Communications Psychology

The emotional content of suicide notes is typically examined using categorical coding, where each labeled passage is treated in isolation from its surrounding language. In contrast, dimensional models of psychopathology propose that affective content varies along continuous gradients. We evaluated this proposition directly. Excerpts from 884 annotated suicide notes were embedded in a semantic space defined solely by their linguistic properties, and we investigated whether human-assigned emotion labels changed smoothly across this space. They did: affective tone showed clear spatial autocorrelation (Moran's $I = 0.18$, $z = 19.68$, $p < 0.001$), an effect that replicated across three different encoders and remained after removing all within-note dependencies. Emotions occupied recognizable yet overlapping regions rather than forming distinct clusters and varied substantially in how tightly they were concentrated: love and hopelessness appeared with similar frequency, but love was far more localized ($z = 15.7$ versus $10.8$). Among all emotions, hopelessness was the most linguistically diffuse, implying that a single categorical label is capturing multiple, qualitatively different manifestations of suicidal distress.

Humans and other animals often act in environments that are at least partly familiar, where aspects of the spatial layout are known. Although such knowledge is known to support navigation and spatial cognition, its role in the online control of action remains unclear. We investigated whether drivers use knowledge of road layout to guide steering in high and low visibility and, if so, the form of such knowledge. In two simulated driving experiments (total N = 90), participants repeatedly drove winding roads containing segments with and without fog. Drivers who repeatedly experienced the same road exhibited more stable steering and lane positioning than drivers encountering novel roads, but only when visibility was reduced. These advantages were accompanied by superior performance on post-tests assessing knowledge of road geometry. We next examined the form of such knowledge by dissociating global knowledge of road layout from local associations between landmarks and road segments. Disrupting landmark-road segment associations produced the largest impairment in steering performance. The benefits of prior experience were largely preserved when road-segment order was scrambled but landmark associations remained intact. These findings show that spatial knowledge can support moment-to-moment steering control when visibility is reduced. Rather than relying on a globally coherent representation of the environment, drivers use local associations between landmarks and upcoming road geometry to anticipate future demands. More broadly, the results elucidate how familiarity with environmental structure contributes to the control of action when visual information is degraded, revealing a close interplay between spatial knowledge and visual control. Significance StatementPeople routinely act within surroundings they have encountered before, from commuting on the same streets to walking familiar hallways. Whether the spatial knowledge acquired from such experience actually shapes online visual control remains an open question. Using a simulated driving task, we show that familiarity with a road improves steering stability specifically when visibility is reduced, and that this benefit depends on learned associations between landmarks and upcoming road geometry rather than a global cognitive map. The results indicate that spatial knowledge plays a key role in moment-to-moment control, letting drivers anticipate road segments they cannot yet see. Unfamiliar roads and impaired spatial learning may compound the risks of poor visibility, suggesting a role for driver-assistance systems that leverage landmarks.

Interception of moving objects requires the nervous system to compensate for sensory delays and uncertainty, yet how behavior is controlled remains debated. Key questions concern whether predictive processes play any role at all and, if so, whether they rely on simple motion extrapolation or incorporate internalized physical priors, such as gravity. Another open question is whether observers adopt a single control strategy or flexibly switch between predictive and reactive control - or between different predictive strategies - depending on task demands. To address these questions, we developed a virtual interception task in which participants intercepted moving targets under systematically varied conditions. We manipulated gravity (1g vs. 0g), visual availability (occluded vs. non-occluded), target velocity, and the initial spatial configuration of the ball and paddle (same vs. opposite side). Results indicate that interception is supported by predictive mechanisms across conditions. Behavioral patterns during occluded 0g trials suggest that participants extrapolate target motion using expectations consistent with gravity. Target velocity, visual occlusion, and task geometry modulated movement strategies, indicating that predictive control is flexibly adapted to task demands. These findings support the view that interception relies on predictive internal models incorporating structured physical priors while revealing flexible, context-dependent adaptations to sensory and task constraints.

