Reply to Letter: Timing and Sex Bias in Prehospital Prediction of Trauma Transfusion Requirements.

BackgroundComposite outcomes, which include mortality and readmission rates, are often used in risk prediction models following hospital discharge when event rates for the primary outcome of interest, mortality, are low. However, greater readmission rates may result in reduced mortality making interpretation of the composite outcome difficult. We assess the usefulness of a composite outcome of post-discharge readmission and mortality as a target outcome in this context.MethodsThis was a secondary analysis of data collected among mothers and their newborn(s) admitted for delivery at two regional referral hospitals in Uganda. Six-week post-discharge mortality (all-cause) and readmission in newborn infants were analyzed using a competing risk framework. The Sub distribution Hazard Ratios (SHRs) were compared across predictor variables to examine the relationship between the two outcomes.ResultsA total of 6040 newborns with complete six-week follow-up were enrolled, of whom 50.6% were male and 64% of mothers delivered via caesarean section. Thirty-five (0.58%) infants died within the six-week follow-up period and 241 (3.99%) were readmitted. Of the 206 predictors, 81 had a consistent association with both outcomes. These include a higher weight (SHRs: 0.14, 0.68) and length of the baby (SHRs: 0.85, 0.91). However, 125 variables depicted an association in opposing directions which may be linked to social and financial barriers to care-seeking. These include a travel time to the hospital of greater than 1 hour (SHRs: 1.4, 0.28).ConclusionWhile mortality is unequivocally a negative outcome, readmission may be a positive outcome, reflecting health seeking, or a negative outcome, reflecting recurrent illness. This directional dichotomy is reflected to varying degrees within different variables. When using a composite outcome for a prediction model, caution should be exercised to ensure that the model identifies individuals at risk of the intended outcomes of interest, rather than merely the proxies used to represent those outcomes. Identifying predictors with a consistent relationship for both outcomes may yield a more optimized and less biased prediction model for use in clinical care.

Psychological influences on forecast bias: The impact of mood, depression, and trading performance on investor expectations.

An N2O emissions model featuring newly integrated abiotic pathways in nitrification.

Enhancing identification confidence in non-targeted screening of emerging contaminants via an ensemble retention time prediction model: Applications in screening and ecological risk assessment.

Toward fair graph neural networks via dual-teacher knowledge distillation.

MDSFD-Net: Alzheimer's disease diagnosis with missing modality via disentanglement learning and feature distillation.

To reveal how different soaking times affect red sandstone's softening characteristics, this study analyzed red sandstone's mineral composition, meso-structure, mechanical properties, and energy evolution laws. A damage constitutive model was established based on mechanical property testing and microstructure determination experiments of rock samples. It considers the initial compaction nonlinear section. The prediction bias in the energy dissipation theory damage model during the compaction stage was corrected based on the correction coefficient. The deterioration of mechanical properties of rock samples is positively correlated with immersion time. The results showed that water soaking caused feldspar, calcite, and other minerals to dissolve. It also reduced clay minerals and made pore development more intense. The mechanical properties of rock samples gradually decrease. This happened as soaking duration increased. When the soaking time reached 150 days, the cumulative deterioration degrees reached 44.25% and 30.78% respectively. The turning point of dissipated energy moved forward. The growth inflection point of the damage variable also advanced. The rock sample damage model fitted well with the experimental curve. It could accurately characterize the softening process. The research results explained the "time-structure-energy-damage" coupling mechanism. This mechanism applies to red sandstone softening under water-rock interactions. The explanation covered both macro and meso perspectives. It provided key theoretical support for red sandstone engineering stability assessment and long-term service safety.

