Fidal

IntroductionAcute kidney injury after cardiac surgery (CSA-AKI) is a common complication after cardiac surgery and an independent predictor of morbidity and mortality. Since evidence-based care is based on risk mitigation and implementation of supportive measures, early risk stratification and consequent adjustment of treatment strategies are elemental. Artificial intelligence screening can aid in pre- and perioperative risk stratification.MethodsThis is a secondary analysis of 130 prospectively recruited patients from one center of a multicenter observational trial that investigated the implementation of a bundle of supportive measures to prevent AKI in patients undergoing cardiac surgery with cardiopulmonary bypass. Machine learning (ML) enabled artificial intelligence (AI) was used to retrospectively analyze patients' electronic health record (EHR) data and generate an AKI risk estimate. The aim of this study was to investigate the feasibility of AI-based risk scores to predict AKI within 72 hours postoperatively and the development of acute kidney disease (AKD) at day 30 after surgery.ResultsOf 130 patients, 33.1% developed CSA-AKI. Of 119 with 30-day follow-up data, 18.5% developed AKD. Day-of-surgery AI risk-scoring was evaluated with an AUROC of 0.79 for occurrence of CSA-AKI, postoperative risk predictions were evaluated with an AUROC of 0.83 for AKD at 30 days postoperatively. ANOVA testing revealed that patients who developed CSA-AKI or AKD had significantly higher predicted risk scores than those who did not, with large effect sizes. Predicted risk also increased significantly over the perioperative period in patients with adverse outcomes.ConclusionsRisk stratification with an ML-based AI approach based solely on EHR data provides a low-effort and high-yield screening method for identifying patients at risk of developing CSA-AKI and AKD. These findings indicate that an EHR (Electronic Health Record)-only model, trained on routine hospital data, provides clinically actionable discrimination in a real-world cardiac surgery cohort, supporting its use for early screening and targeted mitigation.

Cardiac amyloidosis is frequently overlooked, leading to delayed diagnosis and underestimation of its true impact on cardiac mortality. Using CDC WONDER (1999-2020), we identified U.S. decedents aged ≥ 25 years with amyloidosis (ICD-10 E85) as the underlying cause of death and at least one cardiac-related condition as a contributing cause. Age-adjusted mortality rates (AAMRs) per 100,000 population were calculated, and trends were analyzed using Joinpoint regression and autoregressive integrated moving average (ARIMA) forecasting through 2040. Among 16,673 deaths, the AAMR rose from 0.261 in 1999 to 0.608 in 2020, with inflection points in 2012 and 2017 and an overall annual increase of 3.96%. Mortality was higher in men (0.537 vs. 0.210) and Black individuals (0.83 vs. 0.32), with a sharp post-2017 rise, especially among Black decedents. Geographic disparities were also observed, with the highest rates in the Northeast and urban areas. Most deaths occurred in hospitals (43.2%) or at home (31.4%), and mortality rates peaked in those aged 85 and older. Forecasting models project continued increases in AAMR through 2040, reaching approximately 2.0 overall, with especially elevated rates among Black (≈ 5.9) and male (≈ 3.8) decedents. Cardiac mortality in amyloidosis has risen sharply since 2017, likely driven by improved ATTR-CM detection through non-biopsy imaging and guideline adoption. Disparities by sex, race, and geography persist, reflecting uneven diagnostic capacity concentrated in urban and Northeastern centers. Forecasts project continued increases, particularly among Black and male individuals, underscoring the need for more equitable access to diagnosis and treatment.

Quantization is a critical technique for accelerating LLM inference by reducing memory footprint and improving computational efficiency. Among various schemes, 4-bit weight and 8-bit activation quantization (W4A8) offers a strong balance between accuracy and performance. However, existing W4A8 GEMM kernels fall short in practice due to inefficient dequantization on CUDA Cores, which cannot keep pace with the high throughput of Tensor Cores. In this paper, we present LiquidGEMM, a hardware-efficient W4A8 GEMM kernel for efficient LLM serving. LiquidGEMM designs two key techniques: LiquidQuant, a hardware-efficient quantization method that enables fast, overflow-safe dequantization using just two arithmetic instructions per four elements; and an implicit fine-grained pipeline that fully overlaps weight loading, dequantization, and MMA across warp groups without software synchronization or redundant memory traffic. Experimental results show that LiquidGEMM achieves up to 2.90x speedup over state-of-the-art W4A8 kernels and up to 4.94x end-to-end system-level speedup. Compared to various quantized GEMM kernels in NVIDIA TensorRT-LLM, LiquidGEMM delivers 1.12-1.63x performance gains, and achieves up to 1.63x system-level speedup.

Due to the harmful effects of excessive plastic packaging on the environment, there is a growing demand for sustainable alternatives in the food industry. This article provides a comprehensive review of the progress made in the development of biodegradable and active films produced from fruit waste, with a specific focus on the integration of antioxidant and antimicrobial compounds. The research was conducted through the ScienceDirect platform and MDPI in December 2024, leading to the identification of nine relevant studies. The results suggest that agricultural byproducts, including peels, pulps, and seeds, are valuable sources of bioactive compounds capable of providing functionality to packaging, resulting in increased shelf life and reduced use of synthetic additives. The casting technique was widely used in the studies, demonstrating its efficiency in integrating natural additives. Despite significant advances, few studies have tested the direct application of the films to food, highlighting a gap in the practical validation of these technologies. Nevertheless, the findings reinforce the potential of these materials as a viable and environmentally friendly alternative to conventional plastic packaging.

