Faculty Bibliography

Large language models (LLMs) offer promising applications in mental health care to address gaps in treatment and research. By leveraging clinical notes and transcripts as data, LLMs could improve diagnostics, monitoring, prevention, and treatment of mental health conditions. However, several challenges persist, including technical costs, literacy gaps, risk of biases, and inequalities in data representation. In this Viewpoint, we propose a sociocultural-technical approach to address these challenges. We highlight five key areas for development: (1) building a global clinical repository to support LLMs training and testing, (2) designing ethical usage settings, (3) refining diagnostic categories, (4) integrating cultural considerations during development and deployment, and (5) promoting digital inclusivity to ensure equitable access. We emphasise the need for developing representative datasets, interpretable clinical decision support systems, and new roles such as digital navigators. Only through collaborative efforts across all stakeholders, unified by a sociocultural-technical framework, can we clinically deploy LLMs while ensuring equitable access and mitigating risks.

Neuropsychiatric disorders pose a high societal cost, but their treatment is hindered by lack of objective outcomes and fidelity metrics. AI technologies and specifically Natural Language Processing (NLP) have emerged as tools to study mental health interventions (MHI) at the level of their constituent conversations. However, NLP's potential to address clinical and research challenges remains unclear. We therefore conducted a pre-registered systematic review of NLP-MHI studies using PRISMA guidelines (osf.io/s52jh) to evaluate their models, clinical applications, and to identify biases and gaps. Candidate studies (n = 19,756), including peer-reviewed AI conference manuscripts, were collected up to January 2023 through PubMed, PsycINFO, Scopus, Google Scholar, and ArXiv. A total of 102 articles were included to investigate their computational characteristics (NLP algorithms, audio features, machine learning pipelines, outcome metrics), clinical characteristics (clinical ground truths, study samples, clinical focus), and limitations. Results indicate a rapid growth of NLP MHI studies since 2019, characterized by increased sample sizes and use of large language models. Digital health platforms were the largest providers of MHI data. Ground truth for supervised learning models was based on clinician ratings (n = 31), patient self-report (n = 29) and annotations by raters (n = 26). Text-based features contributed more to model accuracy than audio markers. Patients' clinical presentation (n = 34), response to intervention (n = 11), intervention monitoring (n = 20), providers' characteristics (n = 12), relational dynamics (n = 14), and data preparation (n = 4) were commonly investigated clinical categories. Limitations of reviewed studies included lack of linguistic diversity, limited reproducibility, and population bias. A research framework is developed and validated (NLPxMHI) to assist computational and clinical researchers in addressing the remaining gaps in applying NLP to MHI, with the goal of improving clinical utility, data access, and fairness.

BACKGROUND:Stressors for health care workers (HCWs) during the COVID-19 pandemic have been manifold, with high levels of depression and anxiety alongside gaps in care. Identifying the factors most tied to HCWs' psychological challenges is crucial to addressing HCWs' mental health needs effectively, now and for future large-scale events. OBJECTIVE:In this study, we used natural language processing methods to examine deidentified psychotherapy transcripts from telemedicine treatment during the initial wave of COVID-19 in the United States. Psychotherapy was delivered by licensed therapists while HCWs were managing increased clinical demands and elevated hospitalization rates, in addition to population-level social distancing measures and infection risks. Our goal was to identify specific concerns emerging in treatment for HCWs and to compare differences with matched non-HCW patients from the general population. METHODS:We conducted a case-control study with a sample of 820 HCWs and 820 non-HCW matched controls who received digitally delivered psychotherapy in 49 US states in the spring of 2020 during the first US wave of the COVID-19 pandemic. Depression was measured during the initial assessment using the Patient Health Questionnaire-9, and anxiety was measured using the General Anxiety Disorder-7 questionnaire. Structural topic models (STMs) were used to determine treatment topics from deidentified transcripts from the first 3 weeks of treatment. STM effect estimators were also used to examine topic prevalence in patients with moderate to severe anxiety and depression. RESULTS:The median treatment enrollment date was April 15, 2020 (IQR March 31 to April 27, 2020) for HCWs and April 19, 2020 (IQR April 5 to April 27, 2020) for matched controls. STM analysis of deidentified transcripts identified 4 treatment topics centered on health care and 5 on mental health for HCWs. For controls, 3 STM topics on pandemic-related disruptions and 5 on mental health were identified. Several STM treatment topics were significantly associated with moderate to severe anxiety and depression, including working on the hospital unit (topic prevalence 0.035, 95% CI 0.022-0.048; P<.001), mood disturbances (prevalence 0.014, 95% CI 0.002-0.026; P=.03), and sleep disturbances (prevalence 0.016, 95% CI 0.002-0.030; P=.02). No significant associations emerged between pandemic-related topics and moderate to severe anxiety and depression for non-HCW controls. CONCLUSIONS:The study provides large-scale quantitative evidence that during the initial wave of the COVID-19 pandemic, HCWs faced unique work-related challenges and stressors associated with anxiety and depression, which required dedicated treatment efforts. The study further demonstrates how natural language processing methods have the potential to surface clinically relevant markers of distress while preserving patient privacy.

