Search Results (42,730)

Landslides are characterized by sudden occurrence, severe destructiveness, and widespread spatial distribution. Therefore, rapid and accurate landslide detection is essential for reducing infrastructure damage and safeguarding human lives. Conventional landslide monitoring methods are generally time-consuming, labor-intensive, and inefficient, while existing deep learning-based landslide detection methods remain limited in multiscale spatial structure representation, temporal sequence modeling, and spatiotemporal feature fusion. To address these limitations, this study proposes ConFormer-Net, a spatiotemporal landslide detection model that integrates convolutional neural networks (CNNs) with a Transformer architecture for landslide detection from multi-temporal synthetic aperture radar (SAR) imagery. The proposed model adopts a hybrid architecture and incorporates a cross-attention mechanism. Specifically, a dilated convolutional network is employed to extract local spatial features of landslides, while a Transformer encoding module models the temporal dependencies among multi-temporal SAR observations. The extracted spatial and temporal features are subsequently integrated through the cross-attention mechanism to achieve accurate landslide detection. The overall detection performance of ConFormer-Net was first evaluated using samples from different regions in the publicly available Sen12Landslides dataset. The proposed model was then compared with CNN, CNN-LSTM, ConvLSTM, GRU, CNN3D and ResNet50 models. The results demonstrate that ConFormer-Net achieved the best overall performance, with an F1-score of 95.27%, an accuracy of 95.29%, a precision of 96.79%, and a recall of 93.79%. These results indicate that ConFormer-Net enables highly accurate landslide detection while maintaining moderate model complexity, demonstrating its effectiveness for landslide detection from multi-temporal SAR imagery. Full article

In urban slums, adolescent sexual and reproductive health and rights (ASRHR) are shaped by societal structures which contribute to vulnerabilities, health inequities, and adverse ASRHR outcomes. There have been increasing calls for meaningful youth engagement to foster positive change. The Together Project is a youth-led research project implemented in Kallyanpur slum in Dhaka (Bangladesh) aiming to answer the question: How are societal structures shaping ASRHR structural vulnerabilities and agency? Working with 10 adolescent community co-researchers (8 girls, 2 boys), we recruited 22 participants (16 girls, 6 boys) for a photovoice study exploring ASRHR experiences. We collected 58 photos analyzed through Visualizing DEPICT, a participatory method, and co-developed three themes. First, “Navigating limited ASRHR awareness through a broken system” demonstrates how perspectives of ASRHR are structured within a restrictive system. Second, “Facing gender-based inequities from girlhood to womanhood” highlights gendered inequities that influence structural vulnerabilities affecting adolescent girls. Third, “Shaping ASRHR through community and social pressure” summarizes how social structures shape ASRHR. By centering adolescents’ understandings of ASRHR in an urban slum of Dhaka, we present a discussion of how ASRHR are shaped through oppressive structures. Our findings underscore the urgent need for structurally informed ASRHR interventions to support transformative societal changes. Full article

Power mini-grids are being transformed from locally automated electrical systems into cyber-physical ecosystems in which heterogeneous distributed energy resources, storage, converters, flexible demand and uncertain external conditions must be coordinated. Existing surveys commonly treat microgrid control, artificial intelligence (AI), multi-agent systems, communications, digital twins and cybersecurity as separate research streams. In this survey, the cognitive power mini-grid is introduced as a unifying paradigm in which physical and social contexts are perceived by distributed agents, task-relevant semantic information is exchanged, auditable coordination readiness is checked and proposed actions are subject to independent safety checks. A corpus of 247 scholarly publications, standards and technical sources is synthesized across microgrid engineering, distributed AI, semantic communication, language models, federated learning, neuro-symbolic AI, digital twins, causal reasoning, runtime assurance, cybersecurity and community energy markets. A six-layer architecture is proposed in which fast deterministic control is separated from semantic coordination, distributed learning, agentic deliberation and assurance. Representative studies are compared by problem, method, control horizon, information assumptions, validation environment, hardware-in-the-loop (HIL) status, reported outcome, limitations and architectural relevance. Applications are organized into balancing, resilience, demand-response, maintenance and inter-mini-grid markets. Cross-layer challenges are identified in stability under asynchronous interaction, semantic interoperability, hallucination, common-knowledge failure, edge resources, privacy and cyber-physical security constraints. A research agenda is developed around proof-carrying actions, causal digital twins, edge small language models, federated multimodal foundation models, neuromorphic semantic control and human-agent governance. Cognition is therefore positioned as a safety-gated coordination capability above verified physical control loops rather than as a replacement for established control. Full article