Cooling the skin may increase its tolerance to mechanical loading and decrease the risk of developing pressure ulcers. Yet, the mechanisms of action (e.g. cooling-modulation of cytotoxic, post-occlusive hyperaemia), and their individual variability, remain unclear. We investigated the effects of different cooling levels (24{degrees}C and 16{degrees}C) on microvascular, inflammatory and perceptual responses to mechanical loading of the sacrum in healthy young (N=23) and older adults (N=19), and in spinal cord injury patients (SCI; N=10). Healthy participants underwent 45-min loading (~60 mmHg) and 20-min unloading of the sacrum, using an instrumented indenter probe set at either 38{degrees}C (control condition), 24{degrees}C or 16{degrees}C. SCI participants completed a more conservative protocol (i.e. 25min, ~45mmHg loading, 38{degrees}C and 16{degrees}C conditions). Pre-insult skin structure was characterised with optical coherence tomography; skin blood flow (SkBF) at the loading site was continuously measured, alongside thermal acceptability; and post-insult inflammatory responses were determined via skin-sebum cytokines analyses. Compared to control, 24{degrees}C- and 16{degrees}C-cooling induced a similar ~8-fold decrease in peak post-occlusive reactive hyperaemia in healthy participants, with similar temperature-related differences observed in SCI. Pro-inflammatory cytokines decreased post-insult; yet this occurred similarly across all temperatures and groups. The majority of participants ([&ge;]70%) rated both 24{degrees}C- and 16{degrees}C-cooling as thermally acceptable. We conclude that cooling is a potent modulator of the skin microvascular response to mechanical loading in younger, older, and vulnerable skin (SCI). These findings can inform design parameters for thermal technology aimed at preventing the loss of skin integrity (e.g. integrating 24{degrees}C-cooling in support surfaces and skin wearables).

Abstract Background Despite a rising global psychiatric burden, a treatment gap persists where the majority of symptomatic individuals remain unmedicated. Traditional epidemiological analyses treat this untreated population as a single, uniform block, obscuring specific barriers to care. This study uses an unsupervised machine learning pipeline to identify distinct socio-behavioural and biological sub-populations within the untreated cohort to guide targeted public health interventions. Methods Data were pooled from the 2015-2018 National Health and Nutrition Examination Survey (NHANES) cycles (N=11,848 total adult respondents). A symptomatic cohort of 3,075 individuals experiencing daily or weekly anxiety or depression symptoms was isolated, excluding severe liver pathology outliers ("GGT"[&ge;]80" U/L" ). A 22-feature matrix combining continuous clinical biomarkers (systolic blood pressure, waist circumference, HbA1c) and categorical social variables was projected using Factor Analysis of Mixed Data (FAMD). Latent sub-populations were identified via Gaussian Mixture Modelling (GMM), optimized by the Bayesian Information Criterion (BIC). Results The broad baseline population revealed a substantial mental health burden, with 30.4% reporting active psychiatric symptoms, of whom 71.6% were entirely unmedicated. The GMM pipeline successfully isolated three distinct sub-populations (k=3) separated by age, clinical strain, and treatment rates: Cluster 0 (Mature Adults, mean age 55.03): high psychiatric severity (34.1% severe untreated), central obesity, and hypertensive strain (135.82 mmHg), with 64.1% untreated despite frequent primary care contact; Cluster 1 (Working Professionals, mean age 38.38): highly educated, female-dominated (70.5%), with 77.7% untreated driven by moderate distress; Cluster 2 (Emerging Youth, mean age 18.49): a highly vulnerable late-adolescent group with a staggering 90.2% untreated rate. Conclusion The unmedicated symptomatic population is highly diverse and segmented by life stage. These profiles show that the treatment gap is driven by age-specific barriers, specifically workforce-age symptom masking and late-adolescent developmental transitions. Closing this deficit requires shifting from uniform public health approaches toward targeted interventions, such as digital peer support networks for youth and integrated primary care screenings for older adults.

