medRxiv : the preprint server for health sciences最新文献

Human muscle biopsies are often required to study or diagnose diseases. However, traditional approaches are challenging due to limited sample size, quality, or patient discomfort. Fine-gauge needle biopsies (≥14-gauge), present an alternative but yield insufficient sample sizes for histology or function. Ultrasound guidance, coupled with vacuum-assisted, single needle-insertion multiple sampling addresses these challenges. In 19 healthy participants (mean age: 30.1±10 years, 42% male), 2-3 samples were collected from a single needle insertion into the vastus lateralis (VL) and tibialis anterior (TA). Summed VL and TA sample masses averaged 148±38mg and 166±64mg, with dimensions of 15.83±8 x 2.9±0.6mm ² (VL) and 15.07±7 x 3.1±0.9mm ² (TA). VL had a mean fiber cross-sectional area of 4,347±1,931µm ² , with 221±86 fibers quantified. Samples were of sufficient size and quality for thorough analyses from a single biopsy procedure, including mitochondrial respirometry, RT-PCR, collagen content, and biomechanical function. Fibers produced typical isometric stress values of 187kPa with a passive modulus of 239kPa (peak) and 79kPa (stress-relaxed). This procedure was well tolerated, with an average immediate pain rating of 1.5±1 (range:0-4, scale: 1-10) and 24-hour follow-up rating of 1.7±1 (range:0-4). This report describes an approach that yields high-quality muscle samples suitable for histological and biochemical analyses while minimizing discomfort.

Background and aims: Clinical LLM deployment is shifting from feasibility to liability, while current guidance largely treats model behavior as a control problem. We tested whether decision-style system prompts shift clinical action thresholds when clinical facts are held constant, and whether these shifts are consistent across settings and models.

Methods: We defined nine physician personas by crossing three ethical orientations (duty-, care-, utilitarian) with three cognitive styles (intuitive, integrative, analytic). Twenty open-weight LLMs were evaluated on 2,500 simulated ED vignettes and 2,500 MIMIC-IV-Note discharge summaries. For each text, models answered five binary decision items (safety, autonomy, treatment, resource use, follow-up). Each condition was repeated ten times, yielding 5,000,000 total decisions.

Results: Under baseline prompting, models answered "Yes" to 42.8% of decisions. Persona prompts shifted affirmative rates from 36.9% to 46.4%, a 9.5-percentage-point swing under fixed clinical evidence. Effects were largest in autonomy and treatment and were consistent across corpora (85.7% directional agreement; r = 0.82 for effect sizes). Susceptibility varied by model (4.9-16.1 points), with no consistent protection from medical fine-tuning or model size.

Conclusions: Decision-style system prompts reliably change clinical action thresholds in LLMs under fixed evidence. Prompting is a policy-setting layer, not just a communication layer, and should be treated as a first-class deployment configuration.

Objective: Duchenne muscular dystrophy (DMD) is a progressive neuromuscular disorder for which monitoring biomarkers are urgently needed. We aimed to evaluate whether proteins in serum can accurately monitor patients' function within the duration of a clinical trial.

Methods: In this study, we evaluated longitudinal serum proteins of DMD patients participating to the FOR-DMD clinical trial, comparing daily and intermittent corticosteroid regimens in boys aged 4-8 years at baseline. Using the aptamer-based protein platform SomaScan, we profiled 1500 proteins. Associations between protein levels and motor function outcomes, such as Rise from the Floor Velocity (RFV), 10-Meter Run/Walk Velocity (10MRWV), and North Star Ambulatory Assessment (NSAA), were assessed using linear mixed models. In particular, we explored whether patients with higher protein levels also tended to have better functional scores (across-patients analysis), and whether changes in protein levels within the same patient over time were linked to changes in their functional performance (within-patient analysis). Finally, penalized (Lasso) mixed models were applied to evaluate the predictive function of the proteins. The prediction accuracy of the models (evaluated by optimism-corrected Root Mean Squared Error) was compared to that of a simpler model with only age and treatment as predictors.

Results: Across-patients and within-patient analyses revealed consistent associations with three functional tests for a subset of proteins, notably RGMA, ART3, ANTXR2, and CFB. Multivariate models incorporating the proteins significantly associated with at least two tests, improved prediction accuracy for NSAA and RFV by 21% and 8%, respectively. These models also revealed a subset of proteins that were consistently selected. Quantification of CFB, RGMA, ANTXR2, SERPINF1 and ATP5PF using SomaScan showed strong agreement with measurements obtained using orthogonal methods such as ELISA, MRM-MS and an in-house developed bead-based sandwich immunoassay.

Discussion: These findings support the utility of serum protein signatures as objective, quantitative tools for monitoring disease progression and treatment response in DMD during clinical visits and clinical trials.

Objective: This study aimed to design and evaluate an explainable machine learning (ML) framework that integrates sensor-based motor assessments with demographic and clinical data to identify early indicators of mild cognitive impairment (MCI), fall risk, and frailty in older adults.

