medRxiv : the preprint server for health sciences最新文献

The post-acute sequelae of SARS-CoV-2 (PASC), also known as long COVID, remain a significant health issue that is incompletely understood. Predicting which acutely infected individuals will go on to develop long COVID is challenging due to the lack of established biomarkers, clear disease mechanisms, or well-defined sub-phenotypes. Machine learning (ML) models offer the potential to address this by leveraging clinical data to enhance diagnostic precision. We utilized clinical data, including antibody titers and viral load measurements collected at the time of hospital admission, to predict the likelihood of acute COVID-19 progressing to long COVID. Our machine learning models achieved median AUROC values ranging from 0.64 to 0.66 and AUPRC values between 0.51 and 0.54, demonstrating their predictive capabilities. Feature importance analysis revealed that low antibody titers and high viral loads at hospital admission were the strongest predictors of long COVID outcomes. Comorbidities, including chronic respiratory, cardiac, and neurologic diseases, as well as female sex, were also identified as significant risk factors for long COVID. Our findings suggest that ML models have the potential to identify patients at risk for developing long COVID based on baseline clinical characteristics. These models can help guide early interventions, improving patient outcomes and mitigating the long-term public health impacts of SARS-CoV-2.

Objective: The visual system is a significant site of pathology in psychosis spectrum disorders. However, there is limited research investigating human visual cortex (VC) subregions in this population. Using data from the Bipolar-Schizophrenia Network on Intermediate Phenotypes Consortium (BSNIP-1, BSNIP-2, PARDIP), this study examined structural measures in VC subregions in individuals with psychosis spectrum disorders.

Methods: Cortical surface area and thickness in five VC subregions (hOc1, hOc2, hOc3v, hOc4v, MT) were quantified using FreeSurfer v7.1.0 and compared between individuals with psychosis (n=1211) and healthy controls (n=734). Regional specificity was examined by controlling for total surface area or mean cortical thickness. ComBat was used to harmonize scanner effects. Associations between VC measures and symptom severity, cognition, and childhood trauma scores were assessed.

Results: Individuals with psychosis demonstrated smaller surface area in hOc1, hOc2, and hOc3v, and lower cortical thickness in all five VC subregions compared to healthy controls. Thickness reductions in hOc1, hOc4v, and MT were regionally specific. hOc4v and MT were among the top three regions exhibiting the most robust cortical thickness deficits (d = -0.38 to -0.40) across all VC and Desikan-Killiany brain regions. Lower thickness in mid-level visual subregions were associated with greater positive symptoms, poorer cognition, and higher childhood trauma scores.

Conclusions: This study demonstrates that the visual cortex is among the most profoundly affected brain regions in psychotic disorders. Different patterns of area and thickness changes across early and mid-level visual subregions, along with their varying associations with clinical measures, suggest distinct developmental and disease-related influences.

Background: Vectorcardiographic 3D QRS voltage-time integral (VTI_QRS-3D) is a novel marker of ventricular dyssynchrony pertinent for cardiac resynchronization therapy. It may have additional clinical utility but its normal reference ranges have not been established. We sought to define reference ranges for VTI_QRS-3D in healthy individuals.

Methods: We retrospectively analyzed 12-lead ECGs of healthy adults (2010-2014) and compared them to patients with cardiomyopathy with reduced ejection fraction (EF) <50%. Using the Kors matrix, 12-lead ECGs with QRS duration ≤120 ms were converted to vectorcardiographic X, Y, and Z leads. VTI_QRS-3D was calculated as the instantaneous root-mean-square (3D) voltage integrated over the QRS duration. Reference range limits were defined as the 2.5th to 97.5th percentiles respectively for healthy females and males in age groups 18-34, 35-54 and ≥55 years.