AO_SCPLOWBSTRACTC_SCPLOWThere is a gap between neuroscientific theories of learning and the speed of learning observed in many experiments. Since the Cognitive Revolution of the 1950s, compositionality has played a central role in efforts to bridge this gap. Roughly, a compositional system is one where distinct modules are combined according to a set of rules in order to accomplish complex tasks. Recently, significant progress has been made in understanding the emergence of modules in both biological and artificial neural systems. How, and under what conditions, the rules of module recombination are represented in these systems remains an open question. Here we present a neural model that can leverage these rules to dramatically speed up learning. We first show that when faced with multiple tasks which share subcomponents, models learn a low-dimensional representation that captures how subcomponents are reused across the task set. These low-dimensional spaces encode the structure that governs how modules should be recombined. Restricting learning to these subspaces greatly reduces the amount of experience needed to acquire a novel task, even when learning from reinforcement on single trials. In some cases, we can leverage the geometric regularities of these representations to reduce learning to a form of hypothesis testing over a small set of discrete points. Finally, we use this theory to model both behavioral and neural data from non-human primates performing a compositional task, and show that key features in this data are consistent with a model in which exploration during learning is restricted to these low-dimensional spaces. Overall, this work shows that the advantages of modularity in neural systems can be greatly improved upon when models represent the structure of module reuse. Both these features working in tandem lead to learning on timescales similar to biological intelligences, and hence provide a model for how such fast, adaptable behavior can emerge from systems of neurons.

Many neural recordings have revealed low-dimensional sets of behaviorally relevant variables encoded within large-scale neural activity patterns. However, dimensionality reduction analyses alone cannot yield causal explanations for how networks stably implement computations that are resilient to the substantial variability of single neuron dynamics. Further, existing methods for dimensionality reduction often rely on simplifying assumptions about network structure that limit their applicability and explanatory power. To provide a theoretical framework describing the dynamics of low-dimensional computation in high-dimensional neural networks, here we introduce the concept of latent processing units (LPUs), which are architecture-agnostic computational elements operating within biological neural circuitry. Six theorems governing coding and computation by LPUs collectively provide explanations for a range of common biological findings: low-dimensional sets of coding variables can generate high-dimensional neural dynamics; many neurons have activity patterns that represent behaviorally relevant variables but exert little influence on downstream circuits; linear readouts of neural population activity commonly permit near-optimal decoding; the drift of neural representations is often substantial even while network computations remain intact. Overall, our treatment of LPUs, as enacted in network dynamics, unifies the geometric and dynamical views of neural computation under a joint framework and provides systems neuroscience with a causal account of how the brain executes reliable computations.

Large Language Models are increasingly used in consumer-facing mental health tools, many of which claim that prompt engineering alone can ensure safe therapeutic behavior. This study evaluates that assumption by testing 20 proprietary and open-source LLMs on high-risk psychiatric scenarios, using prompts grounded in behavioral therapy principles. Prompt engineering reduced some predictable risks, such as explicit endorsement of self-harm, but consistently failed in ambiguous or clinically nuanced situations. Models frequently validated harmful statements, colluded with hallucinations, minimized symptoms, or used stigmatizing language, including in the newest and largest models. These failures reflect structural limitations such as lack of memory, insufficient contextual reasoning, and training-related biases. Prompt engineering alone is therefore insufficient for safe AI-mediated psychotherapy; clinician-guided fine-tuning, integrated safety mechanisms, and system-level oversight will be required. This work provides early evidence motivating deeper clinician-led evaluation and safety-oriented model development.

Speech perception requires transforming acoustic input into neural representations that support linguistic understanding, yet its underlying computational principles remain unclear. Classical efficient coding theories posit optimal compression of sensory input, whereas alternative accounts propose that neural systems preferentially encode information that supports prediction. A key open question is whether such predictive encoding operates on fixed inputs or on flexible internal representations. We instantiated three hypothesis models of speech processing: (i) optimal compression with deep autoencoders, (ii) predictive reconstruction with predictive autoencoders, and (iii) predictive information representation via latent-space prediction using contrastive learning. We compared resulting speech latent representations to electroencephalographic (EEG) activity during speech listening. Representations learned under the predictive information objective best explained neural latents. Crucially, only representations that selectively compressed predictive information predicted behavioral performance, suggesting that neural speech representations are structured to encode predictive information in latent space rather than to maximize compression or input prediction.

