Critical care medicine plays a pivotal role in addressing life-threatening conditions; however, its development in China is hindered by resource scarcity and inequitable distribution. Existing studies have mostly focused on ICU bed numbers or selected regions, and have rarely provided a comprehensive, nationwide assessment of ICU beds, staffing and equipment using standardized methods, nor have they systematically quantified population-based equity of ICU resources. To address this evidence gap and describe the current landscape of critical care capacity in China, we conducted a nationwide survey of critical care resources. The survey aimed to describe the current status of resources, quantify population-based equity, and explore regional and hierarchical disparities that may affect access to critical care. This cross-sectional survey was conducted from January 1, 2021, to December 31, 2023, encompassing public comprehensive hospitals, private hospitals, and specialty hospitals across 31 provinces (including municipalities) in China. A total of 4,744 hospitals with a comprehensive ICU provided valid responses and were included in the analysis; the overall response rate among eligible hospitals invited through provincial health authorities was approximately 89.80

Deep hashing based on deep neural networks has shown great potential for document image retrieval in Chinese ancient texts. However, existing deep hashing methods for image retrieval are not tailored to the characteristics of ancient Chinese documents, which present several challenges. First, a deep hashing network must capture the complex visual features of a large variety of Chinese characters. Second, each bit of the generated hash code should independently reflect distinctive character features. In addition, due to the large number and diverse types of characters in ancient documents, a deep hashing model needs to reduce manual annotation through unsupervised learning. Finally, an end-to-end architecture is required to improve the efficiency of hash code generation. In this work, we propose an end-to-end deep hashing model named FMAH, designed to efficiently locate document images containing query characters within massive collections of Chinese ancient documents. The model first uses a series of CNN layers to extract stroke-level features, which are then fed into a multi-head self-attention module to capture semantic relationships among strokes. Each attention head is independently mapped to a single bit of the hash code, ensuring that every bit carries distinct semantic information. We introduce group metric learning and group contrastive learning to optimize the model. Based on the generated hash codes, we construct a threaded inverted hash index over the document collection and perform approximate searches using Hamming distance. Extensive experiments demonstrate that the proposed FMAH model achieves state-of-the-art performance in Chinese ancient document retrieval.

Traditional k -means minimizes the sum of squared error (SSE) but may treat data points unequally, as some are assigned to significantly distant centroids. This leads to unfair outcomes in downstream tasks such as facility location planning, where each cluster corresponds to a specific share of limited resources. To address this, we modify the objective of k -means via exponential tilting , which emphasizes the impact of distant data points and yields a new objective: the tilted SSE. We propose TKM, which optimizes via coordinate descent and stochastic gradient descent, and improves fairness by shifting centroids toward underrepresented groups. We adopt the within-cluster variance to quantify fairness among individuals within the same group, which provably reduces extreme disparities in outcomes. To improve efficiency, we propose FastTKM, which uses stochastic dynamics to estimate the tilted SSE with lower computational cost. We theoretically demonstrate that, under our proposed methods, the variance decreases with t , a scaling factor that controls the degree of centroid deviation. Furthermore, our methods exhibit time and space complexities comparable to the classical Lloyd's heuristic. Experimentally, our methods outperform six baselines in terms of clustering utility and fairness across twelve real-world datasets. In terms of efficiency, our methods achieve thousand-fold speedups in running time and reduction in memory usage, with this factor growing as the dataset size increases.

Modern data-intensive applications increasingly require database systems to manage structured records and graph data. This demand gives rise to graph-relational data management, spanning storage, query processing, and optimization across relational and graph data. In response, relational database extensions, multi-model databases, and dedicated graph-relational systems have emerged with diverse architectures. However, evaluation methodologies have not kept pace. Existing relational and graph benchmarks assess the two models largely in isolation, while multi-model benchmarks provide limited coverage of graph-relational workloads. Available graph-relational workloads mainly support functional validation and end-to-end latency measurement, revealing little about how storage, operator, and optimization designs affect performance. To evaluate system capabilities in graph-relational data management, we present GRBench. First, GRBench constructs a linked graph-relational schema from the real-world SciSciNet-v2 dataset and derives scalable instances through consistency-preserving subset extraction. Second, it organizes purpose-built query series for controlled evaluation of query processing and system components. Third, GRBench provides semantically equivalent native query formulations and evaluates representative system architectures through a unified, multidimensional methodology. Based on this evaluation, we analyze design trade-offs and identify open challenges to guide future system design and optimization.