ABSTRACTThis work investigates how correlations between covariates influence the estimation of their effects in pharmacometric models. The focus is on quantifying the impact on conditional and unconditional covariate effect estimates and assessing the consequences for model interpretation, communication, and dosing recommendations. A theoretical framework was used to describe the mathematical relationship between conditional and unconditional coefficients. This was verified by simulations across a wide range of covariate correlation strengths and relative covariate effect sizes. The practical consequences of misinterpreting conditional effects were evaluated in the context of dose selection and a priori dose individualization. As predicted by theory, covariate correlation had a substantial effect on the conditional covariate coefficient estimates, while unconditional estimates remained stable. Interpreting conditional covariate effects in isolation led to incorrect conclusions about dosing needs and introduced bias and imprecision in individual dose predictions. In contrast, both the complete conditional model and the unconditional model gave accurate predictions when applied appropriately. Unconditional covariate effects offer greater interpretability, making them more suitable for communicating individual covariate impacts in drug labels, publications, and forest plots. We demonstrate that conditional effects are highly sensitive to model context and covariate correlation, making them poor proxies for the unconditional effect, which is often the quantity of interest for dosing and communication. To minimize misinterpretation, unconditional effects should be reported when describing the influence of individual covariates, while the complete conditional model should be used for simulations and exposure predictions. This dual approach can improve clarity and reduce the risk of misunderstanding in model‐informed decision‐making.

To investigate how spatial constraints shape cancer metabolism, we devised the spatial Flux Balance Analysis (spFBA) framework for the enrichment of spatial transcriptomics data with relative estimates of metabolic fluxes. Applying spFBA to newly generated high-resolution datasets of paired primary colorectal tumors (CRC) and liver metastases revealed lactate consumption in both primary and metastatic regions. The presence of lactate-consuming niches was confirmed in an independent public dataset, suggesting this may be a recurrent metabolic feature of CRC. Importantly, application to public datasets of renal cancer showed widespread lactate production, consistent with a dominant but heterogeneous Warburg phenotype, ruling out general prediction biases or algorithmic artifacts. spFBA also consistently identified regions of increased proliferation across datasets, supporting the biological validity of its predictions. The framework is applicable to any sequencing-based spatial dataset to effectively uncover metabolic programs that remain invisible to gene expression analysis alone.

Brain tumors are diseases that require early detection and accurate diagnosis. Various studies have applied deep learning methods to classify MRI images of brain tumors, but they still face dataset limitations and imbalanced class distributions that impact model performance. This study aims to evaluate the performance of the transfer learning-based VGG16 model in classifying brain tumors using MRI images. The study used 7,023 MRI images, including glioma, meningioma, pituitary, and no tumor, with a balanced training data distribution. Pre-processing included resizing, data splitting, and augmentation in the form of rotation, width shift, height shift, and zoom to increase data diversity and reduce the impact of class imbalance. The model was trained using several training-validation data splits (70:30, 80:20, and 90:10) with variations of the Adam, RMSprop, and AdamW optimizers and learning rates between 0.1 and 0.0001. The best configuration was obtained in the 80:20 scenario with the Adam optimizer and a learning rate of 0.0001, which was used in the final testing stage using test data that were never used during training and validation. The results showed the highest validation accuracy of 99.89% and a testing accuracy of 98.00%. Confusion matrix analysis showed that all classes could be classified well without prediction bias.