Introduction Auto-Induced Cognitive Trance (AICT) is a non-ordinary state of consciousness (NOSC) that can be accessed by will alone once a standardised self-induction procedure has been learnt. The first research publication on AICT dates back only ten years, meaning that research on this phenomenon is still in its infancy. Previous reports concerning the phenomenology and neurophysiology of AICT revealed similarities with more extensively described NOSCs, as well as unusual features, raising questions about the potential benefits of AICT practice for well-being. Objective This study aimed to gather quantitative descriptive data on features associated with well-being in a large comparative sample of AICT practitioners and non-practitioners. Method This research followed a web-based survey study design which enquired AICT-trained and yet-to-be trained participants to self-report through validated standardised questionnaires on vitality, self-esteem, mental well-being, trait anxiety, life satisfaction, happiness, positive and negative affect, nature-relatedness and connectedness. Data on NOSCs practices, life history events that could have led to spontaneous NOSCs, and demographic data were collected for further inclusion as control variables in statistical models. Results The online questionnaire yielded 607 valid responses, (171 yet-to-be trained participants and 436 AICT-trained participants). AICT practice was found to be associated with increased self-esteem (RSE), overall connectedness (WCS) as well as all subdimensions of connectedness (WCS Self, WCS Others, WCS World). AICT practice Duration exhibited significant effects on global connectedness and all subdimensions of connectedness, self-esteem, trait anxiety (STAIT-5), and positive affect (PANAS+). Conclusions AICT seems to benefit to practitioners well-being shortly after training through increases in self-esteem and in the sense of connectedness. Prolonged AICT practice is associated with added decreased trait anxiety and increased positive affect. Further research is needed to confirm these findings with a sample including AICT-uninterested participants, and to clarify the underlying mechanisms of AICT.

The increasing deployment of artificial intelligence (AI) assistive systems across healthcare, education, and organisational domains necessitates a deeper understanding of dispositional factors shaping trust and acceptance. This study investigated the Big Five personality traits as predictors of trust in and acceptance of AI assistive systems among a large adult sample (N = 380) in Makurdi Benue State. Anchored in the Technology Acceptance Model (TAM) developed by Davis (1989), the study examined both direct and indirect pathways linking personality traits to AI acceptance through trust. Participants completed standardised measures of the Big Five Inventory, Trust in AI Scale, and AI Acceptance Scale. Data were analysed using structural equation modelling (SEM) with maximum likelihood estimation. The hypothesised model demonstrated good fit indices (CFI = .84, TLI = .82, RMSEA = .05). Openness to experience ({beta} = .34, p < .001) and agreeableness ({beta} = .27, p < .01) significantly predicted trust in AI systems, which in turn strongly predicted AI acceptance ({beta} = .62, p < .001). Neuroticism negatively predicted trust ({beta} = -.29, p < .001), while conscientiousness showed a modest positive direct effect on acceptance ({beta} = .18, p < .05). Extraversion was not a significant direct predictor but exerted an indirect effect through trust. Mediation analysis confirmed that trust significantly mediated the relationship between personality traits and AI acceptance. The findings underscore the centrality of dispositional traits in shaping technological trust formation and highlight the psychological architecture underlying human AI interaction. These results contribute to social psychological theory and provide empirical guidance for designing personality sensitive AI systems to enhance user adoption and sustained engagement.

Boredom - a pervasive mental state - promotes the pursuit of novel information by assigning negative value to monotonous conditions. Yet, how the brain extracts and represents the information content of ongoing sensory experience remains poorly understood. Here, we combine behavioral assays, neurophysiological recordings and computational modeling across humans and mice to investigate how sensory information shapes boredom-related behavior. In a cross-species choice task, both humans and mice robustly avoid monotonous sources of sensory stimulation. We formalize perceived monotony using empirical entropy as a measure of information content and show that monotony avoidance scales directly with low entropy and in humans correlates with boredom experience. Human electroencephalography and mesoscopic calcium imaging in mice reveal that the recruitment of neocortical activity tracks stimulus entropy. Two-photon calcium imaging in the auditory cortex of mice further uncovers a stimulus-invariant population code for entropy, supported by neurons tuned to information content. A recurrent network model reproduced this code through an interplay of afferent depression and recurrent facilitation. Together, we demonstrate how the information content of sensory experience is represented in cortical population activity, providing a basis for boredom-related avoidance behavior. Thus, our findings link synaptic and neuronal dynamics to boredom, acting as a safeguard mechanism to ensure high information input to the brain.

