question highlights the discriminative capacity of LLMs in real-world screening applications where traditional assessments may miss early or hidden symptoms. Previous studies used unstructured text data, like social media posts and SMS from open-source forums, both specific or non-specific to mental health topics, to identify stress, depression and/or suicidality risks with zero-shot prompts in LLMs (e.g., ChatGPT-3.5 and Llama2.0), achieving lower to comparable accuracies (42.9%-72.4%) 59,60 . However, the lack of clinically validated measures to assess mental health in these datasets may limit their diagnostic accuracy and applicability in healthcare. A study in perinatal women achieved a low F1-score of 38.0% in predicting PTSD from free-speech, highlighting the potential challenges of using free-speech data to predict complex symptomatology, as well as the non-specialized nature of LLMs for clinical applications 61 . Nevertheless, our results demonstrate the potential of LLMs to assess heterogeneous clinical phenotypes accurately. Using few-shot prompts has shown minor improvements in LLM performance for predicting mental health, as seen in studies by Xu, et al. 60 and Yang, et al. 59 . Current LLMs still face challenges with processing lengthy texts without losing focus 62 . Since the responses to our open-ended, emotion-provoking question were approximately three minutes long, a few-shot approach may not be suitable for our current analysis. Notably, fine-tuned LLMs such as MentaLLaMA by Yang, et al. 59 and Mental-LLM by Xu, et al. 60 have shown a 5–10% performance improvement compared to zero-shot prompting. However, these models were trained on non-clinically validated open-source text data, limiting their clinical relevance and applicability in our study 59 , 60 . Fine-tuning also requires a large amount of high-quality data and can compromise the model's interpretability, particularly in reasoning processes crucial for extracting key domain-specific indicators 59 , 63 . Due to these limitations, we have not yet fine-tuned our LLM but plan to investigate this approach further once more data becomes available. We identified several key domain-specific indicators from the narratives that were most important for the LLM’s reasoning to classify HCWs into the phenotypes. For the High-Symptom Phenotype, the most identified indicators were emotional exhaustion, feeling overwhelmed, emotional numbness, hopelessness, and stress/anxiety. The LLM seems to primarily detect symptom- and behavior-driven language indicative of distress. In contrast, for the Low-Symptom Phenotype, a positive outlook, a sense of control, and overall positive language were the most identified indicators. This suggests that the LLM extracts different aspects from the narratives to predict low symptomatology, which is not necessarily related to a clinical context. Xue, et al. 64 found that nurses with higher burnout levels experienced worse impairments in envisioning positive future events, which can be driven by hope and resilience in facing the future 65 , 66 . Interpreting the subtle key indicators of language used by the LLM for predicting clinical phenotypes could offer valuable insights into risk and protective factors that might drive classification and provide transparency into the “black box” of the LLM’s decision-making process. The ability to identify domain-specific indicators is a key advantage of zero-shot models compared to fine-tuned models 59 , as it allows for nuanced, human-like interpretations of complex symptomatology. Study implications Our study introduces a proof-of-concept for zero-shot LLMs as a potential screener for burnout, depression, and PTSD symptoms among ED HCWs using a single narrative based on an open-ended question regarding work-related future expectations. The ability to accurately predict complex symptomatology from naturalistic speech could be a valuable addition to traditional clinical assessments, which are often hindered by concerns of stigma and confidentiality, and may allow HCWs to discuss confidential evaluations in a supportive and discrete way that reveals underlying strain, which is particularly important in healthcare settings 51 . As ED HCWs operate in a trauma-intensive and unpredictable environment, their risk for burnout 14 , depression 15 , and PTSD 15 is increased. By encouraging self-awareness of well-being among HCWs and reducing barriers to early intervention, objective tools could support in managing the demands of the HCW’s role while proactively addressing mental health needs. Recognizing and promoting HCW well-being and resilience may ultimately result in improved patient outcomes and quality of care. Strengths and limitations This study offers important insights into high-functioning ED HCWs, who, despite their ability to maintain strong emotional control 58 , 67 , 68 , exhibit a high prevalence of mental health disorders. Our study included a heterogeneous ED HCW population by encompassing a broad range of ED positions across various hospital systems, from public to academic, each with unique patient demographics, traumatic events, and working conditions. A methodological strength of this study is that zero-shot LLMs are easy to implement and do not require additional training, allowing accessibility for real-world applications 62 . We leveraged the LLM’s reasoning capabilities to identify domain-specific indicators that may be important for mental health risk and resilience, indicating a step towards uncovering the “black box”. Also, the deterministic nature of Llama3 ensures consistent predictions across runs, making it a reliable tool for clinical screening and intervention planning, while its local deployment enhances data security and HIPAA compliance 69 . Furthermore, we used a short, open-ended question that does not rely on a clinician-administered structured interview, allowing HCWs to share their experiences more freely and possibly increase compliance. Despite using cross-domain clustering, our clinical profiles were still dependent on self-report assessments, which can introduce potential biases, such as underreporting of symptomatology due to social desirability or the ‘'hero complex’' 70 commonly observed among HCWs 40 . Additionally, selection bias may have influenced our results, as HCWs experiencing less burnout, such as early-career clinicians with fewer emotional and time burdens, may have been more likely to participate. Conversely, those with personal burnout experiences may have been motivated to contribute to the research. Furthermore, the cross-sectional design of this sub-study limited the investigation of long-term mental health trajectories. Despite these potential biases, our observed prevalence of burnout (32.3–42.9%) was comparable to other studies 8 , 14 , 51 . We aim to apply longitudinal methodological approaches in the ongoing NIH-funded study Early Signs (R01HL156134) to examine symptom progression and early signs of deterioration. In addition, ED HCWs operating with exceptional emotional control in high-stakes environments often mask symptoms, which complicates the detection of symptoms from transcriptions alone 67 , 68 . We plan to extend our approach by incorporating multimodal measures such as speech features and facial expressivity, which may improve the prediction of clinical phenotypes and enable the capturing of subtle indicators. These will be further evaluated in R01HL156134. Importantly, our model may be overfitted to language patterns specific to an ED context. Future studies should explore fine-tuning with larger, context-specific datasets and enhance the LLM’s adaptability to nuanced, domain-specific language. In addition, our LLM’s limitation to the English language raises questions about its generalizability. Testing across diverse languages, cultures, and populations is needed to ensure global applicability in healthcare settings, especially given the LLM’s hardware dependence. Finally, we were unable to test a hold-out sample in our clustering solution due to sample size constraints, which may have affected the stability of our findings. However, we balanced the number and size of clusters using the elbow method and regularization techniques to reduce the risk of overfitting. We aim to validate our approach on an independent dataset to strengthen the robustness of our clustering results. Conclusion Zero-shot prompts with LLMs have the potential to predict complex clinical phenotypes of mental health symptomatology, such as burnout, depression, and PTSD among ED HCWs based on a single unstructured narrative about work-related future expectations. This could offer a nuanced and objective tool for mental health screening and may foster a supportive, stigma-free environment for routine assessment of mental health in high-stress healthcare environments, empowering HCWs to take charge of their mental health. Future research should incorporate other essential indicators, such as risk and resilience factors, as well as temporal symptom dynamics, to refine and improve this tool, paving the way for scalable, person-centered mental health monitoring across diverse healthcare settings. Declarations Acknowledgments This study was supported by the National Heart, Lung, and Blood Institute [R01HL156134]. The authors would like to thank all clinicians who participated in this study as well as Joseph Lyu and Annie Farnsworth for their assistance with study enrollment and data collection. Ethical declarations Competing interests The authors declare no competing interests. Author information Authors and affiliations Department of Psychiatry, New York University Grossman School of Medicine, New York, New York, USA Victoria Mueller, Sarah B Birnbaum, Charlotte E Hilberdink, Yiwen Zhao, Joseph H Chang, Sapir Gershov, Katharina Schultebraucks Clinic for Psychiatry and Neurosciences, Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt Universität zu Berlin, Berlin, Germany Tolou Maslahati Division of Healthcare Delivery Science, Department of Population Health, New York University Grossman School of Medicine, New York, New York, USA Katharina Schultebraucks Contributions KS conceived and designed the current study; VM and SBB wrote the first draft of the manuscript and did the final edits under supervision of KS and CEH who offered critical feedback and revised the manuscript for important intellectual content; VM carried out the primary data analysis and interpretation under the supervision of KS; SBB and JHC performed data collection; YZ and SG assisted in data analysis; TM offered revisions on the manuscript. All authors approved the final version of the manuscript. Data availability All requests for raw and analyzed data and related materials, including programming code, will be reviewed by our legal departments (New York University Grossman School of Medicine) to verify whether the request is subject to any intellectual property or confidentiality constraints. Any data and materials that can be shared will be released via a material transfer agreement for noncommercial research purposes. Request should be addressed to the corresponding author (K.S.). References CDC. https://www.cdc.gov/nchs/data/nhamcs/web_tables/2021-nhamcs-ed-web-tables-508.pdf (2021). Wong, A. H., Pacella-LaBarbara, M. L., Ray, J. M., Ranney, M. L. & Chang, B. P. Healing the Healer: Protecting Emergency Health Care Workers' Mental Health During COVID-19. Ann Emerg Med 76, 379–384 (2020). https://doi.org/10.1016/j.annemergmed.2020.04.041 Uddin, H., Hasan, M. K., Cuartas-Alvarez, T. & Castro-Delgado, R. Effects of mass casualty incidents on anxiety, depression, and post-traumatic stress disorder among doctors and nurses: a systematic review. Public Health 234, 132–142 (2024). https://doi.org/10.1016/j.puhe.2024.06.001 Marvaldi, M., Mallet, J., Dubertret, C., Moro, M. R. & Guessoum, S. B. Anxiety, depression, trauma-related, and sleep disorders among healthcare workers during the COVID-19 pandemic: A systematic review and meta-analysis. Neuroscience & Biobehavioral Reviews 126, 252–264 (2021). https://doi.org/https://doi.org/10.1016/j.neubiorev.2021.03.024 Labrague, L. J. Psychological resilience, coping behaviours and social support among health care workers during the COVID-19 pandemic: A systematic review of quantitative studies. Journal of Nursing Management 29, 1893–1905 (2021). https://doi.org/https://doi.org/10.1111/jonm.13336 Watson, A. G., Saggar, V., MacDowell, C. & McCoy, J. V. Self-reported modifying effects of resilience factors on perceptions of workload, patient outcomes, and burnout in physician-attendees of an international emergency medicine conference. Psychology, Health & Medicine 24, 1220–1234 (2019). https://doi.org/10.1080/13548506.2019.1619785 García-Izquierdo, M. & Ríos-Rísquez, M. I. The relationship between psychosocial job stress and burnout in emergency departments: An exploratory study. Nursing Outlook 60, 322–329 (2012). https://doi.org/https://doi.org/10.1016/j.outlook.2012.02.002 Adriaenssens, J., De Gucht, V. & Maes, S. Determinants and prevalence of burnout in emergency nurses: a systematic review of 25 years of research. Int J Nurs Stud 52, 649–661 (2015). https://doi.org/10.1016/j.ijnurstu.2014.11.004 Hu, Y. et al. Relationship between workplace violence, job satisfaction, and burnout among healthcare workers in mobile cabin hospitals in China: Effects of perceived stress and