Background: Electronic surveillance for hospital-onset sepsis using CDC’s Adult Sepsis Event definition could provide an efficient and objective method to identify a broad array of serious healthcare-associated infections, many of which are missed through current reporting processes. We developed risk adjustment models of varying complexity to support facility-level comparison of hospital-onset sepsis rates, evaluated trade-offs between model performance and feasibility, and quantified residual inter-facility variation that may reflect gaps in care. Methods: We conducted a retrospective study of adults hospitalized for <3 days within 113 community hospitals between 2022-2023. Hospital-onset Adult Sepsis Events (HO-ASEs) occurring on day 4 or later were identified using updated CDC surveillance criteria. We used logistic regression to develop three risk adjustment models of increasing complexity using covariates from administrative and electronic health record data: basic model (hospital and aggregate patient descriptors), intermediate model (replacing aggregate patient descriptors with patient-level descriptors), and maximal model (adding detailed physiologic and clinical data from hospital days 1-3; Figure 1). We evaluated model performance using Area Under Receiver Operating Curve (AUROC), calculated hospital-level Standardized Infection Ratios (SIRs) for each model and assessed concordance in hospital rankings using Kendall’s tau coefficient (τ). Results: The cohort included 1,557,252 hospitalizations of <3 days, of which 24,169 (1.6%) met HO-ASE criteria and 8,500 (35.2%) died in-hospital. The basic model had limited discrimination (AUROC 0.589, 95% CI, 0.585-0.593). Adding patient-level characteristics to form the intermediate model markedly improved performance (AUROC 0.839, 95% CI 0.836-0.841), with further inclusion of detailed clinical data in the maximal model yielding modest additional improvement (AUROC 0.850, 95% CI 0.848-0.853). Concordance between hospital rankings derived from the crude or basic risk-adjusted HO-ASE rates versus rankings derived from the intermediate or maximal models was moderate (τ 0.40-0.51) whereas concordance between rankings derived from the intermediate vs. maximal models was high (τ 0.86). There was a wide distribution of HO-ASE SIRs across facilities even after risk adjustment using the maximal model (Figure 2), with high signal-to-noise ratios and good calibration. Conclusions: Risk adjustment models incorporating hospital characteristics and patient-level data perform well and might explain substantial variability in HO-ASE rates between facilities. The persistence of residual variability after highly detailed adjustment may reflect differences in care processes, suggesting that risk-adjusted HO-ASE comparisons can help identify gaps and opportunities in the prevention of severe healthcare-associated infections. Our findings support the use of HO-ASE as an electronic, scalable, risk-adjusted metric for facility-level benchmarking to inform quality improvement initiatives.

We investigate the onset of transient natural convection in fluid layers subject to volumetric radiative heating and surface cooling. Linear stability analysis reveals a non-monotonic evolution of stability in deep layers, where the flow undergoes successive stages of initial destabilisation, intermediate suppression and eventual restabilisation. This complex temporal behaviour necessitates the definition of dual critical Rayleigh numbers: a lower bound marking the onset of initial instability and an upper bound required for sustained convection. To efficiently predict these thresholds, we develop a local Rayleigh number model that depends solely on the instantaneous conductive temperature profile. When the Rayleigh number exceeds the lower bound, the critical time $t_c$ for flow onset is determined through transient linear stability analysis, and scaling laws are derived to characterise the dependence of $t_c$ on the Rayleigh number $\textit{Ra}$, Prandtl number $\textit{Pr}$, cooling parameter $\phi$ and layer depth $H$. Two distinct instability triggering mechanisms are identified: a top-triggered regime, where $t_c \sim [\textit{Ra} \textit{Pr} \phi /(2+\textit{Pr})]^{-1/2}$, and a bottom-triggered regime, where $t_c \sim [\textit{Ra} \textit{Pr} \exp (-H)/(2+\textit{Pr})]^{-1/2}$. All theoretical predictions are rigorously validated against direct numerical simulations, providing a unified predictive framework for convective onset in systems governed by coupled effects of radiation absorption and surface cooling.

Recruitment and retention challenges continue to hinder the success of clinical trials. Artificial intelligence (AI) has emerged as a promising means to optimize various clinical trial processes; however, its impact specifically on recruitment and retention has not been comprehensively evaluated. This scoping review utilized the Joanna Briggs Institute framework and adhered to PRISMA-ScR guidelines, systematically searching literature published between January 2018 and June 2024 across multiple databases. Of the 21,573 records screened, 121 studies were included. A meta-analysis was conducted to quantitatively assess the performance of AI-driven tools. AI applications for patient screening demonstrated strong performance, achieving a pooled sensitivity of 0.91 (95% CI: 0.84–0.95) and an area under the curve (AUC) of 0.79 (95% CI: 0.72–0.85). AI tools employed for eligibility identification and classification also exhibited strong outcomes, with pooled sensitivities of 0.80 (95% CI: 0.76–0.84) and 0.92 (95% CI: 0.84–0.96), respectively, and precisions of 0.84 (95% CI: 0.80–0.88) and 0.91 (95% CI: 0.85–0.95). AI tools aimed at identifying patient cohorts showed moderate effectiveness (pooled sensitivity: 0.70 [95% CI: 0.52–0.84]; AUC: 0.74 [95% CI: 0.61–0.84]). Overall, AI presents significant potential for enhancing clinical trial recruitment and retention, with effectiveness varying across specific applications. These findings underscore AI’s valuable role in improving trial efficiency and data quality.

