Reducing mental health struggles in adolescents and young adults is a clinical and public health priority, given the prevalence of neuropsychiatric symptoms and their potential to disrupt key life transitions and the foundation for later independence. Yet, effective early identification and support of vulnerable youth require improved stratification of risk for psychopathology and associated psychosocial difficulties. In adults, the application of machine learning (ML) methods to large-scale electronic health record (EHR) data has shown some promise for improving prediction of psychiatric outcomes. In this focused review, we discuss the small but growing body of work applying ML to EHR data across childhood and adolescence in the context of this adult literature. In doing so, we highlight efforts to augment structured EHR data with dimensional and clinically-relevant constructs based on natural language processing (NLP) and patient-reported outcomes. Finally, we note ongoing initiatives and priorities for advancing the evidence base to promote clinically actionable and developmentally informed risk stratification in youth.

Although efficacious treatments that reduce suicide attempt risk have been developed, signatures that indicate that a patient's suicide risk may be stabilizing over the course of treatment for suicide are not well understood. This study examined whether dynamics in suicide risk and resilience processes over the course of treatment differentiated individuals who did versus did not attempt suicide while in treatment.

Participants were adults who completed up to 12 sessions of suicide-focused therapy as part of a randomized clinical trial. Weekly surveys assessed suicide risk and resilience variables and suicide attempts since the last therapy session. Analyses examined progression of single risk indicators over the course of treatment (n = 70). Dynamical systems models were used to compare symptom dynamics for attempt while in treatment (n = 12) and non-attempt while in treatment (n = 45) groups in the first and last 30 days of treatment using singular value decomposition and vector autoregressive multilevel models.

Participants in the non-attempters group showed increasing stability in suicidal mode dynamics over the course of treatment. Interestingly, there were no significant changes in stability observed for the suicide attempters group. Moreover, differences in the dynamics of these groups emerged in the first 30 days of treatment, potentially indicative of an early signature of treatment response.

These results are consistent with extant theory about how suicide risk networks should change over the course of therapy and have important implications for research to identify indicators of suicide risk stabilization.

Depression is a leading cause of disability worldwide, yet evidence on modifiable factors remains limited and fragmented. Using UK Biobank data, we comprehensively evaluated 218 modifiable factors across eight predefined domains and integrated polygenic risk scores to inform prevention. Among 290,278 participants (mean age 56.9 years; 47.4% male) followed for 14 years, 11,131 developed depression. We identified 154 significant modifiable factors across all eight domains, with the largest numbers observed in the health and medical history, blood assays, and lifestyles domains. Neuroticism, early depressive symptoms, poor health, and low income ranked highest. Mendelian randomization supported potential causal effects for 18 modifiable factors. Population attributable fractions suggested that up to 78.1% of cases could potentially be prevented through improvements across the eight modifiable domains, with the largest contributions from the psychosocial domain (19.3%), the health and medical history domain (16.2%), and the lifestyles domain (13.4%). Domain-level estimates were broadly consistent across genetic risk strata. Our findings highlight psychosocial support, healthy lifestyles, and chronic disease management as key strategies for depression prevention.

Generative artificial intelligence (Gen AI) is becoming increasingly used in mental healthcare. While these systems offer scalable, human-like interactions, concerns persist regarding safety, clinical validity, and the absence of systematic assessment frameworks.

This study systematically reviews existing research on Gen-AI mental-health applications to synthesize current evaluation practices and identify key methodological steps needed for pre-deployment assessment.

A systematic search was conducted across major databases (PubMed, IEEE Xplore, and PsycINFO) following PRISMA guidelines. Studies published between 2017 and 2024 examining Gen-AI tools for mental healthcare were included. The final sample of 72 studies was explored to identify assessment domains.

Four domains of assessment emerged: technical performance, which focuses on model accuracy and reliability on mental-health tasks; clinical validity, which examines alignment with evidence-based guidelines and clinician judgments; expert perceptions, which assess perceived quality and safety of AI-generated outputs; and user perceptions, which capture how end-users experience, interpret, and trust these responses.

Findings highlight the need for unified, multilayered frameworks to guide the safe and responsible deployment of patient-facing Gen-AI mental-health tools. The review proposes a high-level pre-deployment assessment framework to support researchers, developers, clinicians, and regulators.

This longitudinal cohort study aimed to determine whether mental health problems in midlife are linked to frailty after age 70, as measured using two sets of frailty criteria.

We used data from 27,958 adults aged 35-60 years in the Trøndelag Health Study (HUNT1, 1984-1986), with frailty outcomes assessed in 9,956 participants aged 70 years and older (HUNT4, 2017-2019). Mental health at baseline was self-reported. Frailty at follow-up was measured using both the Fried criteria and the 35-item HUNT4-Frailty Index. Linear regression models estimated frailty differences by baseline mental health status, adjusted for sociodemographic and lifestyle factors, stratified by age. We applied inverse probability weighting and multiple imputation to reduce selection bias.