work environment. Prev Med Rep 40, 102667 (2024). https://doi.org/10.1016/j.pmedr.2024.102667 Aljohani, B. et al. Workplace violence in the emergency department: a systematic review and meta-analysis. Public Health 196, 186–197 (2021). https://doi.org/https://doi.org/ 10.1016/j.puhe.2021.02.009 Kuhn, G. Circadian rhythm, shift work, and emergency medicine. Ann Emerg Med 37, 88–98 (2001). https://doi.org/10.1067/mem.2001.111571 Doan–Wiggins, L. et al. Practice Satisfaction, Occupational Stress, and Attrition of Emergency Physicians. Academic Emergency Medicine 2, 556–563 (1995). https://doi.org/https://doi. org/10.1111/j.1553-2712.1995.tb03261.x Aiken, L. H., Clarke, S. P., Sloane, D. M., Sochalski, J. & Silber, J. H. Hospital nurse staffing and patient mortality, nurse burnout, and job dissatisfaction. Jama 288, 1987–1993 (2002). https://doi.org/10.1001/jama.288.16.1987 Hooper, C., Craig, J., Janvrin, D. R., Wetsel, M. A. & Reimels, E. Compassion satisfaction, burnout, and compassion fatigue among emergency nurses compared with nurses in other selected inpatient specialties. J Emerg Nurs 36, 420–427 (2010). https://doi.org/10.1016/j.jen.2009.11.027 Matthews, L. R. et al. PREVALENCE AND PREDICTORS OF POSTTRAUMATIC STRESS DISORDER, DEPRESSION, AND ANXIETY IN PERSONNEL WORKING IN EMERGENCY DEPARTMENT SETTINGS: A SYSTEMATIC REVIEW. The Journal of Emergency Medicine 62, 617–635 (2022). https://doi.org/https://doi.org/10.1016/j.jemermed.2021.09.010 Awan, S. et al. Frontiers | Suicide in Healthcare Workers: Determinants, Challenges, and the Impact of COVID-19. Frontiers in Psychiatry 12 (2022). https://doi.org/10.3389/fpsyt.2021.792925 Marco, C. A., Larkin, G. L., Feeser, V. R., Monti, J. E. & Vearrier, L. Post-traumatic stress and stress disorders during the COVID‐19 pandemic: Survey of emergency physicians. Journal of the American College of Emergency Physicians Open 1 (2020). https://doi.org/10.1002/emp2.12305 Bahadirli, S. & Sagaltici, E. Post-traumatic stress disorder in healthcare workers of emergency departments during the pandemic: A cross-sectional study. The American Journal of Emergency Medicine 50 (2021). https://doi.org/10.1016/j.ajem.2021.08.027 Trudgill, D. I. N., Gorey, K. M. & Donnelly, E. A. Prevalent posttraumatic stress disorder among emergency department personnel: rapid systematic review. Humanities and Social Sciences Communications 7, 89 (2020). https://doi.org/10.1057/s41599-020-00584-x Shanafelt, T. D. et al. Special report: suicidal ideation among American surgeons. Arch Surg 146, 54–62 (2011). https://doi.org/10.1001/archsurg.2010.292 Davidson, J. E., Proudfoot, J., Lee, K., Terterian, G. & Zisook, S. A Longitudinal Analysis of Nurse Suicide in the United States (2005–2016) With Recommendations for Action. Worldviews Evid Based Nurs 17, 6–15 (2020). https://doi.org/10.1111/wvn.12419 Dutheil, F. et al. Suicide among physicians and health-care workers: A systematic review and meta-analysis. PLOS ONE 14, e0226361 (2019). https://doi.org/10.1371/journal.pone.0226361 Machi, M. S. et al. The relationship between shift work, sleep, and cognition in career emergency physicians. Acad Emerg Med 19, 85–91 (2012). https://doi.org/10.1111/j.1553-2712.2011.01254.x John, A. et al. The influence of burnout on cardiovascular disease: a systematic review and meta-analysis. Front Psychiatry 15, 1326745 (2024). https://doi.org/10.3389/fpsyt.2024.1326745 Sumner, J. A. et al. Trauma Exposure and Posttraumatic Stress Disorder Symptoms Predict Onset of Cardiovascular Events in Women. Circulation 132, 251–259 (2015). https://doi.org/10.1161/circulationaha.114.014492 Deligkaris, P., Panagopoulou, E., Montgomery, A. J. & Masoura, E. Job burnout and cognitive functioning: A systematic review. Work & Stress 28, 107–123 (2014). https://doi.org/10.1080/02678373.2014.909545 Sandström, A., Rhodin, I. N., Lundberg, M., Olsson, T. & Nyberg, L. Impaired cognitive performance in patients with chronic burnout syndrome. Biol Psychol 69, 271–279 (2005). https://doi.org/10.1016/j.biopsycho.2004.08.003 Gavelin, H. M. et al. Cognitive function in clinical burnout: A systematic review and meta-analysis. Work & Stress 36, 86–104 (2022). https://doi.org/10.1080/02678373.2021.2002972 Pereira-Lima, K. et al. Association Between Physician Depressive Symptoms and Medical Errors: A Systematic Review and Meta-analysis. JAMA Netw Open 2, e1916097 (2019). https://doi.org/10.1001/jamanetworkopen.2019.16097 Weigl, M. & Schneider, A. Associations of work characteristics, employee strain and self-perceived quality of care in Emergency Departments: A cross-sectional study. International Emergency Nursing 30, 20–24 (2017). https://doi.org/https://doi.org/10.1016/j.ienj.2016.07.002 Lu, D. W., Dresden, S., McCloskey, C., Branzetti, J. & Gisondi, M. A. Impact of Burnout on Self-Reported Patient Care Among Emergency Physicians. West J Emerg Med 16, 996–1001 (2015). https://doi.org/10.5811/westjem.2015.9.27945 De Stefano, C. et al. Effect of emergency physician burnout on patient waiting times. Intern Emerg Med 13, 421–428 (2018). https://doi.org/10.1007/s11739-017-1706-9 Gualano, M. R. et al. The Burden of Burnout among Healthcare Professionals of Intensive Care Units and Emergency Departments during the COVID-19 Pandemic: A Systematic Review. International Journal of Environmental Research and Public Health 2021, Vol. 18, Page 8172 18 (2021). https://doi.org/10.3390/ijerph18158172 Wijeratne, C., Johnco, C., Draper, B. & Earl, J. Doctors’ reporting of mental health stigma and barriers to help-seeking. Occupational Medicine 71, 366–374 (2021). https://doi.org/10.1093/occmed/kqab119 Auth, N. M., Booker, M. J., Wild, J. & Riley, R. Mental health and help seeking among trauma-exposed emergency service staff: a qualitative evidence synthesis. BMJ Open 12, e047814 (2022). https://doi.org/10.1136/bmjopen-2020-047814 Taylor, N. W., Porter, C., Rivera-Rodriguez, M., Miller, I. S. K. & Desmarais, N. Mental Health Disclosure Questions on Medical Licensure Applications: Implications for Medical Students, Residents, and Physicians. Academic Medicine 97 (2022). McGarry, E. & O’Connor, L. Assessing burnout rates and contributing factors in emergency medical services clinicians. Journal of Workplace Behavioral Health 39 (2024). https://doi.org/10.1080/15555240.2023.2292119 Howlett, M. et al. Burnout in emergency department healthcare professionals is associated with coping style: a cross-sectional survey. Emergency Medicine Journal 32 (2015). https://doi.org/10.1136/emermed-2014-203750 Catalogue of Bias Collaboration, S. E., Brassey J, Mahtani K. Recall bias. In: Catalogue Of Bias 2017. https://www.catalogueofbiases.org/biases/recall-bias . Van de Mortel, T. Faking it: Social desirability response bias in self-report research. Thea van de Mortel 25 (2008). Marshall, R. E. et al. Mental health screening amongst police officers: factors associated with under-reporting of symptoms. BMC Psychiatry 2021 21:1 21 (2021-03-08). https://doi.org/10.1186/s12888-021-03125-1 CH, W. et al. Importance of anonymity to encourage honest reporting in mental health screening after combat deployment - PubMed. Archives of general psychiatry 68 (2011 Oct). https://doi.org/10.1001/archgenpsychiatry.2011.112 Maslach, C., Jackson, S. & Leiter, M. Maslach burnout inventory (MBI) Manual (1996–2016). 4th Ed., Menlo Park, CA., Mind Garden, Inc. (1996–2016). Kroenke, K. et al. The PHQ-8 as a measure of current depression in the general population. Journal of affective disorders 114(1–3), 163–173 (2009). Weathers, F. et al. The PTSD Checklist for DSM-5 (PCL-5). Scale available from the National Center for PTSD at www ptsd va gov. (2013). MacQueen, J. in Proceedings of the fifth Berkeley symposium on mathematical statistics and probability. 281–297 (Oakland, CA, USA). Lloyd, S. Least squares quantization in PCM. IEEE Transactions on Information Theory 28, 129–137 (1982). https://doi.org/10.1109/TIT.1982.1056489 AI@Meta. Llama 3 Model Card. (2024). Kuhn, M. & Johnson, K. Applied Predictive Modeling. (2013). https://doi.org/10.1007/978-1-4614-6849-3 Leiter, M. P. & Maslach, C. Latent burnout profiles: A new approach to understanding the burnout experience. Burnout Research 3, 89–100 (2016). https://doi.org/https://doi.org/ 10.1016/j.burn.2016.09.001 Havaei, F., Ma, A., Leiter, M. & Gear, A. Describing the Mental Health State of Nurses in British Columbia: A Province-Wide Survey Study. Healthc Policy 16, 31–45 (2021). https://doi.org/10.12927/hcpol.2021.26500 Önder, Ç. & Basim, N. Examination of developmental models of occupational burnout using burnout profiles of nurses. Journal of Advanced Nursing 64, 514–523 (2008). https://doi.org/ https://doi.org/10.1111/j.1365-2648.2008.04818.x Pei, Y., Li, Y., Wu, L., Xi, J. & Zhou, N. Psychological distress during the COVID-19 pandemic normalization phase in China: A multigroup latent profile analysis among healthcare workers and the general population. International Journal of Disaster Risk Reduction 108 (2024). https://doi.org/10.1016/j.ijdrr.2024.104567 Braule Pinto, A. L. d. C. et al. Longitudinal profile of post-traumatic symptoms in HealthCare Workers during COVID-19 pandemic: A latent transition model. Journal of Psychiatric Research 168, 230–239 (2023). https://doi.org/https://doi.org/10.1016/j.jpsychires.2023.10.031 Ohse, J. et al. Zero-Shot Strike: Testing the generalisation capabilities of out-of-the-box LLM models for depression detection. Computer Speech & Language 88 (2024). https://doi.org/10.1016/j.csl.2024.101663 Danner, M. et al. in 2023 62nd Annual Conference of the Society of Instrument and Control Engineers (SICE). 1290–1296. Lho, S. K. et al. Large Language Models and Text Embeddings for Detecting Depression and Suicide in Patient Narratives. JAMA Network Open 8, e2511922 (2025). https://doi.org/10.1001/jamanetworkopen.2025.11922 Kadović, M. et al. Ability of Emotional Regulation and Control as a Stress Predictor in Healthcare Professionals. International Journal of Environmental Research and Public Health 2023, Vol. 20, Page 541 20 (2022-12-29). https://doi.org/10.3390/ijerph20010541 Yang, K. et al. in Proceedings of the ACM Web Conference 2024 4489–4500 (Association for Computing Machinery, Singapore, Singapore, 2024). Xu, X. et al. Mental-LLM: Leveraging Large Language Models for Mental Health Prediction via Online Text Data. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 8, Article 31 (2024). https://doi.org/10.1145/3643540 Bartal, A., Jagodnik, K. M., Chan, S. J. & Dekel, S. AI and narrative embeddings detect PTSD following childbirth via birth stories. Scientific Reports 14, 8336 (2024). https://doi.org/10.1038/s41598-024-54242-2 Kojima, T., Gu, S. S., Reid, M., Matsuo, Y. & Iwasawa, Y. in Proceedings of the 36th International Conference on Neural Information Processing Systems Article 1613 (Curran Associates Inc., New Orleans, LA, USA, 2022). Pouramini, A. & Faili, H. Matching tasks to objectives: Fine-tuning and prompt-tuning strategies for encoder-decoder pre-trained language models. Applied Intelligence 54, 9783–9810 (2024). https://doi.org/10.1007/s10489-024-05660-2 Xue, B. et al. Phenomenological characteristics of autobiographical future thinking in nurses with burnout: a case-control study. Frontiers in Psychology 14 (2023). https://doi.org/10.3389/fpsyg.2023.1216036 Yıldırım, M., Kaynar, Ö., Chirico, F. & Magnavita, N. Resilience and Extrinsic Motivation as Mediators in the Relationship between Fear of Failure and Burnout. International Journal of Environmental Research and Public Health 20, 5895 (2023). Pharris, A. B., Munoz, R. T. & Hellman, C. M. Hope and resilience as protective factors linked to lower burnout among child welfare workers. Children and Youth Services Review 136, 106424 (2022). https://doi.org/https://doi.org/10.1016/j.childyouth.2022.106424 Malinowska-Lipień, I. et al. Emotional Control among Nurses against Work Conditions and the Support Received during the SARS-CoV-2 Pandemic. International Journal of Environmental Research and Public Health 2021, Vol. 18, Page 9415 18 (2021-09-06). https://doi.org/10.3390/ijerph18179415 Hayward, R. M., Tuckey, M. R. & Renae Maree Hayward, M. R. T. Emotions in uniform: How nurses regulate emotion at work via emotional boundaries. Human Relations 64 (2011-11-08). https://doi.org/10.1177/0018726711419539 Park, Y.