In populations with limited genotyping, single-step genomic best linear unbiased predictions (ssGBLUP) can produce biased or less accurate genomic predictions due to incompatibilities between genomic and pedigree relationship matrices. The study evaluated the impact of five alternative ssGBLUP models for genomic predictions of milk, fat, and protein yield production traits in South African Holstein cattle. The dataset included 696,413 milk production records and pedigrees of 541,325 animals. Production traits were 305-day lactation yields for milk, protein, and fat. Genotype data were based on the Illumina 50K chip v3, with 53,218 SNPs. A total of 1221 animals with genotypes and 41,407 SNP markers were in the final dataset. The five models used to estimate genomic estimated breeding values (GEBVs) were the single-step method (ssGBLUP), ssGBLUP accounting for inbreeding (ssGBLUP_Fx), ssGBLUP with unknown parent groups (ssGBLUP_upg), and two ssGBLUP models with blending, tuning, and scaling parameters set to optimum values in constructing the inverse of the unified relationship matrix (ssGBLUP_adjusted). Realized prediction accuracies were highest for ssGBLUP_adjusted models (6–7% improvements compared to ssGBLUP). Accuracy of GEBVs for milk, protein, and fat yields ranged from 0.23, 0.29, and 0.30 for both ssGBLUP and ssGBLUP_Fx, 0.26, 0.32, and 0.34 for ssGBLUP_upg, and 0.29, 0.35, and 0.37 for ssGBLUP_adjusted models, respectively. Corresponding bias, expressed as regression coefficients, ranged from 0.30, 0.31, and 0.36 for ssGBLUP; 0.31, 0.32, and 0.37 for ssGBLUP_Fx; 0.41, 0.44, and 0.49 for ssGBLUP_upg; and 0.44, 0.47, and 0.53 for ssGBLUP_adjusted models, respectively. The improved accuracy and reduced bias observed with the ssGBLUP_adjusted underscores the importance of optimizing the blending of pedigree- and genome-based relationships to achieve more reliable GEBVs, thereby improving selection decisions in Holstein dairy cattle.

Neural network-compensated satellite clock bias prediction for Low Earth Orbit (LEO) satellites

Reliable machine learning and statistical analysis rely on diverse, well-distributed training data. However, real-world datasets are often limited in size and exhibit underrepresentation across key subpopulations, leading to biased predictions and reduced performance, particularly in supervised tasks such as classification. To address these challenges, we propose Conditional Data Synthesis Augmentation (CoDSA), a novel framework that leverages generative models, such as diffusion models, to synthesize high-fidelity data for improving model performance across multimodal domains, including tabular, textual, and image data. CoDSA generates synthetic samples that faithfully capture the conditional distributions of the original data, with a focus on under-sampled or high-interest regions. Through transfer learning, CoDSA fine-tunes pre-trained generative models to enhance the realism of synthetic data and increase sample density in sparse areas. This process preserves inter-modal relationships, mitigates data imbalance, improves domain adaptation, and boosts generalization. We also introduce a theoretical framework that quantifies the statistical accuracy improvements enabled by CoDSA as a function of synthetic sample volume and targeted region allocation, providing formal guarantees of its effectiveness. Extensive experiments demonstrate that CoDSA consistently outperforms non-adaptive augmentation strategies and state-of-the-art baselines in both supervised and unsupervised settings. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Predictions about M&A success always suffer from the challenge of unbalanced data, which can easily lead to biased predictions. In addition, previous empirical studies have certain limitations when facing high-dimensional relationships, making it difficult to provide a more global perspective. This study constructs and compares several machine learning models to propose an optimal model. This optimal model is lightGBM, which is constructed from the data after SMOTE oversampling. The results of LightGBM come from the CSMAR database of 3672 M&A transactions, revealing the relative importance and directions of 50 predictors on M&A outcomes, which may have contradictory results or do not appear in previous literature. The findings of this study provide new insights into predicting the success of M&A deals of Chinese listed companies and suggest new directions for future research.