The microbiome, as a vast and dynamic community of microbes, is now recognized as a key regulator of host physiology, profoundly influencing health and susceptibility to disease. Accordingly, probiotics are a mainstay of prevention and treatment. But the individual complexity and dynamic specificity of an individual’s microbiome make the previous “one-size-fits-all” research model completely invalid. This review systematically analyzes the applications of artificial intelligence (AI) and machine learning (ML) as transformative computational tools necessary to surmount these challenges. In this article, we detail how these computational methods have been applied throughout the entire research and development pathway of probiotics, including novel strain identification (through multi-omics analysis), formulation and production optimization, and elucidation of complex mechanisms of action in the host. Furthermore, we highlight the emerging frontier of personalized probiotic therapy, demonstrating how AI/ML can be utilized to predict treatment efficacy based on individual host data. The objective of this article is to provide a detailed discourse on the actual and prospective applications of AI and ML in this process, ultimately delineating their revolutionary potential to inform the design of the next generation of probiotics with unprecedented precision, efficacy, and sustainability. Full article

Seismic exploration has progressively expanded into urban fringe regions and areas with intensive human activities, where anthropogenic interference has increased markedly, aggravating the background noise problem and substantially affecting the accuracy of subsequent seismic data processing. Conventional denoising methods struggle to cope with such complex noise patterns and typically rely on manually designed parameter settings, which limits their ability to suppress complex background noise in real seismic data. With the rapid development of deep learning, various neural-network-based methods have been introduced for seismic data denoising. However, existing deep learning approaches are constrained by the high resolution of seismic data and limited computational resources and therefore commonly perform denoising on cropped data patches rather than on the complete seismic section. This limitation weakens the network’s capability to perceive global contextual information, degrading denoising performance and potentially introducing blocking artifacts. To address these issues, we propose a fluctuation-prior-regularized denoising framework that explicitly decomposes the complete seismic data denoising task into global and local denoising subtasks, enabling globally consistent denoising under complex conditions. To overcome the limitations of CNNs and Transformers, we adopt Retentive Networks Meet Vision Transformers (RMT) as the backbone for feature extraction in this work. And a fluctuation-prior-constrained local denoising mechanism is introduced, allowing local patches to indirectly capture global information. In addition, a grouped regularization strategy for global and local tasks is proposed, enabling both optimization tasks to better capture their respective task-specific characteristics. Experimental results on noisy shot gathers constructed from field records and on field-recorded background noise demonstrate that the proposed DFRSeisnet achieves superior denoising performance while effectively alleviating blocking artifacts. Full article

This paper proposes a finite element approach to the numerical solution of boundary value problems of linear elasticity theory formulated directly in terms of the stress tensor components. In contrast to the classical finite element formulation, in which displacements are the primary unknowns, the approach considered here treats the stress components as the sought quantities. Two forms of the boundary value problem are investigated. The first is based on the joint use of the equilibrium equations and the Beltrami–Michell equations, while the second is a transformed system of Poisson-type equations for the stress tensor components. Variational relations based on the Galerkin method are obtained for both formulations. Using linear basis functions on triangular finite elements, local matrices are constructed, and global systems of algebraic equations are formed, whose unknowns are the nodal values of the stresses. The numerical implementation of the developed schemes is carried out in an in-house C++ program and in the FreeFEM++ software environment. To verify the reliability of the proposed mathematical and numerical models, the classical Kirsch problem of a stretched elastic plate with a circular hole—characterized by a pronounced stress concentration near the edge of the hole—is solved. The numerical values of the stress components are compared with the analytical solution obtained using the Airy stress function method, as well as with the results of calculations performed with the FreeFEM++ software package. The comparison shows good agreement between the obtained solutions and confirms the possibility of determining stresses directly, without first computing the displacement field. A mesh-refinement study further showed a systematic reduction in the numerical error: on the finest mesh considered, the relative error decreased to 3.189% for formulation A and to 7.415% for formulation B. The proposed approach extends the applicability of the finite element method to boundary value problems of elasticity theory formulated in terms of stresses and can be used to study problems with complex geometry and local stress concentration. Full article

