medRxiv : the preprint server for health sciences最新文献

Background: As sleep data from wearable devices are increasingly available in health research, there are new opportunities to understand sleep regulation behaviors as modifiable risk factors for disease. At such a large scale (tens of thousands of people over millions of day-level observations), prioritizing and interpreting sleep behaviors is challenging while maintaining biological relevance and modifiability. In this work, we aim to address this challenge by proposing a framework to interpret Fitbit data through a well-known neurobiological framing of sleep regulation, the two-process model.

Methods: We use data from the All of Us Research Program, a national biobank with passively collected Fitbit data for 32,292 people across 15,754,893 total days. We map Fitbit behaviors ( _b ) to either circadian (C) or homeostatic (S) processes. Using iterative exploratory factor analysis to obtain weights, the Fitbit C _b and S _b are then weighted at the level of each day to create C _b and S _b scores.

Findings: C _b and S _b scores were found to align with expected real-world relationships with age, seasonality, shift work, and napping. C _b and S _b scores were interpreted with relation to depression, where it was found that S _b scores are highly associated with likelihood of diagnosis (OR = 1.5, p < 2e-16) while C _b and S _b scores are equally associated with severity (S _b score β = 0.2, C _b score β = 0.21, p < 2e-16).

Interpretation: C _b and S _b scores support longitudinal interpretation (e.g., changes in S _b around treatment), aggregation (e.g., differences in C _b between two groups), and actionable modification (e.g., reduce naps to improve poor S _b ). Overall, our behavior scores allow for interpretation of wearables sleep data and can be utilized across many disease contexts to better understand how sleep influences health.

Funding: This work was supported by NIH training grant T32GM145734 and NIH R21HL172038.

Major depression (MD) is a disorder class that exhibits substantial phenotypic and clinical heterogeneity, yet many large-scale molecular genetic investigations treat MD as a unitary outcome. Here, we applied Genomic Structural Equation Modeling (Genomic SEM) to characterize the genetic variation in two clinically relevant MD subtypes, childhood-onset (child-onset) and treatment-resistant MD, that are independent of the field-standard GWAS of MD in all its forms. In addition, we fit a complementary "boosting" model that leveraged shared signal across the subtype and general MD GWAS to increase power for subtype biological discovery. At the genome-wide level, more than half of the common-variant liability for child-onset and treatment-resistant MD was unique relative to the general MD GWAS, indicating substantial subtype-specific genetic architecture. Unique components of both subtypes showed robust associations with genetic liability for schizophrenia and bipolar disorder, and the child-onset specific component exhibited genome-wide overlap with early developmental outcomes, including autism spectrum disorder and childhood intelligence. Transcriptome-wide analyses implicated upregulation of SMIM19 in liability specific to child-onset MD, while stratified functional enrichment highlighted gene sets involved in limbic and frontal brain systems for the boosted child-onset component. Together, these findings demonstrate that MD contains biologically distinct subtypes that exhibit etiological divergences more akin to separate disorders than subtypes within a concrete diagnostic framework. We find that stratifying MD by biologically distinguishable subtypes may be crucial for enhancing biological discovery and elucidating etiological pathways in molecular genetic studies of depression.

We examined whether a 24-week resistance training program influenced brain amyloid-β (Aβ) and Alzheimer's Disease (AD)-related blood-based biomarkers. Ninety cognitively normal, physically inactive older adults aged 65-80 years were randomly allocated to a 24-week resistance training program (three ∼60-min supervised sessions/week) or a wait-list control group. Primary analyses assessed exercise-induced changes in brain Aβ (Centiloid values) and plasma ptau217/Aβ1-42 IPMS ratio. Secondary analyses examined ptau217/Aβ42 SIMOA ratio, ptau217, ptau181 and Aβ42/40, as well as potential interactions with sex, age, education, apolipoprotein ε4 ( APOE4 ) status, amyloid PET-positivity, and comorbidities. The intervention produced no significant differences on brain Aβ or AD-related blood-based biomarkers (p>0.05) compared to the control group. However, the ptau217/Aβ1-42 IPMS ratio showed a small, non-significant increase in the control group (SMD = 0.162; 95% CI: -0.159 to 0.483) while remaining stable in the exercise group (SMD = 0.01; 95% CI: -0.291 to 0.310) with a similar trend for ptau217/Aβ42 SIMOA. Moderator analyses indicated differential responses by amyloid PET-positivity and APOE4 status on brain Aβ (p for interaction<0.05), with increases observed in APOE4 carriers and amyloid PET-positive individuals in the control group, whereas those allocated to the exercise intervention reduced their levels. The specificity observed within our subgroups suggests that resistance exercise may serve as a targeted intervention to modulate AD pathophysiology, raising new questions regarding its broader role in the delay of the disease in vulnerable populations.

Background and objectives: Progression from mild cognitive impairment (MCI) to Alzheimer's Disease and Related Dementias (AD/ADRD) varies widely across individuals, yet the mechanisms underlying this heterogeneity remain unclear. Identifying clinical and social determinants influencing this transition could enable earlier intervention. While cardiovascular and social risk factors are established contributors to dementia incidence, their role in progression from MCI to dementia may differ. Few studies using real world clinical data have evaluated these potential determinants of MCI progression.

