Major Depressive Disorder (MDD) is a common mental health condition marked by persistent low mood, reduced motivation, and low energy. Persons with MDD often experience large fluctuations in their symptoms over hours and days, which can offer valuable clinical insights, highlighting potential targets for treatment and intervention. Yet, it remains unclear how accurately these acute symptom changes can be predicted. In this preregistered study, we examined whether deep learning models could predict short-term fluctuations in depressive symptoms among individuals diagnosed with MDD. Passively-collected accelerometry data, a ubiquitous, privacy-preserving data stream with empirical ties to MDD symptomatology, was used to predict instability in depressive symptoms. Our sample contains clinically-depressed individuals per a structured clinical interview (N = 179, Mean Age = 38.2 ± 10.0, Women = 81.6%). Participants completed depressive symptom assessments three times daily for 90 days, resulting in an average of 214.49 EMA questionnaires (median: 237, min: 77, max: 270, SD: 54.27) per person. Although our results suggest a small-moderate correspondence between the predictions and observed outcomes of passively-collected movement alone in predicting future high/low depression instability (AUCTest = 0.59 ± 0.06, SensitivityTest = 0.56 ± 0.21, SpecificityTest = 0.59 ± 0.23), our model performed worse than a random intercept only baseline model (AUCTest = 0.83) and predicted intraindividual variability of instability in depressive symptoms below chance (AUCTest = 0.46 ± 0.21). Thus, there is a need to improve predictive performance to enhance clinical utility. This may require the inclusion of additional passive sensing modalities in future research.

Multimodal health sensing offers rich behavioral signals for assessing mental health, yet translating these numerical time-series measurements into natural language remains challenging. Current LLMs cannot natively ingest long-duration sensor streams, and paired sensor–text datasets are scarce. To address these challenges, we introduce LENS, a framework that aligns multimodal sensing data with language models to generate clinically grounded mental-health narratives. LENS first constructs a large-scale dataset by transforming Ecological Momentary Assessment (EMA) responses related to depression and anxiety symptoms into natural-language descriptions, yielding over 100,000 sensor–text QA pairs from 258 participants. To enable native time-series integration, we train a patch-level encoder that projects raw sensor signals directly into an LLM’s representation space. Our results show that LENS outperforms strong baselines on standard NLP metrics and task-specific measures of symptom-severity accuracy. A user study with 13 mental-health professionals further indicates that LENS-produced narratives are comprehensive and clinically meaningful. Ultimately, our approach advances LLMs as interfaces for health sensing, providing a scalable path toward models that can reason over raw behavioral signals and support downstream clinical decision-making.

Major Depressive Disorder (MDD) is one of the most prevalent psychological disorders and frequently co-occurs with alcohol use disorders, increasing the risk of functional impairment. Monitoring alcohol use during depression treatment is therefore critical for early intervention. Passively collected data via devices like smartphones and smartwatches, offers a low-burden method for monitoring behavior in real time. This study investigated whether deep learning models trained on passively collected data (i.e., accelerometer, heart rate, respiratory rate, screen usage, and GPS data) could detect and predict alcohol use in individuals with MDD. Data were collected from 300 clinically depressed individuals who were enrolled in the Tracking Depression Study, a 90-day longitudinal study. Participants self-reported their alcohol use every week by completing the Timeline FollowBack. We trained models to predict same-day and next-day alcohol use. To validate these models, we split the data by participant, so that predictions were made on individuals who were not included in the training set. The models achieved moderate performance (mean AUC = 0.67 for both prediction tasks) when capturing both interindividual (between-person) and intraindividual (within-person) variability. Similar performances were observed when evaluating the model exclusively on predicting intraindividual variability (AUCs = 0.69 same-day, 0.68 next-day). However, model performance remained comparable to a baseline using only the day of week as predictor. These findings suggest that much of the predictive signal derives from temporal patterns. This indicates that interventions aligned with such temporal cues may already be effective, and that the added value of our model appears limited.