Methods: Eighty-three community-dwelling older adults (60 years or older) completed multimodal motor assessments using the Mizzou Point-of-Care Assessment System (MPASS) to capture synchronized gait, balance, and sit-to-stand performance. Sensor-derived motor features were combined with demographic and clinical variables to develop predictive models for MCI, frailty, and fall risk using XGBoost and Decision Tree algorithms. A unified multilabel framework was also developed using XGBoost, Decision Tree, and AdaBoost to predict all three outcomes. Model interpretability was evaluated using SHapley Additive exPlanations (SHAP).

Results: The ML model for MCI achieved the highest performance (94% accuracy, AUC = 0.88, F1 = 0.94), followed by fall risk (94% accuracy, AUC = 0.90) and frailty (82% accuracy, AUC = 0.77). Unified multilabel models showed moderate performance (67-73% accuracy), with XGBoost achieving the highest accuracy (73%), sensitivity, and F1 score, while the Decision Tree showed higher discrimination (AUC = 0.72). SHAP analyses identified stride length and time, center-of-pressure-based balance measures, and knee angular velocity during sit-to-stand as dominant predictors.

Conclusions: This work introduces a novel ML framework using multimodal sensor-based motor assessments to predict MCI, fall risk, and frailty individually and within a unified model. By combining explainable ML with motor-function data, the framework supports transparent early screening of multidomain cognitive and physical decline in aging.

Polygenic scores (PGS) predict complex traits and stratify disease risk but often fail to fully capture individual-level variation. "Misaligned" individuals, whose observed phenotypes deviate from their genetically expected values based on polygenic scores (PGS), provide a powerful model for identifying factors beyond common-variant effects, including additional genetic factors. Here, we apply misalignment classification and enrichment testing frameworks to seven continuous and three dichotomous traits, assessing whether misaligned individuals in the UK Biobank are enriched for rare (minor allele frequency (MAF) < 0.1%) damaging genetic variation. We identify significant enrichment (false discovery rate (FDR)-adjusted P < 0.05) of predicted loss-of-function (pLoF) variants in COPB2 and GORAB among individuals misaligned for lower-than-expected bone mineral density. We refine previously observed grouped-gene enrichment in individuals with misaligned stature to the single-gene level: shorter-than-expected individuals are enriched for pLoF variants in ACAN and IGF1, and taller-than-expected individuals are enriched for predicted damaging missense in FBN1. Using an individual's misalignment classification as a phenotype, we perform an exome-wide scan across seven traits, resulting in 74 FDR-significant genes. We identify KANK1 as a gene associated with later age at menopause, potentially protective against primary ovarian insufficiency. For dichotomous disease status traits, we demonstrate evidence for the liability threshold model in the context of counteracting conditionally-orthogonal common and rare variant pathogenic/protective effects. Among individuals diagnosed with type 2 diabetes, carriers of rare pathogenic pLoF variants in HNF1A and HNF4A had significantly lower polygenic risk than non-carriers (FDR-adjusted one-sided t-test P < 5 × 10^-3). We also show that coronary artery disease controls carrying rare protective pLoF variants in ANGPTL3 had nominally higher polygenic risk (one-sided t-test P = 0.03) than non-carriers. This study highlights the power of misalignment-based analyses in complex continuous phenotypes and disease, with the potential to validate known genetic contributors to traits and identify novel genes. This work paves the way for better molecular diagnoses and targeted therapeutic discovery.

Objectives: To evaluate the clinical performance of a cadmium-zinc-telluride- (CZT-) based photon-counting computed tomography (PCCT) system for low-dose lung cancer screening (LCS-LDCT) using patient-specific 3D-printed lung phantoms, and to compare its image quality and radiomics consistency with a conventional energy-integrating detector CT (EIDCT) system.

Methods: Six 3D-printed lung phantoms, derived from patient CT datasets and representing various lesion types (solid, part-solid, and ground-glass), were imaged on PCCT and EIDCT scanners at matched dose levels (1.6 - 20.4 mGy). Quantitative image metrics, Hounsfield unit (HU) accuracy, image noise, and contrast-to-noise ratio (CNR), were assessed across dose levels. Radiomic features were extracted for each lesion and analyzed via principal component analysis to quantify feature consistency (within-cluster distance) and lesion type separability.

Results: PCCT demonstrated significantly lower image noise and higher CNR compared with EIDCT, particularly at lower dose levels. HU values were consistent across doses for both systems, with reduced variability in PCCT (coefficient of variation < 0.004). Radiomics analysis revealed tighter clustering (reduced within-cluster distances) and comparable lesion type separability between PCCT and EIDCT, indicating enhanced feature stability and lesion differentiation. Qualitative review confirmed superior lesion conspicuity and margin delineation with PCCT.

Conclusions: CZT-based PCCT outperforms conventional EIDCT in quantitative and qualitative imaging metrics for LCS-LDCT, enabling superior image quality and radiomics reproducibility at reduced radiation doses. These findings support the clinical translation of PCCT for lung cancer screening and radiomics-based lesion characterization.