Mental health problems at age 35-49 at baseline were associated with 0.52 (95% CI: 0.35-0.69) higher Fried scores and 0.066 (95% CI: 0.048-0.084) higher HUNT4-FI scores at follow-up. For ages 50-60, Fried scores and HUNT4-FI scores were higher by 0.30 (95% CI: -0.02-0.63) and 0.037 (95% CI: 0.006-0.068), respectively.

We observed that midlife mental health problems were associated with higher frailty levels three decades later, across two commonly used frailty models. These findings underscore the importance of both psychological and physical factors in ageing.

Objectives. To describe health care utilization and diagnostic patterns among adults who died while experiencing homelessness in Harris County, Texas, using linked medicolegal and health information exchange (HIE) data. Methods. Decedent records from the medical examiner (2021-2023) were linked to data from the regional HIE. Demographics, causes of death, health care encounters within 12 months of death, and diagnoses were summarized descriptively. Results. Among 659 decedents, deaths rose from 179 in 2021 to 242 in 2023, despite stable point-in-time counts of homelessness. More than half (58%) had recent health care encounters, with utilization rising sharply in the final 3 months of life. Outpatient visits were the most common, and mental and behavioral disorders predominated across all settings. Toxicity and cardiovascular disease were the leading causes of death. Conclusions. Deaths among people experiencing homelessness rose despite continual and escalating health care contacts before death, providing context for identifying potential prevention strategies. Linking medicolegal and clinical data offers a replicable model for local surveillance and supports targeted interventions integrating harm reduction, behavioral health, and continuity of care. (Am J Public Health. Published online ahead of print October 1, 2026:e1-e9. https://doi.org/10.2105/AJPH.2026.308630).

This study aimed to identify classes of illness perceptions in the Danish general population and examine their association with physical and mental health outcomes.

Latent Class Analysis was conducted on 4,999 participants using the Brief Illness Perception Questionnaire (BIPQ) to identify distinct illness perception classes. Cross-sectional and five-year follow-up analyses explored associations between illness perception class and outcomes including symptom severity, physical functioning, anxiety, depression, and screening for functional somatic disorder (FSD).

Three distinct illness perception classes were identified: untroubled, detached, and distressed. The untroubled class showed positive illness perceptions and favourable outcomes. The detached class showed low perceived control and symptom coherence but did not differ significantly from the untroubled class in outcomes. The distressed class showed negative perceptions of consequences, concern, timeline, and emotional responses and had the poorest health outcomes both at baseline and follow-up, even when adjusting for baseline scores. Class membership was significantly associated with all outcomes cross-sectionally and longitudinally.

Illness perception class membership was associated with long-term physical and mental health. Membership to the distressed class predicted poorer outcomes and higher FSD risk, suggesting that negative illness perceptions may precede negative health outcomes and represent a target for intervention.

The transition from psychiatric inpatient care to community settings is associated with heightened vulnerability and increased risk of adverse mental health outcomes. Understanding age-related differences in mental health status at discharge may inform the targeting of post-discharge interventions and potentially improve longer-term outcomes.

This study examined associations between age and baseline demographic, clinical, and mental health characteristics among participants enrolled in the Text4Support program following discharge from psychiatric inpatient care.

Baseline data from 1,106 participants discharged from inpatient psychiatric units across Alberta, Canada, were analyzed. Participants were grouped by age (18-25 years, 26-40 years, and >40 years). Mental health outcomes were assessed using validated self-report measures. Chi-square tests and one-way ANOVA or Welch's ANOVA were used to examine age-related differences in categorized and continuous outcomes, respectively.

Significant age-related differences were observed across several mental health outcomes. Younger participants (18-25 years) reported higher levels of depression, anxiety, and suicidal ideation, along with lower resilience and recovery scores, compared with participants aged >40 years. Differences in quality of life were also observed, with younger adults reporting greater anxiety/depression and participants aged >40 years reporting more mobility-related problems.

Marked age-related disparities in mental health status were evident at the point of psychiatric discharge, with younger adults demonstrating greater vulnerability. These findings underscore the importance of age-informed risk stratification and tailored transitional care strategies following inpatient psychiatric treatment and suggest that different strategies may be required depending upon age.

To identify multidomain predictors of functional decline among nursing home residents with type 2 diabetes (T2D).

A secondary analysis of longitudinal Minimum Data Set 3.0 data from Iowa nursing home residents with T2D was conducted. Three machine learning models were developed to predict functional decline, defined as worsening Activities of Daily Living Long-Form scores between first and last assessments. Predictor variables were guided by the Functional Consequences Theory and represented biological, psychological, and environmental factors.

The sample included 5,440 residents with T2D from 430 nursing homes (mean age 81; 63% female). The multi-class classification model demonstrated the best fit to predict functional decline (AUC = .93). Lower baseline function, poorer cognition, higher depressive symptoms, older age, and a greater number of assessments were the most influential predictors. Residents' belief in improving function, coexisting chronic diseases, and rurality were prominent predictors.

Functional decline in residents with T2D reflects interrelated biological, psychological, and environmental vulnerabilities, several of which are modifiable.

Routine assessment of baseline function, cognition, and depressive symptoms may support early risk identification. Interdisciplinary interventions supporting mental health and residents' confidence in functional improvement may improve outcomes, particularly in rural and medically complex populations.