-J. et al. Assessing the research landscape and clinical utility of large language models: a scoping review. BMC Medical Informatics and Decision Making 2024 24:1 24 (2024-03-12). https://doi.org/10.1186/s12911-024-02459-6 Cox, C. L. ‘Healthcare Heroes’: problems with media focus on heroism from healthcare workers during the COVID-19 pandemic. Journal of Medical Ethics 46, 510–513 (2020). https://doi.org/10.1136/medethics-2020-106398 Tables Tables 1 and 2 are available in the Supplementary Files section. Additional Declarations The authors have declared there is NO conflict of interest to disclose Supplementary Files Table1and2.docx NatureTransPsychHCWSupplements.docx Supplementary Materials Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7087333","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":485937607,"identity":"7af75f86-1260-49cb-8fa8-41e4846c14b0","order_by":0,"name":"Katharina Schultebraucks","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0001-5085-8249","institution":"NYU Grossman School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Katharina","middleName":"","lastName":"Schultebraucks","suffix":""},{"id":485937608,"identity":"7335df8c-5f63-462d-abd9-f5faed214318","order_by":1,"name":"Victoria Mueller","email":"","orcid":"https://orcid.org/0009-0005-2913-7718","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Victoria","middleName":"","lastName":"Mueller","suffix":""},{"id":485937609,"identity":"2bef69e1-5e84-40c9-a97f-b7b7a512c8ef","order_by":2,"name":"Sarah Birnbaum","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Sarah","middleName":"","lastName":"Birnbaum","suffix":""},{"id":485937610,"identity":"165c9620-a919-464f-a62a-0cb999d55120","order_by":3,"name":"Charlotte Hilberdink","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Charlotte","middleName":"","lastName":"Hilberdink","suffix":""},{"id":485937611,"identity":"6d5195c8-686e-4499-a649-181e500b8658","order_by":4,"name":"Yiwen Zhao","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Yiwen","middleName":"","lastName":"Zhao","suffix":""},{"id":485937612,"identity":"3d79c9b6-a10e-4a85-a37c-f2f85e7fa4d3","order_by":5,"name":"Joseph Chang","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Joseph","middleName":"","lastName":"Chang","suffix":""},{"id":485937613,"identity":"1e60a5dc-2a09-4eb2-9336-90b70368c828","order_by":6,"name":"Tolou Maslahati","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Tolou","middleName":"","lastName":"Maslahati","suffix":""},{"id":485937614,"identity":"ba691d55-e5e6-4af3-a253-11255a9b87b1","order_by":7,"name":"Sapir Gershov","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Sapir","middleName":"","lastName":"Gershov","suffix":""}],"badges":[],"createdAt":"2025-07-09 21:35:39","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7087333/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7087333/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87061181,"identity":"e2395f70-7c1b-4cb8-a68a-86e661b856ee","added_by":"auto","created_at":"2025-07-18 16:58:32","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":75283,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMethodological framework for assessing and classifying clinical phenotypes in HCWs using an LLM approach.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe figure illustrates the comprehensive methodology of the study, consisting of assessment, clinical phenotyping, LLM approach, and evaluation. It details the collected data, the clustering analysis for identifying clinical phenotypes, and the application of a pre-trained LLM with zero-shot prompt engineering to classify these phenotypes based on work-related narratives, as well as the final evaluation of the clinical phenotypes and the LLM-predicted phenotypes.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7087333/v1/1181fdc0f03e3fea7e7170c9.png"},{"id":87061182,"identity":"598c59a4-9275-4547-988b-482854f28d74","added_by":"auto","created_at":"2025-07-18 16:58:32","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":14435,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClinical Phenotypes of burnout, depression, and PTSD symptomatology.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eProfiles of clinical phenotypes based on psychological symptomatology for burnout, depression and PTSD in ED HCWs. Scores are standardized psychological symptom levels on questionnaire item scores of the MBI-9 for burnout symptoms that measure Personal Accomplishment (PA), Emotional Exhaustion (EE) and Depersonalization (DEP), item scores of the PHQ-8 for depressive symptoms, and subdomain scores of the PCL-5 for PTSD-related DSM-5 Cluster B Intrusions, Cluster C Avoidance, Cluster D Negative Cognitions and Mood, and Cluster E Arousal and Reactivity Alterations. Error bars indicate standard errors.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7087333/v1/863af5144e04db6d29264eda.png"},{"id":87061441,"identity":"a504eacb-59f0-4dd4-acf1-83f806b97697","added_by":"auto","created_at":"2025-07-18 17:06:32","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":17332,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePredictive LLM performance to discriminate clinical phenotypes based on narratives. a) Confusion Matrix. \u003c/strong\u003eThis confusion matrix shows the LLM's classification performance based on the actual and predicted classification by the model. Darker cells indicate better classification performance.\u003cstrong\u003e b) Receiver Operating Characteristic (ROC) curve. \u003c/strong\u003eThe LLM’s ability to distinguish between clinical phenotypes based on the True Positive Rate versus False Positive Rate at various classification thresholds showing an AUC of 71.61.\u003cstrong\u003e c) Best-performing zero-shot LLM prompt predictions. \u003c/strong\u003ePerformance of the LLM based on established measures used for machine-learning classification tasks.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7087333/v1/efa7feddf4152f6e12e2e6b0.png"},{"id":88772900,"identity":"7ad6d46d-e5c8-47e1-8697-2b717cf4c6a6","added_by":"auto","created_at":"2025-08-11 09:55:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1058561,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7087333/v1/ad5e2185-6e9c-4ed6-9f9a-5801f80bdedf.pdf"},{"id":87061183,"identity":"4ba2a317-6f58-402c-98bd-6b01fb16a182","added_by":"auto","created_at":"2025-07-18 16:58:32","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":38713,"visible":true,"origin":"","legend":"","description":"","filename":"Table1and2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7087333/v1/e304ad2683aac3bad91a5e22.docx"},{"id":87061185,"identity":"85476cbc-7a4b-4fa3-83e9-12e31bcb55fc","added_by":"auto","created_at":"2025-07-18 16:58:32","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":111489,"visible":true,"origin":"","legend":"Supplementary Materials","description":"","filename":"NatureTransPsychHCWSupplements.docx","url":"https://assets-eu.researchsquare.com/files/rs-7087333/v1/f859a49412f5bd476c0e5db4.docx"}],"financialInterests":"The authors have declared there is \u003cb\u003eNO\u003c/b\u003e conflict of interest to disclose","formattedTitle":"Large Language Models for Accurate Mental Health Screening: Identifying Clinical Phenotypes in ED Healthcare Workers","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe emergency department (ED) is the backbone of the healthcare system, with over 139.8\u0026nbsp;million visits in the US annually\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Frontline ED clinicians and staff serve a crucial societal role providing a vital health safety net. To provide the highest quality of care, ED healthcare workers (HCWs) must maintain a constant state of vigilance, focus, and resilience to operate effectively in a demanding and unpredictable environment on a daily basis\u003csup\u003e\u003cspan additionalcitationids=\"CR3 CR4 CR5\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe stressful environment of the ED poses a serious, ongoing risk to the psychological and physical well-being of HCWs. Alongside demanding work conditions in the ED, such as fast-paced work with high psychological demands\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e, excessive workload\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, patient violence\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, and shift work\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, ED HCWs are regularly exposed to stressful and traumatic situations\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Accumulation of several long-term stressors can be extremely burdensome and leave many HCWs dissatisfied, leading to high attrition rates\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Studies have reported an increased prevalence of mental health issues, such as increased burnout rates in 26\u0026ndash;82% of ED HCWs\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, and around 20% of ED personnel met criteria for PTSD symptoms\u003csup\u003e\u003cspan additionalcitationids=\"CR16 CR17 CR18\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Up to 77.1% have depression\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e and HCWs in demanding roles, including ED providers, are at greater risk for suicide and suicidal ideation\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Burnout in HCWs can also be linked to physical consequences such as increased fatigue\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, cardiovascular issues\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e, and disrupted sleeping patterns that can significantly affect emotional processing and other difficulties in cognitive domains, such as decision-making\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, attention, and memory\u003csup\u003e\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThese consequences for ED HCWs can directly affect the delivery of high-quality healthcare. Poorer well-being can lead to increased medical errors\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e and negatively affect perceptions of patient safety\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, ultimately compromising the quality of care\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Barriers to seeking psychological support, such as feelings of stigma, shame, fear of perceptions of weakness, and concerns about breaches of confidentiality that could affect career and licensing, hinder HCWs in general from seeking help\u003csup\u003e\u003cspan additionalcitationids=\"CR34 CR35\" citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe current traditional tools to assess well-being of HCWs mostly rely on clinical interviews and psychometric assessments that evaluate simplified dichotomous categorical diagnoses without comprehensive consideration of multiple symptoms and potential comorbidities. They can be cost-intensive and time-consuming for large-scale use. They are often burdensome and have low response rates\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e, potentially excluding those who experience more stress and burnout as they may be less likely to participate\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e. Furthermore, self-reported symptoms are susceptible to recall bias\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e and social desirability\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. Qualitative methods may effectively address these biases by capturing valuable individual and complex experiences and emotions through techniques such as semi-structured interviews, providing a holistic, in-depth, and contextual view of these experiences, but are labor- and time-intensive. Studies have shown that mental health symptomatology may be underreported in many high-performing professions, such as healthcare workers\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e, as was found in police officers\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e and military personnel\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e. Therefore, it is crucial to shift away from traditional psychometric assessments that focus on classical diagnoses. Instead, there is an urgent need to adopt more low-burden, objective, and person-centered measures to identify clinical phenotypes that account for heterogeneous, interdependent symptoms and their severity. Going forward, this will facilitate the early detection of stress-related symptoms and the targeted implementation of interventions to strengthen the mental well-being and resilience of HCWs.\u003c/p\u003e\u003cp\u003eThis study aims to identify relevant profiles of clinical phenotypes based on burnout, depression, and PTSD symptomology in ED HCWs by analyzing narratives of work-related stressful situations using large language models (LLMs). We hypothesize that LLMs can effectively classify these heterogeneous clinical phenotypes, thereby providing a more nuanced understanding of mental health risks. This approach can potentially replace traditional assessments, offering a scalable, time-efficient, and unbiased alternative that healthcare providers are more likely to engage with.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cp\u003eAn overview of the methodological framework employed in this study, including participant assessment, clinical phenotyping, the LLM approach, and evaluation, is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Participants\u003c/h2\u003e\u003cp\u003eParticipants were enrolled as part of an ongoing NIH-funded observational longitudinal study (Early Signs, R01HL156134) with the primary aim of investigating burnout and well-being in N\u0026thinsp;=\u0026thinsp;350 ED and Trauma Unit HCWs (clinicians and staff) in the greater New York City (NYC) area. Participants were eligible if they were \u0026ge;18 years old, fluent in English, had direct patient contact, and typically worked full-time clinical hours (including those who worked temporarily part-time due to burnout or temporary leave). Medical students and temporary HCWs were excluded. Participants were recruited through online advertising (Facebook, iConnect (research trial database)) and recruitment visits by research staff to EDs in the NYC area. All participants signed the informed consent and were informed that participation was voluntary and would not affect their employment status or performance reviews. This study was approved by the Institutional Review Board of NYU Grossman School of Medicine.