Longitudinal passive sensing enables continuous health prediction, yet models often fail under cross-dataset distribution shifts. Traditional ML overfits cohort-specific artifacts, while Large Language Models (LLMs) struggle to reason reliably over long, heterogeneous time-series. We introduce TimeSRL, a two-stage LLM framework that routes predictions through an explicit semantic bottleneck. The model first abstracts raw signals into high-level natural language, then predicts behavioral outcomes from these abstractions alone. This forces the model to reason over semantic concepts that we argue generalize better than raw numbers. We optimize this process end-to-end using Group Relative Policy Optimization (GRPO) with Reinforcement Learning from Verifiable Rewards (RLVR), learning outcome-aligned abstractions without gold intermediate annotations. Instantiated on mental-health prediction, TimeSRL achieves state-of-the-art performance on a benchmark designed to stress-test cross-cohort generalization under a rigorous leave-one-dataset-out (LOSO) protocol, reducing mean absolute error (MAE) over strong non-LLM ML and LLM baselines by 3.1–10.1

Major depressive disorder (MDD) is characterized by high levels of heterogeneity in symptom presentation across individuals. While previous research has identified distinct MDD subtypes using self-reported symptoms, few studies have leveraged objective data from smartphones and wearable devices to phenotype MDD symptoms. Passive sensing data from these devices can capture objective behavioral and physiological patterns, potentially revealing distinct digital phenotypes of MDD. We identified latent profiles based on digital biomarkers of depression collected from smartphones and Garmin smartwatches among 297 individuals with MDD. Digital biomarkers included sleep patterns, physical activity, screen time, social engagement, and heart rate variability. An exploratory aim examined whether identified profiles were associated with MDD severity and social and occupational functioning. A two-profile solution demonstrated best fit with the data: Profile 1 (85.7% of the sample) 'average in every way,' and Profile 2 (14.3%) 'deficient sleep, chronically low heart rate variability, and low social engagement.' While profiles did not significantly differ on MDD symptom severity (est = 0.322, S.E. = 0.767, p = 0.444), Profile 2 had lower social and occupational functioning compared to Profile 1 (est = -5.309, S.E. = 2.321, p = 0.023), though this was no longer statistically significant after correcting for type I error. Sleep dysregulation, low heart rate variability, and low social engagement seem to be important indicators of potential social and occupational impairments. Future research should incorporate additional digital biomarkers to refine the identification of digital phenotypes of MDD and validate these profiles against other clinical severity metrics in larger, more diverse samples.

LLMs are increasingly used to explain personal sensing data, translating traces of activity and mood into natural-language accounts of why an anomalous day may have occurred. However, such explanations can sound coherent and personally meaningful even when the underlying evidence is sparse or missing. We introduce epistemic overreach (EO) as a measure for cases where a generated explanation implies more than the available sensing evidence can justify. To audit how often and in what forms EO occurs, we obtained anomalous-day scenarios from three longitudinal sensing datasets of college students: StudentLife, GLOBEM, and CollegeExperience. Across activity, sleep, and affect anomalies, we generated 14,922 explanations using three LLM families – Llama, Qwen, and GPT – under two prompting conditions: one minimally constrained prompt and another prompt explicitly instructing models to bound claims to the data. For each scenario, we varied the amount of behavioral evidence available to the model to examine whether more evidence reduces EO. We evaluated each explanation using a structured rubric, decomposing EO into the dimensions of unsupported causal attribution, unacknowledged data gaps, overconfident language, temporal inconsistency, and diagnostic inference. We find that LLMs routinely attribute anomalous days to causes without sufficient support from the data, and that this pattern replicates across datasets, anomaly types, and model families. Further, providing richer context does not reliably reduce EO; bounded prompting helps but does not eliminate it. These findings suggest that evidential grounding should be a first-order evaluation criterion for LLM-generated personal sensing explanations, alongside fluency and plausibility. We argue that personal sensing explanations require evidential discipline: systems must distinguish what is observed, what is inferred, and what remains unknown.

