Background: Depression is one of the most prevalent mental health disorders globally, affecting approximately 280 million people and frequently going undiagnosed or misdiagnosed. The growing ubiquity of wearable devices enables continuous monitoring of activity levels, providing a new avenue for data-driven detection and severity assessment of depression. However, existing machine learning models often exhibit lower performance when distinguishing overlapping subtypes of depression and frequently lack explainability, an essential component for clinical acceptance.
Objective: This study aimed to develop and evaluate an interpretable machine learning framework for detecting depression and classifying its severity using wearable-actigraphy data, while addressing common challenges such as imbalanced datasets and limited model transparency.
Methods: We used the Depresjon dataset and applied Adaptive Synthetic Sampling (ADASYN) to mitigate class imbalance. We extracted multiple statistical features (eg, power spectral density mean and autocorrelation) and demographic attributes (eg, age) from the raw activity data. Five machine learning algorithms (logistic regression, support vector machines, random forest, XGBoost, and neural networks) were assessed via accuracy, precision, recall, F1-score, specificity, and Matthew correlation constant. We further used Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) to elucidate prediction drivers.
Results: XGBoost achieved the highest overall accuracy of 84.94% for binary classification and 85.91% for multiclass severity. SHAP and LIME revealed power spectral density mean, age, and autocorrelation as top predictors, highlighting circadian disruptions' role in depression.
Conclusions: Our interpretable framework reliably identifies depressed versus nondepressed individuals and differentiates mild from moderate depression. The inclusion of SHAP and LIME provides transparent, clinically meaningful insights, emphasizing the potential of explainable artificial intelligence to enhance early detection and intervention strategies in mental health care.
Background: Cross-sector collaboration is increasingly recognized as essential for addressing complex health challenges, including those in mental health. Industry-academic partnerships play a vital role in advancing research and developing health solutions, yet differing priorities and perspectives can make collaboration complex.
Objective: This study aimed to identify key principles to support effective industry-academic partnerships, from the perspective of industry partners, and develop this into actionable guidance, which can be applied across sectors. Mental health served as a motivating example due to its urgent public health relevance and the growing role of digital innovation.
Methods: Using a 3-stage, mixed-methods approach, we conducted a web-based survey of UK-based digital mental health companies (N=22) to identify key barriers and facilitators to industry-academic partnerships. This was followed by 2 focus groups (n=5) that explored emerging themes from the survey using thematic analysis. Finally, we conducted a workshop with industry representatives, researchers, clinicians, and PPI members to co-develop the Principles of Industry-Academic Partnerships (PIP) guidance.
Results: Survey findings highlighted that industry partners valued academic collaboration for enhancing credibility, facilitating knowledge transfer, and gaining access to PPI networks. However, key barriers included high costs, slow academic timelines, and complex contracting processes. The 4 major themes that emerged from the focus groups were: advantages of collaboration, cultural differences between organizations, collaboration models, and structural barriers within universities. Through informed discussions in the workshop, these themes were explored, leading to the development of 14 actionable strategies. These strategies reflect industry perspectives and formed the PIP guidance, categorized under project initiation, defining the scope and agreements, project execution, and promoting sustainability.
Conclusions: The PIP guidance provides a practical framework to support more effective and mutually beneficial collaborations between industry and academia. Developed through the lens of mental health research, the strategies identified are broadly applicable across disciplines where cross-sector partnerships are essential. Industry partners valued academic collaborations for their credibility and scientific rigor, but highlighted persistent structural and cultural barriers within universities. Addressing these challenges by aligning expectations and timelines, adopting flexible collaboration models, and streamlining operational processes can help foster impactful and sustainable partnerships in mental health and beyond.
Background: Youth mental health issues have been recognized as a pressing crisis in the United States in recent years. Effective, evidence-based mental health research and interventions require access to integrated datasets that consolidate diverse and fragmented data sources. However, researchers face challenges due to the lack of centralized, publicly available datasets, limiting the potential for comprehensive analysis and data-driven decision-making.
Objective: This paper introduces a curated directory of publicly available datasets focused on youth mental health (less than 18 years old). The directory is designed to serve as critical infrastructure to enhance research, inform policymaking, and support the application of artificial intelligence and machine learning in youth mental health research.