Methods: Using electronic health records (EHR) from patients with incident MCI at UCSF Health (2010-2024), we evaluated cardiovascular (blood pressure [BP], body mass index [BMI], and type II diabetes) and social (marital status, language preference, race/ethnicity, and neighborhood disadvantage) risk factors for rate of progression from MCI to AD/ADRD. Covariate-adjusted Cox proportional hazards models estimated hazard ratios for incident AD/ADRD, with evaluation of interactions by sex.

Results: Among 6,529 patients, higher systolic BP was associated with AD/ADRD incidence (HR per 10 mmHg: 1.09, 95% CI: 1.05-1.14). BMI was inversely associated with incidence in both males (HR: 0.94; 95% CI: 0.92-0.97) and females (HR:0.98; 95% CI: 0.96-0.99). Compared to married individuals, widowed patients had a higher hazard of progression (HR: 1.15; 95% CI: 1.00-1.32). Spanish-speaking (HR: 1.38; 95% CI: 1.04-1.81), Chinese-speaking (HR: 1.19; 95% CI: 1.00-1.42), and "Other non-English" speaking patients (HR:1.24; 95% CI: 1.03-1.51) had a higher hazard of progression compared to English speakers. Latinx (HR:1.22; 95% CI: 1.01-1.48) and Asian patients (HR:1.14, 95% CI: 1.00-1.30; p=0.04) also had higher hazards of progression compared to White patients. Neighborhood disadvantage was not significantly associated with disease progression.

Discussion: Cardiovascular and social factors independently influence dementia progression, with some sex-specific patterns. Integrating clinical and social indicators highlights the potential of EHR data to identify high-risk patients earlier in the care continuum and support equitable dementia prevention.

Computational growth and remodeling (G&R) models have been extentively used to investigate abdominal aortic aneurysm (AAA) progression and to support clinical decision-making. However, the development of robust predictive models is often limited by the scarcity of large-scale longitudinal imaging datasets. In this study, we propose a physics-based G&R framework to simulate AAA shape evolution and generate a virtual cohort of aneurysms, thereby addressing data limitations and enabling integration with data-driven machine learning approaches for growth prediction. The proposed arterial G&R model incorporates key mechanisms influencing aneurysm progression, including elastin degradation and stress-mediated collagen production. A modified elastin degradation formulation was introduced to generate realistic aneurysm geometries exhibiting clinically relevant features such as asymmetry and tortuosity. By systematically varying parameters governing elastin damage and collagen production, 200 distinct G&R simulations were performed to produce a diverse set of AAA geometries. The dataset was further expanded using kriging-based spatial interpolation to construct a large in silico cohort. The synthetic dataset, combined with longitudinal imaging data from 25 patients, was used to train and validate four machine learning models: Deep Belief Network (DBN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). A two-step training strategy was adopted to predict maximum aneurysm diameter and growth rate based on prior geometric characteristics. The LSTM model achieved the highest performance for maximum diameter prediction (R² = 0.92), while the RNN demonstrated strong overall performance (R² = 0.90 for maximum diameter and 0.89 for growth rate). The DBN and GRU models also showed competitive predictive capability. Overall, this study demonstrates that integrating physics-based G&R simulations with machine learning enables accurate prediction of AAA growth and maximum diameter. The proposed framework provides a scalable strategy for augmenting limited clinical datasets and offers a promising tool to support personalized risk assessment and treatment planning.

Background: Visceral adipose tissue (VAT) has been associated with cardiovascular disease (CVD) mortality. However, the comparative performance of VAT-related clinical surrogates remains poorly characterized.

Objectives: To evaluate the performance of seven VAT-related clinical surrogates for predicting CVD and cause-specific CVD mortality.

Methods: We analyzed data from the Mexico City Prospective Cohort, a population-based prospective cohort study, with baseline recruitmetn between 1998 - 2004 and ongoing mortality follow-up. CVD mortality included deaths from cardiac, stroke-related, and other vascular causes. Seven VAT-related surrogates (METS-VF, CVAI, EVA, DAAT, LAAP, VAI, and DAI) were estimated using clinical, biochemical, and anthropometric data at baseline. Associations with outcomes were evaluated using Cox regression models to estimate adjusted hazard ratios (aHRs). Discrimination was assessed with Harrell's C-statistic (Cs) and fixed-point at 10-years receiver operating characteristic (ROC) curves, and calibration with slope plots.

Results: In a subsample of 102,385 participants (median age: 47 years; 67% female), 4,068 (3.97%) died from any CVD causes. METS-VF (Cs: 0.722; aHR: 1.17, 95% CI: 1.12-1.23), EVA (Cs: 0.72; 1.14, 1.12-1.23), CVAI (Cs: 0.70; 1.13, 1.09-1.18), and DAAT (Cs: 0.626; 1.13, 1.09-1.18) were positively associated with CVD mortality and showed the highest predictive capacity among the surrogates. Adding METS-VF to a CVD risk score among individuals classified as intermediate risk improved discrimination for CVD mortality.