Although numerous population pharmacokinetic (popPK) models of voriconazole have been developed, their predictive performance and extrapolation capacity remain uncertain. This study aimed to systematically evaluate the predictive performance of previously published voriconazole popPK models using an independent external clinical dataset, and to explore potential factors influencing model performance. Additionally, the study sought to identify candidate models with favorable extrapolation capability and clinical utility. The external evaluation dataset was derived from critically ill patients receiving extracorporeal membrane oxygenation (ECMO) support. Published voriconazole popPK models were selected and reconstructed. Predictive performance was assessed using both prediction- and simulation-based diagnostic tools. Furthermore, Bayesian forecasting was applied to evaluate the impact of prior information on the model's ability to predict individual plasma concentrations. A total of 76 voriconazole plasma concentrations from 19 patients were included, and 14 published models were included for evaluation. Diagnostic analyses revealed that none of the models fully satisfied the predefined thresholds for clinical acceptability. Both visual predictive checks (VPC) and normalized prediction distribution errors (NPDE) demonstrated varying degrees of predictive bias across all models, with individual models showing adequate performance only in specific diagnostic aspects. Bayesian forecasting indicated that increasing the number of prior concentrations did not consistently enhance predictive performance. In most models, the use of a single prior concentration was sufficient to markedly improve predictive accuracy and precision, achieving optimal performance. Notably, models incorporating allometric scaling exhibited the greatest improvement in Bayesian predictive performance. Whether the time interval between the prior and predicted concentrations affects predictive performance requires further investigation. Published voriconazole popPK models exhibited considerable variability in predictive performance. Multiple factors, including covariates, patient characteristics, and ethnic variability, may serve as critical determinants of model predictive performance and extrapolation capability. Integrating popPK modeling with maximum a posteriori Bayesian (MAPB) estimation and prior concentration information may serve as an effective approach to achieving model-informed precision dosing (MIPD) for voriconazole. Furthermore, incorporating allometric scaling into models may further enhance the performance of Bayesian prediction.

In response to the critical demand for improved characterization of atmospheric stability effects in wind turbine wake prediction, this study proposes and systematically validates a new analytical wake model that incorporates atmospheric stability effects. In recent years, research on wake models with atmospheric stability effects has primarily followed two approaches: incorporating stability through high-fidelity numerical simulations or modifying classical analytical wake models. While the former offers clear mechanical insights, it incurs high computational costs, whereas the latter improves efficiency yet often suffers from near-wake prediction biases under stable stratification, lacks a unified framework covering the entire wake region, and relies heavily on case-specific calibration of key parameters. To overcome these limitations, this study introduces a stability-dependent turbulence expansion term with a square of a cosine function and the stability sign parameter, enabling the model to dynamically respond to varying atmospheric conditions and overcome the reliance of traditional models on neutral atmospheric assumptions. It achieves physically consistent descriptions of turbulence suppression under stable conditions and convective enhancement under unstable conditions. A newly developed far-field decay function effectively coordinates near-wake and far-wake evolution, maintaining computational efficiency while significantly improving prediction accuracy under complex stability conditions. The Present model has been validated against field measurements from the Scaled Wind Farm Technology (SWiFT) facility and the Alsvik wind farm, demonstrating superior performance in predicting wake velocity distributions on both vertical and horizontal planes. It also exhibits strong adaptability under neutral, stable, and unstable atmospheric conditions. This proposed framework provides a reliable tool for wind turbine layout optimization and power output forecasting under realistic atmospheric stability conditions.

Abstract Organic aerosols (OAs) exhibit non‐ideal behaviors that challenge conventional models assuming ideal equilibrium partitioning. This study integrates a unified kinetic framework into WRF‐Chem model to handle non‐ideal evolution of OAs with considering kinetic mass transfer process with multidirectional interactions (particle surface area, volume, molecular weight) governed by Fick's second law. Simulations in winter of the North China Plain (NCP) reveal that non‐ideal treatment enhances condensation of organics species and water vapor, amplifies interactions between OA, aerosol liquid water content (ALWC), and secondary inorganic aerosols (SIA, pSO 4 2− , pNO 3 − and pNH 4 + ). The revised framework reduces mean bias in OA and SIA predictions from normalized mean bias (NMB) of −18.4% to −2.9%, −33.4% to −23.0%, −2.0% to −0.3%, and −35.4% to −30.2%, respectively, achieves better performance in reproducing ALWC with better correlation (from 0.81 to 0.88), and improves PM 2.5 modeling accuracy (NMB from −18.0% to −9.5%) in “2 + 26” city cluster among the NCP. The framework enhances predictions without modifying chemical mechanisms and suggests a potential reductions in direct radiative forcing estimation (−0.77 W/m 2 among the NCP). The findings advocate urgent integrating non‐ideal behavior of OA into air quality models to advance aerosol prediction.