CLIP-based anomaly detectors have markedly advanced training-free and zero-shot industrial anomaly detection and localization, yet their predictions remain dominated by patch-wise vision–language similarity or anomaly-aware feature scoring. This formulation is intrinsically limited for logical anomalies, in which every visible part can appear locally normal while its count, position, arrangement, or co-occurrence violates a normal configuration. We introduce a training-free relation prototype re-scoring module that reuses the semantic–spatial relations already encoded by the visual transformer. Because patch tokens contain positional embeddings, the self-attention graph is position-aware as well as content-dependent; we use it as a message-passing operator over detector-specific patch features. Normal images define category-wise, spatially indexed relation prototypes, and each test patch is scored by the Euclidean deviation of its attention-aggregated feature from the corresponding normal prototype. The resulting map supports both dense localization and map-derived image detection. Although we also report the feature-relation map alone to analyze its intrinsic behavior, the final method fuses this map with the original anomaly map so that relation-sensitive evidence is added without discarding the baseline detector’s local appearance cues. We develop the method on AnomalyCLIP and verify its generality on WinCLIP and AA-CLIP. On the logical split of MVTec LOCO AD, the proposed fusion improves AnomalyCLIP from 53.9 to 74.4 pixel AUROC and from 28.7 to 52.7 pixel AUPRO, while image AUROC rises from 50.5 to 71.7. Similar improvements are observed for WinCLIP and AA-CLIP. On CAD-SD, the proposed fusion reaches 92.0 and 93.8 image AUROC for AnomalyCLIP and WinCLIP, respectively, on co-occurrence anomalies. Experiments on the structural split of MVTec LOCO AD and the MVTec AD benchmark show why the baseline map must be retained: fusion preserves substantially more local-defect evidence than relation-only scoring, although its benefit remains detector- and category-dependent. These results identify semantic–spatial relation deviation as a missing cue in CLIP-based logical anomaly detection and localization, without requiring explicit component models, symbolic rules, additional training, or modification of the baseline architecture. Full article

Background/Objectives: Recovery from schizophrenia-spectrum disorders is increasingly recognized as achievable, yet no synthesis has simultaneously examined long-term outcome rates, intervention effectiveness, cognitive predictors, social determinants, personal-recovery trajectories, and mediating mechanisms within a unified framework. This systematic review aimed to address these six domains and to propose an integrative theoretical model of functional recovery, applying meta-analysis where the evidence permitted and structured narrative synthesis elsewhere. Methods: Seven databases (PubMed/MEDLINE, PsycINFO, Embase, Cochrane CENTRAL, Web of Science, Scopus, CINAHL) were searched through December 2025 following PRISMA 2020 and MOOSE guidance. We extracted eligible records (longitudinal designs, ≥6 months, adults with schizophrenia-spectrum disorders reporting functional outcomes) into two structured databases, then consolidated and de-duplicated them, yielding 467 unique papers. After topical screening and a duplicate audit, we retained 368 studies (1980–2025) and grouped them into six thematic clusters. A pre-specified rule pooled only clusters with at least five independent primary studies. Proportions were pooled with the Freeman–Tukey transformation and associations with Fisher’s z under DerSimonian–Laird random-effects models; certainty was appraised with GRADE. Results: Three pools met the threshold. Long-term functional recovery was 35.4% (95% CI 22.6–49.4; k = 11) and symptomatic remission 39.6% (95% CI 29.3–50.5; k = 7), both with very high heterogeneity (I² = 92–98%) and low certainty. The pooled association between psychological mediators and functional outcome was weak and imprecise (r = 0.23; 95% CI −0.09–0.51; k = 7; very low certainty), reflecting facilitators (positive mental health, social support, self-efficacy, hope) and barriers (internalized stigma, substance-use comorbidity, longer duration of untreated psychosis, persistent negative symptoms) acting in opposite directions. The questions concerning specific interventions, cognitive prediction, rehabilitation, and personal recovery were synthesized narratively: structured psychosocial and combined interventions and integrated rehabilitation were associated with small-to-medium functional gains; cognition, particularly social cognition and motivation, predicted later functioning; and personal recovery followed a course partly independent of clinical status. Conclusions: Recovery and remission are attainable for a substantial minority but remain heterogeneous and modestly certain. The proposed Multi-Pathway Dynamic Recovery Model organizes the evidence into five testable principles—an ordered recovery hierarchy, social-cognitive mediation of the cognition–function link, an early-phase intervention window, social-ecological embedding, and self-reinforcing feedback loops, offered as a hypothesis-generating framework rather than a validated structure. Services should prioritize comprehensive early intervention, target social cognition and stigma, and address structural determinants to optimize long-term recovery. Full article