Methods: Unlike a systematic review, this paper offers a brief overview of open data resources, addressing the challenges of fragmented health data in youth mental health research. We conducted a structured search using 3 approaches: targeted searches on reputable health organization websites (eg, National Institutes of Health [NIH] and Centers for Disease Control and Prevention [CDC]), librarian consultation to identify hard-to-find datasets, and expert knowledge from prior research. Identified datasets were curated with key details, including name, description, components, format, access information, and study type, with a focus on freely available resources.
Results: A curated list of publicly available datasets on youth mental health and school policies was compiled. While not exhaustive, it highlights key resources relevant to youth mental health research. Our findings identify major national survey series conducted by organizations such as the NIH, CDC, Substance Abuse and Mental Health Services Administration (SAMHSA), and the U.S. Census Bureau, which focus on youth mental health and substance use. In addition, we include data on state and school health policies, offering varying scopes and granularities. Valuable health data repositories such as ICPSR, Data.gov, Healthdata.gov, Data.CDC.gov, OpenFDA, and Data.CMS.gov host a wide range of research data, including surveys, longitudinal studies, and individual research projects.
Conclusions: Publicly accessible health data are essential for improving youth mental health outcomes. Compiling and centralizing these resources streamlines access, enhances research impact, and informs interventions and policies. By improving data integration and accessibility, it encourages interdisciplinary collaboration and supports evidence-based interventions.
Background: Suicide-related internet use encompasses various web-based behaviors, including searching for suicide methods, sharing suicidal thoughts, and seeking help. Research suggests that suicide-related internet use is prevalent among people experiencing suicidality, but its characteristics among mental health patients remain underexplored.
Objective: This study aimed to examine the sociodemographic, clinical, and suicidality-related characteristics of suicidal mental health patients who engage in suicide-related internet use compared with those who do not.
Methods: A cross-sectional survey was conducted from June to December 2023, recruiting participants aged 18 years and older with recent contact with secondary mental health services in the United Kingdom. The survey assessed sociodemographic characteristics, psychiatric diagnoses, suicidal thoughts and behaviors, and engagement in suicide-related internet use. Statistical analyses included chi-square tests, Wilcoxon tests, and multivariable logistic regression to identify predictors of engaging in suicide-related internet use.
Results: Of 696 participants, 75% (522) engaged in suicide-related internet use in the past 12 months. Those who engaged in suicide-related internet use were almost 3 times as likely to have attempted suicide in the past year (32.5% vs 9.2%, P<.001). They were more likely to have a diagnosis of personality disorder (34.4% vs 18.5%, P<.001) and to disclose suicidal thoughts to someone (87.8% vs 72.8%, P<.001). They also reported higher levels of suicidal ideation intensity (median =6.6 vs 5.1, P<.001). There were no significant sociodemographic differences between groups, including age.
Conclusions: The findings suggest that suicide-related internet use is a common behavior among suicidal mental health patients across various age groups, challenging the notion that it is primarily a concern for younger populations. The association between suicide-related internet use and increased suicidality highlights the need for clinicians to incorporate discussions about web-based behaviors in suicide risk assessments. Given the high rate of disclosure of suicidal thoughts among suicide-related internet users, clinicians may have an opportunity to engage in open, nonjudgmental discussions about their patients' internet use.
Background: Although contingency management has shown some efficacy in substance use disorder treatment, digital contingency management (DCM) needs more evidence supporting its value in treating substance misuse.
Objective: This study aimed to evaluate the effectiveness of DCM in treating substance use disorder by examining 2 key outcome variables-abstinence and appointment attendance.
Methods: A 12-month quasi-experimental design was conducted by enrolling patients into 2 groups using an alternating assignment process: one group receiving treatment-as-usual plus DCM and the other receiving treatment as usual with no contingency management. Propensity score matching was conducted to match groups on covariates. After matching, t tests were conducted to examine the difference between groups on urine abstinence and appointment attendance rates.