\u003c/p\u003e\u003cp\u003eFor this sub-study, 259 assessments (199 assessments at baseline; 66 assessments at 6-month follow-up) of N\u0026thinsp;=\u0026thinsp;199 participants, who were enrolled from April 2022 through August 2024, were included.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Procedures\u003c/h2\u003e\u003cp\u003eData collected during the baseline and 6-month follow-up assessments were included in this sub-study. Participants provided demographic and work-related information, and completed self-report questionnaires to measure current psychological symptom levels for burnout, depression, and PTSD at home prior to or during the assessment. For weekly alcohol intake and daily smoking behavior, the Smoking Alcohol Intake Questionnaire was used. During the assessment, a semi-structured interview was conducted to capture emotion-driven narratives on stressful work experiences through a HIPAA-compliant video platform (WEBEX).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Measures\u003c/h2\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e2.3.1 Psychological Symptomatology\u003c/h2\u003e\u003cp\u003eBurnout symptoms were assessed using the 9 items of the Maslach Burnout Inventory (MBI-9; range 0\u0026ndash;6, Maslach, et al. \u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e across the domains Personal Achievement (PA; items 1, 6, 9), Emotional Exhaustion (EE; items 3, 4, 7), and Depersonalization (DEP; items 2, 5, 8). Depressive symptoms during the past 2 weeks were assessed using the 8-item Patient Health Questionnaire (PHQ-8; range 0-3\u003csup\u003e44\u003c/sup\u003e). PTSD symptomatology in the past month, with regards to COVID-19 or another traumatic experience from working in the ED, was measured using the PTSD Checklist for DSM-5 (PCL-5; 20 items, range 0-4\u003csup\u003e45\u003c/sup\u003e), for which subdomain scores for PTSD\u0026rsquo;s DSM-5 diagnostic symptom clusters were calculated; Cluster \u003cem\u003eB\u003c/em\u003e Intrusions, Cluster \u003cem\u003eC\u003c/em\u003e Avoidance, Cluster \u003cem\u003eD\u003c/em\u003e Negative Cognitions and Mood, and Cluster \u003cem\u003eE\u003c/em\u003e Arousal and Reactivity Alterations. Higher scores on all items and subdomains indicated higher self-reported symptom levels, except for PA, with lower scores indicating higher symptom levels.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\u003ch2\u003e2.3.2 Work-related Narratives\u003c/h2\u003e\u003cp\u003eDuring the semi-structured interview, HCWs were asked the following emotion-provoking question verbatim to capture an unstructured narrative on work-related future expectations: \u0026ldquo;What are your expectations about your role and your work as an ED clinician in the future?\u0026rdquo;. Participants had a predetermined time of three minutes to answer the open-ended question.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Statistical Approach\u003c/h2\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003e\u003cb\u003e2.4.1 Clustering analysis\u003c/b\u003e\u003c/h2\u003e\u003cp\u003eTo identify heterogeneous profiles of clinical phenotypes, we performed clustering analysis using the k-Means algorithm\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e,\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e in Python (v3.9, \u0026lsquo;scikit-learn\u0026rsquo; library v1.5.0). Item scores of the MBI-9 and PHQ-8, and PTSD symptom subdomain scores for PCL-5 Cluster B, C, D, and E were included in the clustering analysis. Data were preprocessed including checking for missingness (\u0026ge;30% missing values, applied when calculating subdomain scores) and near-zero variance (threshold\u0026thinsp;=\u0026thinsp;0.1) for each feature. We excluded n\u0026thinsp;=\u0026thinsp;1 participant due to missing questionnaire data and removed PHQ-8 item 8 from the analysis due to near-zero variance. Multivariate outliers of 9 assessments (3.4%; 8 at baseline and 1 at 6-month follow-up) from n\u0026thinsp;=\u0026thinsp;7 participants were identified and removed based on the Mahalanobis Distance (threshold\u0026thinsp;=\u0026thinsp;0.001), resulting in a total of N\u0026thinsp;=\u0026thinsp;254 assessments (190 at baseline; 64 at 6-month follow-up). In total, N\u0026thinsp;=\u0026thinsp;191 participants were included in the final clustering analysis.\u003c/p\u003e\u003cp\u003eFor clustering, the \u003cem\u003ek\u003c/em\u003e-means algorithm partitions data into \u003cem\u003ek\u003c/em\u003e clusters by minimizing the sum of squared distances between data points and their assigned cluster centroids. The variables were standardized by centering them around the mean and scaling to unit variance. Model evaluation was based on pre-identified indicators of the best-fitting model by testing the fit of a 1-profile model and subsequently increasing the profile number by 1 until the addition of a profile was no longer optimal or improved. Indicators were the Silhouette score (higher value: better fit), Davies-Bouldin score (lower value: better fit), Calinski-Harabasz score (higher value: better fit), Dunn Index (higher value: better fit) and the Elbow Method, which involved plotting the within-cluster sum of squares (WCSS) against the number of clusters to identify the \u0026lsquo;elbow point\u0026rsquo;, where the slope rate of decrease is reduced and is sharpest (see Supplementary Material S1 on best-fitting model evaluation).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\u003ch2\u003e2.4.2 LLM zero-shot prompts\u003c/h2\u003e\u003cp\u003eTo determine whether LLMs can effectively predict and identify clinical phenotypes, we employed a pre-trained LLM with zero-shot prompt engineering. In a zero-shot setting, the model generates predictions without having seen any labeled examples related to the specific task beforehand. Instead, it relies entirely on its pre-existing knowledge from training and the information provided in the prompt. The LLM analyzed narratives about work-related future expectations from the participants whose video recordings of the interview were available to distinguish and classify clinical phenotypes. In total, 126 narratives were utilized from a subgroup of N\u0026thinsp;=\u0026thinsp;90 participants with available video recordings.\u003c/p\u003e\u003cp\u003e Using a three-step approach, narratives were transcribed by 1) manually segmenting the video into separate clips per question, 2) applying speaker diarization (partitioning speech recordings according to the identity of each speaker) using PyAnnote (v3.1.1), and 3) transcribing audio using Whisper (v20231117) and checking manually for quality.\u003c/p\u003e\u003cp\u003eTo evaluate the capability of pre-trained LLM in predicting clinical phenotypes, we developed zero-shot prompt templates composed of four components:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\text{P}\\text{r}\\text{o}\\text{m}\\text{p}\\text{t}=\\:Task\\:+\\:Clinical\\:Phenotypes\\:+\\:Further\\:Information\\:+\\:Narrative$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe \u0026lsquo;\u003cem\u003eTask\u003c/em\u003e\u0026rsquo; component specifies the primary assignment for the LLM to perform. The \u0026lsquo;\u003cem\u003eClinical Phenotypes\u0026rsquo;\u003c/em\u003e component describes the characteristics of the distinct clinical phenotypes derived from the cluster analysis. The \u0026lsquo;\u003cem\u003eAdditional Information\u0026rsquo;\u003c/em\u003e component provides supplementary context (e.g., details about symptom comorbidities). Lastly, the \u0026lsquo;\u003cem\u003eNarrative\u0026rsquo;\u003c/em\u003e component contains participants' narratives about their work-related future expectations. Example prompts are provided in Supplementary Material S2.\u003c/p\u003e\u003cp\u003eWe locally employed Meta\u0026rsquo;s open-source, HIPAA-compliant, Llama3-8B-Instruct model\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e as LMM on an NVIDIA GPU using the \u0026ldquo;transformers\u0026rdquo; Python package (v4.44.0). Additional implementation details are provided in Supplementary Material S3. It is important to note that we disabled Llama-3 sampling in order for the model to consistently generate the same response for a given input. Instead of randomly sampling from the model's probability distribution during generation, the model always selects the highest probability token at each step, resulting in deterministic predictions. Predictive model performance was measured by precision, recall, F1-score and ROC-AUC, which are established performance measures used for machine-learning classification tasks\u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Clinical phenotypes\u003c/h2\u003e\n \u003cp\u003eA descriptive overview of demographic information, psychological symptomatology, and behavioral characteristics is presented in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eThe best-fitting clustering model consisted of two clinical phenotypes (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). Their descriptive profile labels were based on means of total and subdomain scores (Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). We interpreted the profiles as follows: \u0026ldquo;High-Symptom Phenotype\u0026rdquo; (39.0%, 99 assessments from n\u0026thinsp;=\u0026thinsp;79 distinct participants) with higher symptom levels on items for burnout with regards to emotional exhaustion and depersonalization, as well as depressive symptoms and PTSD symptoms across all clusters, re-experiencing, avoidance, negative alterations in cognition and mood, and hyperarousal; and \u0026ldquo;Low-Symptom Phenotype\u0026rdquo; (61.0%, 155 assessments from n\u0026thinsp;=\u0026thinsp;120 participants) with lower symptom levels on items for burnout with regards to emotional exhaustion and depersonalization, as well as depressive symptoms and PTSD symptom clusters for re-experiencing, avoidance, negative alterations in cognition and mood, and hyperarousal. No significant difference was observed in burnout-related personal accomplishment between the two clinical phenotypes.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 LLM zero-shot prompting\u003c/h2\u003e\n \u003cp\u003eThe mean word count of the narratives was 360.90 (SD\u0026thinsp;=\u0026thinsp;124.31, for a descriptive overview of both profiles in the LLM participant sample, see Supplementary Material S4). The Llama3-8B-Instruct model predicted both clinical phenotype profiles using work-related narratives with a sensitivity of 77.1%, accuracy of 70.9%, and F1-score of 71.8% (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eBased on the zero-shot prompt, the LLM identified several key domain-specific indicators extracted from the narratives that served as reasoning for classifying participants into clinical phenotypes (Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). Classification into the High-Symptom Phenotype was most frequently predicted based on the following five domain-specific indicators: emotional exhaustion (68.1%), feeling overwhelmed (62.3%), emotional numbness and avoidance (33.3%), hopelessness/helplessness (30.4%), and stress/anxiety (27.5%); and for the Low-Symptom Phenotype: positive outlook (94.6%), clear future vision (65.5%), sense of control (58.2%), positive language (56.4%), and coping mechanisms (27.3%). See Supplementary Material S5 for an example of such domain-specific indicators.\u003c/p\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eKey domain-specific indicators extracted from the narratives for reasoning of the LMM to predict the High-Symptom or Low-Symptom Phenotype.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eHigh-Symptom Phenotype\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eLow-Symptom Phenotype\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCategories\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003en (%)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCategories\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003en (%)\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEmotional exhaustion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e48 (67.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePositive outlook\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e52 (94.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFeeling overwhelmed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e45 (63.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eClear future vision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e36 (65.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEmotional numbing and avoidance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e25 (35.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSense of control\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e32 (58.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHopelessness/helpless\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e21 (29.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePositive language\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e31 (56.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStress/anxious\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e19 (26.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCoping mechanisms\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15 (27.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCognitive issues\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e13 (18.