The longitudinal relationship between loneliness and depressive symptoms is well documented. However, few studies have considered the bidirectional relationship between loneliness and anhedonia, a cardinal symptom of depression. Moreover, the associations between temporal dynamics of loneliness and depression, and how dampening of positive affect may impact these associations, are under-examined. This study included a sample of clinically depressed individuals who completed baseline measures of dampening and depression, as well as ecological momentary assessments of loneliness and depression during a 90-day study period (N = 150). Multilevel models revealed that within-person loneliness predicted decreased next-week anhedonia when accounting for within-person and between-person variation in anhedonia. Within-person anhedonia predicted greater next-week loneliness when accounting for autoregressive effects. Moderation results revealed that at higher levels of loneliness, the association between baseline dampening and depression assessed after 90 days was positive. At lower levels of loneliness, the association between dampening and depression 90 days later was negative. Our findings imply that loneliness may motivate short-term reaffiliation efforts and that anhedonia may decrease motivation for social activities. Further, social connection may buffer against the effect of dampening on depressive symptoms. These results suggest targeting positive affect and social reward may mitigate depressive symptomatology.

Young people are increasingly turning to conversational AI agents for support in figuring out what they want to do with their lives. Reflecting on career goals is characteristic of emerging adulthood (ages 18 to late twenties), a critical time for exploration and commitment to one’s personal and vocational identity. Yet it is unclear whether and how conversational AI agents benefit or hinder the development of positive and coherent identities, which has potential consequences for well-being. Our two pre-registered field experiments (Study 1: N= 165; Study 2: N = 248) tested the efficacy of digital intervention treatments designed to promote adaptive identity processes via career-focused reflection activities. Using an active-controlled RCT design, we assigned participants to either a Self-Reflection or an AI-Mediated Reflection condition. We found that both conditions promoted within-person adaptive change in daily identity exploration and commitment processes on treatment days, compared to no treatment days. However, when comparing the relative efficacy of the two treatments, we found that self-reflection may better facilitate adaptive changes (e.g., increasing identity commitment, decreasing ruminative exploration), compared to AI-mediated reflection. Exploratory analyses also showed that the intervention had positive effects on daily affect and post-intervention eudaimonic well-being, although AI-mediated reflection was less beneficial. Linguistic analysis of the reflection content suggest that the conditions may shape the cognitive and affective experience of reflection in meaningfully different ways. Our discussion contextualizes these findings within the literature on conversational AI agents as a tool for self-development and highlights future directions for research on human-AI interactions.

The modern workplace is undergoing a radical transformation, driven by technological advances that blur the boundaries between human capability and digital augmentation. At the forefront of this evolution is passive sensing technology - a suite of tools that quietly monitor and interpret human behavior without active user engagement. This paper examines how these technologies are reshaping our understanding of workplace dynamics, with a particular focus on employee wellbeing and productivity. Through a comprehensive review of recent research, we explore both the transformative potential and inherent challenges of passive sensing in professional environments. Our analysis reveals emerging patterns in how these technologies can support worker health and performance, while also highlighting critical gaps in current research and opportunities for future innovation. We conclude by outlining a roadmap for integrating passive sensing into future workplaces in ways that enhance human potential while preserving dignity and autonomy.

Despite major strides in conceptualizing and modeling the multifaceted nature of suicidal thought and behavior (STB) over the past few decades, the overall predictability of STB has not improved. This may be partly due to the dynamic nature of suicidal ideation (SI), which often fluctuates over hours, yet is largely overlooked in studies. Bolstered by the application and promise of natural language processing (NLP) across the mental health field, efforts toward richer operationalization of acute SI may include analyses on written data that occur alongside changes in SI, thus offering a better understanding of STB as it unfolds. Ecological momentary assessment (EMA) data from 268 participants with major depressive disorder (MDD) were utilized to investigate acute changes in SI. Data consisted of thrice-daily SI severity scores measured through self-report responses to item 9 of the Patient Health Questionnaire mobile version (MPHQ-9) as well as free-form diary text. Using difference scores and probability of acute change thresholds, eleven acute SI phase trajectory types were defined to label change in SI over three consecutive EMAs. In total, 5,938 acute SI trajectories were paired with the temporally centered diary entries. The Sentiment Analysis and Cognition Engine (SEANCE) tool was applied to quantify the written content of each diary entry across eight established lexica. Entry results were grouped based on phase trajectory type, and the Kruskal-Wallis test was employed with post-hoc multiple hypothesis correction to statistically compare SEANCE features between all group pairs. There were 131 statistically significant (adjusted p-value < 0.05) pairwise differences between acute SI phase trajectory groups, implicating 31 NLP features. Consistent with the literature, results highlighted qualities of writing that are generally associated with heightened SI, including personal pronoun usage, passivity, and negative valence. Patterns of significance also uncovered novel contextual nuance in terms of how characteristics such as verbosity, hostility, anger, and pleasantness present in relation to SI over short change trajectories. This work provides an accessible exploratory framework that capitalizes on the benefits of dense EMA sampling and NLP to profile and quantify acute SI trajectories. The use of the MPHQ’s item 9 to quantify SI is an important limitation as it is designed to also capture precursory SI, passive SI, and SI-adjacent behaviors, potentially overestimating the SI expressed by participants. Nonetheless, future research should continue to focus on short timeframes as there are likely important signals and interpretative nuances to SI expression that have yet to be fully detailed.

