Integrating advanced artificial intelligence (AI) into clinical decision-support often requires the sharing of sensitive patient data with external services, raising privacy concerns. Homomorphic encryption (HE) allows computing directly on encrypted data, without revealing the underlying patient information.

To develop a large language model (LLM)-assisted diagnosis framework while preserving patient privacy in the clinical text analysis, by leveraging HE and using rare disease (RD) diagnosis as a representative application. To demonstrate HE does not hinder the system performance.

Texts from patient histories and a RD knowledge base were embedded by LLMs into vectors, then encrypted using HE to obscure private information while retaining the semantic nuances. Diagnostic recommendations were generated by computing and ranking the similarities between the patient history and RD vectors in the encrypted space. The system was evaluated using 50 synthetic case reports (5 RDs, each with 10 reports).

Applying HE did protect private information from reverse-embedding attacks. HE imposed little disruption to the diagnostic accuracy, with normalized discounted cumulative gains (nDCG) of 0.6108 ± 0.3412 (encrypted) versus 0.6083 ± 0.3415 (unencrypted). The accuracy and computational performance were tunable and consistent, as demonstrated across five different LLMs.

Our privacy-preserving framework opens tremendous opportunities toward hosting and serving powerful AI solutions across institution boundaries, which would remove the need for local deidentification and incentivize users to access secure external decision-support services.

Integrating HE with LLM retrieval can promote the dissemination of nonredundant, high-capacity AI services by preserving both privacy and accuracy.

Depression is associated with pathological dysregulations affecting both the brain and the body, with the latter being reflected in plasma proteins. While plasma protein signatures of depression have been increasingly recognized, a holistic examination of interactions with brain features is lacking.

Leveraging data from 3,966 UK Biobank participants, we identified a multimodal neuroimaging-plasma protein component of depression (NeuroPro-Dep) by integrating plasma proteins and five brain modalities via an ICD-10 diagnosis-constrained multimodal fusion approach.

Notably, NeuroPro-Dep demonstrates detectable associations with depression symptoms across datasets from diverse populations, underscoring its clinical potential. This capability is anchored in its five brain modalities alterations, including hippocampal atrophy, reduced cortical sensorimotor network functional connectivity, and impaired internetwork structural connectivity of the frontoparietal network. The multimodal neuroimaging-derived plasma protein modality of NeuroPro-Dep is enriched in metabolic pathways, as further supported by association analysis linking this modality to body mass index (BMI), type 2 diabetes, and other metabolic indicators. Crucially, two-step Mendelian randomization analysis revealed that the NeuroPro-Dep plasma protein modality exerts a causal effect on depression through BMI (plasma protein to BMI: or=0.28, p=0.035; BMI to depression: or=1.14, p=4.37×10−11).

Overall, this study underscores metabolic dysfunction as a bridge between brain changes, depression, and physical diseases, while providing a novel multimodal biological signature and valuable insights that may inform future treatment strategies.

Intolerance of uncertainty (IU) – a dispositional inability to react effectively to uncertain situations – has been increasingly conceptualized as a transdiagnostic risk factor for internalizing problems such as generalized anxiety and depression. However, evidence for its temporal role in the development of these conditions remains limited, particularly in adolescents, a group at heightened risk for psychopathology.

A total of 5,291 adolescents (46.2% boys; M age = 14.40 ± 1.56, range = 10–18 years) completed self-report measures of IU, generalized anxiety and depressive symptoms at baseline, 6 months and 12 months. Linear and logistic regression analyses examined whether baseline IU predicted subsequent symptom severity and elevated (above-cut-off) symptom levels over time.

Higher baseline IU significantly predicted increases in generalized anxiety and depressive symptoms, as well as higher odds of elevated generalized anxiety and depressive symptom levels at both 6- and 12-month follow-ups, even after adjusting for baseline symptom severity or baseline elevated symptom status. Baseline IU also predicted the new-onset and persistence of elevated symptoms across both intervals. Stratified analyses revealed developmental and sex differences: IU’s predictive effects were strongest in early adolescence for girls and in middle-to-late adolescence for boys.