Discrepancies in iodised salt coverage rate (ISCR) between household salt and that used in catering establishments may significantly compromise the accuracy of dietary iodine intake assessments. To evaluate this impact, we analysed data from the 2023 Shanghai Diet and Health Survey, a cross-sectional study involving 2920 adults. Dietary intake was assessed using three 24-h dietary recalls and an FFQ, while condiment intake was collected using the weighed inventory method. Additionally, salt samples from 960 canteens and restaurants were tested to determine the ISCR in dining establishments. Results showed that the ISCR was 85·9 % in dining establishments, markedly higher than the 53·3 % observed in households. Among employed participants in Shanghai, 51·7 %, 56·1 % and 18·7 % reported consuming breakfast, lunch and dinner outside the home at least once during the 3-d study period, respectively. The estimated daily iodine intake was 101 μg/d when dining-out salt was assumed to have the same ISCR as household salt, but it increased to 118 μg/d after accounting for the ISCR discrepancy. In conclusion, the rising prevalence of eating out has reshaped residents’ dietary habits, rendering traditional household-centric survey methods inadequate for iodine intake estimation in Shanghai. Incorporating ISCR differences between household and dining settings is essential for more accurate dietary iodine assessments.

Clay minerals in the Yishu fault zone, as typical products of hydrothermal activity, carry critical information regarding subsurface geothermal processes. To clarify their hyperspectral remote sensing characteristics and their guiding value for geothermal resource exploration, this study takes the Yishu fault zone and three key subareas (Tangtou, Tongjing and Fangzi) as the research objects. Based on ZY1-02D hyperspectral remote sensing data, combined with principal component analysis and the Crósta method, we analysed the spectral features of typical clay minerals (illite, montmorillonite and chlorite) and extracted their spatial distribution patterns. In addition, ZY1-02E thermal infrared data were used to invert the land surface temperature so as to explore the spatial correlation between clay mineral distribution and geothermal anomalies. The results show that the spatial distribution of clay minerals is closely associated with fault structures, being mainly concentrated along the four major fault belts of the Yishu fault zone and their surrounding areas. High-intensity hydrothermal alteration zones dominated by clay minerals exhibit significant spatial overlap with surface temperature anomalies. The consistent distribution of clay minerals and known geothermal fields further confirms that the hyperspectral characteristics of clay minerals can effectively indicate the upwelling channels and accumulation zones of geothermal fluids. This study provides a rapid and accurate technical method for geothermal resource prospecting in the Yishu fault zone and other tectonically similar regions.

Understanding fluid-elastic instabilities in slender bodies is crucial for predicting and controlling flow-induced vibration (FIV) in engineering and biological systems. The FIV of a prolate spheroid with an aspect ratio of $\epsilon = 3$, a mass ratio of $m^* = 3$ and a damping ratio of $\zeta = 0$, elastically mounted in a uniform flow at ${\textit{Re}} = 600$, are investigated using direct numerical simulations and a reduced-order model (ROM). As reduced velocity $U_r$ increases, five vibration states emerge: quasi-steady (QS), periodic (PM), large-amplitude chaotic (LAC), quasi-periodic (QP) and small-amplitude chaotic (SAC) modes. These mode regimes form two categories of response branches, namely synchronised branches (SB) and desynchronised branches (DB). In SB, three synchronisation mechanisms are identified, i.e. conventional lock-in (CLI), secondary-component lock-in (SCLI) and superharmonic lock-in (SHLI), corresponding to PM, LAC and QP modes, respectively. In contrast, DB comprises two types. The flow-dominated desynchronised (FDD) branch corresponds to QS mode, where flow instability dominates while structural vibrations remain weak. The dual-mode competition desynchronised (DMCD) branch corresponds to the SAC mode, where fluid and structural instabilities coexist but fail to synchronise. Analysis of wake dynamics identifies spanwise, transverse and high-frequency spanwise shedding patterns that are closely correlated with vibration regimes. The overlap of the response branches produces three distinct hysteresis zones, emphasising the sensitivity of spheroidal FIV to initial conditions and its inherently path-dependent behaviour. Dynamic mode decomposition (DMD) and an ERA-based ROM, which together resolve mode-specific spatial structures and frequency evolution, show that the vibration dynamics is governed by a persistently unstable wake mode (WM) and a structural mode (SM) whose stability alternates across branches. This clarifies the resulting sequence of vibration modes and provides insight into how different branches compete and transition.

