Physiology-informed machine learning for patient-level surgical decision-making in severe traumatic brain injury: Multicenter model development with external validation.

Galectin-9 in cancer: Unraveling its diverse roles beyond immune regulation.

Hybrid fermatean Z-AHP-TOPSIS model for efficient health insurance policy decision making

The ninth TNM edition distinguishes single-station (N2a) from multistation (N2b) disease, increasing concern for cross-contamination between N2 lymph nodes during EBUS staging when ROSE is unavailable. The absolute probability and clinical impact of this phenomenon are uncertain. We estimated a theoretical upper bound on false upstaging and the number of needle changes required to prevent one false upstage. A Monte Carlo decision model simulated mediastinal staging in 100,000 virtual patients across published estimates of occult N2 prevalence, probability of a single-station involvement, and contamination rates. Outcomes were the probability of false N2b upstaging and the number of additional needle changes needed to prevent one false upstage (NNC). Sensitivity analyses varied the contamination rate, single-station probability, and sampling order. False upstaging in central cN0 disease was ∼0.5% to 0.8% in the base case, increasing to 1.4% in higher-risk scenarios. cN1 disease showed slightly higher probabilities. Needle-change efficiency was low; preventing one false upstage required roughly 250 to 350 extra needle changes in central cN0 and >1000 in peripheral cN0. Sensitivity analysis showed comparable influence of contamination rate, single-station probability, and sampling order. The risk of spurious upstaging during EBUS staging without ROSE is small but not negligible, concentrated in central cN0 and cN1 tumors. When ROSE is not available, routine needle change between N2 stations provides minimal benefit for most patients. In cases where ROSE is not available and resources are limited, a selective strategy targeting high-risk contexts may offer the best balance between accuracy and procedural efficiency.

In this study, we address the inventory decision problem of ameliorating goods by explicitly incorporating a demand spillover effect between product categories – an interaction that has received little attention in operations management. We first empirically demonstrate the existence of this spillover using multi-year sales data from 11 Port wine brands across 86 markets. Building on these insights, we integrate the spillover effect into a stochastic inventory decision model for a (Port) wine seller who must decide whether to sell existing inventory or continue aging it to offer higher-quality products in the future. The problem is formulated as a Markov Decision Process and solved using a forecast-based Deterministic Lookahead (DLA) approach and a Proximal Policy Optimization (PPO) algorithm. Our results show that accounting for the spillover effect can increase profits by up to 1.31%, and that both proposed solution methods outperform the myopic strategy currently applied by producers. While the DLA policy performs best under high forecast accuracy, the PPO algorithm proves more robust when uncertainty is high. The study contributes to bridging marketing and operations perspectives by quantifying the economic impact of spillover effects and providing decision-support tools for managing aged inventory under demand uncertainty.

Psychological safety and teacher voice behavior: Power distance orientation in Chinese schools.

An online learning-driven risk assessment and response decision model for digital twin projects in water conservancy

Understanding the reasons behind human choices under risk is a central goal of decision scientists, but traditional methods relying on behavioral data are limited by strict invariance assumptions. We introduce a scalable analytical framework using large language models (LLMs) to analyze verbal reports and identify articulated reasons for choice between monetary lotteries. A validated LLM accurately identified predefined decision reasons in participants' free-text reports, aligning with their actual choices in 95% of trials. Our analysis reveals that the reasons behind people's decisions vary systematically and are driven more by the structure of the choice problem than by individual differences. Crucially, reasons identified from verbal reports yield more parsimonious and informative representations of decision processes compared to those inferred from choices alone; furthermore, problem-specific reason profiles achieve out-of-sample prediction accuracy that is competitive with established computational models. This work demonstrates that verbal reports are a rich data source and our analytical framework can unlock their potential, delivering results that challenge the field's foundational invariance assumptions and pave the way for more context-sensitive and interpretable models of human decision making.

By using three or more ordered response categories and varying the stimulus feature being judged over a large range, it is possible to generate a family of psychometric functions (PMFs), each based on a different partition of the responses. An earlier paper showed how, when it is treated as a probability distribution, the traditional single PMF based on binary-choice data can be decomposed into sensory and decision components, expressed as two independent random variables that are summed to create the PMF. Here we extend this development to the multiple-response procedure, and use it to elucidate the relations among the spreads and shapes of the resulting family of PMFs, which can be described by their first four cumulants. For example, we determine conditions under which the PMFs can have the same spread and shape, differing only by translation on the stimulus axis. Whereas PMFs depend on both sensory and decision processes, differences among the PMFs in a family depend only on the decision processes. Application of this multiple-PMF method to several decision models, whose evaluations depend on the PMF cumulants, shows it to have greater power than the single-PMF method for understanding the perceptual process. Although this work was inspired by experiments on the perception of temporal order, it can be applied to experiments where features of stimuli other than their occurrence times are being compared, such as the pitch of tones or the brightness of lights.