Results: The study included 468 healthy adults (age 44.6 ± 17.0 years; 63.9% female) and 314 patients with cardiomyopathy (age 62.1 ± 14.0 years; 34.4% female). VTI_QRS-3D was significantly larger in the cardiomyopathy patients compared to the healthy population (48.2±21.4 vs. 38.1±9.3 μVs, p<0.0001). Increased age and female sex were significant predictors of lower VTI_QRS-3D in the healthy population (both p<0.0001). VTI_QRS-3D reference ranges for respective age groups for healthy females were 23.2-55.0, 23.9-56.4 and 19.6-50.9 μVs, and for healthy males were 29.9-57.2, 28.2-56.7 and 21.4-55.9 μVs.

Conclusion: VTI_QRS-3D is higher at younger age in healthy population, male sex and in patients having cardiomyopathy with reduced EF. Age and sex need to be accounted for using VTI_QRS-3D as a marker for structural heart disease.

Introduction: Emerging evidence suggests that 20-30% of cases of cerebral palsy (CP) may have a genetic cause. Our group previously identified subsets of patients with CP or CP-masquerading conditions who warrant genetic testing, including those with regression or progressive neurological symptoms (CP masqueraders) and those without any known risk factors for CP (cryptogenic CP). Recognition of these subgroups in clinical settings remains challenging.

Methods: To address this challenge, we developed and evaluated a computational phenotyping approach using ICD- 9/ICD-10 billing codes to automatically identify patients with unexplained CP or CP-masquerading conditions who may benefit from genetic testing. We applied this computational phenotyping approach to a cohort of 250 participants from the Boston Children's Hospital CP Sequencing Study, aimed at identifying genetic causes in CP and CP-masquerading conditions.

Results: Manual review served as the gold standard, identifying 8% as CP masqueraders, 42% as cryptogenic CP, and 50% as non-cryptogenic CP. Computational phenotyping based on ICD-9/10 codes achieved a sensitivity of 95%, specificity of 72%, positive predictive value of 77%, and negative predictive value of 94% in identifying cases warranting genetic testing.

Conclusions: Our findings demonstrate the feasibility of using computational phenotyping to identify patients with CP or CP- masquerading conditions who warrant genetic testing. Further studies are needed to evaluate the effectiveness and real-world application of this tool in larger healthcare systems. Nonetheless, the computational phenotyping approach holds promise as a possible clinical decision support that could be integrated into electronic health record systems, enhancing clinical workflows and facilitating actionable genetic diagnoses.

Introduction: Frontotemporal dementia involves progressive atrophy in deep gray matter nuclei, including the thalamus and basal ganglia (such as the caudate, putamen, nucleus accumbens, and globus pallidus), which are critical for cognition and behavior. This study examined cross-sectional and longitudinal atrophy using a state-of-the-art multi-atlas segmentation method sTHOMAS.

Methods: T1-weighted MRI scans from 274 participants at baseline and 237 at follow-up obtained from the Frontotemporal Lobar Degeneration Neuroimaging Initiative database were analyzed using sTHOMAS. Group differences were assessed using ANCOVA, adjusting for age, gender and intracranial volume as covariates.

Results: Atrophy was significant in the mediodorsal, pulvinar, anterior ventral nuclei, nucleus accumbens, and claustrum, with bvFTD most affected cross-sectionally. Longitudinally, the nucleus accumbens, mediodorsal, and pulvinar nuclei declined further. Atrophy correlated with naming (mediodorsal), working memory (ventrolateral posterior), and executive dysfunction (nucleus accumbens) neuropsychological tests.

Discussion: These findings highlight progressive, nucleus-specific atrophy in FTD and emphasize the importance of cross-sectional as well as longitudinal imaging and sex-specific analyses in understanding disease progression.

Background: Extracting structured data from free-text medical records is laborious and error-prone. Traditional rule-based and early neural network methods often struggle with domain complexity and require extensive tuning. Large language models (LLMs) offer a promising solution but must be tailored to nuanced clinical knowledge and complex, multipart entities.