BACKGROUND:Depression is prevalent among older adults. Understanding the network structure of depression across diverse cultural contexts is essential to preventing and treating depression. This study evaluated the prevalence and combined network structure of depression among older adults based on national surveys from five countries. METHODS:This study combined data from five national cohort studies. The Center for Epidemiologic Studies Depression (CESD) scale was used to assess depressive symptoms. Meta-analysis was used to estimate the overall prevalence of depression, while network models were constructed using Ising models. The most central depressive symptoms were identified using the Expected Influence (EI) index. RESULTS:In total, 102,202 older adults were included. The pooled prevalence of depression was 18.9% (95% confidence interval (CI):10.3%-27.4%). In the combined network model, the most central symptoms were "Feeling depressed", "Feeling sad", "Lack of happiness" and "loneliness", while the strongest positive edge was "Not enjoy life" (CESD7) - "Lack of happiness" (CESD6). CONCLUSIONS:Our findings indicated depressive symptoms are common among older adults across several countries. Moreover, interventions to address feelings of sadness, depressed mood, and lack of happiness as well as loneliness may be beneficial in alleviating depression across older adults in these countries.

BACKGROUND:Loneliness is a form of psychological distress associated with increased risk of depression, anxiety, and adverse health outcomes across the life span. This study evaluates an online gaming‑based community intervention that combines professionally facilitated groups, commercial video games, and skills‑focused workshops for adults who play video games. OBJECTIVE:This study aimed to examine the feasibility of this health‑supporting gaming community and to characterize 30‑ and 60‑day changes in depression, anxiety, psychological well‑being, and psychological flexibility, as well as heterogeneous trajectories of depressive symptoms. METHODS:In a longitudinal observational study, adults in the United States self‑enrolled in a gaming therapeutics community hosted on Discord. Participants completed baseline, 30‑day, and 60‑day assessments including the Patient Health Questionnaire‑9 (PHQ‑9), Generalized Anxiety Disorder‑7 Scale, World Health Organization-5 Well‑Being Index, and Psychological Flexibility Scale (Psy‑Flex). Of 438 participants with 30‑day data, 403 met inclusion criteria for longitudinal analyses and 157 (35.6%) completed the 60‑day survey. Within‑person change scores and standardized mean differences were calculated, and latent growth mixture modeling was used to identify depressive‑symptom trajectories and baseline predictors of nonresponse versus improvement. RESULTS:At baseline, mean PHQ‑9 score was 13.37 (SD 6.04), decreasing to 10.27 (SD 5.80) at 60 days (Cohen d=0.52). Mean Generalized Anxiety Disorder‑7 scores decreased from 11.23 (SD 5.24) to 8.25 (SD 4.22) (Cohen d=0.60). Psy‑Flex scores increased from 11.51 (SD 4.30) to 12.55 (SD 4.37; Cohen d=0.24), and World Health Organization‑5 Well‑Being Index scores increased from 7.86 (SD 3.82) to 8.08 (SD 4.44; Cohen d=0.24). Latent growth mixture modeling identified 3 depressive‑symptom trajectories: a low group (229/438, 52.3%; baseline PHQ‑9 mean 8.80 (SD 3.67); 60‑day mean 7.64, SD 3.87), a chronic group (118/438, 26.9%; baseline mean 18.30 (SD 4.03); 60‑day mean 16.78, SD 4.10), and an improvers group (91/438, 20.8%; baseline mean 18.52, SD 3.13; 60‑day mean 7.42, SD 4.27). In logistic regression among participants with moderate‑to‑severe baseline depression, a gender identity other than woman was associated with lower odds of belonging to the Improvers versus Chronic group (odds ratio 0.42, 95% CI 0.19-0.94). Post hoc analyses indicated lower odds of improvement for nonbinary participants compared with women (odds ratio 0.25, 95% CI 0.10-0.59). No other baseline characteristics significantly distinguished chronic versus improving trajectories. CONCLUSIONS:A professionally moderated, gaming‑based community intervention was feasible to deliver, engaged a diverse sample of adult gamers, and was associated with medium‑sized reductions in depressive and anxiety symptoms and small improvements in well‑being and psychological flexibility over 60 days. A subgroup moved from moderate‑to‑severe symptoms to subthreshold depressive symptoms. These findings support further controlled evaluation of health‑supporting gaming communities as a scalable support and potential preventive context for adult gamers experiencing distress.