To address the challenges of formation maintenance, collision avoidance, and cooperative navigation for multi-agent systems in shared environments, a spatio-temporal trajectory optimization method under global structural constraints and A attention-enhanced graph neural network (AE-GNN)-based cooperative navigation strategy are proposed. First, a differentiable formation similarity metric based on graph Laplacian theory is constructed to characterize the intrinsic geometric structure of formations, which is invariant to translation, rotation, and scale transformations. Based on this metric, a distributed trajectory optimization framework is developed by jointly considering formation maintenance, obstacle avoidance, inter-agent collision avoidance, and trajectory smoothness. Furthermore, an adaptive formation constraint adjustment mechanism based on environmental passage margin is introduced, allowing agents to relax rigid formation constraints in narrow passages and gradually recover the desired formation after obstacle avoidance. To improve cooperative decision-making in densely cluttered environments, an AE-GNN navigation method is proposed, where an attention mechanism dynamically assigns higher weights to critical neighboring agents during feature aggregation, enhancing cooperative information interaction and collision avoidance capability. Simulation and physical experiments validate the effectiveness of both proposed methods. The trajectory optimization method achieves a success rate of 97.5% in dense-obstacle formation-navigation simulations, while AE-GNN demonstrates effective cooperative navigation and collision avoidance in multi-agent scenarios. Full article

The heat release rate (HRR) indicates the scale of a tunnel fire, and inverting it in real time from ceiling sensors supports fire detection and ventilation control. Purely data-driven (deep learning) models are accurate within the training range but cannot extrapolate to larger fires, whereas a purely physics-based formula is less accurate and fails during the fast-growth transient. This paper proposes a data and knowledge dual-driven HRR inversion method. The data component is an encoder-only Transformer on ceiling thermocouples, and the knowledge component is a slope-corrected plume-scaling inversion. The two are coupled by a training-time soft constraint and an inference-time two-layer gate: a magnitude gate raising the physics weight beyond the training power ceiling, and a steady-state gate down-weighting it during transients. On 24 simulated cases (six slopes × four powers, 0.5–4 MW), data are split by slope and power into mutually exclusive training, validation, and test subsets, the test covering unseen slopes and powers. The method outperforms the physics formula at every power tier; on power extrapolation it far surpasses the pure deep learning model (R² = 0.84), and on slope extrapolation it matches that model (R² = 0.94). The results demonstrate, within the present single-geometry FDS tunnel configuration and the investigated working conditions (0–5% slopes, 0.5–4 MW, t² growth, natural ventilation), that physics-guided gated fusion can improve HRR estimation under a 4 MW single-power extrapolation test while retaining the accuracy of the data-driven model under slope extrapolation; the conclusions are not claimed to be directly transferable to other tunnel configurations. Full article

Background and Objective: Obstructive sleep apnea (OSA) is a fatal widespread sleep-related breathing disorder and a major risk factor for cardiovascular and cerebrovascular diseases, significantly impacting older adults’ health worldwide. Due to the risk of such complications, timely and accurate identification of OSA is crucial. Polysomnography is considered the most accurate technique for detecting OSA; however, it is limited by its complexity and multi-channel requirements. A promising alternative is electrocardiogram (ECG)-based diagnosis, which continuously monitors heart rhythm and captures subtle cardiac changes associated with OSA. Nevertheless, existing ECG-based approaches still face challenges related to complex feature engineering, limited capture of complementary temporal–spectral information and global dependencies, along with inadequate feature recalibration and fusion, which can restrict OSA detection. Thus, further improvements are still required to achieve clinically reliable performance. Methods: To address these challenges, this study proposes SApneaNet, a novel advanced deep learning method for detecting OSA events using ECG signals. The proposed approach employs the continuous wavelet transform (CWT) to convert ECG signals into RGB log-scalograms, enabling the simultaneous analysis of temporal and frequency-domain features. The generated RGB log-scalograms are then fed into a deep CNN encoder with adaptive squeeze-and-excitation (ASE), followed by a transformer and an adaptive gated feature fusion (AGFF) architecture. In this framework, to improve OSA detection performance, the CNN extracts rich local features, the ASE module performs channel-wise recalibration to enhance feature representations, the transformer performs data-parallel processing and captures global contextual dependencies, and the AGFF mechanism adaptively emphasizes informative features while suppressing less relevant ones. Results: The experimental results on the Apnea-ECG dataset showed that the model achieved a sensitivity of 94.7%, specificity of 95.2%, F1-score of 93.5%, accuracy of 95.1%, Cohen’s kappa of 89.4%, and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.989 for per-segment classification. Furthermore, for per-recording classification, the model achieved an accuracy of 100.0%, a mean absolute error (MAE) of 2.025, and a Pearson correlation coefficient (PCC) of 0.992. Overall, the experimental results demonstrated that the proposed model achieved excellent and competitive performance compared with other advanced state-of-the-art methods for OSA classification. Conclusions: The proposed model demonstrates strong efficacy in OSA detection, providing a novel and robust alternative to conventional diagnostic methods. The model’s reliable and consistent diagnostic performance highlights its potential for integration into practical OSA diagnostic systems, including home-based health monitoring devices and clinical decision-support tools. Full article