Methods: We developed a flexible, end-to-end LLM pipeline to extract diagnoses, per-specimen anatomical-sites, procedures, histology, and detailed immunohistochemistry results from pathology reports. A human-in-the-loop process to create validated reference annotations for a development set of 152 kidney tumor reports guided iterative pipeline refinement. To drive nuanced assessment of performance we developed a comprehensive error ontology-categorizing by clinical significance (major vs. minor), source (LLM, manual annotation, or insufficient instructions), and contextual origin. The finalized pipeline was applied to 3,520 internal reports (of which 2,297 had pre-existing templated data available for cross referencing) and evaluated for adaptability using 53 publicly available breast cancer pathology reports.

Results: After six iterations, major LLM errors on the development set decreased to 0.99% (14/1413 entities). We identified 11 key contexts from which complications arose- including medical history integration, entity linking, and specification granularity- which provided valuable insight in understanding our research goals. Using the available templated data as a cross reference, we achieved a macro-averaged F1 score of 0.99 for identifying six kidney tumor subtypes and 0.97 for detecting metastasis. When adapted to the breast dataset, three iterations were required to align with domain-specific instructions, attaining 89% agreement with curated data.

Conclusion: This work illustrates that LLM-based extraction pipelines can achieve near expert-level accuracy with carefully constructed instructions and specific aims. Beyond raw performance metrics, the iterative process itself-balancing specificity and clinical relevance-proved essential. This approach offers a transferable blueprint for applying emerging LLM capabilities to other complex clinical information extraction tasks.

Purpose: This collaborative study, led by the Clinical Genome Resource Severe Combined Immunodeficiency Disease Variant Curation Expert Panel (ClinGen SCID-VCEP), implemented and adapted the American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP) guidelines for interpreting germline variants in genes with established relationships to SCID. The effort focused on the 7 most common SCID-related genes identified by SCID newborn screening in North America: ADA , DCLRE1C , IL2RG , IL7R , JAK3 , RAG1 , and RAG2 .

Methods: The SCID-VCEP conducted a rigorous review of variants that involved database analyses, literature review, and expert feedback to derive gene-specific modifications to the ACMG/AMP guidelines. These specifications were validated using a pilot set of 90 variants. Results: Of these 90 variants, 25 were classified as pathogenic, 21 as likely pathogenic, 14 as variants of uncertain significance (VUS), 18 as likely benign, and 12 as benign. Seventeen variants with conflicting classifications in ClinVar were successfully resolved. The criteria included modifications to 20 of the 28 original ACMG/AMP criteria specific to SCID-related genes.

Conclusion: The SCID-specific variant curation guidelines developed by the SCID-VCEP will enhance the precision of SCID genetic diagnosis and provide a robust framework for interpreting variants in SCID-related genes, contributing to appropriate treatment of SCID.

Background: Genome-wide association studies (GWAS) have identified over 1,000 blood pressure (BP) loci and over 80 loci for Alzheimer's disease (AD). Considering BP is an AD risk factor, identifying pleiotropy in BP and cognitive performance measures may indicate mechanistic links between BP and AD.

Methods: Genome-wide scans for pleiotropy in BP variables-systolic (SBP), diastolic (DBP), mean arterial (MAP), and pulse pressure (PP)-and co-calibrated scores for cognitive domains (executive function, language, and memory) were performed using generalized linear mixed models and 116,075 longitudinal measures from 25,726 participants of clinic-based and prospective cohorts. GWAS was conducted using PLACO to estimate each SNP's main effect and interaction with age, and their joint effect on pleiotropy. Effects of genome-wide significant (GWS) pleiotropic SNPs on cognition as direct or mediated through BP were evaluated using Mendelian randomization. Potential contribution of genes in top-ranked pleiotropic loci to cognitive resilience was assessed by comparing their expression in brain tissue from pathologically confirmed AD cases with and without clinical symptoms.