IU emerged as a transdiagnostic longitudinal predictor of generalized anxiety and depressive symptoms in adolescents, supporting its value as an early screening marker of vulnerability. Interventions targeting IU may offer an effective strategy for reducing broad internalizing risk during this critical developmental period.

Crush syndrome, also known as traumatic rhabdomyolysis, often occurs in disasters like earthquakes, traffic accidents, and building collapses. It results in muscle cell necrosis, leading to symptoms such as hypovolemic shock, hyperkalemia, and acute kidney injury (AKI). The mortality rate of rhabdomyolysis is about 10%. An intelligent auxiliary diagnostic system for rhabdomyolysis based on machine learning is crucial.

This study aimed to develop an intelligent auxiliary diagnostic system, utilizing the MIMIC-III database and electronic medical records from several Chinese hospitals. The system was trained and validated with 70% and 30% of patient data, respectively. A variety of machine learning methods were used to establish injury grading and prognosis assessment models, and statistical methods and feature importance assessment methods were used to analyze and explain the features selected by the model.

The 22 clinical indicators originally included in the modeling were reduced to 11 through feature analysis and feature screening. In internal tests, the auxiliary diagnostic system based on 11 parameters achieved an average accuracy of 0.89. External validation with 360 cases showed an average accuracy of 0.84, with an AKI prediction accuracy of 0.83 and an AUC value of 0.82. Death risk prediction accuracy reached 0.89.

This intelligent auxiliary diagnostic system provides valuable recommendations for the diagnosis and treatment of rhabdomyolysis, improving treatment success rates and optimizing medical resource allocation in disaster situations. By leveraging machine learning, we have established a robust and reliable tool to assist healthcare providers in managing this complex condition.

Schizophrenia is a severe psychiatric disorder characterised by positive symptoms, such as hallucinations and delusions, which are linked to dysregulated striatal connectivity. Although traditional models highlight the limbic striatum’s role in salience processing, emerging evidence suggests that the associative striatum, critical for cognitive control and habit formation, also plays a significant role. However, the structural connectivity underlying striatal subregions and its relationship to symptom severity and treatment response remains poorly understood.

This study aimed to investigate the structural connectivity of striatal subregions in first-episode schizophrenia (FE-SCZ) patients and to evaluate its association with positive symptoms and changes following antipsychotic treatment.

We recruited 80 FE-SCZ patients and 80 healthy controls who underwent diffusion tensor imaging probabilistic tractography to assess white matter tract strength between the striatum and ten cortical targets. Longitudinal analysis was conducted in patients at baseline (within 2 weeks of initial antipsychotic exposure) and after ongoing treatment to evaluate changes in connectivity and their relationship to symptom improvement.

FE-SCZ patients exhibited reduced connectivity between the dorsolateral prefrontal cortex (dlPFC) and associative striatum and increased connectivity between the anterior cingulate cortex (ACC) and associative striatum compared to controls. Longitudinal analysis revealed that antipsychotic treatment increased dlPFC–associative striatum connectivity and decreased ACC–associative striatum connectivity, which correlated with reductions in positive symptom severity.

These findings highlight the critical role of striatal subregions in the pathophysiology of schizophrenia, emphasising the associative striatum’s involvement in cognitive control and salience attribution. Changes in striatal connectivity after continued antipsychotic therapy may serve as a biomarker for symptom improvement, advancing our understanding of schizophrenia and guiding future therapeutic strategies.

This study aimed to investigate self-management experiences at home among gynaecological cancer patients with lower limb lymphoedema.

Lower limb lymphoedema is a common complication following gynaecological tumour treatment, causing physical and psychological distress and significantly impacting patients’ quality of life. Clinical observations reveal that many patients with lower limb lymphoedema following gynaecological tumour treatment exhibit poor compliance with family self-management, leading to complications such as worsening oedema, cellulitis, or erysipelas. This study seeks to gain insight into patients’ actual self-management experiences within their families, offering insights for tailored intervention plans and improved patient self-management compliance in clinical practice.