Results: Two cohorts of propensity-matched patients (66 interventions and 59 controls) were analyzed. Abstinence was significantly higher in the DCM group (mean 0.92, 95% CI 0.88-0.96) than in the treatment-as-usual group (mean 0.85, 95% CI 0.79-0.90; P<.01). Appointment attendance also demonstrated significant differences between the groups, with the DCM group achieving a mean rate of 0.69 (95% CI 0.65-0.74) compared with 0.50 (95% CI 0.45-0.55) in the treatment-as-usual group (P<.001). This notable increase highlights the role of DCM in fostering engagement with care, an essential factor for successful treatment outcomes.
Conclusions: The results suggest that DCM can be an effective treatment modality for substance use disorder.
Background: Nonsuicidal self-injury (NSSI) is common among adolescents and is associated with adverse clinical outcomes, as well as suicidal behavior. Current treatments are resource-intensive and may not be accessible to all adolescents with NSSI. Internet-delivered emotion regulation individual therapy for adolescents (IERITA) with NSSI disorder is a promising treatment option, but its cost-effectiveness is unknown.
Objective: This study aims to evaluate the cost-effectiveness of IERITA for adolescents with NSSI disorder.
Methods: Within-trial cost-effectiveness analysis of a randomized controlled trial at three child and adolescent mental health services in Sweden (n=166). A total of 12 weeks of IERITA plus treatment as usual (TAU) versus TAU only were compared. The primary outcome was the frequency of NSSI at 1-month posttreatment. Secondary outcomes were NSSI remission and quality-adjusted life years (QALYs).
Results: IERITA led to reductions in NSSI frequency, a higher proportion of participants with NSSI remission, and more QALYs at 1-month posttreatment, at additional health care costs of US $3663 (95% CI US $2182-$5002) and societal costs of US $4458 (95% CI US $-577 to $9509). The incremental cost of one additional NSSI remission at 1-month posttreatment was US $18,677, and the incremental cost per QALY gained was US $792,244 for IERITA+TAU relative to TAU. IERITA had an 8% probability of being cost-effective at a societal willingness-to-pay threshold of US $84,000 for one QALY at 1-month posttreatment, which increased to 18% at 3-months posttreatment.
Conclusions: IERITA delivered adjunctive to TAU led to improvements in NSSI frequency, remission, and QALYs, at additional costs compared to TAU only. This study provides an estimate of the additional cost of delivering IERITA; however, future studies should include longer follow-up periods to better assess the magnitude of the effects on QALYs and societal costs.
Background: Just-in-time adaptive interventions (JITAIs) aim to provide psychological support during critical moments in daily life.
Objective: This preregistered study aims to evaluate the feasibility of a social support JITAI for individuals with subclinical and clinical levels of depressive symptoms awaiting psychotherapy. Triggered by ecological momentary assessment (EMA) reports, the intervention encouraged participants to activate their (digital) social support networks.
Methods: A total of 25 participants completed 2689 EMA surveys and received 377 JITAIs over an 18-day intervention period, including a microrandomized trial, to compare 4 strategies to trigger an intervention: fixed cutoff points of distress variables, personalized thresholds (through Shewhart control charts) of distress variables, momentary support need, and no intervention.
Results: The results showed high feasibility, with participants completing 85.37% (2689/3150) of the EMA surveys, exhibiting a low study-related attrition rate (7%; total attrition rate was 17%), and reporting minimal technical issues. Engagement and perceived helpfulness were heterogeneous and moderate, with participants seeking support in one-third of the instances after an intervention was triggered instances. JITAIs triggered by self-reported need for support were rated as more appropriately timed, helpful, and effective for promoting support-seeking behavior compared to those based on distress indicators, despite being triggered less frequently. Barriers, such as time constraints and perceived unavailability of support providers, likely affected support-seeking behavior, as indicated by additional qualitative analyses. Exploratory effectiveness analyses indicated Cohen d effect sizes between 0.06 and 0.14 in reducing distress after JITAIs were received.
Conclusions: The findings of this study demonstrate that a social support JITAI is feasible to implement, with high compliance and minimal technical issues. However, further research is needed to evaluate the JITAI's effectiveness and optimize trigger strategies in addressing individual needs for and barriers to engagement.