Ecological momentary assessment (EMA) offers advantages over retrospective questionnaires by reducing recall bias and capturing rapid symptom dynamics, and it is increasingly used to measure depression symptoms. However, few depression symptom measures are validated for EMA use in the manner expected for traditional questionnaires. Therefore, the current study examined the internal consistency, longitudinal stability, and convergent validity of the Mobile Patient Health Questionnaire-9 (MPHQ-9), a version of the PHQ-9 modified for EMA. Depressed participants (N = 280; Female = 83.93%; White = 79.29%) completed the MPHQ-9 three times daily for 90 days. Data from the first and last 2 weeks were analyzed to align with a pre-study PHQ-9 and post-study PHQ-9 and Inventory of Depression and Anxiety Symptoms - Expanded Version (IDAS-II). The MPHQ-9 demonstrated fair to substantial adjusted item-total correlations (r = 0.42–0.83), often exceeding the PHQ-9 (r = 0.39–0.72), with Cronbach’s alpha coefficients of .91 and .81, respectively. Reliability analyses of the MPHQ-9 using generalizability theory (GT) and multilevel modeling (MLM) to account for repeated measures yielded substantial between-person reliability (~1.0) but mixed within-person reliability estimates of 0.81 (GT) and 0.44 (MLM). The MPHQ-9 showed moderate stability (r = .69, ICC = .58), compared to the slight stability of the PHQ-9 (r = .39, ICC = .37). There was moderate agreement between the MPHQ-9 and both the PHQ-9 (r = .71) and the IDAS-II General Depression subscale (r = .65). This study provides initial validation of the MPHQ-9 and compares its psychometric properties to the traditional PHQ-9.

Information workers increasingly struggle with productivity challenges in modern workplaces, facing difficulties in managing time and effectively utilizing workplace analytics data for behavioral improvement. Despite the availability of productivity metrics through enterprise tools, workers often fail to translate this data into actionable insights. We present a comprehensive, user-centric approach to address these challenges through AI-based productivity agents tailored to users' needs. Utilizing a two-phase method, we first conducted a survey with 363 participants, exploring various aspects of productivity, communication style, agent approach, personality traits, personalization, and privacy. Drawing on the survey insights, we developed a GPT-4 powered personalized productivity agent that utilizes telemetry data gathered via Viva Insights from information workers to provide tailored assistance. We compared its performance with alternative productivity-assistive tools, such as dashboard and narrative, in a study involving 40 participants. Our findings highlight the importance of user-centric design, adaptability, and the balance between personalization and privacy in AI-assisted productivity tools. By building on these insights, our work provides important guidance for developing more effective productivity solutions, ultimately leading to optimized efficiency and user experiences for information workers.

Large language models (LLMs) show promise for health applications when combined with behavioral sensing data. Traditional approaches convert sensor data into text prompts, but this process is prone to errors, computationally expensive, and requires domain expertise. These challenges are particularly acute when processing extended time series data. While time series foundation models (TFMs) have recently emerged as powerful tools for learning representations from temporal data, bridging TFMs and LLMs remains challenging. Here, we present Time2Lang, a framework that directly maps TFM outputs to LLM representations without intermediate text conversion. Our approach first trains on synthetic data using periodicity prediction as a pretext task, followed by evaluation on mental health classification tasks. We validate Time2Lang on two longitudinal wearable and mobile sensing datasets: daily depression prediction using step count data (17,251 days from 256 participants) and flourishing classification based on conversation duration (46 participants over 10 weeks). Time2Lang maintains near constant inference times regardless of input length, unlike traditional prompting methods. The generated embeddings preserve essential time-series characteristics such as auto-correlation. Our results demonstrate that TFMs and LLMs can be effectively integrated while minimizing information loss and enabling performance transfer across these distinct modeling paradigms. To our knowledge, we are the first to integrate a TFM and an LLM for health, thus establishing a foundation for future research combining general-purpose large models for complex healthcare tasks.

