Christian A. Webb

Background: Passive smartphone-based digital phenotyping offers a scalable approach for continuous mental health assessment. Continuous passive monitoring may be particularly valuable for posttraumatic stress disorder (PTSD), given fluctuating symptoms and the limitations of episodic clinical assessment. While previously validated for depression and anxiety, its application to PTSD remains limited. This study evaluated passive smartphone behavioral data for estimating PTSD symptom severity.Methods: Adults (N = 176) completed the PTSD Checklist for DSM-5 (PCL-5) and passively contributed smartphone behavioral metadata via the Behavidence app. A Random Forest classifier using 44 behavioral features was developed to classify none (PCL-5 61) PTSD symptom severity. The final model generated participant-level PTSD-related scores, which were evaluated across the full cohort. Feature importance was examined using Gini impurity and SHAP.Results: In the holdout test set, the classifier achieved 80.3% accuracy, an AUC of 0.82, and 89.5% precision. When applied to the full cohort, participant-level MHSS values demonstrated a moderate positive correlation with continuous PCL-5 scores (r = 0.49, p < .001) and increased significantly across questionnaire-defined PTSD severity groups (F(3,172) = 24.06, p < .001, η² = 0.30). Feature importance analyses identified smartphone engagement, temporal usage patterns, and application-specific behaviors as the strongest contributors to model predictions.Conclusions: Passive smartphone-based digital phenotyping demonstrated promising performance for identifying PTSD symptom severity. Participant-level MHSS values were associated with continuous and categorical measures of PTSD severity, supporting passive behavioral monitoring as a scalable complement to traditional PTSD assessment and longitudinal symptom monitoring.

Background:Mental health symptoms fluctuate over short timeframes, yet clinical assessment typically relies on infrequent self-report, limiting the ability to capture dynamic symptom patterns. Passive digital phenotyping enables continuous, low-burden monitoring of behavioral signals associated with mental health, but the clinical meaning of short-term passive symptom trajectories remains poorly understood. Precision psychiatry calls for approaches that enable individualized characterization of symptom trajectories to inform stratified and responsive treatment.Methods:Anonymized passive digital phenotyping data from 13,098 adults were analyzed using a smartphone application that generated daily Mental Health Similarity Scores (MHSS) for depression, anxiety, stress, and attention-deficit/hyperactivity disorder (ADHD) across seven consecutive days, approximating the interval between routine weekly clinical assessments, using validated machine-learning models anchored to standard clinical questionnaires. Latent class growth analyses identified short-term trajectories, and bivariate models examined depression-anxiety co-fluctuation and stratification by depression burden.Results:Depression and anxiety demonstrated heterogeneous short-term trajectories, including stable low, stable high and fluctuating patterns, suggesting distinct short-term symptom dynamics. In contrast, stress and ADHD were largely stable over time. Bivariate analyses identified four synchronized depression-anxiety trajectory classes, ranging from stable low to fluctuating co-occurring profiles. Higher depression was associated with persistently elevated and less variable anxiety patterns.Conclusions:Passive digital phenotyping offers a scalable, low-burden approach for continuous symptom monitoring, enabling identification of dynamic risk and resilience profiles at the individual level. Trajectory-based digital phenotypes may support precision psychiatry by informing patient stratification, early detection of symptom change, and more adaptive and personalized approaches to monitoring and intervention.Key Words: Passive digital phenotyping; Mental health; Precision Psychiatry

Introduction: Anxiety disorders are the most common youth mental health problems in the USA. Most children do not receive treatment and among the few who do, many do not experience significant improvement. Thus, there is an urgent need to identify novel intervention targets to improve treatments for childhood anxiety. Past research demonstrates that parent anxiety plays a role in the development and maintenance of child anxiety, but empirical evidence for the mechanisms that drive this transmission is lacking. Interpretation bias, or the tendency to perceive threat in ambiguous situations, may lead parents to engage in anxiety-promoting parenting behaviours, which may transfer this cognitive bias to their children, resulting in youth anxiety. Given the current lack of empirical evidence supporting this theorised mechanism, the present study will test effects of parent interpretation bias on parent behaviour and child interpretation bias. Our central hypothesis is that parent interpretation bias influences child interpretation bias through its effects on anxiety-promoting parenting behaviours. The study will also evaluate potential moderators of these theorised relationships, such as parent sex, race and ethnicity, parent anxiety, child age and child sex. Methods and analysis: We will randomise 300 parents of children aged 7–12 to a smartphone app-delivered interpretation bias manipulation condition or a self-assessment control condition. Through a series of study assessments over the course of 3 months, parents and children will complete self-report, interview, behavioural and daily diary measures to assess the primary proximal outcome (anxiety-promoting parenting behaviours), the secondary distal outcome (child interpretation bias) and moderators. To test our central hypothesis, we will first use hierarchical linear modelling to assess whether parents in the interpretation bias manipulation condition show fewer anxiety-promoting parenting behaviours compared with control, and whether children of those parents show improved interpretation bias compared with control. We will then implement mediation models to evaluate whether a reduction in anxiety-promoting parenting behaviours mediates the association between parent and child interpretation bias. Ethics and dissemination: Ethical approval for this study was granted by the Mass General Brigham Institutional Review Board (2022P003245). We will disseminate the results of this study to the scientific community, clinical community and the public through several channels such as peer-reviewed publications in scientific journals, national conferences, hospital lectures, press releases and Psychology Today blog posts. Trial registration number: NCT05665491 .

