Faculty Bibliography

BACKGROUND:Health care workers (HCWs) face sustained psychological demands that place them at heightened risk for burnout and posttraumatic stress disorder (PTSD). However, assessing psychological distress in this population remains challenging because of stigma, underreporting, and the limitations of self-report tools. Although nonverbal behaviors such as facial expressions and gaze hold diagnostic promise, most approaches overlook the fine-grained, temporal fluctuations in these signals. In this study, we focused on microbehavior intervals-brief, involuntary changes in multimodal nonverbal signals-that emerge during emotion-eliciting interviews. OBJECTIVE:This study aimed to determine whether microbehavior intervals improve the discrimination of psychological distress profiles among HCWs with symptoms of burnout and PTSD. METHODS:HCWs participated in a semistructured interview that included 5 work-related, emotionally charged questions and that was recorded via Webex (online video platform). Participants also completed validated questionnaires for burnout (Maslach Burnout Inventory General Survey 9-item) and PTSD (PTSD checklist for Diagnostic and Statistical Manual, 5th edition). Recordings were analyzed with computer vision models to generate time-series data of facial expressions, head movement, gaze, body posture, and hand gestures. An unsupervised anomaly detection model (MOMENT [a Family of Open Time-Series Foundation Models]) isolated microbehavior intervals without requiring manual labels. Features derived from these intervals were used to train a deep learning classifier that predicted 4 symptom classes of psychological distress: "moderate-severe burnout," "subthreshold-provisional PTSD," "burnout+PTSD," and "resilient." We conducted an ablation study by systematically removing one behavioral data stream at a time. Finally, we conducted an explainability analysis to characterize the features driving model predictions. RESULTS:-score. The explainability analysis revealed distinct temporal patterns across symptom classes, with irregularity and variability in microbehaviors emerging as key predictors. CONCLUSIONS:Focusing on microbehavior intervals yields a scalable, interpretable, and annotation-free framework for detecting psychological distress from nonverbal signals. By moving from whole-video features to fine-grained multimodal temporal modeling, we successfully captured subtle, involuntary fluctuations in nonverbal responses to emotion-eliciting questions. This multimodal approach enables an objective, robust, and explainable assessment of psychological distress and offers a promising complement to conventional psychometric assessments.

Understanding trauma-related psychopathology increasingly requires tools capable of capturing how symptoms and underlying mechanisms unfold across real-world contexts. Traditional assessments, which depend on retrospective recall and intermittent clinical contact, are limited in their ability to reflect the dynamic and context-sensitive processes that shape trauma-related mental health outcomes. Emerging digital health technologies and computational approaches address this gap by enabling the continuous measurement of behavior and physiology, whereas biological markers provide mechanistic insights into processes such as stress-related neuroendocrine, autonomic, immune, and sleep-wake functioning that reflect the biological embedding of trauma over time. This review integrates recent advances in digital and biological measurement in posttraumatic stress disorder (PTSD) and outlines principles for combining these approaches within computational frameworks. We emphasize the importance of grounding digital proxies in established biological theory; modeling intraindividual dynamics rather than relying solely on cross-sectional group differences; and integrating multimodal behavioral, physiological, and contextual data. We further highlight essential considerations for ethical, inclusive, and methodologically rigorous implementation, including privacy-by-design, data quality, and clinical interpretability. Together, these approaches support a more mechanistic, dynamic, and personalized understanding of trauma-related risk and recovery, with the potential to enhance early risk stratification and guide preventive and individualized models of care.

Healthcare workers (HCWs) in emergency departments face significant mental health risk due to chronic stressors and repeated trauma, yet symptom underreporting and bias in self-reports hinder accurate assessments. Expressive flexibility, the ability to dynamically modulate and recover from stressor-related changes in emotional arousal as reflected in observable behavior, has been linked to resilience. This NIH-funded study (R01HL156134) utilized digital phenotyping and computer vision to analyze dynamic facial expressivity during video-recorded interviews about work-related stressful situations with 240 HCWs (278 assessments). Participants additionally completed validated questionnaires to assess burnout, PTSD, depression, anxiety, and resilience. Latent profile analysis revealed two clinical phenotypes: At-risk (57.6%) and Resilient/Adaptive (42.4%). Machine learning models demonstrated high classification performance (accuracy = 0.83 ± 0.06, F1-score = 0.87 ± 0.05). Our findings indicate that digital biomarkers of temporal facial dynamics may serve as objective behavioral proxies of expressive flexibility, potentially capturing dynamics consistent with underlying stress-regulatory processes. These findings highlight their potential to improve identification of resilience-related phenotypes and support well-being and mental health in HCWs.