Background: Large-scale sequencing initiatives have generated extensive genomic resources essential for variant interpretation, yet their effective use often requires bioinformatics expertise. To support identification of Parkinson's disease (PD) risk and disease-causing variants, we developed an open-access, summary-level genomic data browser.

Methods: We performed uniform joint variant calling to harmonize whole-genome sequencing (WGS) data from AMP-PD Release 4, GP2 Data Releases, and additional controls from the Alzheimer's Disease Sequencing Project. Clinical exome sequencing (CES) data from GP2 Release 8 was also included.

Results: The integrated dataset includes 31,665 WGS and 9,559 CES samples, spanning eleven ancestries and over 300 million variants.

Conclusion: The GP2 Genome Browser is a lightweight, flexible platform providing intuitive gene- and variant-level summaries with ancestry-stratified allele frequencies and functional annotations. It is open source and freely accessible at https://gp2.broadinstitute.org, enabling broad access to PD genomic data and supporting global research efforts.

Central venous pressure (CVP), a key component of hemodynamic monitoring, is widely used to guide fluid resuscitation in critically ill patients. It is typically measured using central venous line catheterization, which is the gold standard, but this method is invasive, time-consuming, and associated with complications. This study aims to investigate whether machine learning (ML)-based analysis of features extracted from a non-invasive, standard-of-care waveform-the photoplethysmography (PPG) signal-can identify patients with elevated CVP. We trained Light Gradient-Boosting Machine (LightGBM) model using a large perioperative dataset (MLORD), containing 17,327 surgical patients from 2019 to 2022 at UCLA. For this study, we selected 1665 patients with both PPG and CVP waveforms available. A total of 843 PPG features per cardiac cycle (CC) were extracted from the PPG waveforms using a signal processing-based feature extraction tool, along with the simultaneous maximum value calculated from the corresponding CCs in the CVP waveform. Additionally, for each patient, the average and standard deviation of each PPG feature, as well as the mean of the maximum CVP values, were calculated across all cardiac cycles, resulting in 843 averaged PPG features, 843 PPG feature standard deviations, and one averaged maximum CVP value per patient. The average maximum CVP value was used as the ground truth to classify patients as either normal (5 ≤ CVP ≤ 15 mmHg) or elevated (CVP > 15 mmHg). Of the 1,665 patients, 1,182 were normal and 483 were elevated. The dataset was split into 90% for training (1,063 normal and 435 elevated) and 10% for testing (119 normal and 48 elevated). From the 1686 PPG features (843 averaged and 843 standard deviation), 246 were selected for model development using the Recursive Feature Elimination with Cross-Validation (RFECV) approach. To further enhance performance, hyperparameters were tuned through 5-fold cross-validation on the training set. Finally, the best-performing configuration was retrained on the full training data, and its performance was evaluated on the held-out test set. To provide a robust estimate and confidence interval, a bootstrapping procedure with 100 iterations was performed on the test set. The LightGBM classifier achieved a mean area under the receiver operating characteristic curve (AUC) of 0.79 (95% CI: 0.71-0.84) and mean accuracy of 0.71 (95% CI: 0.65-0.77), demonstrating good discriminatory power in distinguishing between patients with normal and elevated CVP. This study highlights the ability of PPG-derived features to discriminate between patients with normal and elevated CVP using ML. These early findings lay the groundwork for future research aimed at developing non-invasive approaches to CVP assessment.

Background: Suicidal thoughts and behaviours (STBs) are traditionally framed as arising from mental health problems, however, emerging evidence highlights the importance of social determinants in contributing to suicide risk. We examined STBs and their socioeconomic, psychosocial and clinical correlates in two peri-urban communities in Cape Town.

Methods: We conducted psychiatric diagnostic interviews (MINI) at three public-sector facilities (2023-2024). Adults aged ≥18 years, with and without HIV, were recruited in a 2:1 ratio. We examined STB prevalence and associations between past 30-day suicidal ideation and sociodemographic factors, violence exposure, perceived stress, mental disorders, and HIV.

Results: We enrolled 613 participants (63.9% female; 65.4% HIV-positive; median age 39). The prevalence of past 30-day suicidal ideation was 14.0%, while 22.2% reported a lifetime suicide attempt. Odds of past 30-day ideation were higher in females (OR 2.07, 95% CI 1.21-3.54), those who had experienced violence in their community (1.93, 1.09-3.41) or family (3.00, 1.55-5.81), those with high perceived stress (4.57, 1.93-10.81), and in those with depression (6.62, 3.39-12.92), post-traumatic stress disorder (6.69, 2.97-15.04), and an alcohol use disorder (2.27, 1.23-4.17). Associations with high perceived stress and community violence persisted after adjustment for mental disorders. HIV and other sociodemographic factors were not significantly associated.

Conclusion: STB prevalence was high in peri-urban communities in Cape Town and strongly associated with mental disorders, violence exposure, and perceived stress. These findings underscore the role of structural and psychosocial stressors in shaping suicide risk in low-income communities.