Background: Bidirectional next-day associations between sleep disturbances and affective symptoms have been shown in previous research, yet the consecutive day effects between these factors remains poorly understood. Methods: We analysed longitudinal ecological momentary assessment (EMA) data obtained from a subsample of young persons in the Brazilian High-Risk Cohort (BHRC) study collected in 2020-2021. Participants reported sleep quality each morning and rated affective symptoms relating to mood, anxiety, and energy four times daily for 28 days. We selected 88 individuals (17.83{+/-}1.74 years, 56 [63.6%] female gender) with at least one instance where individuals were observed three-days in a row. Within-person bidirectional next-day effects between sleep quality and affective symptoms were estimated using mixed-effects regression analysis adjusting. We then applied g-estimation approaches to estimate the effect that lagged sleep quality and consecutive improvements in sleep quality had on affective symptoms. Results: Sleep quality and affective symptoms had bidirectional next-day effects, with sleep quality tending to have greater influence on affective symptoms than the reverse. Improved lagged sleep quality had positive effects on affective symptoms incrementally above the prior night's sleep quality. Also, improvement of sleep quality across consecutive days had incremental and approximately equal effects on affective symptoms. Conclusions: Sleep quality and affective symptoms exhibit a feedback loop, whereby poor sleep quality influences affective symptoms over consecutive days. Breaking these feedback loops, by improving sleep quality across several consecutive nights should improve affective symptoms. This supports interventions that target sustained improvement in sleep and possibly circadian regulation to improve affective symptoms.

The increased vulnerability to mental health problems in adolescence is frequently reported but poorly understood, hampered by a rigid diagnostic system which fails to capture intertwining symptoms and only loosely aligns with biological axes of variability. Here, we reconceptualised the mental health symptoms of young adolescents in the ABCD cohort (N=11862) as a latent topology of overlapping symptom dimensions, using an unsupervised machine learning algorithm to establish how transdiagnostic dimensions co-occur and overlap within individuals. Combining this with a novel classification approach, we delineated zones within this landscape, within which specific profiles of symptoms were robustly represented. These data-driven profiles were leveraged to establish associated resting-state functional connectivity and genetic characteristics. In doing so we recaptured the commonly reported p-factor axis as well as further symptom-subtype dimensions. Gene ontology analysis revealed that shared neurobiological and cellular mechanisms embedded in both the genome and transcriptome may confer risk for psychopathology.

Collective animal behavior occurs in high-stakes contexts where failing to coordinate effectively with group-mates can spell disaster for individuals. Yet, identifying instances of coordination failure is challenging, meaning their evolutionary effects remain mysterious. Synchronous calls in alternating frog choruses (i.e., inadvertent signal collisions) are unambiguous failure events that impose steep attractiveness costs. We modeled tungara frog chorusing dynamics to reveal the sensorimotor and social mechanisms underpinning synchrony. Ultimately, inter-male variation in two key sensorimotor attributes, the periods of male calling rhythms and call latencies, generated divergent synchrony engagement patterns. Modeling female preferences revealed that these varied behavioral outcomes then yielded disparate attractiveness consequences. By mechanistically linking the causes and consequences of coordination failure, we demonstrate that non-random failure patterns in collectives generate selection gradients that refine sensorimotor tuning.