Federated Learning (FL) allows distributed model training without sharing raw data, but suffers when client participation is partial. In practice, the distribution of available users (\emph{availability distribution} $q$) rarely aligns with the distribution defining the optimization objective (\emph{importance distribution} $p$), leading to biased and unstable updates under classical FedAvg. We propose \textbf{Fereated AVerage with Optimal Transport (\textbf{FedAVOT})}, which formulates aggregation as a masked optimal transport problem aligning $q$ and $p$. Using Sinkhorn scaling, \textbf{FedAVOT} computes transport-based aggregation weights with provable convergence guarantees. \textbf{FedAVOT} achieves a standard $\mathcal{O}(1/\sqrt{T})$ rate under a nonsmooth convex FL setting, independent of the number of participating users per round. Our experiments confirm drastically improved performance compared to FedAvg across heterogeneous, fairness-sensitive, and low-availability regimes, even when only two clients participate per round.

Aims: Being able to communicate, through language or other means is an essential pre-requisite in the development of friendships. This study investigated the friendships of students aged 11 to 19 with complex communication needs (CCN), who had limited speech and used Augmentative and/or Alternative Communication (AAC) to communicate. This paper outlines the methods used for engaging students with CCN while findings relating to the development of friendships have been published in a separate paper (Ho et al., 2024). Methods: This study used a Mosaic Approach, which combines traditional methods of interviewing and observation with participatory methods, to elicit students’ views. Participants included six students, eight of their parents and six of their teachers or teaching assistants. Findings It is possible to elicit the views of students with CCN using a Mosaic approach, which involved adapted versions of participatory approaches including a Pyramid Ranking, Card Sorting, School Tours, Using Photos and Objects, and Book Making, in addition to traditional methods of data collection. There needs to be considerable preparation to ensure that the approaches used are personalised for each student knowing how they prefer to communicate, what to include and making appropriate use of communication partners. Limitations This study aimed to understand friendships in and out of school. Information about friendships out of school was limited as the communication partners were from school and it was more difficult for them to elaborate on students’ responses when they spoke about home. Students who might not understand the concept of friendship, or could not communicate about friendships through AAC were excluded from this study. More innovative approaches could be trialled in future research to elicit their views. Conclusion Students with CCN are not passive responders to their environment. They have agency and can choose who they want to be friends with, what activities they want to engage in with different friends and how friends adapt to the other’s needs to create meaningful friendships.

Background: The shortage of healthcare professionals and demanding working conditions have prompted healthcare organizations to implement effective retention strategies. Leadership practices largely shape the organizational climate and influence occupational well-being and related retention. Research has identified warmth and competence as the two fundamental dimensions by which leaders are evaluated.Purpose: This study explores how perceptions of leadership-related warmth and competence translate into hospital employees' loyalty, both directly and indirectly, through occupational well-being, that is, through the mechanisms of positive and negative work-related affect.Research Design: A cross-sectional survey design was used.Study Sample: The study sample consisted of N = 1907 employees from a U.S. hospital.Data Collection and Analysis: Data were collected via a structured survey instrument assessing leadership perceptions, affective well-being at work, and employee loyalty. Structural equation modeling was used to assess the hypothesized relationships and mediating mechanisms.Results: Both leadership-related warmth and competence perceptions directly influence employee loyalty. We identified different mechanisms underlying the indirect effects of occupational well-being on loyalty. Only leadership-related warmth perceptions had a positive indirect effect on loyalty through both positive and negative work-related affect. Leadership-related competence perceptions increased both positive and negative work-related affect.Conclusions: An empathic leadership style that leads to favorable leadership-related warmth perceptions is an effective approach for promoting occupational well-being and loyalty among hospital employees. Leaders should foster leadership-related competence perceptions in periods with more favorable working conditions while avoiding them in stressful working conditions.

This paper proposes a Digital Oath framework to engineers and digital actors designing AI systems. The framework consists of an actual oath with defined sets of pledges, as well as a supporting ground in form of a hub of collaborators. The oath with the proposed structure and content, inspired by the Hippocratic Oath, aims to be universal and encourage the digital actor to commit to principles related to the design of the digital systems and values of the human being, the peoples and the communities as well as the environment. It is structured in three sets of pledges, the first has the actor commit to participate and be aware of human rights and of the system being built. The second set contains guardrail pledges that inspire to do no harm. The third set promotes positive ethics, or the design of systems that do good. Furthermore, to address the inevitable complexity of translating principles into practice, the digital oath proposes a hub structure, a network of interdisciplinary professionals, sharing knowledge and situational experiences seeking to document good practices, and define methodologies for ethical AI designs. Within this framework, the digital oath links together the positive ethical commitment of the singular person and the collaborative deliberation to make sure AI design is centered on the human, the peoples and the communities. Through motivation, intention and collaboration, the digital oath framework is a step-by-step all-responsible end-user-oriented approach to building positive human-centered positive AI systems.