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSelf-awareness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12 (21.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFrustration\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e13 (18.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWork-life balance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11 (20.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFuture uncertainty\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11 (15.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAcknowledgement of the problem\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10 (14.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFatigue/drained\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e9 (12.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDisappointment\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e8 (11.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFeeling guilt or self-doubt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e8 (11.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSeeking a change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 (9.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDissatisfaction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7 (9.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePhysical exhaustion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e5 (7.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDistress\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4 (5.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eImpact personal life or relationships\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3 (4.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLack of sustainability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2 (2.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSleep problems\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1 (1.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003en (%): number (percentage) of narratives in which the key domain-specific indicator was identified as predictive of either the High- or Low-Symptom Phenotype. The percentage was calculated within the two phenotypes separately.\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis study is the first to utilize LLMs to predict heterogeneous clinical phenotypes in ED HCWs. The clinical phenotypes were identified using a person-centered clustering approach based on self-reported burnout, depression, and PTSD symptomatology. These mental health issues are highly prevalent in this profession, given the high-stress and trauma-intensive nature of the ED environment. Using a zero-shot LLM approach with a single narrative response to an open-ended question about work-related future expectations, we accurately classified ED HCWs into these two phenotypes. In addition, we identified several key domain-specific indicators used by the LLM for clinical phenotype classification.\u003c/p\u003e\u003cp\u003eUsing a person-centered clustering approach, we identified a High- and Low-Symptomatology Phenotype in ED HCWs, capturing the entire range, severity, and interdependence of symptoms rather than focusing on dichotomous categorical diagnoses. These two heterogeneous phenotypes, symptomatic and resilient, are consistent with previous studies\u003csup\u003e\u003cspan additionalcitationids=\"CR51 CR52 CR53\" citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e. The prevalence of resilient individuals was higher, aligning with other studies that report high resilience in HCWs despite their high workloads and high-performing nature\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e,\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e. Yet, a significant portion still reported mental issues, which is also aligned with previous findings\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e,\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e. Our results extend previous findings\u003csup\u003e\u003cspan additionalcitationids=\"CR51 CR52 CR53\" citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e by including individual symptom levels across various mental health domains, revealing that our High-Symptom Phenotype encompasses complex comorbidities across burnout, depression, and PTSD, which could influence and exacerbate symptom development, outcomes, and diagnosis\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. Pei, et al. \u003csup\u003e\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e conducted one of a few studies in COVID-19-affected Chinese HCWs that also included multiple symptom types (depression, anxiety, insomnia, and resilience) and identified four distinct profiles: low, mild, moderate, and high severity. Importantly, they did not include measures of occupational burnout in their phenotyping approach, neglecting a top-of-mind concern that may lead to attrition from Emergency Medicine and even poorer patient care\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eLeiter and Maslach \u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e identified five profiles in HCWs based on MBI subdomain scores, which were later replicated in an online survey study of Canadian nurses\u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e. The profiles include: burnout (high symptomatology across Emotional Exhaustion (EE), Depersonalization (DEP) and Personal Accomplishment (PA)); engagement (low symptomatology across EE, DEP and PA); and three intermediate profiles with varying patterns of low, medium, and high symptomatology across subdomains. In our study, we found two phenotypes that differed significantly in EE and DEP severity but not in PA. The consistent PA levels across phenotypes suggest that ED HCWs maintain work-related self-efficacy despite their exposure to trauma and demanding work conditions, including high work pace and shift work, that may lead to experiencing burnout in other domains. This seems to align with \u0026Ouml;nder and Basim \u003csup\u003e\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e, who highlighted that PA in nurses had a non-linear relationship with DEP and EE and may exist independently of the other domains. The role of PA remains poorly understood and should be further investigated\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eRegarding comorbid symptoms of burnout, mild depressive symptoms were observed in the High-Symptom Phenotype, and roughly one-third of HCWs in this phenotype reported PTSD symptom levels above the probable diagnostic threshold. These high severity levels may reflect the mental health risks of vicarious trauma exposure in the ED, and possibly the impact of working during the COVID-19 pandemic. For example, two latent profiles of post-traumatic stress symptoms (High- and Low-PTSS) were identified from an online survey of Brazilian HCWs affected by COVID-19\u003csup\u003e54\u003c/sup\u003e, underscoring the necessity to include PTSD symptomatology in analyses of HCW well-being.\u003c/p\u003e\u003cp\u003eUsing zero-shot prompting in an LLM, we accurately predicted these complex phenotypes from a single unstructured narrative about work-related future expectations. Our results show promising potential in predicting mental health outcomes using zero-shot prompting and are comparable to other studies. For instance, Ohse, et al. \u003csup\u003e\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e used GPT-4 to predict self-reported PHQ-9 scores from 84 transcribed clinical interviews in a general population with an F1-score of 76.0%. In addition, Danner, et al. \u003csup\u003e\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e aimed to identify risk for PTSD and depression using GPT-3.5 on semi-structured clinical interviews from N\u0026thinsp;=\u0026thinsp;193 veterans and general population individuals, predicting probable depression with an F1-score of 78.0%. Remarkably, our study uniquely classified mental health symptomatology based on a three-minute free-speech text about work-related future expectations rather than structured or semi-structured clinical interviews. We also predicted complex, heterogeneous symptomatology with reliable accuracy, adding complexity to our model.\u003c/p\u003e\u003cp\u003eIn a recent study, Lho et al.\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e applied multiple LLMs and embedding-based models to analyze self-concept narratives derived from sentence completion tests of psychiatric patients, achieving ROC-AUCs of 0.72\u0026ndash;0.75 for depression and 0.72\u0026ndash;0.73 for suicide risk using zero-shot LLMs. While this demonstrates the effectiveness of LLMs on clinically enriched data from individuals already engaged in psychiatric evaluation, our study extends this potential into a more subtle and challenging context. Specifically, we focused on high-functioning ED professionals who are trained to suppress or mask emotional distress\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e. The ability to identify latent psychopathology from unstructured narratives in response to one open-ended question highlights the discriminative capacity of LLMs in real-world screening applications where traditional assessments may miss early or hidden symptoms.\u003c/p\u003e\u003cp\u003ePrevious studies used unstructured text data, like social media posts and SMS from open-source forums, both specific or non-specific to mental health topics, to identify stress, depression and/or suicidality risks with zero-shot prompts in LLMs (e.g., ChatGPT-3.5 and Llama2.0), achieving lower to comparable accuracies (42.9%-72.4%)\u003csup\u003e59,60\u003c/sup\u003e. However, the lack of clinically validated measures to assess mental health in these datasets may limit their diagnostic accuracy and applicability in healthcare. A study in perinatal women achieved a low F1-score of 38.0% in predicting PTSD from free-speech, highlighting the potential challenges of using free-speech data to predict complex symptomatology, as well as the non-specialized nature of LLMs for clinical applications\u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e. Nevertheless, our results demonstrate the potential of LLMs to assess heterogeneous clinical phenotypes accurately.\u003c/p\u003e\u003cp\u003eUsing few-shot prompts has shown minor improvements in LLM performance for predicting mental health, as seen in studies by Xu, et al. \u003csup\u003e\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e and Yang, et al. \u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e. Current LLMs still face challenges with processing lengthy texts without losing focus\u003csup\u003e\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e. Since the responses to our open-ended, emotion-provoking question were approximately three minutes long, a few-shot approach may not be suitable for our current analysis. Notably, fine-tuned LLMs such as MentaLLaMA by Yang, et al. \u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e and Mental-LLM by Xu, et al. \u003csup\u003e\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e have shown a 5\u0026ndash;10% performance improvement compared to zero-shot prompting. However, these models were trained on non-clinically validated open-source text data, limiting their clinical relevance and applicability in our study\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e,\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e. Fine-tuning also requires a large amount of high-quality data and can compromise the model's interpretability, particularly in reasoning processes crucial for extracting key domain-specific indicators\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e,\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e. Due to these limitations, we have not yet fine-tuned our LLM but plan to investigate this approach further once more data becomes available.\u003c/p\u003e\u003cp\u003eWe identified several key domain-specific indicators from the narratives that were most important for the LLM\u0026rsquo;s reasoning to classify HCWs into the phenotypes. For the High-Symptom Phenotype, the most identified indicators were emotional exhaustion, feeling overwhelmed, emotional numbness, hopelessness, and stress/anxiety. The LLM seems to primarily detect symptom- and behavior-driven language indicative of distress. In contrast, for the Low-Symptom Phenotype, a positive outlook, a sense of control, and overall positive language were the most identified indicators. This suggests that the LLM extracts different aspects from the narratives to predict low symptomatology, which is not necessarily related to a clinical context.\u003c/p\u003e\u003cp\u003eXue, et al. \u003csup\u003e\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e found that nurses with higher burnout levels experienced worse impairments in envisioning positive future events, which can be driven by hope and resilience in facing the future\u003csup\u003e\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e,\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e\u003c/sup\u003e. Interpreting the subtle key indicators of language used by the LLM for predicting clinical phenotypes could offer valuable insights into risk and protective factors that might drive classification and provide transparency into the \u0026ldquo;black box\u0026rdquo; of the LLM\u0026rsquo;s decision-making process.