Citations

Abstract INTRODUCTION Although cerebrospinal fluid (CSF) biomarkers reflect neurodegeneration in Alzheimer's disease (AD), it remains unclear whether these biomarkers track neurodegeneration in early‐onset Alzheimer's disease (EOAD). METHODS In 80 EOAD patients, we examined correlations between eight CSF biomarker levels and cortical thickness decreases within the EOAD cortical atrophy signature. Significant correlations were then entered into a multiple regression analysis. We also examined contributions of EOAD atrophy and CSF biomarkers to cognitive impairment. RESULTS The levels of four CSF biomarkers correlated to EOAD signature atrophy. Stepwise regression analyses revealed that CSF neurofilament light chain (NfL) levels best predicted EOAD signature atrophy. Multiple regression showed EOAD signature atrophy combined with CSF NfL levels explained more variance in cognitive impairment than either factor alone. DISCUSSION Within EOAD patients, CSF NfL levels relate to the magnitude of cortical atrophy, and a combination of EOAD signature atrophy and CSF NfL levels most robustly predict cognitive impairment.

Abstract Objective This study evaluated sequential multi-site fine-tuning of large language models (LLMs), simulating incremental deployment across institutions for extracting mobility functional status from unstructured clinical notes. Materials and Methods We assembled a corpus of 600 clinical notes from 3 institutions, yielding 3810 annotated note sections (1200 training; 2610 testing). Building upon our previously established LLM mobility extraction pipeline, which utilized optimized prompt engineering, we applied parameter-efficient low-rank adaptation (LoRA) fine-tuning. We evaluated sequential multi-site fine-tuning, where the model is trained locally and only the updated LoRA weights are transferred to the next site. We compared this against joint fine-tuning, which requires pooling all multi-site patient data centrally before training. Performance was measured by micro F1-scores for 2 distinct subtasks: mobility extraction and impairment classification. Results Sequential multi-site fine-tuning yielded comparable results to centralized joint fine-tuning for mobility extraction (F1, 0.861 vs 0.894) and impairment classification (F1, 0.912 vs 0.900). Both sequential and joint fine-tuning outperformed an untuned 70B model. Additionally, single-site LoRA models showed robust cross-site generalization. Discussion Sequential multi-site fine-tuning offers a resource-friendly approach to adapt LLMs to clinical environments. By retaining previously acquired knowledge without severe interference, this strategy allows for local fine-tuning at each institution without the need to pool or share sensitive patient data required for joint fine-tuning. Conclusion Sequential multi-site fine-tuning provides a viable solution for the scalable, incremental deployment of LLMs across institutions, showing promising generalizability for clinical information extraction from electronic health record narratives.

BACKGROUND: Randomized trials comparing radial and femoral access report access site crossover rates of 3% to 9% for radial access and 1% to 3% for femoral access. However, contemporary real-world crossover rates are not well characterized. The aim was to evaluate contemporary trends and predictors of vascular access site crossover during percutaneous coronary intervention (PCI). METHODS: We analyzed index PCI procedures recorded in the National Cardiovascular Data Registry CathPCI Registry from 2018 to 2024. Access site crossover was identified using documented access data, with radial-to-femoral crossover inferred when the final successful access site was femoral. Temporal trends and clinical predictors of crossover were examined. Operator experience was assessed using cumulative total and radial PCI volumes from 2011 to 2024 as captured in CathPCI Registry versions 4 and 5. RESULTS: Among 3 861 790 index PCIs, 192 902 (5.0%) involved access site crossover, with a modest increase over time. Radial-to-femoral crossover accounted for 81.2% of crossover events. Factors independently associated with crossover included older age, female sex, higher body mass index, and peripheral vascular disease. Operators with the lowest total PCI volume and the lowest radial PCI volume experienced the highest crossover rates. Notably, operators with low radial PCI volume demonstrated a substantial decline in radial-to-femoral crossover over the study period. CONCLUSIONS: In contemporary clinical practice, access site crossover rates are consistent with those reported in randomized trials. Although crossover occurs more frequently among lower-volume operators, rates decline with increasing operator experience, supporting the presence of a radial access learning curve and improvement in procedural performance over time.

