Clinical large language models (LLMs) are increasingly used for documentation, diagnosis, and decision support, but their opaque reasoning can limit clinician trust, regulatory assessment, and safe deployment. Explainability research has expanded rapidly, yet existing reviews largely address traditional machine learning or general-domain LLMs. We conducted a PRISMA-ScR scoping review to map explainability approaches for decoder-only clinical LLMs with over one billion parameters, searching PubMed, Scopus, Web of Science, ACM Digital Library, and arXiv through early 2026. Among 69 included studies, LLM-native generative and interactive methods dominated (58.0%, n = 40), spanning chain-of-thought rationales, retrieval-augmented evidence citation, and agentic decomposition. Intrinsic by-design methods accounted for 24.6% (n = 17); post-hoc XAI methods accounted for 17.4% (n = 12). General medicine was the most represented clinical domain, and diagnosis was the dominant task. Proprietary models were used in 75.4% of studies, yet every mechanistic analysis relied on open-source models, revealing a transparency asymmetry: most deployed models are the least transparent. Although 59.4% of studies quantitatively evaluated explanations, metrics remain non-standardized and rarely assess faithfulness. Local explanations predominated, and no study prospectively evaluated explanations in live clinical workflows. These findings show that clinical LLM explainability has shifted toward fluent generative rationales, but evidence that such explanations reflect model reasoning remains limited. To support trustworthy deployment, we highlight three regulatory priorities: prioritizing explanations that enable independent verification or logic auditing over plausibility-only rationales; preferring inspectable models where regulatory documentation is required; and prospectively validating explanations in clinical workflows before scaling.

Data on effectiveness of COVID-19 vaccinations during the 2024-2025 respiratory season are limited, particularly among those with underlying medical conditions (UMC). We estimated BNT162b2 KP.2 vaccine effectiveness (VE) against COVID-19-associated hospital admission, emergency department (ED), and urgent care (UC) visits in two U.S. states.

Retrospective cohort study of non-immunocompromised adults living in Louisiana or California, with ≥1 year prior continuous enrollment in insurance plans contributing to the HealthVerity claims database beginning August 22, 2024. The effectiveness of BNT162b2 KP.2 vaccine (2024-2025 formulation, hereafter referred to as BNT162b2), measured as a time-varying exposure against hospital admission, ED, or UC encounters with International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) code U07.1 was calculated as 1 - adjusted hazard ratio using Cox proportional hazard models adjusted for age group, sex, state, insurance payor, presence or absence of UMCs, and pre-index healthcare utilization. Stratifications included those aged 65 years and older, those aged 18-64 years with UMCs, and those aged 18-64 years without UMCs.

The cohort included 6,256,421 individuals (93% California, 7% Louisiana); 330,565 (5%) received the BNT162b2 vaccine. Vaccinated individuals were older and had more comorbidities, wellness visits, and prior influenza vaccination. Overall, 66% of the study population had ≥1 UMC; the most prevalent conditions were obesity (25%), history of immunocompromised conditions (23%), and mental health conditions (19%). COVID-19-related encounter rates for ED, UC or hospitalization were lower among vaccinated compared to unvaccinated persons (25.1 vs 36.3 per 100,000 person-months). Among all adults, VE was 37% against hospitalization, 12% against ED/UC encounters, and 16% against ED/UC/hospitalization encounters. Results were similar across age groups and UMCs.

BNT162b2 provided protection against COVID-19-associated outcomes of ED, UC or hospitalization among non-immunocompromised U.S. adults, including those with UMCs, over the course of the 2024-2025 respiratory virus season, supporting continued vaccine recommendations.

This study was posted on clinicaltrials.gov prior to analyses (NCT06923137).

Loneliness can have profound health consequences for cancer patients, yet age-related patterns of loneliness remain inconsistent and mainly cross-sectional. We describe loneliness in young (<40 years), middle-aged (40-65 years), and older (≥65 years) patients over time during early cancer survivorship, identify biopsychosocial risk factors for loneliness, and examine its impact on health outcomes across age groups.

This multicenter prospective longitudinal study included cancer patients within two months of diagnosis and at 6-, 12-, and 18-month follow-ups. Loneliness was measured using the UCLA loneliness scale. Generalized-linear-mixed-models examined changes over time. Mixed-effects regression models identified risk factors for loneliness and examined the predictive value of loneliness on physical and mental health outcomes over time.

In total, 994 patients (53% men, 60.5 years) were included in this analysis. Older patients reported lowest and most stable loneliness over time (15-20%), whereas middle-aged (19-31%) and young (25-38%) patients reported higher loneliness values, particularly shortly after diagnosis. Risk factors for being lonely included lower social support, absence of a partner, lower socioeconomic status, advanced disease, and greater comorbidity burden (all p < 0.05). Higher loneliness at diagnosis predicted mental disorders and psychosocial care needs at follow-ups, in addition to distinct age-specific relationships of reduced physical and mental QoL for middle-aged, and non-adherence for older patients (all p < 0.05).