Conclusions: In this large cohort of Mexican adults, four VAT-related clinical surrogates, particularly METS-VF, demonstrated good discriminatory performance for long-term CVD mortality. These indices could help to identify individuals with high VAT accumulation and high CVD risk in resource-limited settings.

Rare diseases affect over 300 million people worldwide, yet patients often endure years-long diagnostic delays that limit timely intervention and trial opportunities. Computational rare disease recognition (RDR) remains constrained by knowledge resources that are often incomplete, heterogeneous, and dependent on extensive multi-disciplinary expert curation that cannot scale. Large language models (LLMs) applied directly for end-to-end diagnosis or disease discrimination face similar knowledge bottlenecks while also raising concerns around cost, reproducibility, and data governance. Here, we introduce GEN-KnowRD, a knowledge-layer-first framework that leverages LLMs to generate schema-guided rare disease profiles, systematically assesses their quality, and constructs a computable knowledge base (PheMAP-RD) for local deployment. GEN-KnowRD integrates this knowledge into lightweight inference pipelines for both general-purpose disease screening and specialized early discrimination from longitudinal electronic health records. Across six public benchmarks for general-purpose screen (9,290 patients spanning 798 rare diseases), GEN-KnowRD significantly improves disease ranking compared to a state-of-the-art, HPO-centered diagnostic framework (up to 345.8% improvement in top-1 success), advanced end-to-end LLM reasoning (up to 129.1% improvement), and a variant of GEN-KnowRD instantiated with expert-curated knowledge rather than LLM-generated profiles. In two real-world cohorts for early diagnosis of idiopathic pulmonary fibrosis (511 patients) as a use case, GEN-KnowRD also demonstrates robust discrimination performance gains, supporting effective RDR during the pre-diagnostic window. These findings demonstrate that repositioning LLMs from diagnostic reasoning to the knowledge layer-decoupling knowledge construction from patient-level inference-yields stronger RDR, while providing scalable, continuously updatable, and reusable infrastructure for diagnosis, screening, and clinical research across the rare disease landscape.

Objective: Cannabis use during pregnancy is increasing; associations with neonatal growth may be confounded by nicotine. We evaluated prenatal cannabis exposure (PreCE) and neonatal outcomes in a prospective cohort with biochemical control for nicotine exposure.

Methods: In the Cannabis Use During Early Life and Development (CUDDEL) study, pregnant women with a lifetime history of cannabis use were classified as PreCE if they self-reported use or had urine THC-COOH positivity at any trimester (n=297) and as unexposed if they reported no use and tested negative (n=151). Linear regression and modified Poisson models estimated associations with birthweight and small for gestational age (SGA; <10th and <5th percentiles), adjusting for sociodemographic factors, gestational age, maternal age and BMI, and urinary cotinine. Analyses stratified by cannabis use frequency (>weekly vs

Results: Participants (N=448; 18-41 years; 85.3% non-Hispanic Black) had lower birthweight with PreCE in adjusted models (Beta=-0.08; padj=0.041). High-frequency PreCE was associated with lower birthweight compared with unexposed pregnancies (Beta=-0.13; padj=0.03), whereas low-frequency PreCE was not. Cotinine-positive PreCE showed the greatest birthweight reduction versus unexposed (Beta=-0.20; padj<0.001). PreCE was also associated with higher likelihood of SGA <5th percentile; risk was highest in PreCE+Nicotine compared with both unexposed and PreCE-Nicotine groups.

Conclusions: Prenatal cannabis exposure was associated with reduced birthweight and SGA in this cohort. Nicotine co-exposure intensified these associations, yet effects persisted without cotinine, supporting cannabis as an independent perinatal risk factor and emphasizing the value of cotinine assessment in populations where blunt use or secondhand exposure is common.

Polygenic risk scores (PRSs) have emerged as a valuable tool for genetic risk prediction and stratification in human diseases. Over the past decade, extensive methodological efforts have focused on improving the predictive power of PRS, leading to the development of numerous methods for PRS construction. Benchmarking these various methods thus becomes an essential task that is crucial for guiding future PRS applications. While studies have benchmarked subsets of these methods on specific phenotypes and cohorts, the resulting evidence remains fragmented, with a lack of work that comprehensively assess the relative performance of the various PRS methods. In this study, we addressed this gap by systematically constructing a PRS method benchmarking database synthesizing published results from 2009 to 2025. We applied a spectral ranking inference framework with uncertainty quantification to rank 14 PRS methods that had been adequately compared against each other in the literature. We constructed rankings using two complementary sources: original method-development studies and applications/benchmarking studies. While the highest-ranked methods (LDpred2 and AnnoPred) and the lowest-ranked method (C+T) were consistently identified from both sources, the relative ordering of most methods showed moderate variability. We further constructed phenotype-specific rankings, providing more detailed insights into the robustness and phenotype-specific strengths of individual methods. Collectively, the overall and phenotype-specific rankings of the PRS methods, along with the curated benchmarking data from the literature, provide a dynamic and practical reference database that can continuingly be updated with emerging new PRS methods and published benchmarking results to guide future PRS applications.