Abstract Motor speech disorders (MSDs) are early markers of neurological disease, but expert perceptual analysis is rarely available outside specialized centers. Automated speech analysis offers a scalable alternative, yet prior studies have not systematically compared modeling approaches or assessed clinically relevant metrics in independent datasets. This study compared static acoustic features, articulatory informed Phonet features, and self-supervised pretrained models for binary and multi label MSD classification. We trained and evaluated models on 583 speech samples using speaker level splits. Baseline models included logistic regression and Gated Recurrent Units (GRUs) trained on eGeMAPS and MFCCs. We extracted three types of Phonet derived features and evaluated pretrained HuBERT and SSAST models in frozen, partially fine-tuned, and fully fine-tuned configurations. Binary classification distinguished MSDs from controls, while multi label classification identified six MSD subtypes. Models were assessed using validation AUC, and cut points were tested on two independent datasets. Pretrained and Phonet based models substantially outperformed static acoustic features. In binary classification, HuBERT achieved the highest AUC (0.95), while compact Phonet derived GRUs achieved comparable performance (up to 0.94). These models generalized well to independent datasets, maintaining high sensitivity (0.94) and specificity (0.97). In multi label classification, Phonet models achieved the highest macro average AUC (0.86), but threshold-based subtype performance declined on unseen data. Automated MSD detection is feasible and clinically promising. Binary classification generalized well, whereas multi label classification showed limited threshold stability across datasets.

Hepatic resection (HR) and liver transplantation (LT) are curative-intent treatments for hepatocellular carcinoma (HCC). Patients without cirrhosis are typically analyzed as a single group, obscuring potentially important differences between those with histologically normal liver and those with non-cirrhotic parenchymal disease. We aimed to characterize long-term outcomes following HR stratified by underlying parenchymal quality, with a contemporary LT cohort as an external benchmark. Retrospective cohort study of 662 adults undergoing curative-intent surgery for HCC at 2 tertiary centers (2010–2024). Resection patients (n=380) were stratified by non-tumor parenchymal histology into HR-Normal (n=84), HR-Diseased non-cirrhotic (n=159), and HR-Cirrhotic (n=137). A total of 282 LT recipients served as an external reference but were not entered into comparative models. Overall survival (OS) was analyzed by Kaplan–Meier and multivariable Cox regression, and recurrence by Fine–Gray competing-risks regression. Ninety-day mortality was low across resection strata (1.2%–3.2%). OS differed significantly across parenchymal groups (log-rank p =0.01), driven by superior outcomes in HR-Normal patients compared with both HR-Cirrhotic ( p =0.008) and HR-Diseased ( p =0.005). HR-Cirrhotic and HR-Diseased did not differ ( p =0.97). Five-year OS was 74.4%, 59.8% and 57.2%, respectively. On multivariable Cox regression, both cirrhosis (HR 2.03, 95% CI 1.16–3.54, p =0.01) and diseased non-cirrhotic parenchyma (HR 2.04, 95% CI 1.21–3.43, p =0.008) independently predicted death versus normal liver. Fine–Gray analysis confirmed higher recurrence risk in cirrhotic (sHR 2.09, p =0.01) and diseased non-cirrhotic (sHR 1.80, p =0.04) groups. The LT benchmark cohort achieved a 5-year OS of 82.4%. Histologically normal liver identifies a distinct HR subgroup with favorable oncologic outcomes. The conventional cirrhotic/non-cirrhotic dichotomy underestimates the prognostic impact of underlying parenchymal disease.

Background: We describe an investigation of a cluster of eight Corynebacterium amycolatum sternal surgical site infections (SSIs) that occurred among cardiac surgery patients from February to October 2025. Although C. amycolatum is a member of the skin microbiota, clinically significant infections caused by this organism may occur in cardiac surgery patients. Methods: The investigation included retrospective reviews of SSIs, clinical cultures and medical records, surgical observations, and operating room (OR) assessment. Line lists of surgical case criteria and infection prevention practices were created to assess common factors among cases (e.g., staffing, instrumentation, equipment, skin preparation and antisepsis, nasal decolonization, antibiotic protocols, and closure). Whole genome sequencing of isolates was performed and analyzed using SNIPPY for single-nucleotide polymorphism (SNP)-based phylogenetic analysis. Results: Phylogenetic analysis indicated isolates from five of eight case patients who had cardiac surgeries occurring over a six-week period from February to April 2025 were genetically related (?7 SNPs across the genome), suggesting a common source of infection. These five isolates were ceftriaxone, meropenem, and penicillin resistant. The other three case patient isolates were unrelated (<3181 to <15,000 SNPs); two were ceftriaxone, meropenem and penicillin resistant (one also doxycycline intermediate resistant), and the third penicillin resistant. Use of sternal tape for closure was identified as a recent practice change. Four of five genetically related cases involved the same surgeons and suture tape for sternal closure (wire for all other cases). Sternal tape was discontinued with three cases occurring afterwards. Seven patients completed chlorhexidine gluconate bathing and nasal decolonization. Hair was removed in the OR with reusable surgical clippers. All received the same skin antiseptic, and antibiotic prophylaxis was appropriate. No trends were identified in instrument or equipment use. Broad infection control measures were implemented, including surgical clipper replacement, OR cleaning and ultraviolet light disinfection, headlamp cleaning, and environmental repairs. Leadership was engaged through frequent meetings and audits. Infection control practices such as minimizing OR traffic, double gloving, glove changes, and strict adherence to hand hygiene and attire were reinforced. Two of eight cases occurred after these recommendations were completed but were not genetically related to the primary cluster. Conclusion: We describe a cluster of invasive C. amycolatum sternotomy SSIs. Only two genetically unrelated cases of C. amycolatum occurred following infection control investigation and interventions, underscoring the effectiveness of a comprehensive, multidisciplinary approach. This organism has rarely been reported as a cause of SSIs and may be an emerging pathogen.