Secchi disk depth (SDD) is a widely critical indicator of water transparency. However, existing retrieval models often suffer from limited transferability and biased predictions when applied to optically diverse waters. Here, we compiled a dataset of 6218 paired in situ SDD and remote sensing reflectance (Rrs) measurements to evaluate model generalization. We benchmarked nine machine learning (ML) models (RF, KNN, SVM, XGB, LGBM, CAT, RealMLP, BNN-MCD, and MDN) under three validation scenarios with progressively decreasing training-test overlap: Random, Waterbody, and Cross-Optical Water Type (Cross-OWT). Furthermore, SHAP analysis was employed to interpret feature contributions and relate model behaviors to optical properties. Results revealed a distinct scenario-dependent generalization gradient. Random splits yielded minimal bias. In contrast, Waterbody transfer consistently shifted predictions toward underestimation (SSPB: −16.9% to −3.8%). Notably, Cross-OWT extrapolation caused significant error inflation and a bias reversal toward overestimation (SSPB: 10.7% to 88.6%). Among all models, the Mixture Density Network (MDN) demonstrated superior robustness with the lowest overestimation (SSPB = 10.7%) under the Cross-OWT scenario. SHAP interpretation indicated that engineered indices, particularly NSMI, functioned as regime separators, with substantial shifts in feature attribution occurring at NSMI values between 0.4 and 0.6. Accordingly, feature sensitivity analysis showed that removing band ratios and indices improved Cross-OWT robustness for several classical ML models. For instance, KNN exhibited a significant reduction in Median Symmetric Accuracy (MdSA) from 96% to 40% after feature reduction. These findings highlight that model applicability must be evaluated under scenario-specific conditions, and feature engineering strategies require rigorous testing against optical regime shifts to ensure generalization.

Surface deformation induced by the extraction of natural resources constitutes a non-stationary spatiotemporal process. Modeling surface deformation time series obtained through Interferometric Synthetic Aperture Radar (InSAR) technology using deep learning methods is crucial for disaster prevention and mitigation. However, the complexity of model hyperparameter configuration and the lack of interpretability in the resulting predictions constrain its engineering applications. To enhance the reliability of model outputs and their decision-making value for engineering applications, this study presents a workflow that combines a Tree-structured Parzen Estimator (TPE)-based Bayesian optimization approach with ensemble inference. Using the Rhineland coalfield in Germany as a case study, we systematically evaluated six deep learning architectures in conjunction with various spatiotemporal coding strategies. Pairwise comparisons were conducted using a Welch t-test to evaluate the performance differences across each architecture under two parameter-tuning approaches. The Benjamini–Hochberg method was applied to control the false discovery rate (FDR) at 0.05 for multiple comparisons. The results indicate that TPE-optimized models demonstrate significantly improved performance compared to their manually tuned counterparts, with the ResNet+Transformer architecture yielding the most favorable outcomes. A comprehensive analysis of the spatial residuals further revealed that TPE optimization not only enhances average accuracy, but also mitigates the model’s prediction bias in fault zones and mineralize areas by improving the spatial distribution structure of errors. Based on this optimal architecture, we combined the ten highest-performing models from the optimization stage to generate a quantile-based susceptibility map, using the ensemble median as the central predictor. Uncertainty was quantified from three complementary perspectives: ensemble spread, class ambiguity, and classification confidence. Our analysis revealed spatial collinearity between physical uncertainty and absolute residuals. This suggests that uncertainty is more closely related to the physical complexity of geological discontinuities and human-disturbed zones, rather than statistical noise. In the analysis of super-threshold probability, the threshold sensitivity exhibited by the mining area reflects the widespread yet moderate impact of mining activities. By contrast, the fault zone continues to exhibit distinct high-probability zones, even under extreme thresholds. It suggests that fault-controlled deformation is more physically intense and poses a greater risk of disaster than mining activities. Finally, we propose an engineering decision strategy that combines uncertainty and residual spatial patterns. This approach transforms statistical diagnostics into actionable, tiered control measures, thereby increasing the practical value of susceptibility mapping in the planning of natural resource extraction.