Background To develop and internally and externally validate a model for predicting in-hospital mortality risk among patients with cirrhosis and sepsis, using routine clinical data from the first 24 hours after ICU admission. Methods This was a retrospective cohort study. The development cohort was drawn from MIMIC-IV (2008–2019; n = 2463) and randomly split 8:2 into a training set (n = 1970) and an internal validation set (n = 493). The external validation cohort comprised 237 patients from Zhongnan Hospital of Wuhan University (2015–2025). Adults with a diagnosis of cirrhosis who met Sepsis-3 criteria were included; patients with an ICU length of stay ≤ 24 hours or critical missing data were excluded. The primary endpoint was all-cause in-hospital mortality. Seven algorithms were trained: Logistic Regression, Ridge, Lasso, Decision Tree, Random Forest, XGBoost, and LightGBM. Missing values were handled with IterativeImputer. The primary criterion for selecting the optimal model was the arithmetic mean of AUROCs across cross-validation, internal validation, and external validation. Accuracy, precision, recall/sensitivity, specificity, F1 score, and Brier score were also reported. Performance was compared with conventional scores including SOFA, SAPS II, OASIS, LODS, MELD, and ABIC. The final model was interpreted with SHAP. Results A total of 2700 patients were included (2463 from MIMIC-IV and 237 from Zhongnan Hospital). XGBoost showed the best overall performance, with AUROCs of 0.940 (training), 0.833 (internal validation), and 0.776 (external validation); the three-set mean AUROC was 0.814. Discrimination of XGBoost exceeded that of the compared conventional scores (DeLong test, P < 0.05). The leading SHAP contributors included liver transplantation status, total bilirubin, prothrombin time, bloodstream infection, and urine output. Conclusions An XGBoost model based on first-day ICU data can stratify in-hospital mortality risk in patients with cirrhosis and sepsis, with discrimination superior to conventional scores and with clinically interpretable predictors.

INTRODUCTION:The impact of continuous renal replacement therapy (CRRT) with oXiris hemofilter (oXiris CRRT) on sepsis outcomes remain controversial. We aimed to investigate the association between the oXiris CRRT and subsequent outcomes in adult patients with septic shock. METHODS:This single-center, retrospective cohort study included adult patients with septic shock, who were admitted to the intensive care unit (ICU) of our tertiary referral hospital between 2019 and 2023, and underwent at least one CRRT session. Patients were categorized into two groups based on the initially hemofilter used (oXiris group vs. M100 group). A 1:1 propensity score matching was performed to compare the primary outcome 28-day vasopressor-free days. RESULTS:Of 465 patients who met all eligibility criteria, 69 cases in the oXiris group were matched with 69 individuals in the M100 group. Compared with the M100 group, the use of oXiris CRRT was associated with longer 28-day vasopressor-free days (24 vs. 19 days, P = 0.04), shorter CRRT duration (72 vs. 117 h, P = 0.03), and a greater reduction in SOFA scores (-2 vs. 1, P = 0.004). However, competing risk analysis indicated no significant between-group difference in the cumulative incidence of vasopressor weaning after accounting for the competing risk of ICU mortality. Furthermore, no significant between-group differences were observed in changes in lactate, PCT, or IL-6 levels, fluid balance, ICU or hospital length of stay (LOS), or short-term mortality. CONCLUSIONS:In this cohort of septic shock patients undergoing CRRT, unadjusted analyses demonstrated potential improvements in hemodynamic stability and organ function following oXiris therapy. Nevertheless, these preliminary signals were not confirmed by competing risk analysis of the primary outcome. Therefore, the clinical impacts of oXiris CRRT observed in the present study must be viewed cautiously, well-designed prospective investigations are warranted to verify its definitive therapeutic value.

