Communications medicine最新文献

Background: Medulloblastoma (MB), the most common malignant pediatric brain tumor, lacks prognostic tools integrating clinical, molecular, and treatment-related characteristics for individualized management.

Methods: We developed machine learning models using multicenter data from 729 Chinese patients (2001-2023), of whom 509 were assigned to the training set and 220 to the testing set, and further validated the models on 201 patients from international MB consortia. To accommodate patients and researchers with varying datatypes, four application scenarios were established, including clinical-molecular-radiotherapy (CMR), clinical-molecular (CM), clinical-radiotherapy (CR), and clinical-only (CO).

Results: We construct four model scenarios and assess their predictive performance in the testing set: an XGBoost-based CMR model (incorporating 11 features, including molecular subgroup, radiotherapy dose, and key gene expression) with a C-index of 0.612; an XGBoost-based CM (C-index = 0.609); a GBM-based CR (C-index = 0.637); and a GBM-based CO (C-index = 0.635). External validation demonstrates robust performance, with radiotherapy and molecular data contributing significantly to enhanced efficacy. In addition, interactive web-based Shiny applications have been launched to facilitate dynamic risk assessment and treatment optimization.

Conclusions: By integrating multidimensional data, our framework enables the tailored prognostication and clinical decision to meet the multidimensional requirements of research and medicine.

Background: Severe COVID-19 is associated with dysregulated immune responses. Immune responses heterogeneity was previously reported during the first waves of the pandemic. We aimed to characterise mucosal transcriptomic profiles in critically-ill patients during the Omicron era.

Methods: This prospective multicentre study included 94 critically-ill COVID-19 patients between May 2022 and August 2023. Upper respiratory tract mucosal transcriptomes were obtained from nasopharyngeal swabs and clustered based on KEGG cytokine-cytokine receptor interaction pathways using an unsupervised algorithm. Differential transcript expression, cell population abundance and gene set enrichment analyses were performed.

Results: Here we show that in 56 critically ill COVID-19 patients, transcriptomic clustering reveals two distinct COVID-19 Immune Transcriptomic Respiratory Profiles (CITRP), including CITRP-1 and CITRP-2, characterised by differential expression of cytokine and immune response pathways. Patients in the CITRP-2 group display a more pronounced immune and inflammatory response, involving specific innate immune pathways, neutrophil degranulation and T-helper 2 cytokines (e.g., IL-1, IL-4 and IL-13), and a significantly higher proportion of neutrophils than patients in the CITRP-1 group. No significant differences are observed between the two transcriptomic clusters in clinical, biological and virological characteristics at ICU admission or in patient outcomes.

Conclusions: This study highlights the heterogeneity of the immune response in critically-ill COVID-19 patients in the Omicron era, identifies two endotypes from the analysis of upper airway mucosal transcriptomics. Our findings suggest the existence of two distinct pathogenic mechanisms and the detrimental role of neutrophil and Th2 helper cell-mediated inflammation in a subset of patients with severe disease. They support the need for personalised treatment strategies targeting neutrophil-mediated lung damage and/or specific cytokine production in a subset of critically-ill COVID-19 patients.

Background: The current knowledge about medication adherence is based on studies focusing only on few health conditions and little is known about how strongly adherence is shaped by person-specific behaviour. The aim of the cohort study is to 1) evaluate the effect of multiple factors affecting medication adherence in a consistent manner across 137 active substances, and 2) calculate individual medication adherence score (IMAS), evaluate its predictive power, stability over time, and impact on health outcomes. In essence, IMAS describes persons' medication-taking "baseline".

Methods: We utilised a representative dataset with electronic health records, claims, and dispensed medications across 137 active substances and applied continuous multiple interval measures of medication availability (CMA). To assess the effect of various demographic, health, and medication-related variables on CMA, we employed linear mixed models.

Results: Here we show that the medication adherence ranged from 0.423 (albuterol, 95% CI 0.414-0.432) to 0.922 (warfarin, 95% CI 0.917-0.926). The demographic, health- and medication-related factors explained 11.6% and IMAS 22.0% of the variation in adherence. IMAS predicted adherence across medication classes, reduced the risk of overall hospitalisation (hazard ratio = 0.76, 95% CI 0.60-0.97, p < 0.05) and cause-specific incidence for 17 conditions.

Conclusions: Thus, IMAS represents a person-level metric that captures baseline medication-taking behaviour across therapeutic classes and predicts both medication adherence as well as health outcomes. Our analysis suggests that medication-taking behaviour represents a broader patient-level phenomenon manifesting consistently across medications, suggesting its potential for personalised interventions in clinical practice and more efficient public health strategies and policies.

Background: Circulating nucleic acids in blood plasma form an attractive, minimally invasive resource to study human health and disease. In this study, we aimed to identify cell-free RNA alterations that can distinguish cancer patients from cancer-free individuals.