Sex differences in fear memory are well documented, yet the role of menstrual cycle phases and ovarian hormone dynamics remains unclear. Here, we investigated the individual and combined effects of progesterone and estradiol on fear acquisition and within-session fear extinction in both humans and mice. Human participants included men, women using oral contraceptives, and naturally cycling women across three menstrual phases. Our animal experiment included males and naturally cycling females tested across all estrous stages. Results in humans show an association between high estradiol levels and enhanced within-session extinction, whereas progesterone shows no direct effect. To further explore the relationship between sex hormones and within-session fear extinction, a machine-learning approach (Histogram-based Gradient Boosting Regression Tree) was followed. This showed that the facilitating effect of estradiol on within-session fear extinction is potentiated by its interactions with progesterone. Specifically, a high progesterone to estradiol ratio measured before extinction predicted enhanced extinction in both humans and mice. These findings identify the progesterone to estradiol ratio as a potential translational biomarker of fear extinction, with relevance to sex-informed treatments for fear-related disorders, and uncover a key memory process shared across species.

BACKGROUND:While previous literature has established a link between psychopathology symptom trajectories and psychosocial risk, most studies have not extended into adulthood or incorporated comprehensive risk measures. We aimed to identify symptom trajectories for externalizing and internalizing symptoms over a 25-year span and predict them using machine learning. METHODS:We used data from a longitudinal birth cohort study (N = 317, 171 females). Psychopathology symptoms were measured between age 8 and 33. Symptom trajectories were identified using growth mixture models. An extensive battery of risk and protective measures over the lifespan (e.g., polygenic scores, psychosocial risk factors, temperament) was used to train machine learning-based classification models. RESULTS:We identified three trajectories for both externalizing and internalizing symptoms (entropy > 0.8, likelihood ratio test p < .05): Low, increasing (late-adolescent emergence), and decreasing (childhood-limited) symptoms classes. Both genetic (e.g., polygenic scores for cognition and mental disorders) and environmental factors (e.g., adverse family characteristics, poor social functioning) predicted symptom trajectories. For externalizing symptoms trajectories, random forest achieved fair discriminatory power with a multiclass macro-average AUC of 0.77 and a multiclass micro-average of 0.88. For internalizing symptoms trajectories, logistic regression achieved modest discriminatory power with a multiclass macro-average AUC of 0.75 and a multiclass micro-average of 0.83. CONCLUSIONS:Psychopathology risk trajectories vary across development and are influenced by diverse risk factors. Understanding these risk factors can support early identification of at-risk individuals and inform targeted prevention and intervention efforts.

Many individuals hospitalized due to severe viral infections develop mental and physical sequelae, which could potentially be prevented by targeted interventions for those at risk. Our goal was to develop and externally validate an algorithm for predicting mental and physical symptoms after SARS-CoV-2 hospitalization utilizing routinely collected clinical data. Participants were included from two independent samples of the German National Pandemic Cohort Network (NAPKON): a model development sample (SUEP; N = 451; mean age: 55.6 ± 15.3; 36.2% female) and an external validation sample (HAP: N = 158; mean age: 55.1 ± 12.1; 39.9% female). Machine learning models leveraging demographic, clinical and biological variables collected at the time of admission were employed to predict Patient-Reported Outcomes Measurement Information System scores (PROMIS) across 7 domains (physical function, anxiety, depression, fatigue, sleep disturbance, ability to participate in social roles and activities, and pain) three months after SARS-CoV-2 hospitalization. Shapley Additive exPlanation values were used to provide interpretable information about key predictive factors. Approximately 15-20% of participants reported moderate to severe impairment in at least one PROMIS domain three months after hospitalization. For the mental health composite score, the best-performing model achieved RMSE = 1.833 ± 0.341 and R² = 0.927 ± 0.031 in SUEP and RMSE = 3.131 and R² = 0.893 in HAP. For the physical health composite, the best-performing model achieved RMSE = 2.908 ± 0.703 and R² = 0.824 ± 0.052 in SUEP and RMSE = 3.019 and R² = 0.850 in HAP. Furthermore, the models achieved high predictive performance across all individual PROMIS domain scores in both samples. We provide an externally validated methodology for accurately predicting mental and physical symptomatology following hospitalization due to a severe viral infection. This approach may facilitate the development of a brief risk stratification tool at the point of hospitalization, enabling early identification of at-risk patients, improving the prediction accuracy of subsequent psychological and physical sequelae, and supporting timely preventive interventions.

Artificial intelligence (AI) and machine learning (ML) have seen remarkable growth in mental health applications over the past few decades, demonstrating significant potential to transform psychiatric care. Despite these advancements, the translation of AI systems into clinical practice remains fraught with challenges. This Perspective examines critical hurdles in psychiatric AI research, emphasizing limitations in research rigor, model reliability, interpretability, clinical utility, and ethical considerations. We argue that a human-assisted AI framework-incorporating incremental feedback, self-adaptation, and dynamic collaboration-can address biases, enhance transparency, and build trust in AI systems. Moreover, initiatives in clinical education, cultural adaptation, and data/software sharing are essential to fostering public engagement, data transparency, and research reproducibility. By focusing on these areas, we aim to bridge the gap between AI potential and its successful, ethical implementation in mental health care, guiding the development of trustworthy, effective, and culturally adaptive AI-powered psychiatric tools.