Diverse types of edge data, such as 2D geo-locations and 3D point clouds, are collected by sensors like lidar and GPS receivers on edge devices. On-device searches, such as k-nearest neighbor (kNN) search and radius search, are commonly used to enable fast analytics and learning technologies, such as k-means dataset simplification using kNN. To maintain high search efficiency, a representative approach is to utilize a balanced multi-way KD-tree (BMKD-tree). However, the index has shown limited gains, mainly due to substantial construction overhead, inflexibility to real-time insertion, and inconsistent query performance. In this paper, we propose UnIS to address the above limitations. We first accelerate the construction process of the BMKD-tree by utilizing the dataset distribution to predict the splitting hyperplanes. To make the continuously generated data searchable, we propose a selective sub-tree rebuilding scheme to accelerate rebalancing during insertion by reducing the number of data points involved. We then propose an auto-selection model to improve query performance by automatically selecting the optimal search strategy among multiple strategies for an arbitrary query task. Experimental results show that UnIS achieves average speedups of 17.96x in index construction, 1.60x in insertion, 7.15x in kNN search, and 1.09x in radius search compared to the BMKD-tree. We further verify its effectiveness in accelerating dataset simplification on edge devices, achieving a speedup of 217x over Lloyd's algorithm.

Approximate nearest neighbor search is a core operation in modern information systems, supporting large-scale similarity search tasks. While partition-based indexes accelerate this process, the routing process becomes increasingly expensive as the number of partitions increases, which remains an often-overlooked bottleneck in existing optimizations. To address this problem, we propose LorIndex (Low-rank Index), the first routing framework to exploit low-rank structure for efficient partition selection. Moreover, we develop a novel optimization algorithm that transforms the intractable discrete routing problem into a sequence of convex sub-problems with closed-form solutions, with formal proofs of index efficiency. Experimental results on three public datasets show that LorIndex consistently outperforms seven state-of-the-art methods, including graph-based approaches, with superior throughput across all recall levels. Specifically, across MNIST, Fashion, and Gist datasets, LorIndex achieves state-of-the-art throughput performance for the majority of recall rates from 0.8 to 1.0 compared to nine baseline methods, including 67% improvement over HNSW at 0.9 recall on MNIST.

Graph-centric cross-model data integration and analytics (GCDIA) refer to tasks that leverage the graph model as a central paradigm to integrate relevant information across heterogeneous data models, such as relational and document, and subsequently perform complex analytics such as regression and similarity computation. As modern applications generate increasingly diverse data and move beyond simple retrieval toward advanced analytical objectives (e.g., prediction and recommendation), GCDIA has become increasingly important. Existing multi-model databases (MMDBs) struggle to efficiently support both integration (GCDI) and analytics (GCDA) in GCDIA. They typically separate graph processing from other models without global optimization for GCDI, while relying on tuple-at-a-time execution for GCDA, leading to limited performance and scalability. To address these limitations, we propose GredoDB, a unified MMDB that natively supports storing graph, relational, and document models, while efficiently processing GCDIA. Specifically, we design 1) topology- and attribute-aware graph operators for efficient predicate-aware traversal, 2) a unified GCDI optimization framework to exploit cross-model correlations, and 3) a parallel GCDA architecture that materializes intermediate results for operator-level execution. Experiments on the widely adopted multi-model benchmark M2Bench demonstrate that, in terms of response time, GredoDB achieves up to 107.89 times and an average of 10.89 times speedup on GCDI, and up to 356.72 times and an average of 37.79 times on GCDA, compared to state-of-the-art (SOTA) MMDBs.