Methods: We first performed mRNA capture sequencing on 266 blood plasma samples from cancer patients and controls, including a discovery set of 208 donors across 25 cancer types and a replication set of 58 donors across three cancer types. We first conducted group-level comparisons and then compared individual patient profiles to a reference control population in a one-versus-many approach. This approach was further evaluated in independent cohorts: a prostate cancer plasma cohort (n = 180), a non-malignant disease plasma cohort (n = 125), a lymphoma plasma cohort (n = 65), and a bladder cancer urine cohort (n = 24), each including both patients and controls.

Results: Here we show that cancer patients exhibit both cancer type-specific and general cell-free RNA alterations. However, differentially abundant RNAs vary widely among patients and across cohorts, hampering robust biomarker identification. By comparing individual patient profiles to control populations, we identify so-called biomarker tail genes, which strongly deviate from controls. The number of these genes per sample distinguishes cancer patients from control samples. Independent cohorts also confirm the potential of this approach.

Conclusions: Our findings demonstrate substantial heterogeneity in cell-free RNA alterations among cancer patients and propose that patient-specific changes can be exploited for classification.

Background: Despite the identification of numerous genetic loci associated with autoimmune diseases (ADs) through genome-wide association studies (GWAS), elucidating the mechanisms underlying these associations remains challenging.

Methods: We integrated GWAS results with multi-omics data across diverse immune cell types to investigate both the shared and disease-specific association signals across 15 common ADs.

Results: Our analyses reveal a high prevalence of locus-sharing (50.8%) across these diseases when defined by physical proximity, but a substantially lower proportion of shared association signals (14.7%) when defined by linkage disequilibrium. This suggests that loci shared across diseases often harbor distinct association signals and mechanisms. We demonstrate that within individual loci, association signals frequently exhibit regulatory activity in different cell types and, less commonly, target different genes. Notably, for several loci, disease-specific associations appear to be mediated through regulatory activity in distinct cell types. Overall, we identify 1,554 genes associated with ADs. Further pathway enrichment and protein-protein interaction network analyses unveil both shared functions and disease-specific pathways among these genes.

Conclusions: By integrating GWAS and multi-omics data, our study delineates the genetic and regulatory architecture underlying autoimmunity, suggesting potential therapeutic targets and opportunities for drug repurposing.

Background: Artificial intelligence enhances pathology screening efficiency, yet clinical adoption remains limited because most systems operate as opaque black boxes. We aim to resolve this opacity by establishing a framework that generates transparent, evidence-linked reasoning to support diagnostic auditing.

Methods: We present a framework that shifts off-the-shelf multimodal large language models from passive pattern recognition to active diagnostic reasoning. Using small labeled subsets from breast and prostate cancer datasets, we employ a two-phase self-learning process to derive diagnostic criteria without updating model weights. We integrate expert feedback from board-certified pathologists to ensure the generated descriptions align with established medical standards.

Results: Here we show that our framework produces audit-ready rationales while achieving over 90% accuracy in distinguishing normal tissue from invasive carcinoma. Beyond binary classification, the model effectively differentiates complex subtypes like ductal carcinoma in situ by autonomously identifying hallmark histological features, including nuclear irregularities and structural disruption. These computer-generated descriptions closely match expert assessments. Our approach delivers substantial performance gains over conventional baselines and adapts effectively across diverse tissue types and independent foundation models.

Conclusions: By uniting visual understanding with reasoning, our framework provides a promising approach for clinically trustworthy artificial intelligence. This framework helps bridge the gap between opaque classifiers and auditable systems, suggesting a viable path toward evidence-linked interpretation in medical workflows.

Background: Pancreas transplantation remains the only definitive treatment for diabetes mellitus. However, the global number of pancreas transplants and utilisation of pancreas grafts is declining. We aimed to identify significant donor factors associated with pancreas non-use.

Methods: Population-cohort study using United States (US) data from the Organ Procurement and Transplant Network (OPTN) registry (2010-2024). Multivariable regression models were constructed to assess associations between donor characteristics and pancreas utilisation. Restricted cubic splines were used to preserve non-linear relationships and interaction terms with donation date were performed, to capture evolving decision-making behaviours.

Results: We identify 23 donor factors significantly associated with utilisation (n = 14,612 transplants from 133,986 donors). The most important continuous donor factors are age, BMI and peak creatinine; all showing significant non-linear relationships with utilisation (all P < 0.001). Donor type is the most important categorical variable, with donation after circulatory death (DCD) having 92% lower odds of utilisation (aOR=0.078, 95% CI = 0.070 to 0.087, P = < 0.001). Interaction analyses reveal increasing reluctance to use DCD donors or older donors over the study period (both interaction P < 0.001). Conversely, clinicians have become more comfortable transplanting pancreases from Hepatitis C positive donors and IV drug use (IVDU) donors over time (both interaction P < 0.001).

