82837 research outputs found
Sort by
Outcomes in severe aortic stenosis patients with echocardiogram characteristics of cardiac amyloid and new onset left bundle branch block after transcatheter aortic valve insertion
No full textTranscatheter aortic valve implantation (TAVI) has become a cornerstone therapy for patients with severe aortic stenosis (AS), yet heterogeneity in outcomes remains, with prognosis influenced by both myocardial function and conduction system complications. Occult cardiac amyloid (CA) is present in up to one in seven TAVI patients, echocardiographic parameters linked to CA have been associated with poor outcomes and may provide prognostic insight for the broader severe AS population. At the same time, cardiac conduction disturbances—particularly new left bundle branch block (LBBB)—represent one of the most frequent and clinically challenging post-procedural complications. This thesis evaluates prognostic markers derived from transthoracic echocardiography (TTE) and characterizes conduction disease management in order to identify strategies for improved patient selection, monitoring, and post-TAVI care.
In the first study, 34 patients with severe AS and preserved ejection fraction undergoing TAVI were retrospectively analyzed. Echocardiographic parameters previously associated with CA and adverse outcomes were assessed in relation to survival, hospital utilization, and quality of life. The cohort had an average age of 85 years and was 38% female. A stroke volume index (SVi) <35 mL/m² was present in 21% of patients and emerged as the strongest predictor of mortality at two years (RR 3.08, 95% CI 1.11–8.55). Left ventricular S’ (LV S’) <6 cm/s, observed in 70%, was most predictive of health resource use, correlating with increased hospital days both before (20 vs. 4 days, p=0.01) and after TAVI (30 vs. 8 days, p=0.005). These findings suggest that echocardiographic measures of impaired systolic function beyond LVEF provide independent prognostic information in TAVI patients. Routine incorporation of SVi and LV S’ into pre-procedural assessment may allow for better risk stratification, targeted surveillance, and potentially earlier intervention, although larger studies are needed to confirm this.
The second study assessed the incidence, management, and outcomes of new LBBB following TAVI. A retrospective review of 355 patients without baseline conduction disease at the Mazankowski Alberta Heart Institute (2010–2020) revealed that 27% developed new-onset LBBB, with 54% of these being persistent at discharge (NOP-LBBB). The average age was 83 years and 42% were female. Of patients with new LBBB, 22.4% underwent permanent pacemaker implantation before discharge, with 60% of these implants being prophylactic, reflecting concern for progression to high-grade AV block. However, neither QRS duration >150 ms nor PR interval >240 ms was associated with prophylactic device implantation (X² p=0.053; p=0.067). At follow-up, average pacing burdens were modest (12.6% at first follow-up, 21% at 12 months), and 41.6% of prophylactically paced patients were paced <1%, suggesting limited clinical need. Moreover, a significant temporal increase in prophylactic pacemaker use for NOP-LBBB was observed between early (2010–2015) and contemporary (2019–2020) practice (X² adjusted p=0.04). These findings highlight practice variability and the lack of robust markers for identifying which patients with NOP-LBBB truly require device implantation. In settings without ready access to prolonged outpatient monitoring or electrophysiological testing, conservative strategies may lead to overtreatment. Ongoing randomized trials are needed to establish standardized management pathways.
Together, these two studies underscore complementary challenges in TAVI care: risk stratification of patients with severe AS and the management of conduction complications post-procedure. Echocardiographic markers such as SVi and LV S’ appear to provide independent prognostic value, supporting their use in pre-procedural assessment. Concurrently, the high incidence but variable clinical significance of new LBBB emphasizes the importance of refining post-TAVI monitoring and device implantation strategies. Future research should aim to develop algorithms to optimize outcomes, reduce unnecessary interventions, and personalize care in this growing patient population
SynthSQL: a framework for generating synthetic text-to-SQL benchmarks with tunable query difficulty
No full textEvaluating LLMs on tasks such as text-to-SQL is challenging because strong performance might reflect genuine reasoning or merely familiarity with training-set patterns. To address this, we introduce SynthSQL: a framework for generating synthetic text-to-SQL benchmarks with tunable query difficulty, allowing precise control over the structural and semantic characteristics of synthesized SQL. By calibrating SQL query complexity from easy to hard, SynthSQL offers a more rigorous and flexible means of evaluating LLM-based text-to-SQL systems. SynthSQL has both a query generator and a natural language conversion module that produces a question that can be answered by the SQL it generated.
Our experiments on the SynthSQL benchmark show that state-of-the-art methods (e.g., DIN-SQL and DAIL-SQL) underperform the chain-of-thought (CoT) prompting approach we introduced—particularly as query complexity increases. We attribute this gap to two factors: limitations of our natural language question generator and the methods’ difficulty generalizing to SynthSQL’s more challenging, novel queries. To further validate the dataset’s quality and difficulty calibration, we plan a human evaluation study.