Abstract Kidney-function assessment relies on blood urea as a clinically informative metabolic marker; however, its dependence on venipuncture and centralised laboratory testing limits high-frequency monitoring and delays timely clinical intervention. Here, we report an integrated platform combining a wearable buffered microfluidic patch with a physiology-informed, data-driven calibration framework for real-time, non-invasive estimation of blood urea from microlitre-scale sweat volumes (4.79 μL). By precisely regulating the release kinetics of internal buffer salts, the device stabilises the local reaction microenvironment, mitigating variability in sweat pH and flow to ensure reproducible measurement. The resulting signals are processed through an artificial intelligence (AI)-enabled analysis pipeline that integrates sweat urea with patient-specific physiological information to generate clinically interpretable outputs. In multicentre studies, sweat urea shows a strong association with blood urea across diverse cohorts, but with nonlinear and time-lagged relationships that limit direct use. The AI-enabled calibration model compensates for these effects, enabling high-fidelity estimation of blood urea (r = 0.945 versus gold-standard measurements) at clinically relevant concordance levels. The platform further identifies kidney injury with 89.1% accuracy and stratifies disease severity with 83.2% accuracy. Notably, these results demonstrate that the integration of physicochemical stabilisation and AI-enabled data-driven translation establishes sweat as a clinically actionable surrogate for renal monitoring, supporting population-level estimation and highlighting the potential for personalised longitudinal assessment, and enabling a scalable, non-invasive strategy for high-frequency kidney disease management.

In the digital age, social networks have become critical platforms for information dissemination, but they also pose significant risks due to the rapid spread of rumors and misinformation. Existing approaches to rumor control often rely on models that assume a single exposure to anti-rumor information is sufficient to mitigate its impact, overlooking the necessity of multiple impressions for effective behavior change. In this work, we address the Rumor Control with Impression Counts problem by proposing the first-ever Machine Learning (ML)-based solution. Our approach leverages Graph Neural Networks combined with a greedy algorithm to efficiently manage large-scale social networks. To further enhance computational efficiency, we incorporate Evolutionary Optimization, resulting in a method that not only addresses the effectiveness challenges but also scales efficiently with network size. Extensive experiments on real-world datasets demonstrate that our approach outperforms existing methods in both effectiveness and scalability, improving computational efficiency by 1 to 2 orders of magnitude.

The k-means algorithm can simplify large-scale spatial vectors, such as 2D geo-locations and 3D point clouds, to support fast analytics and learning. However, when processing large-scale datasets, existing k-means algorithms have been developed to achieve high performance with significant computational resources, such as memory and CPU usage time. These algorithms, though effective, are not well-suited for resource-constrained devices. In this paper, we propose a fast, memory-efficient, and cost-predictable k-means called Dask-means. We first accelerate k-means by designing a memory-efficient accelerator, which utilizes an optimized nearest neighbor search over a memory-tunable index to assign spatial vectors to clusters in batches. We then design a lightweight cost estimator to predict the memory cost and runtime of the k-means task, allowing it to request appropriate memory from devices or adjust the accelerator's required space to meet memory constraints, and ensure sufficient CPU time for running k-means. Experiments show that when simplifying datasets with scale such as 10^6, Dask-means uses less than 30MB of memory, achieves over 168 times speedup compared to the widely-used Lloyd's algorithm. We also validate Dask-means on mobile devices, where it demonstrates significant speedup and low memory cost compared to other state-of-the-art (SOTA) k-means algorithms. Our cost estimator estimates the memory cost with a difference of less than 3% from the actual ones and predicts runtime with an MSE up to 33.3% lower than SOTA methods.

The efficiency of spatial queries is pivotal for the analysis of geometry data in the fields such as computational simulation, point cloud processing and digital engineering. Utilizing the computational capabilities of modern hardware, such as GPUs, offers a promising avenue for accelerating spatial query processing. However, conventional tree-based indexing methods are not optimized for maximal exploitation of GPU resources. To address this problem, we introduce BLAEQ, a multigrid index designed to maximize the potential of GPUs. BLAEQ adopts a multigrid strategy, which represents an index tree with vectors as layers and matrices as connectors. Although BLAEQ shares conceptual similarities with traditional tree-based indexes, its innovative multigrid architecture facilitates effective parallelization on GPUs during the query phase. To optimize GPU utilization, BLAEQ is entirely constructed using BLAS (Basic Linear Algebra Subprograms), leveraging the efficiency of hardware-tuned BLAS libraries like CuBLAS. This design confers BLAEQ with enhanced performance over existing spatial query methods. Our study assesses BLAEQ's performance against state-of-theart spatial query techniques using a range of both real-world and synthetic datasets. The experimental outcomes demonstrate that BLAEQ outperforms the benchmark approaches in terms of query efficiency on geometry data.