Conclusions: This large population cohort study demonstrates significant shifts in utilisation decision-making over time. Growing reluctance to use DCD, despite evidence of favourable outcomes, highlights a valuable area to focus US pancreas utilisation efforts. Meanwhile, previously underused groups such as Hepatitis C positive and IVDU donors show growing acceptance, supporting expansion of these donor populations globally.

Background: Human-to-human transmission of pathogens fundamentally depends on interactions among infectious and susceptible individuals, yet traditional population-scale models often overlook the stochastic, behaviour-driven, and highly heterogeneous nature of these interactions.

Methods: Here, we develop a large-scale actor-based model capturing early epidemic dynamics of a novel respiratory pathogen on dynamic contact networks. We build these networks upon explicitly integrating detailed demographic and residential registry data from the Netherlands. The model simulates the Dutch population characterised by age, residency and mobility patterns, with actors interacting stochastically across households, workplaces and schools.

Results: We show how the geographic and demographic profiles of initial cases impact transmission trajectories, with densely populated municipalities in the country's western core acting as key hubs driving epidemic spread. The framework enables rigorous assessment of intervention strategies incorporating behavioural adaptations. As case studies, we quantify the effects of symptomatic self-isolation and travel restrictions to and from major urban centres, highlighting their potential to modulate epidemic outcomes.

Conclusions: Our findings underscore the necessity of integrating fine-scale human-to-human contact realism and population scale in epidemic forecasting and control.

Background: Multimorbidity is linked to systemic low-grade inflammation, poor glycaemic control, dyslipidaemia, and hypertension, yet evidence on effective interventions is limited. We evaluated the impact of a 12-week personalised exercise therapy and self-management support programme, in addition to usual care, on these outcomes in individuals with multimorbidity.

Methods: This was a pre-planned secondary analysis of the MOBILIZE multicentre randomised controlled trial (NCT04645732). Participants (n = 228) had at least two of the following conditions: knee/hip osteoarthritis, chronic obstructive pulmonary disease, heart disease, hypertension, type 2 diabetes, or depression. The intervention included 24 supervised 60-minute group-based exercise sessions and 24 self-management sessions over 12 weeks. Outcomes were assessed at baseline and 4 months, including interleukin-1 receptor antagonist (IL-1ra), high-sensitivity C-reactive protein (hs-CRP), tumour necrosis factor-alpha (TNF-α), interleukin-6 (IL-6), glycated Hemoglobin (HbA1c), fasting glucose, insulin, High-Density Lipoprotein (HDL), Low-Density Lipoprotein (LDL), triglycerides, and blood pressure.

Results: Compared to usual care, the intervention group shows a statistically significant reduction in systolic blood pressure (mean difference: -4.7 mmHg, 95% CI: -8.8 to -0.6). No significant between-group differences are observed for other biomarkers, although favouring the intervention. Sensitivity analyses-excluding participants with low adherence, those receiving supervised exercise in the control group, or undergoing surgery-support the primary findings.

Conclusions: A 12-week personalised exercise and self-management programme reduces systolic blood pressure in people with multimorbidity. These findings support incorporating exercise therapy into multimorbidity care guidelines as a non-pharmacological adjunct.

Background: Ultra-processed foods (UPF) dominate modern food systems and contribute significantly to early-life diets. However, the multilevel predictors of UPF consumption in early childhood, from family factors to neighbourhood environments, remain underexplored.

Methods: We leveraged data from a subset of the Canadian CHILD Cohort Study (n = 2,411), to assess UPF intake in three-years-old children using the NOVA classification system. A machine-learning variable selection algorithm and mixed-effect models identified independent predictors of UPF spanning family behaviours to neighbourhood environments.

Results: Here we show parental factors including prenatal maternal UPF intake (β = 2.8 % daily energy from UPF, [95%CI 2.3,3.2]) and greater paternal adherence to a Western-like dietary pattern (β = 1.1, [95%CI 0.6,1.6]) are associated with higher UPF intake. Other factors such as shorter breastfeeding duration, longer daily screen time, and having older siblings are also associated with a higher proportion of daily energy intake from UPF at three years of age (all p-values < 0.05). In contrast, children residing in neighbourhoods with better access to employment opportunities (β = -1.9, [95%CI -3.0,-0.9]) and higher density of fresh food markets (β = -2.0, [95%CI -3.4,-0.5]) are associated with lower proportion of daily energy intake from UPFs.

Conclusions: These findings indicate that the early childhood UPF intake reflects the convergence of family behaviours and structural features of the built environment. Interventions to reduce UPF intake must go beyond individual food choice and address food systems design, including how the interrelated factors of daily time demands, travel distance requirements and public infrastructure constrain access to healthier options that shape children's diet.