BACKGROUND:Given the increased use of network analysis in sleep studies, this systematic review and statistical evaluation aimed to aggregate network studies to identify the most central symptoms in composite network models of sleep-related symptoms. METHODS:A systematic search of cross-sectional network studies focused exclusively on sleep within community or clinical samples was conducted across PubMed, Web of Science (WOS), PsycINFO, and EMBASE databases up to March 5, 2025. Studies were categorized by topic and measurement instruments. Statistical evaluations extracted the most central symptoms across network models. RESULTS:The review included 23 studies of 84,510 participants and 29 network models. Explored topics included insomnia/sleep disturbances, sleep quality, sleep attitudes and behaviors, daytime function, and dream content. Regarding main analyses, key central symptoms in network models of insomnia were "Difficulty staying asleep" [median rank:1.5, Interquartile range (IQR): 1-2], "Distress caused by the sleep difficulties" (median rank:2, IQR: 2-3) and "Interference with daytime functioning" (median rank:3.5, IQR: 2.25-4). For sleep quality, "Subjective sleep quality" (median rank:1, IQR: 1-1), "Daytime dysfunction" (median rank:3, IQR: 2-5.25) and "Sleep disturbance" (median rank:3.5, IQR: 2-4.5) were the most central experiences. CONCLUSIONS:Identified central symptoms offer plausible targets for intervention across populations and guide future research directions.

BACKGROUND/UNASSIGNED:Late-life depression (LLD) is a significant global public health challenge among older adults. Exploring central/influential symptoms with longitudinal study designs can enhance the efficacy of detection, early prevention, and interventions for LLD. This study aimed to identify key symptoms of LLD using a panel graphical vector autoregression (panel-GVAR) model based on longitudinal national survey data. METHODS/UNASSIGNED:Data from the China Health and Retirement Longitudinal Study (CHARLS) between 2013 and 2020, encompassing four waves, were utilized to construct a longitudinal depressive symptom network. Depressive symptoms were assessed using the 10-item Center for Epidemiological Studies Depression Scale (CESD-10). In expected influence (in-EI) and out expected influence (out-EI) were identified to characterize the interaction of symptoms within the temporal network, while expected influence (EI) was used to examine the interaction of symptoms in both the contemporaneous network and the between-subjects network. RESULTS/UNASSIGNED:A total of 1393 older adults were assessed. A persistently significant increase in the prevalence of depression was observed over time. In the temporal network, "restless sleep" (CESD7) and "could not get going" (CESD10) were the most influential symptom and most influenced symptom, respectively. In both the contemporaneous network and the between-subjects network, "felt depressed" (CESD3) emerged as the most central symptom within the community of depressive symptoms. CONCLUSIONS/UNASSIGNED:Given the challenges associated with treating LLD and its adverse effects on daily life for older adults, timely interventions targeting identified key symptoms may help prevent and mitigate depression in this population.