Background/Objectives: Patients with ulcerative colitis (UC) carry a substantially elevated risk of developing colorectal cancer (CRC); yet, the transcriptional intermediates that bridge chronic mucosal inflammation and colorectal malignancy remain poorly characterized. A related question is whether any genes shared between the two conditions actually help drive transformation, or whether they represent downstream responders. Methods: We integrated two UC microarray cohorts (totalling 227 samples) and the TCGA-COAD dataset (n = 524), combining weighted gene co-expression network analysis (WGCNA) with directional differential expression to define shared candidate genes. Three machine-learning approaches—LASSO, random forest, and SVM-RFE—were applied to that candidate set, and the stability of their intersection was assessed by bootstrap resampling. The resulting panel was then examined by replication in an independent UC cohort, two-sample Mendelian randomization, immune deconvolution, survival modelling, and Visium spatial transcriptomics. Results: WGCNA and directional differential expression together yielded 91 shared candidate genes, and the three algorithms converged on four: CXCL1, S100P, THY1, and TRIM29. Bootstrap resampling showed that this convergence is a property of one particular fit rather than a reproducible selector (all four recovered together in 0.3% of 1000 resamples), so we treated the ensemble as a hypothesis-generating step and made our case for the panel based on independent downstream evidence. This evidence was consistent, as, in an independent UC cohort (GSE92415, n = 183), all four genes moved in the same direction as in discovery, and a four-gene model separated inflamed from control mucosa with a cross-validated AUC of 0.986 (95% CI 0.965–1.000). Two-sample Mendelian randomization using strong blood cis-eQTL instruments (F = 345 for CXCL1, F = 2423 for S100P) returned null estimates; for S100P, the design had 80% power to detect an odds ratio of 1.07 per standard deviation, so this is an informative null rather than an absence of data, while THY1 and TRIM29 had no cis-eQTLs in eQTLGen and remain untestable. Each gene showed a distinct pattern of immune cell correlation in TCGA-COAD. LASSO-Cox retained only CXCL1, whose higher expression was associated with a better rather than worse overall survival (HR 0.79 per SD, p = 0.020); its discrimination was weak (optimism-corrected C-index 0.56 for the score alone) and did not reach significance in an external cohort (GSE39582, n = 573; p = 0.11), so we reported no prognostic model. In Visium spatial data from four colonic sections, a three-compartment structure was reproducible between UC sections—CXCL1 near the neutrophil and interferon signals, THY1 within the stroma and anticorrelated with the epithelium, and S100P and TRIM29 alongside the epithelium—although, with two sections per group, the difference in module scores between the UC and control mucosa was not resolvable. Conclusions: Collectively, these four genes describe a spatially structured epithelial–stromal–immune module of the inflamed mucosa; however, the findings are hypothesis-generating and are not sufficient for prognostic or stratification use. Full article