Background Norepinephrine (NE) is the first-line vasopressor for septic shock, but high doses may exacerbate microcirculatory impairment and organ dysfunction. This study aimed to investigate the relationship between NE dose and microcirculation, as well as its impact on prognosis, in ICU septic patients. Methods A prospective observational study was conducted on adult septic patients admitted to the ICU of Zhongnan Hospital of Wuhan University between January and September 2025. Sublingual microcirculatory parameters (microvascular flow index (MFI), total vessel density (TVD), perfused vessel density (PVD), proportion of perfused vessels (PPV), heterogeneity index (HI)) were monitored using a handheld imaging system within 24 hours of admission and on Day 3. Clinical data were simultaneously collected. Multiple regression models were used to analyze associations between NE dose, microcirculation and clinical outcomes (28-day/90-day mortality), with preliminary exploration of their dose-response relationships. Result Of 144 screened adult septic patients, 66 were enrolled. NE dose was significantly correlated with multiple microcirculatory parameters: positive correlations were observed with Lac, mottling score, and HI (Lac: r = 0.583, p < 0.001; mottling score: r = 0.364, p = 0.003; HI: r = 0.444, p < 0.001); negative correlations with MFI and PPV (MFI: r = -0.492, p < 0.001; PPV: r = -0.420, p < 0.001). After adjusting for covariates including APACHE II scores, heart rate, and IL-6 levels, the generalized additive model (GAM) analysis revealed significant nonlinear relationships between NE dose and microcirculatory parameters (all p < 0.05). When NE dose exceeds the threshold range of 0.71–0.80 \(\:\mu\:\)g/kg/min, microcirculation in septic patients deteriorates significantly. Multivariable Cox regression showed high NE dose (cut-off value = 0.80 \(\:\mu\:\)g/kg/min) was associated with increased mortality (HR = 1.14, 95% CI: 0.48 to 2.68, p = 0.035). Conclusions This study demonstrates that excessive NE use is independently associated with worsening microcirculatory perfusion and increased mortality in ICU septic patients. When NE dose pumping over 0.71–0.80 \(\:\mu\:\)g/kg/min, microcirculatory dysfunction should be noted. This provides important evidence for the precise regulation of NE dose in sepsis management, emphasizing the significance of integrating microcirculation monitoring to avoid further deterioration and improve patient prognosis.

Adult trauma patients with refractory acute cardiopulmonary failure suffer from high morbidity and mortality. In the past decade, a growing body of researches has shown survival benefits of extracorporeal membrane oxygenation (ECMO) in trauma patients who fail to respond to optimal damage control resuscitation (DCR), and there is an opportunity to formulate clinical practice guidelines to guide clinicians in implementing trauma ECMO at the bedside. The Chinese Society of Extracorporeal Life Support (CSECLS) convened a domestic panel of interdisciplinary experts to develop this guideline, adhering to the principles of the World Health Organization (WHO) Manual for Guideline Development and the policy of conflict of interest. Clinical key questions pertaining to trauma ECMO use were informed from expert interviews and literature reviews, and formulated as PICO (Population/Intervention/Comparison/Outcome) format for literature retrieval of original studies supporting the question. Then, panelists were assigned to address specific clinical questions, synthesize evidence, formulate recommendations and determine their strength, following the Recommendations Assessment, Development and Evaluation (GRADE) framework. The guideline steering committee and stakeholders approved the final document. Eleven recommendations regarding trauma ECMO use in adult patients were formulated, focusing on the following topics: (1) indications; (2) patient screening; (3) timing of initiation; (4) multidisciplinary approach; (5) trauma ECMO management; and (6) complication prevention. Supporting evidences are elaborated in detail, and expert opinions on clinical application and future research provided. Although the quality of the body of evidence is low to very-low, most researches have shown that ECMO improves the survival of adult trauma patients with varied injury mechanisms. However, decision-making should consider the individual characteristics, benefits and potential harms, patients’ values and preferences, and long-term outcomes.