Purpose: Suicide mortality is underestimated due to misclassification under undetermined and accidental deaths. This study examined national trends in suicide and related external causes of death in Portugal from 2002 to 2023, by sex and age group, assessing potential shifts suggesting masked suicide and quantifying the relationship between undetermined, suicide, and accident death rates through ratio indices. Methods: Using official mortality data from Portugal's Statistics Institute (INE) for 2002-2023, we calculated age-standardised (SDR) and age-specific death rates (ASDR) for suicide (X60-X84), undetermined intent deaths (Y10-Y34), and unintentional deaths (V01-X59), disaggregated by sex and four age groups (15-24, 25-44, 45-64, 65+). We estimated undetermined-to-suicide (UnD:Suic) and undetermined-to-accidents (UnD:Accs) rate ratios for SDRs and ASDRs. Trends were analysed using joinpoint regression (APC/AAPC) and structural breakpoint analysis (Chow test, BIC). Results: Suicide SDRs declined across the period for males (AAPC: -2.25%) and females (AAPC: -1.32%), with the sharpest reductions among males aged 25-44 (AAPC: -2.56%) and females aged 65+ (AAPC: -2.44%). Deaths of undetermined intent rose steeply from 2002 to 2005-2006 and declined thereafter. Unintentional deaths declined in most age groups, except females aged 65+ (AAPC: +1.41%). Both ratio series peaked around 2005-2009, declined progressively through the 2010s, and reached their lowest values in 2021-2022. Age-specific analyses revealed a significant and sustained increase in both ratios among females aged 45-64. Structural breakpoints clustered around 2004, 2013-2015, and 2019-2020. Conclusion: Suicide mortality declined in Portugal from 2002 to 2023, but divergent trends in undetermined and accidental deaths across sex and age subgroups highlight ongoing misclassification. Age- and sex-specific ratio analyses identify the population subgroups where misclassification is most concentrated, providing a foundation for future imputation-based estimates of probable suicide burden.

Importance: Patients who achieve remission from major depressive disorder (MDD) often face a preference-sensitive decision between continued antidepressant maintenance and discontinuation with active monitoring. Quantifying the tradeoff between depression burden and long-term medication exposure may support more individualized shared decision-making. Objective: To quantify tradeoffs between continuous antidepressant maintenance and active monitoring after MDD remission, and to identify preference thresholds favoring each strategy across relapse-risk strata. Design: Individual-level decision-analytic health-state transition model calibrated to randomized maintenance-discontinuation trials and a longitudinal first depressive episode cohort, with a 5-year time horizon. Setting: Outpatient clinical decision after completion of an 8-month continuation phase following remission from MDD. Participants: Adults in remission from MDD, represented across 4 clinically anchored relapse-risk strata ranging from very low risk after a first mild episode to high risk after highly recurrent depression. Exposures: Continuous antidepressant maintenance vs discontinuation with active monitoring and antidepressant restart after detected relapse. Main Outcomes and Measures: Severity-weighted depression-months, antidepressant medication-years, medication-years per depression-month averted, and net benefit across preference thresholds defined as the maximum additional medication-years a patient would be willing to accept to avert 1 depression-month. Results: Continuous maintenance reduced depression burden but required substantially more medication exposure, with efficiency strongly dependent on relapse risk. Medication-years per depression-month averted ranged from 11.8 (95% uncertainty interval [UI], 7.8-19.6) in the very low-risk group to 1.5 (95% UI, 0.8-3.0) in the high-risk group. At a preference threshold of 3 medication-years per depression-month averted, maintenance was preferred for moderate- and high-risk patients; at a threshold of 2, only for high-risk patients; and at a threshold of 1, for no risk group. Conclusions and Relevance: In this decision-analytic model, the value of continuous antidepressant maintenance depended strongly on baseline relapse risk and patient preferences regarding long-term medication exposure. These findings provide a quantitative framework for shared decision-making about antidepressant maintenance after remission from MDD.