Anhedonia and avolition are core clinical features of schizophrenia, bipolar disorder, and major depressive disorder, which have been traditionally assessed using clinical rating scales. However, recent developments in mobile technology allow for measurement of anhedonia and amotivation using passive sensors (e.g., global positioning system and actigraphy) and surveys completed in daily life (i.e., ecological momentary assessment [EMA]). The current study examined associations between clinical rating scales assessing anhedonia and amotivation and passive sensing measures. We aimed to determine the added value of passive sensing measures in explaining variability in clinical interviews, compared to models using EMA alone. We recruited a transdiagnostic sample (schizophrenia = 41, bipolar disorder = 47, and major depressive disorder = 48) to complete an in-person assessment session, as well as a 2-week EMA and passive sensing protocol. Passive sensing measures included physical distance traveled, number of phone calls sent/received, and number of texts sent/received. EMA included the assessment of interest and enjoyment in daily activities. We found that reports of interest/enjoyment in daily activities significantly predicted gold standard, clinical rating scales of anhedonia and avolition across diagnostic groups (standardized β = -0.208, p = .015, model R2 = .04). Including participant distance traveled into this model aided our ability to explain variance (standardized β = -0.280, p < .001, model R² = .12). Finally, adding call (standardized β = -0.170, p = .039) and text (standardized β = -0.198, p = .022) data further improved variance explained (model R² = .18). These data suggest that passive sensor streams strengthen the associations between assessments in daily life and gold standard ratings of anhedonia and avolition, suggesting "added value" in using these approaches to understand motivational experience in people with psychotic and mood pathology. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Individuals with major depressive disorder (MDD) experience fewer positive and more negative emotions and use fewer positive words to describe themselves. Natural language processing techniques have been used to predict depression, with pronoun and emotion usage being identified as important features. However, it is unclear how depressed individuals use positive and negative words when writing about themselves. Individuals with MDD (N = 258) completed ecological momentary assessments three times a day (including the Patient Health Questionnaire-9 [PHQ-9] and a free-text diary entry) and weekly ecological momentary assessments (including a free-text response to a life events prompt) over a 90-day study period. Using natural language processing techniques, we generated 20 model features to detect and predict averages of and changes in weekly depression from diary entries. Four regression models detected and predicted total PHQ-9 and changes in PHQ-9, and two classification models detected and predicted moderate to severe depression. The models classified current (area under the receiver operating curve [AUC] = 0.68) and future depression (AUC = 0.63), and suggest that lower valence increased usage of "I"/"me"/"my," and lower valence of passages with "I"/"me" as the subject, influenced model predictions toward more severe depression, supporting prior research. These findings highlight that depressed individuals use less positive and more negative words when referring to themselves. Treatments targeting positive affect and digital interventions with written components may be beneficial for targeting MDD. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Negative rumination and emotion regulation difficulties have been consistently linked with depression. Despite anhedonia-the lack of interest in pleasurable experiences-being a cardinal symptom of depression, emotion regulation of positive emotions, including dampening, are considered far less in the literature. Given that anhedonia may manifest through blunted responses to previously positive or enjoyable experiences, it is vital to understand how different positive emotion regulation strategies impact anhedonia symptom severity and how it can vary or change over time. Moreover, understanding the detrimental or protective nature of positive emotion regulation on anhedonia can aid with future anhedonia-focused treatments. Therefore, the current study examined the temporal association between anhedonia dynamics and two different emotion rumination strategies in response to positive emotions: dampening and positive rumination. Depressed persons (N = 137) completed baseline measures of positive emotion regulation, difficulties regulating negative emotions, and anxiety, and completed ecological momentary assessments three times per day for 90 days regarding their depressive symptoms, including anhedonia. We assessed baseline dampening and amplifying scores to predict anhedonia dynamics through four linear models with interactions. Providing partial support for our hypotheses, results indicate that amplifying positivity is positively associated with fluctuations, instability, and acute changes in anhedonia over the course of 90 days; however, neither dampening, difficulties regulating negative emotions, nor anxiety were related to anhedonia dynamics. The current findings suggest that amplifying positivity may be able to predict changes in anhedonia over time and should further be examined as a potential protective factor of anhedonia.