Sepsis-associated acute kidney injury (SA-AKI) is a severe condition with high mortality rates and a lack of specific treatments. Dendrobine (DEN) has shown diverse pharmacological effects across different diseases. Nonetheless, its impact on SA-AKI remains unexplored. This study aimed to investigate DEN's therapeutic potential in SA-AKI and elucidate its mechanism of action. In vivo, SA-AKI models were induced through cecal ligation and puncture or lipopolysaccharide (LPS) administration, while in vitro model was established using LPS-stimulated HK-2 cells. We found that pre-treatment with DEN reduced levels of inflammation-related cytokines, including tumor necrosis factor-alpha (TNF-α), interleukin-1 beta (IL-1β), and interleukin-6 (IL-6), and improved kidney function in SA-AKI both in vitro and in vivo. RNA-seq analysis unveiled the critical role of mitophagy in DEN treatment for SA-AKI. We observed an initial increase in mitophagy-related proteins such as PINK1, PARKIN, and LC3B/A, peaking at 8 h post-LPS stimulation, followed by a subsequent decline. Additionally, we demonstrated that DEN upregulated the expression of mitophagy-associated proteins in both in vitro and in vivo SA-AKI models. Notably, we found that carbonyl cyanide 3-chlorophenylhydrazone (CCCP) increased LC3B/A levels in DEN treatment for SA-AKI, whereas Mdivi-1 counteracted the effect of DEN on PINK1, PARKIN, and LC3B/A. These findings demonstrated that DEN enhances mitophagy through the activation of PINK1/PARKIN-mediated pathways, thus mitigating SA-AKI.

Recommending suitable items to a group of users, commonly referred to as the group recommendation task, is becoming increasingly urgent with the development of group activities. The challenges within the group recommendation task involve aggregating the individual preferences of group members as the group's preferences and facing serious sparsity problems due to the lack of user/group-item interactions. To solve these problems, we propose a novel approach called Dependency Relationships-Enhanced Attentive Group Recommendation (DREAGR) for the recommendation task of occasional groups. Specifically, we introduce the dependency relationship between items as side information to enhance the user/group-item interaction and alleviate the interaction sparsity problem. Then, we propose a Path-Aware Attention Embedding (PAAE) method to model users' preferences on different types of paths. Next, we design a gated fusion mechanism to fuse users' preferences into their comprehensive preferences. Finally, we develop an attention aggregator that aggregates users' preferences as the group's preferences for the group recommendation task. We conducted experiments on two datasets to demonstrate the superiority of DREAGR by comparing it with state-of-the-art group recommender models. The experimental results show that DREAGR outperforms other models, especially HR@N and NDCG@N (N=5, 10), where DREAGR has improved in the range of 3.64% to 7.01% and 2.57% to 3.39% on both datasets, respectively.

With the development of modern cloud platforms, an increasing number of users are migrating their data analysis tasks to the cloud. Cloud platforms offer a “pay-as-you-go” model, prompting users to focus on both performance and resource costs. Existing query optimization methods primarily address query performance while neglecting resource costs. Mapping queries to their resource consumption is a complex task. To tackle this challenge, we propose a novel learning-based query resource recommendation method called LORE. LORE efficiently and accurately estimates the optimal resources for queries by leveraging dual information from SQL query statements and query execution plans. We model SQL queries and execution plans as directed acyclic graphs and utilize graph neural networks to derive comprehensive representations. To capture the dependencies among all nodes involved in data transmission within an execution plan, we assign path weights to the dependency edges of each node. Our approach integrates data distribution information and captures both direct and indirect dependencies among plan nodes while avoiding unnecessary redundant computations. Experimental results demonstrate that, compared to traditional and other learning-based methods, the LORE model achieves higher accuracy in predicting the optimal resources for queries.

Acute respiratory distress syndrome (ARDS) is a highly lethal non-cardiogenic pulmonary edema. In recent years, with the rapid development of critical care medicine technology and the global pandemic of coronavirus infectious disease-2019 (COVID-19), the medical community has gained new insights into the diagnosis and treatment of ARDS. Lung-protective mechanical ventilation remains its primary and widely accepted treatment approach. However, there is still a lack of systematic summary of non-mechanical ventilation treatment strategies for ARDS. Therefore, the special committee of critical care medicine of the Chinese Research Hospital Association organized domestic experts in related disciplines, followed the guidelines of the GRADE international framework, systematically reviewed, analyzed, and discussed relevant domestic and international research, ultimately producing this guideline.