\u003c/p\u003e\u003cp\u003eThe ability to identify domain-specific indicators is a key advantage of zero-shot models compared to fine-tuned models\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e, as it allows for nuanced, human-like interpretations of complex symptomatology.\u003c/p\u003e\u003cp\u003e\u003cb\u003eStudy implications\u003c/b\u003e\u003c/p\u003e\u003cp\u003eOur study introduces a proof-of-concept for zero-shot LLMs as a potential screener for burnout, depression, and PTSD symptoms among ED HCWs using a single narrative based on an open-ended question regarding work-related future expectations. The ability to accurately predict complex symptomatology from naturalistic speech could be a valuable addition to traditional clinical assessments, which are often hindered by concerns of stigma and confidentiality, and may allow HCWs to discuss confidential evaluations in a supportive and discrete way that reveals underlying strain, which is particularly important in healthcare settings\u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAs ED HCWs operate in a trauma-intensive and unpredictable environment, their risk for burnout\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, depression\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, and PTSD\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e is increased. By encouraging self-awareness of well-being among HCWs and reducing barriers to early intervention, objective tools could support in managing the demands of the HCW\u0026rsquo;s role while proactively addressing mental health needs. Recognizing and promoting HCW well-being and resilience may ultimately result in improved patient outcomes and quality of care.\u003c/p\u003e\u003cp\u003e\u003cb\u003eStrengths and limitations\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThis study offers important insights into high-functioning ED HCWs, who, despite their ability to maintain strong emotional control\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e,\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e,\u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e, exhibit a high prevalence of mental health disorders. Our study included a heterogeneous ED HCW population by encompassing a broad range of ED positions across various hospital systems, from public to academic, each with unique patient demographics, traumatic events, and working conditions. A methodological strength of this study is that zero-shot LLMs are easy to implement and do not require additional training, allowing accessibility for real-world applications\u003csup\u003e\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e. We leveraged the LLM\u0026rsquo;s reasoning capabilities to identify domain-specific indicators that may be important for mental health risk and resilience, indicating a step towards uncovering the \u0026ldquo;black box\u0026rdquo;. Also, the deterministic nature of Llama3 ensures consistent predictions across runs, making it a reliable tool for clinical screening and intervention planning, while its local deployment enhances data security and HIPAA compliance\u003csup\u003e\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e\u003c/sup\u003e. Furthermore, we used a short, open-ended question that does not rely on a clinician-administered structured interview, allowing HCWs to share their experiences more freely and possibly increase compliance.\u003c/p\u003e\u003cp\u003eDespite using cross-domain clustering, our clinical profiles were still dependent on self-report assessments, which can introduce potential biases, such as underreporting of symptomatology due to social desirability or the \u0026lsquo;'hero complex\u0026rsquo;'\u003csup\u003e\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e\u003c/sup\u003e commonly observed among HCWs\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. Additionally, selection bias may have influenced our results, as HCWs experiencing less burnout, such as early-career clinicians with fewer emotional and time burdens, may have been more likely to participate. Conversely, those with personal burnout experiences may have been motivated to contribute to the research. Furthermore, the cross-sectional design of this sub-study limited the investigation of long-term mental health trajectories. Despite these potential biases, our observed prevalence of burnout (32.3\u0026ndash;42.9%) was comparable to other studies\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e. We aim to apply longitudinal methodological approaches in the ongoing NIH-funded study Early Signs (R01HL156134) to examine symptom progression and early signs of deterioration. In addition, ED HCWs operating with exceptional emotional control in high-stakes environments often mask symptoms, which complicates the detection of symptoms from transcriptions alone\u003csup\u003e\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e,\u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e. We plan to extend our approach by incorporating multimodal measures such as speech features and facial expressivity, which may improve the prediction of clinical phenotypes and enable the capturing of subtle indicators. These will be further evaluated in R01HL156134.\u003c/p\u003e\u003cp\u003eImportantly, our model may be overfitted to language patterns specific to an ED context. Future studies should explore fine-tuning with larger, context-specific datasets and enhance the LLM\u0026rsquo;s adaptability to nuanced, domain-specific language. In addition, our LLM\u0026rsquo;s limitation to the English language raises questions about its generalizability. Testing across diverse languages, cultures, and populations is needed to ensure global applicability in healthcare settings, especially given the LLM\u0026rsquo;s hardware dependence. Finally, we were unable to test a hold-out sample in our clustering solution due to sample size constraints, which may have affected the stability of our findings. However, we balanced the number and size of clusters using the elbow method and regularization techniques to reduce the risk of overfitting. We aim to validate our approach on an independent dataset to strengthen the robustness of our clustering results.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eZero-shot prompts with LLMs have the potential to predict complex clinical phenotypes of mental health symptomatology, such as burnout, depression, and PTSD among ED HCWs based on a single unstructured narrative about work-related future expectations. This could offer a nuanced and objective tool for mental health screening and may foster a supportive, stigma-free environment for routine assessment of mental health in high-stress healthcare environments, empowering HCWs to take charge of their mental health. Future research should incorporate other essential indicators, such as risk and resilience factors, as well as temporal symptom dynamics, to refine and improve this tool, paving the way for scalable, person-centered mental health monitoring across diverse healthcare settings.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis study was supported by the National Heart, Lung, and Blood Institute [R01HL156134].\u0026nbsp;The authors would like to thank all clinicians who participated in this study as well as Joseph Lyu and Annie Farnsworth for their assistance with study enrollment and data collection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical declarations\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors and affiliations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDepartment of Psychiatry, New York University Grossman School of Medicine, New York, New York, USA\u003c/p\u003e\n\u003cp\u003eVictoria Mueller, Sarah B Birnbaum, Charlotte E Hilberdink, Yiwen Zhao, Joseph H Chang, Sapir Gershov, Katharina Schultebraucks\u003c/p\u003e\n\u003cp\u003eClinic for Psychiatry and Neurosciences, Charit\u0026eacute; \u0026ndash; Universit\u0026auml;tsmedizin Berlin, corporate member of Freie Universit\u0026auml;t Berlin and Humboldt Universit\u0026auml;t zu Berlin, Berlin, Germany\u003c/p\u003e\n\u003cp\u003eTolou Maslahati\u003c/p\u003e\n\u003cp\u003eDivision of Healthcare Delivery Science, Department of Population Health, New York University Grossman School of Medicine, New York, New York, USA\u003c/p\u003e\n\u003cp\u003eKatharina Schultebraucks\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eKS conceived and designed the current study; VM and SBB wrote the first draft of the manuscript and did the final edits under supervision of KS and CEH who offered critical feedback and revised the manuscript for important intellectual content; VM carried out the primary data analysis and interpretation under the supervision of KS; SBB and JHC performed data collection; YZ and SG assisted in data analysis; TM offered revisions on the manuscript. All authors approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll requests for raw and analyzed data and related materials, including programming code, will be reviewed by our legal departments (New York University Grossman School of Medicine) to verify whether the request is subject to any intellectual property or confidentiality constraints. Any data and materials that can be shared will be released via a material transfer agreement for noncommercial research purposes. Request should be addressed to the corresponding author (K.S.).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eCDC. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cdc.gov/nchs/data/nhamcs/web_tables/2021-nhamcs-ed-web-tables-508.pdf\u003c/span\u003e\u003cspan address=\"https://www.cdc.gov/nchs/data/nhamcs/web_tables/2021-nhamcs-ed-web-tables-508.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWong, A. H., Pacella-LaBarbara, M. L., Ray, J. M., Ranney, M. L. \u0026amp; Chang, B. P. Healing the Healer: Protecting Emergency Health Care Workers' Mental Health During COVID-19. \u003cem\u003eAnn Emerg Med\u003c/em\u003e 76, 379\u0026ndash;384 (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.annemergmed.2020.04.041\u003c/span\u003e\u003cspan address=\"10.1016/j.annemergmed.2020.04.041\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUddin, H., Hasan, M. K., Cuartas-Alvarez, T. \u0026amp; Castro-Delgado, R. Effects of mass casualty incidents on anxiety, depression, and post-traumatic stress disorder among doctors and nurses: a systematic review. \u003cem\u003ePublic Health\u003c/em\u003e 234, 132\u0026ndash;142 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.puhe.2024.06.001\u003c/span\u003e\u003cspan address=\"10.1016/j.puhe.2024.06.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMarvaldi, M., Mallet, J., Dubertret, C., Moro, M. R. \u0026amp; Guessoum, S. B. Anxiety, depression, trauma-related, and sleep disorders among healthcare workers during the COVID-19 pandemic: A systematic review and meta-analysis. \u003cem\u003eNeuroscience \u0026amp; Biobehavioral Reviews\u003c/em\u003e 126, 252\u0026ndash;264 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.neubiorev.2021.03.024\u003c/span\u003e\u003cspan address=\"10.1016/j.neubiorev.2021.03.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLabrague, L. J. Psychological resilience, coping behaviours and social support among health care workers during the COVID-19 pandemic: A systematic review of quantitative studies. \u003cem\u003eJournal of Nursing Management\u003c/em\u003e 29, 1893\u0026ndash;1905 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1111/jonm.13336\u003c/span\u003e\u003cspan address=\"10.1111/jonm.13336\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWatson, A. G., Saggar, V., MacDowell, C. \u0026amp; McCoy, J. V. Self-reported modifying effects of resilience factors on perceptions of workload, patient outcomes, and burnout in physician-attendees of an international emergency medicine conference. \u003cem\u003ePsychology, Health \u0026amp; Medicine\u003c/em\u003e 24, 1220\u0026ndash;1234 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/13548506.2019.1619785\u003c/span\u003e\u003cspan address=\"10.1080/13548506.2019.1619785\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGarc\u0026iacute;a-Izquierdo, M. \u0026amp; R\u0026iacute;os-R\u0026iacute;squez, M. I. The relationship between psychosocial job stress and burnout in emergency departments: An exploratory study. \u003cem\u003eNursing Outlook\u003c/em\u003e 60, 322\u0026ndash;329 (2012). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.outlook.2012.02.002\u003c/span\u003e\u003cspan address=\"10.1016/j.outlook.2012.02.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAdriaenssens, J., De Gucht, V. \u0026amp; Maes, S. Determinants and prevalence of burnout in emergency nurses: a systematic review of 25 years of research. \u003cem\u003eInt J Nurs Stud\u003c/em\u003e 52, 649\u0026ndash;661 (2015). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijnurstu.2014.11.004\u003c/span\u003e\u003cspan address=\"10.1016/j.ijnurstu.2014.11.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHu, Y. \u003cem\u003eet al.\u003c/em\u003e Relationship between workplace violence, job satisfaction, and burnout among healthcare workers in mobile cabin hospitals in China: Effects of perceived stress and work environment. \u003cem\u003ePrev Med Rep\u003c/em\u003e 40, 102667 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.pmedr.2024.102667\u003c/span\u003e\u003cspan address=\"10.1016/j.pmedr.2024.102667\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAljohani, B. \u003cem\u003eet al.