Tracking emotion fluctuations in adolescents' daily lives is important for understanding mood dynamics and identifying early markers of affective disorders. This study examines whether natural language processing (NLP) methods, including large language models (LLMs), can approximate adolescents' concurrent self-reported negative affect from ecological momentary assessment narratives. We compared nomothetic (group-level) and idiographic (individualized) models as well as different NLP techniques for capturing within-person emotion fluctuations. Ecological momentary assessment text responses from 97 adolescents (ages 14-18, 77.3% female, 22.7% male, N_EMA = 7,680) were analyzed. Text features were extracted using dictionary-based approaches, topic modeling, and emotion ratings derived from a LLM (GPT-4o). Random Forest and Elastic Net Regression models predicted negative affect from text features, and SHapley Additive exPlanations values were used to assess feature importance. Key findings, interactive visualizations, and model comparisons are available via the website: https://emotracknlp.streamlit.app/. Idiographic models combining text features from different NLP approaches performed comparably to nomothetic models and to Generative Pre-Trained Transformer 4 Omni (GPT-4o) alone in R² (nomothetic: .06-.11, idiographic: .06-.10, GPT-4o: .06-.10) but yielded lower root-mean-squared error (nomothetic: .52-.88, idiographic: .47-.76, GPT-4o: .85-1.36), improving within-person precision. Importantly, there were substantial between-person differences in model performance and predictive linguistic features. When selecting the best-performing model for each participant (e.g., nomothetic or idiographic), significant correlations between predicted and observed emotion scores were found for 90.7%-94.8% of participants. Findings suggest that although nomothetic models and LLMs offer scalability, idiographic models provide greater predictive precision with sufficient within-person data. A personalized approach selecting optimal models for each individual may enhance emotion monitoring and inform targeted interventions through contextual text insights. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

Mind‑body practices, such as meditation and yoga, involve paying attention to breathing sensations. During these practices, individuals report “

Importance: Low-intensity interventions can help to widen access to treatment for common mental disorders.Objective: To develop a data-driven model for personalized assignment to one of three available low intensity interventions, and to evaluate the clinical outcomes of model-matched cases.Design: Analysis of electronic health records for an observational cohort. A multi-arm causal forest estimated conditional treatment effects in a training sample of early cases (~70%) and evaluated the potential benefit of personalized treatment selection in a hold-out test sample of cases treated at a later date (~30%).Setting: Sixteen NHS Talking Therapies services offering low intensity psychological interventions in England.Participants: N= 89,467 patients receiving one of three interventions.Interventions: cCBT (computerized CBT), iGSH (individual guided self-help), and gGSH (group guided self-help).Main Outcomes and Measures: Reliable recovery on psychometric measures of depression (PHQ-9) and anxiety (GAD-7). Personalized intervention benefit (PIB) indexed recommendation strength as the predicted advantage of the recommended treatment over alternatives.Results: Observed assignment in the test sample was highly concentrated in iGSH (87.9%), with 9.0% receiving gGSH and 3.1% cCBT. The causal forest recommended cCBT for 64.1%, iGSH for 33.5%, and gGSH for 2.4%. Only 30.7% received their model-recommended treatment. Overall reliable recovery was 41.1%. Receiving the recommended treatment was associated with higher reliable recovery, with the benefit increasing across PIB quartiles: Q1 = 0.5%, Q2 = 2.5%, Q3 = 6.8%, Q4 = 11.3% (p < .001). Predictors of observed assignment and model-recommended treatment modestly overlapped (Spearman ρ = 0.36).Conclusions and Relevance: Causal forest recommendations diverged substantially from observed care, with current practice markedly underutilizing cCBT. Personalized treatment selection could improve reliable recovery rates by up to 11%.

Mind–body practices like meditation and yoga, which are widely used to support mental health, involve paying attention to internal bodily sensations like the breath. During these practices, individuals often report "interoceptive lapses," moments when attention drifts away from the body. While lapses in attention to the external world have been widely studied, little is known about the physiological and neural mechanisms of interoceptive lapses. Interoceptive lapses may share markers with exteroceptive attention lapses — such as reaction time variability and changes in default–mode network (DMN) connectivity—but may also depend on distinct brain systems and breathing physiology. Here we examined behavioral, physiological and neural signals preceding lapses in a sample of 93 adolescents enriched for GAD and depression symptoms. Participants performed a 20–minute breath counting task in the fMRI scanner with simultaneous breath recordings. Lapses were defined as moments when counting errors occurred. The sample was split into a training and validation sample, and machine learning models predicting attentional lapses were tested. The strongest predictors were timing and variability of button responses (AUCs > 0.75). Breathing variability and breathing–behavior coupling showed smaller but generalizable predictive value (AUCs < 0.65). Whole–brain connectivity models also predicted lapses (AUC ≈ 0.65), incorporating the DMN, dorsal and ventral attention, and somatomotor networks. Further, models that included brain connectivity marginally outperformed behavior–only models. Comparisons to previous exteroceptive findings indicate some common markers (e.g., reaction time variability) and some unique markers (e.g., selective perceptual coupling with attentional networks). Although limited by the clinical sample and lack of a control task, these results highlight brain–body markers of interoceptive attention that may inform real–time monitoring during mind–body interventions.