Anhedonia and depressed mood are two cardinal symptoms of major depressive disorder (MDD). Prior work has demonstrated that cannabis consumers often endorse anhedonia and depressed mood, which may contribute to greater cannabis use (CU) over time. However, it is unclear (1) how the unique influence of anhedonia and depressed mood affect CU and (2) how these symptoms predict CU over more proximal periods of time, including the next day or week (rather than proceeding weeks or months). The current study used data collected from ecological momentary assessment (EMA) in a sample with MDD (N=55) and employed mixed effects models to detect and predict weekly and daily CU from anhedonia and depressed mood over 90 days. Results indicated that anhedonia and depressed mood were significantly associated with CU, yet varied at daily and weekly scales. Moreover, these associations varied in both strength and directionality. In weekly models, less anhedonia and greater depressed mood were associated with greater CU, and directionality of associations were reversed in the models looking at any CU (compared to none). Findings provide evidence that anhedonia and depressed mood demonstrate complex associations with CU and emphasize leveraging EMA-based studies to understand these associations with more fine-grained detail.

As concerns about employee burnout and skilled staff shortages in cybersecurity grow, our study aims to better understand the contributing factors to burnout in this field. Utilizing a mixed-methods approach, we analyze self-reported job and personal characteristics, along with digital activity data from 35 incident responders, identifying several factors such as high workload, time pressure, and lack of support from management. Our findings reveal that over half of the participants experience burnout (N=19), which is linked to increased workload, limited control, poor teamwork, and inadequate recognition. Burned-out responders often work more than 40 hours per week, have poor sleep quality, and engage in more email activities, meetings, and after-hour collaborations. Through our research, we also identify coping strategies individuals use to mitigate these stressors. Based on our findings, we provide practical recommendations to help organizations better support their cybersecurity incident response teams. While our study acknowledges limitations and suggests future research directions, it contributes significantly to understanding the challenges faced by cybersecurity incident responders. Our insights offer a comprehensive understanding of burnout factors in this domain and have broader implications for other high-stress work environments consistent with the interdisciplinary nature of CSCW.

Social isolation is a common problem faced by individuals with serious mental illness (SMI), and current intervention approaches have limited effectiveness. This article presents a blended intervention approach, called mobile Social Interaction Therapy by Exposure, to address social isolation in individuals with SMI. The approach combines brief in-person cognitive-behavioral therapy (CBT) with context-triggered mobile CBT interventions that are personalized using mobile sensing data. Our approach targets social behavior and is the first context-aware intervention for improving social outcomes in SMI.

Understanding the dynamics of mental health among undergraduate students across the college years is of critical importance, particularly during a global pandemic. In our study, we track two cohorts of first-year students at Dartmouth College for four years, both on and off campus, creating the longest longitudinal mobile sensing study to date. Using passive sensor data, surveys, and interviews, we capture changing behaviors before, during, and after the COVID-19 pandemic subsides. Our findings reveal the pandemic's impact on students' mental health, gender based behavioral differences, impact of changing living conditions and evidence of persistent behavioral patterns as the pandemic subsides. We observe that while some behaviors return to normal, others remain elevated. Tracking over 200 undergraduate students from high school to graduation, our study provides invaluable insights into changing behaviors, resilience and mental health in college life. Conducting a long-term study with frequent phone OS updates poses significant challenges for mobile sensing apps, data completeness and compliance. Our results offer new insights for Human-Computer Interaction researchers, educators and administrators regarding college life pressures. We also detail the public release of the de-identified College Experience Study dataset used in this paper and discuss a number of open research questions that could be studied using the public dataset.