INTRODUCTION/BACKGROUND:Poststroke depression affects approximately 30% of stroke survivors and is linked to worse functional outcomes, cognitive decline, reduced quality of life and increased mortality. While early treatment of poststroke mental health conditions is critical, current pharmacological options offer limited efficacy. Music listening interventions are a promising, low-risk, accessible and affordable alternative that may enhance recovery through engagement of reward-related brain circuits. However, most music listening studies have focused on the acute stage of stroke, lack objective measures of music engagement and rarely assess underlying neural mechanisms. To address these gaps, we propose a feasibility study of a remotely delivered music-listening intervention for individuals with chronic stroke, incorporating objective tracking of music exposure and multimodal assessments of mental health, cognitive, neural and physiological changes. METHODS AND ANALYSIS/METHODS:We will conduct a parallel group randomised controlled feasibility trial enrolling 60 patients with chronic stroke from a well-characterised stroke registry in New York City. Participants will be randomised to either an intentional music listening (IML) group or an active control group that listens to audiobooks. The study includes a 4-week preintervention period during which no treatment is administered; this phase is designed to assess the stability of outcome measures. Following this, participants will engage in 1-hour daily listening sessions over a 4-week intervention period. All listening activity (ie, track identity, duration and engagement) will be continuously tracked using custom open-source software, providing a measure of treatment dose. Behavioural outcomes related to mental health will be assessed at baseline, preintervention, postintervention and 3-month follow-up. Multimodal biomarkers (functional and structural MRI, electrodermal activity and heart rate) will be collected preintervention and postintervention. The primary objective is to establish feasibility, defined by rates of retention and adherence, treatment fidelity, feasibility, acceptability and participant burden. Secondary outcomes include recruitment and randomisation rates. This trial will provide essential data to inform the design of future large-scale clinical studies of IML for poststroke mental health recovery. ETHICS AND DISSEMINATION/BACKGROUND:The study was approved by New York University's Institutional Review Board (FY2024-8826). All human participants will provide written informed consent prior to participation and will be adequately compensated for their time. Results will be reported in peer-reviewed journals. TRIAL REGISTRATION NUMBER/BACKGROUND:NCT07127159.

We introduce varying and irregular sampling time-series analysis (VISTA), a clustering approach for multivariate and irregularly sampled time series based on a parametric state-space mixture model. VISTA is specifically designed for the unsupervised identification of groups in data sets originating from healthcare and psychology where such sampling issues are commonplace. Our approach adapts linear Gaussian state-space models (LGSSMs) to provide a flexible parametric framework for fitting a wide range of time series dynamics. The clustering approach itself is based on the assumption that the population can be represented as a mixture of a fixed number of LGSSMs. VISTA's model formulation allows for an explicit derivation of the log-likelihood function, from which we develop an expectation-maximization scheme for fitting model parameters to the observed data samples. Our algorithmic implementation is designed to handle populations of multivariate time series that can exhibit large changes in sampling rate as well as irregular sampling. We evaluate the versatility and accuracy of our approach on simulated and real-world data sets, including demographic trends, wearable sensor data, epidemiological time series, and ecological momentary assessments. Our results indicate that VISTA outperforms most comparable standard times series clustering methods. We provide an open-source implementation of VISTA in Python. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

BACKGROUND:Although evidence-based Prolonged Grief Disorder treatments (PGDT) exist, pretreatment characteristics associated with differential improvement remain unidentified. To identify clinical factors relevant to optimizing PGDT outcomes, we used unsupervised and supervised machine learning to study treatment effects from a double-blinded, placebo-controlled, randomized clinical trial. METHODS:Patients were randomized into four treatment groups for 20 weeks: citalopram with grief-informed clinical management, citalopram with PGDT, pill placebo with PGDT, or pill placebo with clinical management. The trial included 333 PGD patients aged 18-95 years (M = 53.9; SD = 14.4). Symptom trajectories were assessed using latent growth mixture modeling based on Inventory for Complicated Grief scores collected every 4 weeks. The relationship between patient-level characteristics and assigned trajectories was examined using logistic regression with elastic net regularization based on the administration of citalopram, PGDT, and risk factors for developing PGD. RESULTS:Three response trajectories were identified: lesser severity responders (60 %, n = 200), greater severity responders (18.02 %, n = 60), and non-responders (21.92 %, n = 73). Significant differences between greater severity responders and non-responders emerged by Week 8, persisting through the 6-month follow-up assessment. The elastic net model (AUC = 0.702; F1 = 0.777) indicated that higher baseline depression severity, grief-related functional impairment, and not receiving PGDT were associated with a decreased probability of response. LIMITATIONS/CONCLUSIONS:An independent validation cohort of PGDT patients is needed to further study generalizability of findings. CONCLUSIONS:Differential PGDT courses and the role of depression severity and grief-related functional impairment in treatment non-response were identified. These findings underscore the importance of determining clinical factors relevant to optimizing individual treatment strategies.