Histone lysine-specific demethylase 4A (KDM4A) plays a critical role in the embryonic development of mammals such as mouse and goat, however its function in zebrafish (Danio rerio) embryogenesis remains poorly understood, due to the existence of two kdm4a paralogs (kdm4aa and kdm4ab) in zebrafish. The current study revealed that kdm4aa–/– embryos exhibited dramatically increased mortality during gastrulation. RT-qPCR showed that RNA-binding motif protein 46 (rbm46) was downregulated after kdm4aa knockout. CUT&Tag-qPCR revealed that kdm4aa knockout significantly decreased H3K4me3, while increasing H3K9me3 and H3K36me3 at rbm46 promoter. kdm4aa–/– embryos displayed elevated reactive oxygen species (ROS) and reduced adenosine triphosphate (ATP). And rbm46 mRNA injection alleviated ROS accumulation and increased ATP level, thereby rescuing the lethal phenotype in kdm4aa–/– embryos. Our findings demonstrate that kdm4aa knockout disturbs zebrafish embryogenesis by suppressing rbm46 mRNA expression.

Fine-grained mortality forecasting has gained momentum in actuarial research due to its ability to capture localized, short-term fluctuations in death rates. This paper introduces MortFCNet, a deep-learning method that predicts weekly death rates using region-specific weather inputs. Unlike traditional Serfling-based methods and gradient-boosting models that rely on predefined fixed Fourier terms and manual feature engineering, MortFCNet automatically learns patterns from raw time-series data without needing explicitly defined Fourier terms or manual feature engineering. Extensive experiments across over 200 NUTS-3 regions in France, Italy, and Switzerland demonstrate that MortFCNet consistently outperforms both a standard Serfling-type baseline and XGBoost in terms of predictive accuracy. Our ablation studies further confirm its ability to uncover complex relationships in the data without feature engineering. Moreover, this work underscores a new perspective on exploring deep learning for advancing fine-grained mortality forecasting.

Accurate mortality forecasting is crucial for actuarial pricing, reserving, and capital planning, yet the traditional Lee-Carter model struggles with non-linear age and cohort patterns, coherent multi-population forecasting, and quantifying prediction uncertainties. Recent advances in deep learning provide a range of tools that can address these limitations, but actuarial surveys have not kept pace. This paper provides the first concise view of deep learning in mortality forecasting. We cover six deep network architectures, namely Recurrent Neural Networks, Convolutional Neural Networks, Transformers, Autoencoders, Locally Connected Networks, and Multi-Task Feed-Forward Networks. We discuss how these architectures tackle cohort effects, population coherence, interpretability, and uncertainty in mortality forecasting. Evidence from the literature shows that carefully calibrated deep learning models can consistently outperform the Lee-Carter baselines; however, no single architecture resolves every challenge, and open issues remain with data scarcity, interpretability, uncertainty quantification, and keeping pace with the advances of deep learning. This review is also intended to provide actuaries with a practical roadmap for adopting deep learning models in mortality forecasting.

To fully understand resilience and to inform resilience-promoting interventions, it is important to explore how resilience develops and the factors that influence it. Using a multidimensional approach that considers both well-being resilience (higher than expected wellbeing after adversity) and depression resilience (lower than expected depression after adversity), this study examined resilience trajectories among Chinese 0adolescents and the associations of gratitude and perceived stress with resilience trajectories. Data from a four-wave longitudinal study were analyzed from 563 Chinese adolescents (mean age at Time 1 = 12.83 years, 51.87% boys). Parallel-process latent class growth modeling identified four distinct trajectories of resilience development: flourishing resilience (increasing resilience; 21.67%), increasing wellbeing resilience but decreasing depression resilience (28.24%), declining resilience (29.48%), and increasing depression resilience but decreasing wellbeing resilience (20.61%). Gratitude was associated with greater odds of being in the flourishing resilience group. Furthermore, perceived stress was associated with lower odds of being in the flourishing resilience group and higher odds of being in the declining resilience group. The findings suggest that resilience is a dynamic and multidimensional construct with highly heterogeneous developmental trajectories. Gratitude and perceived stress may be effective targets for interventions to enhance adolescent resilience.

This paper introduces a high single-pulse energy, narrow-linewidth mid-infrared self-optical parametric oscillator (mid-IR SOPO) with a cavity length of 120 mm and a Nd:MgO:PPLN crystal. To achieve high single-pulse energy and high peak power in mid-IR light sources, a LiNbO3 electro-optic Q-switch (EOQ) is introduced for the first time in a mid-IR SOPO. A narrow-linewidth EOQ-SOPO rate equation is formulated, and experiments are conducted using a single Fabry–Pérot etalon. At a 500 μs pump pulse width, a 4.71 mJ single-pulse idler light at 3838.2 nm is achieved, with a linewidth of 0.412 nm, single-pulse width of 4.78 ns and peak power of 985 kW. At 200 μs, the idler light at 3845.2 nm exhibits a minimum linewidth of 0.212 nm.