Results: Pleiotropy GWAS identified GWS associations with APOE and 11 novel loci. In the total sample, pleiotropy was identified for SBP and language with JPH2 ( P _Joint =6.09×10 ^-9 ) and GATA3 ( P _G×Age =1.42×10 ^-8 ), MAP and executive function with PAX2 ( P _G×Age =4.22×10 ^-8 ), MAP and language with LOC105371656 ( P _G×Age =1.75×10 ^-8 ), and DBP and language with SUFU ( P _G =2.10×10 ^-8 ). In prospective cohorts, pleiotropy was found for SBP and language with RTN4 ( P _G×Age =1.49×10 ^-8 ), DBP and executive function with ULK2 ( P _Joint =2.85×10 ^-8 ), PP and memory with SORBS2 ( P _G =2.33×10 ^-8 ), and DBP and memory with LOC100128993 ( P _G×Age =2.81×10 ^-8 ). In clinic-based cohorts, pleiotropy was observed for PP and language with ADAMTS3 ( P _G =2.37×10 ^-8 ) and SBP and memory with LINC02946 ( P _G×Age =3.47×10 ^-8 ). Five GWS pleiotropic loci influence cognition directly, and genes at six pleiotropic loci were differentially expressed between pathologically confirmed AD cases with and without clinical symptoms.

Conclusion: Our results provide insight into the underlying mechanisms of high BP and AD. Ongoing efforts to harmonize BP and cognitive measures across several cohorts will improve the power of discovering, replicating, and generalizing novel associations with pleiotropic loci.

Many difficult to understand clinical features characterize COVID-19 and Post-Acute Sequelae of COVID-19 (PASC or Long COVID, LC). These can include blood pressure instability, hyperinflammation, coagulopathies, and neuropsychiatric complaints. The pathogenesis of these features remains unclear. The SARS-CoV-2 Spike protein Receptor Binding Domain (RBD) binds Angiotensin Converting Enzyme 2 (ACE2) on the surface of host cells to initiate infection. We hypothesized that some patients may produce anti-RBD antibodies that resemble ACE2 sufficiently to have ACE2-like catalytic activity, that is they are ACE2-like proteolytic abzymes that may help mediate the pathogenesis of COVID-19 and LC. In previous work, we showed that some acute COVID-19 patients had immunoglobulin-associated ACE2-like proteolytic activity, suggesting that some COVID-19 patients indeed produced ACE2-like abzymes. However, it remained unknown whether ACE2-like abzymes were seen only in acute COVID-19 patients or whether ACE2-like abzymes could also be identified in convalescent COVID-19 patients. Here we show that some convalescent COVID-19 patients attending a clinic for patients with persistent pulmonary symptoms also have ACE2-like abzymes and that the presence of ACE2-like catalytic activity correlates with alterations in blood pressure in an exercise test.

Nearly 7.5% U.S. adults have long COVID. Recent epidemiological studies indicated that long COVID, is significantly associated with subsequent brain structure changes. However, it remains unknown if long COVID is causally associated with brain structure change. Here we applied two Mendelian Randomization (MR) methods - Inverse Variance Weighting MR method (IVW) for correlated instrument variables and Component analysis-based Generalized Method of Moments (PC-GMM) - to examine the potential causal relationships from long COVID to brain structure changes. The MR study was based on an instrumental variable analysis of data from a recent long COVID genome-wide association study (GWAS) (3,018 cases and 994,582 controls), the Enhancing NeuroImaging Genetics through Meta Analysis (ENIGMA) (Global and regional cortical measures, N = 33,709; combined hemispheric subcortical volumes, N = 38,851), and UK Biobank (left/right subcortical volumes, N = 19,629). We found no significant causal relationship between long COVID and brain structure changes. As we gain more insights into long COVID and its long-term health outcomes, future works are necessary to validate our findings and understand the mechanisms underlying the observed associations, though not causal, of long COVID with subsequent brain structure changes.