These findings underscore the need for benchmarks that span a broader range of complexities than existing public datasets, enabling more accurate assessments of LLMs’ true text-to-SQL capabilities
Localization of Mobile Nodes in GNSS-Denied Environments
No full textWe consider the problem of localizing mobile nodes. The mobile nodes operate as transceivers using wireless or free-space optical communication, fulfilling their primary task of communicating with each other and/or of sending data
to particular collection nodes. We assume that it is important to know the location of each node at each point in time. In many environments, relying on Global Navigation Satellite Systems (GNSS) such as the Global Positioning System (GPS) is either infeasible, e.g., in underground facilities or in-side buildings, or is under attack using various countermeasures, e.g., during military operations. We therefore consider localization schemes where many,
usually the majority, of the nodes do not have access to authoritative location information. The few nodes with authoritative information are called “anchors”. We review several algorithms and approaches proposed to solve this
problem. We assume that it is difficult for an adversary to jam the signals on all nodes, all of the time. Thus, it is possible for some nodes to receive signals from other, primarily nearby, nodes despite jamming. . The common characteristic of the approaches relevant to our assumptions is that the locations of the nodes are determined, indirectly, from the coordinates of the few anchor nodes. Localization is achieved by calculating global distance metrics between nodes and between nodes and the anchor nodes. We pay attention to the 3- dimensional version of the problem where mobile, ground and airborne, e.g., Unmanned Aerial Vehicles (UAVs), assets are combined. We review the process for generating appropriate datasets to test such algorithms and example application of current approaches to the particular application
Subpopulation delineation of Canadian polar bears (Ursus maritimus) in the eastern Beaufort Sea
No full textWildlife management often delineates a species into units to improve monitoring, population estimation, and status assessment. Polar bears (Ursus maritimus) are delineated in 20 subpopulations based on an International Union for the Conservation of Nature definition. This definition requires subpopulations to be geographically distinct groups of individuals with low demographic or genetic exchange. I examined whether the Southern Beaufort Sea and Northern Beaufort Sea, were spatially separated using polar bear telemetry data collected between 2007-2014. To assess possible methods of subpopulation delineation, I grouped 75 adult and sub-adult bears into spatial groups using three classification methods: an observed space-use, a capture location, and an agglomerative hierarchical clustering. I then estimated the overlap between spatial groups during the harvest (February – June) and non-harvest (July – January) periods for each classification method. My results found that polar bears within the eastern Beaufort Sea are not geographically separated based on any of the classification methods, and that 61 bears crossed a subpopulation boundary. This assessment suggests that the entire eastern Beaufort Sea region represents one subpopulation. Any boundary in the eastern Beaufort Sea that separates polar bears into groups would best be considered as delineating wildlife management units rather than unique subpopulations based on biological separation
Identification of Key Genes and Pathways in Esophageal Cancer Using Weighted Gene Co-expression Network Analysis
No full textBackground:
Esophageal squamous cell carcinoma (ESCC) is an aggressive malignancy with poor five-year survival rates largely due to late diagnosis and limited targeted therapies. To improve molecular understanding, identify biological pathways and potential biomarkers, we applied an integrated network-based analysis combining weighted gene co-expression network analysis (WGCNA), protein–protein interaction (PPI) network topology, and microRNA–mRNA integration.
Methods:
We analyzed gene expression data from the GSE161533 dataset to identify differentially expressed genes (DEGs) between esophageal cancer and normal tissues. WGCNA revealed co-expression modules; the module most strongly correlated with tumor stage underwent Gene Ontology and KEGG enrichment, and was mapped to a PPI network using STRING and Cytoscape. Hub genes were identified by intersecting top-ranked nodes from CytoHubba’s EPC, MNC, Degree, and EcCentricity algorithms. External validation of hub gene expression was performed in the independent GSE225178 dataset. Finally, we identified regulatory microRNAs by intersecting differentially expressed miRNAs from GSE112264 (serum) and GSE114110 (tissue), predicting their targets with miRWalk, and overlapping these with yellow-module DEGs.
Results:
The yellow module showed a strong positive correlation with the stage of the disease (r = 0.53, p = 2.3 × 10⁻¹¹). Functional enrichment of the yellow module highlighted involvement in cancer related pathways including extracellular matrix organization, collagen metabolism, and PI3K–Akt signaling. Network analysis identified three hub genes: BGN, THBS2, and COL5A1; of which THBS2 and COL5A1 were significantly upregulated in the validation cohort. Integrative miRNA–mRNA analysis revealed that downregulation of hsa-miR-369-5p and hsa-miR-376a-3p likely contributes to the overexpression of THBS2 and COL5A1, respectively.