INTRODUCTION/BACKGROUND:Suicidal thoughts and behaviours (STB) are a critical public health concern, with 700 000 deaths by suicide each year. The period immediately following hospital discharge is associated with an elevated risk for suicide. Monitoring suicidal ideations throughout this period is therefore critical. However, its highly dynamic nature limits the utility of traditional risk assessments through infrequent outpatient visits. Recent advancements in ambulatory assessments and multimodal predictive approaches offer a promising new avenue. Hence, the present study aims to examine how psychological, linguistic, neurobiological and smartphone-based characteristics relate to suicidal ideation and to improve STB monitoring through a deep phenotyping approach. METHODS AND ANALYSIS/METHODS:In this interdisciplinary, multicentre, prospective observational study, we plan to recruit a total of 200 inpatients with current and/or past STB. The study comprises the following components: (1) a baseline assessment, conducted while participants are still in the hospital. This includes interviews, an electroencephalography recording, a video-recorded verbal task and self-report questionnaires; (2) data collection through a smartphone application during the first 4 weeks after hospital discharge with two active collection weeks of five daily ecological momentary assessments and two 1 min video diaries every other day, as well as smartphone passive sensing for 28 consecutive days and (3) two follow-up assessments, 4 weeks and 3 months after discharge. The primary outcome is self-reported suicidal ideation after hospital discharge. ETHICS AND DISSEMINATION/BACKGROUND:The Ethics Committee of the Faculty of Arts and Social Sciences of the University of Zurich, Switzerland, approved the study for the Zurich and Basel sites (Ref: 22.09.19). Approval for the New York Site was granted by the Institutional Review Board of NYU Langone Health (i23-00366). Study findings will be disseminated via peer-reviewed, open-access publications, conference presentations, patient and public events, and dedicated social media outlets. TRIAL REGISTRATION NUMBER/BACKGROUND:CRSII5_205913.

Frontline caregivers, such as nurses and physicians, face heightened risk of intrusive memories, a core symptom of posttraumatic stress disorder. Understanding intrusions is key to protecting workers' mental health and sustaining effective patient care. Emergency department (ED) staff (N = 331) were assessed before starting ED work and three and six months later. Baseline measures included demographics, prior trauma, cognitive and emotional processing, and polymorphisms of the memory-related NR3C1, KIBRA, and ADRA2B genes as predictors of intrusions at three months, i.e., peak ED stress exposure. We examined intrusion occurrence, frequency, distress, and content, and investigated whether experiencing intrusions moderated changes in mental health from baseline to follow-ups. Intrusions were prevalent and often concerned potential patient care errors. Being a carrier of the G allele of the BCL1 polymorphism of the NR3C1 gene and ruminating were associated with intrusions at three months, while cognitive flexibility, emotional suppression, and emotion-focused positive rumination appeared protective. From baseline to three and six months, participants' anxiety increased, and those experiencing intrusions reported significantly lower work performance and engagement. Our results underline the psychological costs of high-stakes ED settings and suggest improvements to error culture and emotion regulation training as promising avenues to support staff and ultimately patients.

INTRODUCTION/UNASSIGNED:Traumatic experiences are thought to alter memory acquisition and consolidation. Cognitive models of PTSD suggest that voluntary and involuntary memories after trauma can be independently addressed through interventions. The administration of oxytocin before exposure to a trauma film led to more intrusive (involuntary) memories than placebo. The effect of oxytocin on voluntary memory of the traumatic film, however, remains unclear. The current study aimed to assess whether intranasal oxytocin administration facilitates forced-choice recognition memory after a trauma film paradigm. MATERIAL & METHODS/UNASSIGNED:We performed a pooled analysis of two randomized, double-blind, placebo-controlled studies (N = 437) to assess the impact of intranasal oxytocin administration on declarative memory. Participants received 24 I.U. of oxytocin, either 40 minutes before a trauma film paradigm or immediately afterward. We applied a forced-choice recognition task seven days after the trauma film paradigm. The task comprised pre-, peri, and post-trauma film scene details. RESULTS/UNASSIGNED:<.001). DISCUSSION/UNASSIGNED:Although the exogenous oxytocin administration before a trauma film has been shown to influence the acquisition of intrusive memories, it does not seem to affect the recognition memory of trauma film details. That aligns with cognitive models of PTSD, suggesting that voluntary and involuntary memory after trauma can be independently addressed through experimental interventions.