Employing a phenomenological approach in qualitative research, one-on-one semi-structured interviews were conducted to gather face-to-face data from participants. A total of 16 gynaecological cancer patients with lower extremity lymphoedema were selected via purposive sampling from a tertiary cancer hospital. Semi-structured in-depth interviews took place between February and July 2021, with data analysed via the Colaizzi 7-step analysis method.

Five key themes emerged: inadequate and uneven availability of medical resources for patients with lymphoedema, limited support for patients, deficient home self-management skills, considerable psychological stress during home management, and variations in self-management behaviours.

Based on the study findings, increased investment in lymphoedema-related medical care is recommended. Additionally, healthcare professionals can consider promoting family and social support, enhancing patient health education, offering remote psychological counselling, encouraging positive coping behaviours among gynaecological cancer patients with lower limb lymphoedema, and ultimately enhancing their self-management at home.

Studies highlight the thalamus as a key region distinguishing early- from late-onset obsessive-compulsive disorder (OCD). While structural thalamic correlates with OCD onset age are well-studied, resting-state functional connectivity (rsFC) remains largely unexplored. This study examines thalamic subregional rsFC to elucidate pathophysiological differences in OCD based on different onset times.

The study comprised 85 early-onset OCD (EO-OCD) patients, 94 late-onset OCD (LO-OCD) patients, and 94 age- and sex-matched healthy controls (HCs). rsFC analysis was conducted to assess thalamic connectivity across seven subdivisions among the groups.

Both EO-OCD and LO-OCD patients exhibited increased rsFC between the primary motor thalamus and the posterior central gyrus and between the thalamic premotor and the supplementary motor areas. EO-OCD patients showed significantly stronger rsFC between the prefrontal thalamus (Ptha) and the middle frontal gyrus (MFG) compared to both LO-OCD patients and HCs. In contrast, LO-OCD patients demonstrated reduced rsFC between the Ptha and the inferior parietal lobule (IPL) compared to EO-OCD patients and HCs. Additionally, the rsFC between the Ptha and both the MFG and IPL was negatively correlated with age of onset, with earlier onset linked to stronger connectivity.

These findings reveal both shared and distinct thalamic connectivity patterns in EO-OCD and LO-OCD patients. Sensory-motor networks exhibiting thalamic hyperconnectivity are critical for the manifestation of OCD, regardless of age of onset. The frontal–parietal network and thalamic hyperconnectivity may present a compensatory mechanism in EO-OCD patients, while hypoconnectivity with the frontoparietal network may reflect a neural mechanism underlying LO-OCD.

Electronic Health Record (EHR) data are critical for advancing translational research and AI technologies. The ENACT network offers access to structured EHR data across 57 CTSA hubs. However, substantial information is contained in clinical narratives, requiring natural language processing (NLP) for research. The ENACT NLP Working Group was formed to make NLP-derived clinical information accessible and queryable across the network.

We established the ENACT NLP Working Group with 13 sites selected based on criteria including clinical notes access, IT infrastructure, NLP expertise, and institutional support. We divided sites into five focus groups targeting clinical tasks within disease contexts. Each focus group consisted of two development sites and two validation sites. We extended the ENACT ontology to standardize NLP-derived data and conducted multisite evaluations using the Open Health Natural Language Processing (OHNLP) Toolkit.

The working group achieved 100% site retention and deployed NLP infrastructure across all sites. We developed and validated NLP algorithms for rare disease phenotyping, social determinants of health, opioid use disorder, sleep phenotyping, and delirium phenotyping. Performance varied across sites (F1 scores 0.53–0.96), highlighting data heterogeneity impacts. We extended the ENACT common data model and ontology to incorporate NLP-derived data while maintaining Shared Health Research Informatics NEtwork (SHRINE) compatibility.

This demonstrates feasibility of deploying NLP infrastructure across large, federated networks. The focus group approach proved more practical than general-purpose approaches. Key lessons include the challenge of data heterogeneity and importance of collaborative governance. This work also provides a foundation that other networks can build on to implement NLP capabilities for translational research.