Loneliness is common during early cancer survivorship, particularly among young and middle-aged patients shortly after cancer diagnosis. Age-related mechanisms, vulnerabilities and consequences of loneliness may inform clinical communication, e.g. about non-adherence or recommendation of tailored psychosocial support.

This study was registered in the International Clinical Trials Registry (NCT04620564, https://clinicaltrials.gov/).

Polygenic risk scores for Alzheimer's disease (AD-PRS) are widely used to estimate genetic susceptibility to AD, but their relationship with the rate of cognitive decline (CD) after clinical onset remains insufficiently characterized.

To examine the association between AD-PRS and longitudinal CD across the AD spectrum and to evaluate the predictive contribution of individual AD-PRS variants.

Large longitudinal observational study in a single-center cohort, with an external cohort to assess generalizability.

Memory clinic cohort from Ace Alzheimer Center Barcelona (Ace) with external cohort using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

The study included 7,233 patients from Ace and 863 from ADNI, with a mean follow-up of 5.4 years in Ace and 3.6 years in ADNI. A biomarker sub-cohort included 1075 participants from Ace and 569 from ADNI.

CD was quantified as the annual change in Mini-Mental State Examination (MMSE) scores estimated using linear mixed-effects models. Associations between AD-PRS and longitudinal MMSE trajectories were tested adjusting for clinical and sociodemographic (CSD) variables and APOE genotype. Machine learning models and SHapley Additive exPlanations (SHAP) were used to evaluate the predictive relevance of individual variants.

Higher AD-PRS was associated with faster CD in the full clinical cohort and in biomarker subset, independently of APOE genotype. AD-PRS was not associated with baseline MMSE. APOE ε4 was associated with lower baseline MMSE and faster CD only in the full clinical sample. Genetic predictors provided limited improvement beyond CSD variables, and model performance showed limited reproducibility across cohorts.

AD-PRS is associated with longitudinal CD across the AD spectrum. Although polygenic burden contributes to variability in cognitive trajectories, its added predictive value beyond routinely available clinical variables remains modest.

A growing body of literature leverages large language models (LLMs) to make mental health predictions. However, these models are prone to bias, and studies to validate their clinical utility are lacking.

This scoping review aims to uncover bias and clinical utility limitations stemming from the methodological design of LLM-based mental health predictive systems. In addition, it intends to document the level of self-reflection about bias and clinical challenges reported by authors in their own work.

This work follows the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines and was registered online. Eligible studies were original research articles in English published between 2019 and 2024, using LLMs to detect mental health conditions in nonsynthetic textual data. The search was conducted in 5 scientific databases (PubMed, Web of Science, IEEE Xplore, ACM Digital Library, and ACL Anthology) with queries associating keywords related to "Mental Health," "Large Language Models," and "Prediction." We extracted both methodological information about the included studies and authors' statements relevant to issues of bias and clinical utility. This extraction was based on a framework screening the entire pipeline of development of LLMs with applications in mental health: research design and selection, data collection, outcome definition, model development, and postdeployment considerations. Statistical description of the retrieved entities, as well as thematic coding, was performed for analysis.

A total of 2472 articles were identified, of which 263 (10.6%) were assessed for eligibility, and 201 (8.1%) were included in the review. Included studies were mostly recent, indicating a growing interest in the use of LLMs for mental health predictions. Our analysis revealed that a majority of studies share similar methodological choices along their development pipeline: most of them focus on depressive disorders identified via processing user texts on social media, mainly with the use of nonspecialist LLMs derived from BERT (Bidirectional Encoder Representations from Transformers). Following previous works on these matters, we highlighted how these choices may hinder the clinical relevance and fairness of the envisioned systems. Similarly, we found that 164 (81.6%) studies mention themes related to bias and clinical utility; however, most of the discussion revolves around data-centered issues. Only 41 (20.4%) articles mention themes associated with at least 3 out of 5 pipeline steps, suggesting a limited appropriation of the notions of bias and clinical utility in such a sensitive context as mental health analysis.

Bias and clinical utility are lightly covered in the field of LLM-based mental health prediction research as of 2019-2024. In-depth approaches involving interdisciplinary teams of clinicians and natural language processing specialists are needed to ensure technical soundness, clinical relevance, and fair outcomes for potential users.

Pulmonary arterial hypertension (PAH) is relatively uncommon yet life-threatening morbidity in neonates and infants. South Asian countries including India may record a relatively higher incidence of pediatric PAH. The etiology of pediatric PAH is heterogeneous, and little evidence is available on the clinical spectrum of pediatric PAH from low- and middle-income countries. This descriptive study aims to describe the incidence and underlying causes of pediatric PAH at a tertiary care referral center in coastal Karnataka, India.

This is a retrospective study which involved complete enumeration of pediatric PAH cases admitted at a tertiary care referral center in coastal Karnataka between January 2018 and December 2021 after prior approval from the Institutional Ethics Committee. Clinical data were retrieved from the electronic medical records.