Most genetic variants associated with complex traits are hypothesized to regulate gene expression. To understand the genetics underlying gene expression variability, we characterized 14,324 RNA-sequencing samples from the Trans-Omics for Precision Medicine program and performed expression and splicing quantitative trait locus (e/sQTL) analyses in six tissues and cell types, including whole blood ( n = 6454) and lung ( n = 1291). We detected tens of thousands of secondary cis-e/sQTLs, showing that secondary cis-e/sQTL discovery remains unsaturated. We fine-mapped UK Biobank–derived genome-wide association study (GWAS) signals from 164 traits and identified e/sQTL colocalizations for 10,611 GWAS signals, including 7096 that colocalize with secondary e/sQTLs. Our results suggest that even larger e/sQTL analyses will uncover additional secondary e/sQTLs, further benefiting GWAS interpretation.

Background: Platelet-rich plasma (PRP) is an emerging therapy that demonstrates potential benefits for knee osteoarthritis (OA); however, the accessibility of PRP remains limited, in part, because its use is not covered by insurance. Purpose: To identify factors contributing to health disparities in PRP treatment for knee OA using machine learning models. Study Design: Cross-sectional study; Level of evidence, 3. Methods: The authors collected data from electronic medical records at 2 upper Midwest medical centers (Mayo Clinic and Detroit Medical Center) from 2022 to 2023 encompassing demographic, socioeconomic, and clinical factors. Data collection encompassed age, sex, race, ethnicity, area deprivation index, marital status, income, rurality, insurance, employment, language, religion, treatment site, and region. Logistic regression and random forest models were applied to 3667 samples (80% training, 20% testing), comparing PRP recipients against a standard care group. Models were evaluated using 10-fold cross-validation to ensure robust performance assessment. Results: Logistic regression revealed that increased age, widowed status, lower income, rural residence, unemployment/disability, and Medicaid/Medicare insurance were associated with lower odds of receiving PRP. Upper-middle income, contracted insurance, and treatment at the Mayo-Arizona and Mayo-Florida sites were associated with higher odds. The model achieved areas under the curve of 0.83 (training set) and 0.82 (testing set). Random forest analysis identified distance to care, age, area deprivation index, income, and treatment site as top predictors, with areas under the curve of 0.92 (training) and 0.87 (testing). Conclusion: The current study demonstrated that both models discerned key barriers (income, distance to care, retirement status) and facilitators (regional differences) affecting PRP accessibility. These findings provide potential policy initiatives aimed at improving PRP access for underserved demographics, offering a quantitative blueprint to foster equitable access to regenerative therapies for knee OA.

We demonstrate the structural and functional similarities of organoid-derived gastric mucus and native mucus collected from human patients. The bioengineered gastric mucus mimics its native counterpart in its proteomic profile and physical architecture. This work highlights the translational potential of organoid-derived mucus for functional investigations of the human gastric mucus layer.

Abstract: Extracorporeal membrane oxygenation (ECMO) is increasingly used to support patients with severe cardiac or respiratory failure. Neurological complications occur across ECMO configurations and represent a major contributor to mortality and long-term disability. Major neurological complications include ischemic and hemorrhagic stroke, hypoxic ischemic brain injury, seizures, delirium, spinal cord and peripheral nervous system injury, and brain death. The mechanisms underlying neurological injury during ECMO are multifactorial and are influenced by ECMO configuration, reflecting differences in cerebral perfusion, gas exchange, anticoagulation exposure, and precannulation physiologic insults. This review synthesizes neurological complications associated with ECMO, neurological monitoring, and outcomes.