Conclusions:
Our systems biology approach elucidates key gene modules and regulatory miRNAs implicated in ESCC progression. THBS2, COL5A1, and their regulatory miRNAs emerge as potential biomarkers and therapeutic targets. These findings provide a framework for future functional validation and the development of early-detection and precision-medicine strategies in ESCC
Project 42 Remembered but Invisible Landmarks Enhance Path Integration (Supplementary files, Excel format)
No full textMachine Learning Approaches to Enhance Resilience of Power Systems
No full textThe increasing frequency and intensity of high-impact low-probability (HILP) events, specifically wildfires at wildland-urban interfaces, pose significant challenges to the resilience of power systems. These HILP events threaten infrastructure, cause power outages, and disrupt communities. The main goal of this research is developing advanced methodologies to overcome the obstacles of power system resilience enhancement in the face of HILP events. A key aspect of this thesis is the application of machine learning techniques. Indeed, machine learning-based strategies address the complicated nature of the proposed challenge to achieve relatively accurate and fast solutions.In the initial phase of this research, the benefits of strategically deploying gas turbines are discussed to improve the resilience of integrated distribution and gas systems during HILP events. To address the challenges provided by HILP events, a tri-layer two-level resilience problem is developed, aiming to reduce load shedding as a resilience index during post-event outages. An adaptive distributionally robust optimization technique is proposed to deal with the inherent unpredictability of HILP events through a multicut Benders decomposition. Additionally, a diurnal variant of the long-term short-term memory network is trained to handle uncertainties raised by the high penetration rate of renewable energy resources. This approach provides a novel perspective on the construction of a robust and adaptable framework to enhance the resilience of integrated systems.In the next research, a rank transactive energy framework is introduced for managing energy in networked microgrids, focusing on optimal dispatch of local resources. The energy trading framework is a systematic method for addressing energy exchange challenges, resulting in more resilient and efficient power distribution. It includes a ranking structure that leads to optimal energy allocation among various types of microgrids. To achieve optimal dispatch within the framework, a quantile-based chance-constrained method is proposed to address unpredictabilities in networked microgrids. This probabilistic approach relies on an extreme learning machine that establishes probabilistic constraints.Then, a resilience enhancement approach against wildfires is developed through coordination among three optimization methods: Priori, posteriori, and interactive. This coordination is achieved by leveraging a learning-based surrogate model to simulate wildfire effects on the preferences of decision makers in wildland-urban areas. The proposed surrogate model consists of primary, current, and refined models that are used for pre-traning, training, and fine-tuning. As a result, the model adapts to changing circumstances and identifies optimal operation strategies that strike a balance between operational efficiency and social convenience during HILP events. The main novelty of the proposed surrogate model is an interactive optimization framework that takes the preferences and priorities of decision makers into account
Conservation planning for forests, tree species, and their genetic populations under climate change in Canada and the USA
No full textTrailing edge tree populations at the warm or dry margins of a species’ range often contain genetic traits that confer tolerance to environmental extremes. These traits may be valuable for supporting adaptation to future climates in other parts of the species’ range, yet the populations that hold them are at heightened risk of loss under projected climate change if not actively conserved. This study presents a continental-scale analysis to identify trailing edge populations of the 100 most common North American tree species within the United States and Canada, systematically prioritize collection of at-risk populations, and to evaluate regions suitable for their long-term conservation through assisted migration.
Using a climate envelope modeling approach and 11 bioclimatic variables, we matched ecosystems historically occupied by a species (1960s baseline) with those projected to have similar climates under 2050s conditions (SSP2-4.5 scenario). Trailing edge populations were defined as those ecosystems where species lose suitable climate habitat by the 2050s. Conservation priorities were assessed using three criteria: (1) forest cover loss, indicating potential local extirpation due to fundamental niche limits; (2) climate velocity, estimating the geographic distance needed to track suitable conditions; and (3) the number of species with at-risk populations per ecosystem. These criteria were combined to identify jurisdictions where seed collections for assisted migration may have the greatest long-term value.
Our results show that trailing edge populations are concentrated in ecozones across the Appalachian region (in number of species with populations at risk), as well as the temperate mixed forests of Midwest and the southern boreal forest (proportional to local species richness). Summaries by jurisdiction with high predicted climate velocity and forest cover loss, such as states and provinces with forested areas bordering the central plains, are expected to have limited capacity for in situ persistence, highlighting a potential need for human intervention. Regions such as the Great Lakes basin and north-eastern Canada emerge as major prospective recipients of assisted migration due to high climate matching with trailing-edge populations and relatively stable forest potential under projected climates.
These findings are integrated in an online Protected Area Selection Tool for North America (http://tinyurl.com/PAST-NAm), which enables users to identify climatically suitable recipient protected areas or ecosystems for a source ecosystem and time period of interest. Limitations include the exclusive use of macroclimatic variables, ecosystem-level resolution, and the absence of projected uncertainty or non-analogue climate filters. The study provides a first assessment to support seed collection, in situ conservation, and climate-informed reforestation planning, with the understanding that species- and site-specific evaluations remain necessary for implementation
Stacey Waters - Abstract 4 - Innovate Conference 2025
No full textFaculty are challenged to meet the individual needs of students in undergraduate nursing. Incorporating Universal Design for Learning (UDL) principles in blended lecture delivery is an essential skill for nursing faculty to meet the needs of diverse learners. However, resources for application of UDL principles to blended lecture have been elusive. We have integrated a range of UDL principles into digital tools and hands-on approaches to teaching and learning