The center documented a total of 165 cases of pediatric PAH, with an incidence of 1.19% among the total neonatal intensive care unit admissions within the study period. Persistent fetal circulation was the most common sub-classification (77.44%), with underlying inherited metabolic disorders (30%), infectious pathologies, and malnutrition being the final diagnosis in this group. Neonates subclassified under neonatal cardiac failure majorly showed underlying cardiac anomalies (80%), while 4% of the cases were attributed to other causes including hyperbilirubinemia-induced encephalopathy and COVID-induced multisystem inflammatory syndrome, viral endocarditis, etc.

Inherited metabolic and cardiac anomalies, antenatal and postnatal maternal morbidities, nutritional deficiencies, and secondary infections are the major causes of neonatal and infantile PAH. Multilevel screening could substantially reduce the incidence and the morbidities associated with infantile PAH.

Adult burn survivors are at risk for developing anxiety and reduced health-related quality of life. The purpose of the study was to investigate predictors of anxiety at one- and two-years post injury. Data from 266 patients were queried from the Burn Model System National Database between 2015-2020. Patient-reported outcomes at one- and two-years post injury were analyzed using the Patient-Reported Outcomes Measurement Information System (PROMIS-29) for anxiety severity, and the Mental Component Summary (MCS) and Physical Component Summary (PCS) for mental and physical health-related quality of life respectively. Logistic models were computed to determine predictors of anxiety at one- and two-years post injury. Results showed that at one-year post injury, burns to specific body regions and pre-existing diagnoses of mental health disorders predict high anxiety severity scores and low MCS scores, whereas burns at perineum area predict low PCS scores. At two-years post injury, younger age predicted low MCS scores, and burns to the leg area, older age, and pre-existing diagnosis of diabetes predict low PCS scores. Findings from this study showed that burns to visible areas, pre-existing mental health disorders, and younger age predict increased anxiety and lower mental-health related quality of life, whereas burns to the perineum and lower limbs, and older age predict lower physical-health related quality of life. Therefore, allocating mental and physical health-related resources to burn survivors based on burn area, age, and comorbidities is crucial for quality of life after burn injury.

Convergent thinking is a critical aspect of creative thinking, and it has a substantial influence on the learning and work of college students. Emotion regulation not only enhances emotional well-being but also plays a vital role in the creative process. Additionally, various social and cognitive factors contribute to the completion of complex convergent thinking tasks, with cognitive load being one of the primary determinants. This study recruited 93 undergraduates as participants and randomly divided them into three groups: cognitive reappraisal, expressive suppression, and free viewing. All participants were required to complete a convergent creativity test (Chinese compound remote associate problems, CCRA) under high and low cognitive load conditions. The results showed a significant main effect of cognitive load on convergent thinking scores, with participants achieving higher scores under low cognitive load than under high cognitive load. However, the interaction between cognitive load and emotion regulation strategy on convergent thinking scores was not significant. Response time analyses revealed a significant interaction between cognitive load and regulation strategy: under high cognitive load, cognitive reappraisal was associated with faster responses than free viewing, whereas under low cognitive load, expressive suppression led to faster responses than both cognitive reappraisal and free viewing. This study explored the impact of emotion regulation on convergent thinking under different cognitive loads, which has certain implications for the optimization of college students' mental and physical health and academic performance.

Close relational partners play a critical role in shaping each other's emotions during social interactions, a process known as interpersonal emotion regulation. We examined how different forms of dementia (i.e., Alzheimer's disease, behavioral variant frontotemporal dementia, primary progressive aphasia) affect interpersonal emotion regulation. We assessed how caregivers perceived their partners' prosocial regulatory efforts and tracked real-time changes in caregivers' affect during a dyadic interaction. We also investigated how these processes relate to caregivers' mental health. Informal caregivers (N = 62) reported on their own mental health symptoms (depression, anxiety) and how their care recipients, who had different forms of dementia, use prosocial interpersonal emotion regulation (i.e., efforts to make their partner feel better). Dyads then engaged in a 10-min unrehearsed discussion about a conflict in their relationship. Following this, caregivers viewed a video recording of the interaction and used a rating dial to provide continuous ratings of their own affective valence (negative-neutral-positive) during the conversation. Results revealed that among care recipients with Alzheimer's disease, primary progressive aphasia, and behavioral variant frontotemporal dementia, those with behavioral variant frontotemporal dementia showed the lowest use of prosocial interpersonal emotion regulation, and their caregivers experienced the greatest increases in negative affect during the interaction. Across diagnoses, greater increases in caregiver negativity during the interaction were associated with greater caregiver depression (but not anxiety). These findings provide new information about interpersonal emotion regulation in dementia and highlight how the emotional changes that occur during dyadic interactions are tied to caregiver depressive symptoms. (PsycInfo Database Record (c) 2026 APA, all rights reserved).