Automated interpretation of neuroimaging studies requires simultaneous assessment of multiple imaging evidence variables, each tied to distinct anatomical structures. Vision-language models (VLMs) offer a unified framework for multi-task analysis, but adapting pre-trained VLMs remains challenging. Full fine-tuning is computationally prohibitive, and joint multi-task training requires simultaneous access to all task data, which is often infeasible in clinical settings. Although model merging enables multi-task composition without joint re-training, existing methods focus on post-hoc algorithms with limited extension to VLMs and minimal application to neuroimaging. Here, we present GRadient-guided Adapter Merging (GRAM), a layer-selective low-rank adaptation (LoRA)-based fine-tuning and merging framework for multi-task neuroimaging visual question-answering (VQA). GRAM uses a gradient ratio that contrasts class-specific gradients to identify task-discriminative layers, and applies subspace-constrained projected gradient descent to restrict LoRA updates to directions consistent with the geometry of the pre-trained model. We leveraged a structured VQA benchmark, developed from the National Alzheimer's Coordinating Center (NACC) dataset, that pairs multi-sequence brain MRI studies with question-answer pairs across clinically relevant imaging evidence variables. Experiments on the VQA benchmark showed that GRAM outperformed or matched all-layer LoRA fine-tuning and a standard merging baseline while reducing inter-task interference during merging, and approached or surpassed the performance of joint multi-task training without joint re-training.

Background. UK clinical guidance recommends that structured risk prediction tools and risk stratification should not be used in self-harm, to predict suicide or determine who is offered treatment. Underpinning this position is the premise that routinely collected health data contain no useful predictive signal, which has received little direct scrutiny. Objective. To test whether routinely collected electronic health record data can distinguish groups at higher and lower risk of severe outcomes following paracetamol overdose. Methods. We analysed 4,095 adults presenting to NHS Lothian emergency departments with paracetamol overdose (2017-2023). Elastic-net logistic regression was fitted to 37 routinely collected electronic health record features to predict a composite of death or mental health inpatient admission at 0-7, 8-30 and 31-365 days following attendance, evaluated on a held-out 20% test set with bootstrapping. Findings. Events occurred in 5.5% of patients at 0-7 days, 2.0% at 8-30 days and 7.9% at 31-365 days, dominated by mental health admission. Bootstrap AUROC 95% confidence intervals lay above 0.5 in every window (0.65-0.82, 0.63-0.90, 0.71-0.85): models ranked patients better than chance. Calibration slopes (1.04, 1.14, 1.07) were close to one. Ranking drew primarily on mental health-related features. Conclusions. Routinely collected health data carried predictive signal for severe outcomes after paracetamol overdose, although discrimination fell short of what is needed for individual-level clinical use. Clinical implications. These models are not proposed for clinical deployment; however, treating risk prediction as a settled question will redirect research efforts, potentially excluding this patient population from machine learning advances driving improvements in care in other medical specialties.

Introduction: Group interpersonal psychotherapy may improve outcomes beyond depressive symptoms, but evidence on functioning, social support, and household welfare remains limited. We examined these outcomes using mixed methods in a pilot cluster-randomized controlled trial of a six-week intervention in rural Uganda. Methods: Twenty-four villages were randomized to group interpersonal psychotherapy or enhanced care as usual. Eligible participants were female, aged at least 13 years, with Patient Health Questionnaire-9 scores of 10 or more. Outcomes were assessed at baseline, two weeks, and three months after treatment. Regression models used village-clustered standard errors. Exploratory sensitivity analyses compared controls with 33 intervention participants assessed independently of facilitators and after an honesty and confidentiality prompt. Fourteen interviews and three focus group discussions with 30 intervention participants were analysed thematically and integrated by outcome domain. Results: Of 292 randomized participants, 263 completed the three-month assessment. Intervention participants had lower anxiety scores than controls at three months (mean difference -7.18; p<0.001), higher subjective wellbeing (mean difference 3.70; p<0.001), lower disability scores (mean difference -1.01; p<0.001), greater perceived social support (difference 23.4 percentage points; p<0.001), and more meals reported for children in the previous 24 hours (mean difference 0.62; p<0.001). Household food insecurity was lower in the full-sample analysis (difference -23.3 percentage points; p=0.008), but not in the exploratory sensitivity analysis (difference 1.5 percentage points; p=0.883). Qualitative accounts described recovery as restored capacity to work, manage relationships, care for children, and respond to hardship despite material constraints. Conclusions: These preliminary findings suggest that group interpersonal psychotherapy may improve outcomes beyond depressive symptoms, although household welfare findings were less consistent. Larger trials with independent outcome assessment and longer follow-up are needed. Trial registration: The trial was retrospectively registered with the Pan African Clinical Trials Registry (PACTR202606549854263) on 29 June 2026.