\u003c/em\u003e Workplace violence in the emergency department: a systematic review and meta-analysis. \u003cem\u003ePublic Health\u003c/em\u003e 196, 186\u0026ndash;197 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/\u003c/span\u003e\u003cspan address=\"https://doi.org/https://doi.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.puhe.2021.02.009\u003c/span\u003e\u003cspan address=\"10.1016/j.puhe.2021.02.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKuhn, G. Circadian rhythm, shift work, and emergency medicine. \u003cem\u003eAnn Emerg Med\u003c/em\u003e 37, 88\u0026ndash;98 (2001). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1067/mem.2001.111571\u003c/span\u003e\u003cspan address=\"10.1067/mem.2001.111571\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDoan\u0026ndash;Wiggins, L. \u003cem\u003eet al.\u003c/em\u003e Practice Satisfaction, Occupational Stress, and Attrition of Emergency Physicians. \u003cem\u003eAcademic Emergency Medicine\u003c/em\u003e 2, 556\u0026ndash;563 (1995). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.\u003c/span\u003e\u003cspan address=\"https://doi.org/https://doi.\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003eorg/10.1111/j.1553-2712.1995.tb03261.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1553-2712.1995.tb03261.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAiken, L. H., Clarke, S. P., Sloane, D. M., Sochalski, J. \u0026amp; Silber, J. H. Hospital nurse staffing and patient mortality, nurse burnout, and job dissatisfaction. \u003cem\u003eJama\u003c/em\u003e 288, 1987\u0026ndash;1993 (2002). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jama.288.16.1987\u003c/span\u003e\u003cspan address=\"10.1001/jama.288.16.1987\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHooper, C., Craig, J., Janvrin, D. R., Wetsel, M. A. \u0026amp; Reimels, E. Compassion satisfaction, burnout, and compassion fatigue among emergency nurses compared with nurses in other selected inpatient specialties. \u003cem\u003eJ Emerg Nurs\u003c/em\u003e 36, 420\u0026ndash;427 (2010). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jen.2009.11.027\u003c/span\u003e\u003cspan address=\"10.1016/j.jen.2009.11.027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMatthews, L. R. \u003cem\u003eet al.\u003c/em\u003e PREVALENCE AND PREDICTORS OF POSTTRAUMATIC STRESS DISORDER, DEPRESSION, AND ANXIETY IN PERSONNEL WORKING IN EMERGENCY DEPARTMENT SETTINGS: A SYSTEMATIC REVIEW. \u003cem\u003eThe Journal of Emergency Medicine\u003c/em\u003e 62, 617\u0026ndash;635 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.jemermed.2021.09.010\u003c/span\u003e\u003cspan address=\"10.1016/j.jemermed.2021.09.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAwan, S. \u003cem\u003eet al.\u003c/em\u003e Frontiers | Suicide in Healthcare Workers: Determinants, Challenges, and the Impact of COVID-19. \u003cem\u003eFrontiers in Psychiatry\u003c/em\u003e 12 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyt.2021.792925\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2021.792925\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMarco, C. A., Larkin, G. L., Feeser, V. R., Monti, J. E. \u0026amp; Vearrier, L. Post-traumatic stress and stress disorders during the COVID‐19 pandemic: Survey of emergency physicians. \u003cem\u003eJournal of the American College of Emergency Physicians Open\u003c/em\u003e 1 (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/emp2.12305\u003c/span\u003e\u003cspan address=\"10.1002/emp2.12305\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBahadirli, S. \u0026amp; Sagaltici, E. Post-traumatic stress disorder in healthcare workers of emergency departments during the pandemic: A cross-sectional study. \u003cem\u003eThe American Journal of Emergency Medicine\u003c/em\u003e 50 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ajem.2021.08.027\u003c/span\u003e\u003cspan address=\"10.1016/j.ajem.2021.08.027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTrudgill, D. I. N., Gorey, K. M. \u0026amp; Donnelly, E. A. Prevalent posttraumatic stress disorder among emergency department personnel: rapid systematic review. \u003cem\u003eHumanities and Social Sciences Communications\u003c/em\u003e 7, 89 (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/s41599-020-00584-x\u003c/span\u003e\u003cspan address=\"10.1057/s41599-020-00584-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShanafelt, T. D. \u003cem\u003eet al.\u003c/em\u003e Special report: suicidal ideation among American surgeons. \u003cem\u003eArch Surg\u003c/em\u003e 146, 54\u0026ndash;62 (2011). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/archsurg.2010.292\u003c/span\u003e\u003cspan address=\"10.1001/archsurg.2010.292\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDavidson, J. E., Proudfoot, J., Lee, K., Terterian, G. \u0026amp; Zisook, S. A Longitudinal Analysis of Nurse Suicide in the United States (2005\u0026ndash;2016) With Recommendations for Action. \u003cem\u003eWorldviews Evid Based Nurs\u003c/em\u003e 17, 6\u0026ndash;15 (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/wvn.12419\u003c/span\u003e\u003cspan address=\"10.1111/wvn.12419\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDutheil, F. \u003cem\u003eet al.\u003c/em\u003e Suicide among physicians and health-care workers: A systematic review and meta-analysis. \u003cem\u003ePLOS ONE\u003c/em\u003e 14, e0226361 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0226361\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0226361\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMachi, M. S. \u003cem\u003eet al.\u003c/em\u003e The relationship between shift work, sleep, and cognition in career emergency physicians. \u003cem\u003eAcad Emerg Med\u003c/em\u003e 19, 85\u0026ndash;91 (2012). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1553-2712.2011.01254.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1553-2712.2011.01254.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJohn, A. \u003cem\u003eet al.\u003c/em\u003e The influence of burnout on cardiovascular disease: a systematic review and meta-analysis. \u003cem\u003eFront Psychiatry\u003c/em\u003e 15, 1326745 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyt.2024.1326745\u003c/span\u003e\u003cspan address=\"10.3389/fpsyt.2024.1326745\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSumner, J. A. \u003cem\u003eet al.\u003c/em\u003e Trauma Exposure and Posttraumatic Stress Disorder Symptoms Predict Onset of Cardiovascular Events in Women. \u003cem\u003eCirculation\u003c/em\u003e 132, 251\u0026ndash;259 (2015). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/circulationaha.114.014492\u003c/span\u003e\u003cspan address=\"10.1161/circulationaha.114.014492\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDeligkaris, P., Panagopoulou, E., Montgomery, A. J. \u0026amp; Masoura, E. Job burnout and cognitive functioning: A systematic review. \u003cem\u003eWork \u0026amp; Stress\u003c/em\u003e 28, 107\u0026ndash;123 (2014). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02678373.2014.909545\u003c/span\u003e\u003cspan address=\"10.1080/02678373.2014.909545\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSandstr\u0026ouml;m, A., Rhodin, I. N., Lundberg, M., Olsson, T. \u0026amp; Nyberg, L. Impaired cognitive performance in patients with chronic burnout syndrome. \u003cem\u003eBiol Psychol\u003c/em\u003e 69, 271\u0026ndash;279 (2005). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.biopsycho.2004.08.003\u003c/span\u003e\u003cspan address=\"10.1016/j.biopsycho.2004.08.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGavelin, H. M. \u003cem\u003eet al.\u003c/em\u003e Cognitive function in clinical burnout: A systematic review and meta-analysis. \u003cem\u003eWork \u0026amp; Stress\u003c/em\u003e 36, 86\u0026ndash;104 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02678373.2021.2002972\u003c/span\u003e\u003cspan address=\"10.1080/02678373.2021.2002972\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePereira-Lima, K. \u003cem\u003eet al.\u003c/em\u003e Association Between Physician Depressive Symptoms and Medical Errors: A Systematic Review and Meta-analysis. \u003cem\u003eJAMA Netw Open\u003c/em\u003e 2, e1916097 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jamanetworkopen.2019.16097\u003c/span\u003e\u003cspan address=\"10.1001/jamanetworkopen.2019.16097\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWeigl, M. \u0026amp; Schneider, A. Associations of work characteristics, employee strain and self-perceived quality of care in Emergency Departments: A cross-sectional study. \u003cem\u003eInternational Emergency Nursing\u003c/em\u003e 30, 20\u0026ndash;24 (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.ienj.2016.07.002\u003c/span\u003e\u003cspan address=\"10.1016/j.ienj.2016.07.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLu, D. W., Dresden, S., McCloskey, C., Branzetti, J. \u0026amp; Gisondi, M. A. Impact of Burnout on Self-Reported Patient Care Among Emergency Physicians. \u003cem\u003eWest J Emerg Med\u003c/em\u003e 16, 996\u0026ndash;1001 (2015). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5811/westjem.2015.9.27945\u003c/span\u003e\u003cspan address=\"10.5811/westjem.2015.9.27945\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDe Stefano, C. \u003cem\u003eet al.\u003c/em\u003e Effect of emergency physician burnout on patient waiting times. \u003cem\u003eIntern Emerg Med\u003c/em\u003e 13, 421\u0026ndash;428 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11739-017-1706-9\u003c/span\u003e\u003cspan address=\"10.1007/s11739-017-1706-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGualano, M. R. \u003cem\u003eet al.\u003c/em\u003e The Burden of Burnout among Healthcare Professionals of Intensive Care Units and Emergency Departments during the COVID-19 Pandemic: A Systematic Review. \u003cem\u003eInternational Journal of Environmental Research and Public Health 2021, Vol. 18, Page 8172\u003c/em\u003e 18 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijerph18158172\u003c/span\u003e\u003cspan address=\"10.3390/ijerph18158172\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWijeratne, C., Johnco, C., Draper, B. \u0026amp; Earl, J. Doctors\u0026rsquo; reporting of mental health stigma and barriers to help-seeking. \u003cem\u003eOccupational Medicine\u003c/em\u003e 71, 366\u0026ndash;374 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/occmed/kqab119\u003c/span\u003e\u003cspan address=\"10.1093/occmed/kqab119\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAuth, N. M., Booker, M. J., Wild, J. \u0026amp; Riley, R. Mental health and help seeking among trauma-exposed emergency service staff: a qualitative evidence synthesis. \u003cem\u003eBMJ Open\u003c/em\u003e 12, e047814 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmjopen-2020-047814\u003c/span\u003e\u003cspan address=\"10.1136/bmjopen-2020-047814\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTaylor, N. W., Porter, C., Rivera-Rodriguez, M., Miller, I. S. K. \u0026amp; Desmarais, N. Mental Health Disclosure Questions on Medical Licensure Applications: Implications for Medical Students, Residents, and Physicians. \u003cem\u003eAcademic Medicine\u003c/em\u003e 97 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMcGarry, E. \u0026amp; O\u0026rsquo;Connor, L. Assessing burnout rates and contributing factors in emergency medical services clinicians. \u003cem\u003eJournal of Workplace Behavioral Health\u003c/em\u003e 39 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/15555240.2023.2292119\u003c/span\u003e\u003cspan address=\"10.1080/15555240.2023.2292119\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHowlett, M. \u003cem\u003eet al.\u003c/em\u003e Burnout in emergency department healthcare professionals is associated with coping style: a cross-sectional survey. \u003cem\u003eEmergency Medicine Journal\u003c/em\u003e 32 (2015). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/emermed-2014-203750\u003c/span\u003e\u003cspan address=\"10.1136/emermed-2014-203750\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCatalogue of Bias Collaboration, S. E., Brassey J, Mahtani K. Recall bias. In: Catalogue Of Bias 2017. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.catalogueofbiases.org/biases/recall-bias\u003c/span\u003e\u003cspan address=\"https://www.catalogueofbiases.org/biases/recall-bias\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVan de Mortel, T. Faking it: Social desirability response bias in self-report research. \u003cem\u003eThea van de Mortel\u003c/em\u003e 25 (2008).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMarshall, R. E. \u003cem\u003eet al.\u003c/em\u003e Mental health screening amongst police officers: factors associated with under-reporting of symptoms. \u003cem\u003eBMC Psychiatry 2021 21:1\u003c/em\u003e 21 (2021-03-08). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12888-021-03125-1\u003c/span\u003e\u003cspan address=\"10.1186/s12888-021-03125-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCH, W. \u003cem\u003eet al.\u003c/em\u003e Importance of anonymity to encourage honest reporting in mental health screening after combat deployment - PubMed. \u003cem\u003eArchives of general psychiatry\u003c/em\u003e 68 (2011 Oct). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/archgenpsychiatry.2011.112\u003c/span\u003e\u003cspan address=\"10.1001/archgenpsychiatry.2011.112\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMaslach, C., Jackson, S. \u0026amp; Leiter, M. Maslach burnout inventory (MBI) Manual (1996\u0026ndash;2016). 4th Ed., Menlo Park, CA., Mind Garden, Inc. (1996\u0026ndash;2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKroenke, K. \u003cem\u003eet al.\u003c/em\u003e The PHQ-8 as a measure of current depression in the general population. \u003cem\u003eJournal of affective disorders\u003c/em\u003e 114(1\u0026ndash;3), 163\u0026ndash;173 (2009).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWeathers, F. \u003cem\u003eet al.\u003c/em\u003e The PTSD Checklist for DSM-5 (PCL-5). Scale available from the National Center for PTSD at www ptsd va gov. (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMacQueen, J. in \u003cem\u003eProceedings of the fifth Berkeley symposium on mathematical statistics and probability.\u003c/em\u003e 281\u0026ndash;297 (Oakland, CA, USA).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLloyd, S. Least squares quantization in PCM. \u003cem\u003eIEEE Transactions on Information Theory\u003c/em\u003e 28, 129\u0026ndash;137 (1982). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TIT.1982.1056489\u003c/span\u003e\u003cspan address=\"10.1109/TIT.1982.1056489\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAI@Meta. Llama 3 Model Card. (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKuhn, M. \u0026amp; Johnson, K. Applied Predictive Modeling. (2013). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-1-4614-6849-3\u003c/span\u003e\u003cspan address=\"10.1007/978-1-4614-6849-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLeiter, M. P. \u0026amp; Maslach, C. Latent burnout profiles: A new approach to understanding the burnout experience. \u003cem\u003eBurnout Research\u003c/em\u003e 3, 89\u0026ndash;100 (2016). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/\u003c/span\u003e\u003cspan address=\"https://doi.org/https://doi.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.burn.2016.09.001\u003c/span\u003e\u003cspan address=\"10.1016/j.burn.2016.09.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHavaei, F., Ma, A., Leiter, M. \u0026amp; Gear, A. Describing the Mental Health State of Nurses in British Columbia: A Province-Wide Survey Study. \u003cem\u003eHealthc Policy\u003c/em\u003e 16, 31\u0026ndash;45 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.12927/hcpol.2021.26500\u003c/span\u003e\u003cspan address=\"10.12927/hcpol.2021.26500\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e\u0026Ouml;nder, \u0026Ccedil;. \u0026amp; Basim, N. Examination of developmental models of occupational burnout using burnout profiles of nurses. \u003cem\u003eJournal of Advanced Nursing\u003c/em\u003e 64, 514\u0026ndash;523 (2008). https://doi.org/\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1365-2648.2008.04818.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1365-2648.2008.04818.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePei, Y., Li, Y., Wu, L., Xi, J. \u0026amp; Zhou, N. Psychological distress during the COVID-19 pandemic normalization phase in China: A multigroup latent profile analysis among healthcare workers and the general population. \u003cem\u003eInternational Journal of Disaster Risk Reduction\u003c/em\u003e 108 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijdrr.2024.104567\u003c/span\u003e\u003cspan address=\"10.1016/j.ijdrr.2024.104567\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBraule Pinto, A. L. d. C. \u003cem\u003eet al.\u003c/em\u003e Longitudinal profile of post-traumatic symptoms in HealthCare Workers during COVID-19 pandemic: A latent transition model. \u003cem\u003eJournal of Psychiatric Research\u003c/em\u003e 168, 230\u0026ndash;239 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.jpsychires.2023.10.031\u003c/span\u003e\u003cspan address=\"10.1016/j.jpsychires.2023.10.031\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOhse, J. \u003cem\u003eet al.\u003c/em\u003e Zero-Shot Strike: Testing the generalisation capabilities of out-of-the-box LLM models for depression detection. \u003cem\u003eComputer Speech \u0026amp; Language\u003c/em\u003e 88 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.csl.2024.101663\u003c/span\u003e\u003cspan address=\"10.1016/j.csl.2024.101663\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDanner, M. \u003cem\u003eet al.\u003c/em\u003e in 2023 \u003cem\u003e62nd Annual Conference of the Society of Instrument and Control Engineers (SICE).\u003c/em\u003e 1290\u0026ndash;1296.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLho, S. K. \u003cem\u003eet al.\u003c/em\u003e Large Language Models and Text Embeddings for Detecting Depression and Suicide in Patient Narratives. \u003cem\u003eJAMA Network Open\u003c/em\u003e 8, e2511922 (2025). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jamanetworkopen.2025.11922\u003c/span\u003e\u003cspan address=\"10.1001/jamanetworkopen.2025.11922\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKadović, M. \u003cem\u003eet al.\u003c/em\u003e Ability of Emotional Regulation and Control as a Stress Predictor in Healthcare Professionals. \u003cem\u003eInternational Journal of Environmental Research and Public Health\u003c/em\u003e 2023, \u003cem\u003eVol. 20, Page 541\u003c/em\u003e 20 (2022-12-29). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijerph20010541\u003c/span\u003e\u003cspan address=\"10.3390/ijerph20010541\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang, K. \u003cem\u003eet al.\u003c/em\u003e in \u003cem\u003eProceedings of the ACM Web Conference 2024\u003c/em\u003e 4489\u0026ndash;4500 (Association for Computing Machinery, Singapore, Singapore, 2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu, X. \u003cem\u003eet al.\u003c/em\u003e Mental-LLM: Leveraging Large Language Models for Mental Health Prediction via Online Text Data. \u003cem\u003eProc. ACM Interact. Mob. Wearable Ubiquitous Technol.\u003c/em\u003e 8, Article 31 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3643540\u003c/span\u003e\u003cspan address=\"10.1145/3643540\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBartal, A., Jagodnik, K. M., Chan, S. J. \u0026amp; Dekel, S. AI and narrative embeddings detect PTSD following childbirth via birth stories. \u003cem\u003eScientific Reports\u003c/em\u003e 14, 8336 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-024-54242-2\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-54242-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKojima, T., Gu, S. S., Reid, M., Matsuo, Y. \u0026amp; Iwasawa, Y. in \u003cem\u003eProceedings of the 36th International Conference on Neural Information Processing Systems\u003c/em\u003e Article 1613 (Curran Associates Inc., New Orleans, LA, USA, 2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePouramini, A. \u0026amp; Faili, H. Matching tasks to objectives: Fine-tuning and prompt-tuning strategies for encoder-decoder pre-trained language models. \u003cem\u003eApplied Intelligence\u003c/em\u003e 54, 9783\u0026ndash;9810 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10489-024-05660-2\u003c/span\u003e\u003cspan address=\"10.1007/s10489-024-05660-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXue, B. \u003cem\u003eet al.\u003c/em\u003e Phenomenological characteristics of autobiographical future thinking in nurses with burnout: a case-control study. \u003cem\u003eFrontiers in Psychology\u003c/em\u003e 14 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyg.2023.1216036\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2023.1216036\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYıldırım, M., Kaynar, \u0026Ouml;., Chirico, F. \u0026amp; Magnavita, N. Resilience and Extrinsic Motivation as Mediators in the Relationship between Fear of Failure and Burnout. \u003cem\u003eInternational Journal of Environmental Research and Public Health\u003c/em\u003e 20, 5895 (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePharris, A. B., Munoz, R. T. \u0026amp; Hellman, C. M. Hope and resilience as protective factors linked to lower burnout among child welfare workers. \u003cem\u003eChildren and Youth Services Review\u003c/em\u003e 136, 106424 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://doi.org/10.1016/j.childyouth.2022.106424\u003c/span\u003e\u003cspan address=\"10.1016/j.childyouth.2022.106424\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMalinowska-Lipień, I. \u003cem\u003eet al.\u003c/em\u003e Emotional Control among Nurses against Work Conditions and the Support Received during the SARS-CoV-2 Pandemic. \u003cem\u003eInternational Journal of Environmental Research and Public Health\u003c/em\u003e 2021, \u003cem\u003eVol. 18, Page 9415\u003c/em\u003e 18 (2021-09-06). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijerph18179415\u003c/span\u003e\u003cspan address=\"10.3390/ijerph18179415\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHayward, R. M., Tuckey, M. R. \u0026amp; Renae Maree Hayward, M. R. T. Emotions in uniform: How nurses regulate emotion at work via emotional boundaries. \u003cem\u003eHuman Relations\u003c/em\u003e 64 (2011-11-08). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0018726711419539\u003c/span\u003e\u003cspan address=\"10.1177/0018726711419539\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePark, Y.-J. \u003cem\u003eet al.\u003c/em\u003e Assessing the research landscape and clinical utility of large language models: a scoping review. \u003cem\u003eBMC Medical Informatics and Decision Making\u003c/em\u003e 2024 \u003cem\u003e24:1\u003c/em\u003e 24 (2024-03-12). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12911-024-02459-6\u003c/span\u003e\u003cspan address=\"10.1186/s12911-024-02459-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCox, C. L. \u0026lsquo;Healthcare Heroes\u0026rsquo;: problems with media focus on heroism from healthcare workers during the COVID-19 pandemic. \u003cem\u003eJournal of Medical Ethics\u003c/em\u003e 46, 510\u0026ndash;513 (2020). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/medethics-2020-106398\u003c/span\u003e\u003cspan address=\"10.1136/medethics-2020-106398\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 and 2 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Clinical Phenotypes, Healthcare Workers (HCWs), Large Language Model (LLM), Emergency Department (ED), Burnout","lastPublishedDoi":"10.21203/rs.3.rs-7087333/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7087333/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe emergency department (ED) is crucial to the healthcare system. ED healthcare workers (HCWs) play a vital role in providing essential healthcare under demanding conditions. Traditional mental health assessment tools often rely on clinical interviews and psychometric assessments, which can be time-consuming, costly, and subject to biases. We aimed to identify heterogenous clinical profiles using a person-centered clustering approach based on burnout, depression, and PTSD symptomatology. We additionally investigated whether these phenotypes could be predicted by narratives about work-related future expectations using a Large Language Model (LLM). Based on n\u0026thinsp;=\u0026thinsp;199 ED HCWs from an ongoing NIH-funded study (R01HL156134), \u003cem\u003ek\u003c/em\u003e-means clustering revealed a High- and Low-Symptom Phenotype, significantly differentiating severity levels. Zero-shot LLM prompt engineering accurately predicted those clinical phenotypes from work-related narratives (accuracy\u0026thinsp;=\u0026thinsp;70.9%; F1-score\u0026thinsp;=\u0026thinsp;71.8%; sensitivity\u0026thinsp;=\u0026thinsp;77.1%) based on key domain-specific indicators identified in the LLM\u0026rsquo;s reasoning. Our approach leverages LLMs based on unstructured narratives, offering an objective, time-efficient alternative that enhances early risk stratification and fosters a stigma-free environment for mental health assessment in high-stress healthcare settings, revealing subtle but meaningful variations in symptomatology. Future research should incorporate indicators such as risk factors and symptom dynamics to refine this tool for scalable, person-centered mental health monitoring across diverse healthcare settings.\u003c/p\u003e","manuscriptTitle":"Large Language Models for Accurate Mental Health Screening: Identifying Clinical Phenotypes in ED Healthcare Workers","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-18 16:58:27","doi":"10.21203/rs.3.rs-7087333/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fad86167-4340-4e17-a6f5-d952fe57cff4","owner":[],"postedDate":"July 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":51588187,"name":"Health sciences/Diseases/Psychiatric disorders"},{"id":51588188,"name":"Biological sciences/Psychology/Human behaviour"},{"id":51588189,"name":"Health sciences/Biomarkers/Predictive markers"}],"tags":[],"updatedAt":"2025-08-11T09:47:19+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-18 16:58:27","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7087333","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7087333","identity":"rs-7087333","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}