Estimating multiscale ocean-surface states from sparse observations is challenging because the state is high-dimensional, sampling is irregular, and posterior distributions can be strongly non-Gaussian. We develop Latent-LWETKF, a structured latent-space implementation of the localized weighted ensemble transform Kalman filter. A convolutional autoencoder maps sea surface height (SSH) fields to spatially organized latent tensors, and nonlinear ensemble analysis updates one active block at a time. The active analysis dimension is the number of scalar latent coordinates updated jointly in that block. Each block proposal is reinserted into the complete member-specific latent state and decoded before its likelihood is evaluated through the original physical-space observation operator, reducing the active dimension while retaining large-scale and cross-block context. The method is evaluated in a coupled fast-slow Lorenz-63 system and a GLORYS12V1-based closed-loop SSH reconstruction experiment over the Luzon Strait. Relative to a local particle filter, it more accurately recovers empirical marginals, low-probability states, and innovation-increment relationships. Relative to the learning-based SSH reconstruction benchmark 4DVarNet-SSH, it reduces RMSE by 10.2%, 12.6%, 8.1%, 7.5%, and 3.4% at observation ratios of 1%, 3%, 5%, 10%, and 20%, respectively. In the Luzon Strait experiments, gains increase with distance from observations and decreasing local coverage, supporting tractable nonlinear ensemble analysis with 40 members while retaining physical observation geometry and complete-field decoding context. Full article

Background: U-Net-style encoder–decoder architectures are widely used for 3D medical image segmentation, but their conventional skip connections usually transfer encoder features to the decoder through static concatenation. Such direct skip fusion may inadequately address the semantic gap between low-level encoder features and high-level decoder representations, especially in heterogeneous volumetric medical images. This study aims to improve skip-connection fusion by making encoder–decoder feature transfer adaptive, spatially expressive, and scale-aligned. Methods: We propose Capsule-Expert Routing UNet (CER-UNet), a skip-enhanced encoder–decoder architecture for 3D medical image segmentation. CER-UNet replaces static skip fusion with a Capsule-based Mixture-of-Experts (CapMoE) routing mechanism that dynamically selects and combines capsule-inspired spatial experts for skip-feature refinement. The routed features are further reorganized using ALIGNER-based multi-scale alignment before being injected into the decoder. The model also incorporates 2.5D Inception-style factorized convolutions and parameter-free Statistical Attention CBAM (SCBAM) to support efficient volumetric representation learning. Experiments were conducted on three public benchmarks: Synapse multi-organ CT, BTCV abdominal CT, and ACDC cardiac MRI. Results: CER-UNet achieved average Dice scores of 86.70% on Synapse, 84.94% on BTCV, and 92.52% on ACDC, using approximately 32M parameters in the proposed 2.5D configuration. These results show competitive segmentation performance compared with CNN-based, Transformer-based, and recent hybrid methods while maintaining a compact parameter budget. Conclusions: The findings suggest that adaptive skip-connection enhancement through capsule-expert routing and scale-aligned feature fusion can improve the effectiveness of encoder–decoder feature transfer for volumetric medical image segmentation. Full article

Objective: To compare clinical and laboratory markers for 30-day mortality in older patients with perforated peptic ulcer and assess biochemical markers beyond age and American Society of Anesthesiologists (ASA) class. Methods: Patients aged ≥65 years undergoing emergency surgery for perforated peptic ulcer from January 2015 to February 2026 were retrospectively analyzed. Discrimination was assessed by the area under the receiver operating characteristic curve (AUC), and incremental value by likelihood-ratio tests and Brier scores. Six marker-specific incremental tests underwent Benjamini–Hochberg false discovery rate (FDR) adjustment. Models incorporating blood urea nitrogen-to-albumin ratio (BAR) or log₂-transformed creatinine underwent bootstrap internal validation. Results: Thirty-day mortality was 37.7% (26/69). Age and ASA class yielded AUCs of 0.785 and 0.755. BAR had the highest univariable biochemical discrimination (AUC, 0.777). Inflammation-based indices showed limited discrimination (AUC, 0.424–0.492); Prognostic Nutritional Index (PNI) had the highest discrimination among immune-nutritional/composite markers (AUC, 0.636; p = 0.061). BAR improved the age- and ASA-based model nominally (p = 0.023), but not after FDR adjustment (adjusted p = 0.068). Creatinine provided the strongest incremental information (AUC, 0.905; Brier score, 0.123; adjusted p = 0.008). In the joint model, creatinine improved fit beyond BAR (p = 0.024), whereas BAR did not improve fit beyond creatinine (p = 0.773). Conclusions: Prognostic information was concentrated in age, ASA class, and renal-protein markers. Creatinine provided the most consistent incremental information; BAR showed high univariable discrimination but did not retain significant incremental value after FDR adjustment. Inflammation-based indices remained limited. Independent validation is required. Full article