Abstract Diagnostic MRI evaluation of temporal lobe epilepsy (TLE) depends on the subjective visual interpretation of MRI images. These interpretations could be enhanced by quantitative artificial intelligence (AI) support tools. Humans often make sequential and conditional decisions during their radiological interpretations, such as whether an abnormality is present and if present, characterizing the abnormality. It is not known whether it is superior to train AI to treat every decision separately in a similar step-wise manner or to train a model holistically on all decisions simultaneously. Here, we analyzed three large epilepsy MRI datasets [n=3,676, 2,320 people with epilepsy and 1,356 healthy controls (HC)] to perform two tasks: 1) establish the presence of a TLE pattern on MRI and 2) determine TLE pattern lateralization. We compared Step-wise models that independently classify TLE versus HC and lateralize patients as left TLE (L-TLE) or right TLE (R-TLE), against a Simultaneous model trained to distinguish all three classes in a single step. To do this, 3D volumetric T1-weighted images were input into an EfficientNetV2 model multiple times to ensure reproducibility of results. Class prediction, model classification confidence, and saliency maps were output for interpretability. Step-wise models outperformed the Simultaneous model on both tasks (both ps<.001), with an average ∼2.8% accuracy increase for discriminating HC from TLE and an average 12.7% accuracy increase for distinguishing L-TLE from R-TLE. For both the Step-wise and Simultaneous models, important features discriminating TLE from HC included the known TLE limbic pattern involving the hippocampus, para-hippocampal cortical regions, cingulate cortex, and lateral temporal regions. However, there was less concordance between the Step-wise and Simultaneous models for the L-TLE versus R-TLE task (all Fisher’s Zs>10.5, ps<.001); the Step-wise model focused less on subcortical regions such as the thalamus and hippocampus and focused more on distributed cortical pathology. Across the two Step-wise models, 95.1% of TLE patients had accurate classifications in either HC versus TLE and/or L-TLE versus R-TLE tasks. These results included 69.6% of patients being both correctly labeled as TLE and lateralized, 13.9% being correctly labeled TLE but lateralized incorrectly, and 11.6% being lateralized correctly but not detected as TLE. These findings provide evidence that diagnostic tasks with simpler, Step-wise AI models may enhance diagnostic performance and interpretability in clinical workflows. Future AI clinical support tools can leverage this step-wise approach in the early identification of TLE-related structural patterns, supporting timely diagnosis and treatment decisions.

ABSTRACT Against the backdrop of increasing uncertainty and sustainability pressures, sustainable supply chain management (SSCM) has become critical for balancing economic, environmental, and social performance. Failure mode and effect analysis (FMEA) is widely used in SSCM, yet traditional FMEA is limited in handling uncertain linguistic information, deriving reasonable risk factor weights, and capturing intrinsic correlations among failure modes. To address these gaps, this study proposes a novel three‐stage FMEA‐based decision model for SSCM risk assessment. In the first stage, trapezoidal interval type‐2 fuzzy sets (TrIT2FSs) are employed to represent uncertain expert assessments. In the second stage, an integrated deck of cards with decision‐making trial and evaluation laboratory (DOC‐DEMATEL) method is developed to determine risk factor weights by considering dual interactions among risk factors and experts. In the third stage, a TrIT2FS‐based grey relational analysis (Tr‐GRA) method is constructed to rank failure modes while capturing their intrinsic relationships. Finally, an SSCM case is analyzed, followed by sensitivity and comparative analyses to validate the model. Results show that operation complexity, opportunity loss, and lack of trust are the highest‐priority failure modes. The proposed model outperforms traditional FMEA and multi‐criteria decision‐making methods in robustness and rationality. These findings provide clear managerial insights to help enterprises strengthen risk detection, improve supply chain collaboration, and optimize operational processes toward sustainable development.

Ensuring defensible policy inferences and robust decision reliability is fundamental for multi-criteria decision-making activities, particularly in transport safety engineering. This study developed a hybrid preference function-nested and machine learning-embedded decision model, i.e., EXPROM II-K-means with a linear discriminant analysis, to provide a robust support system for policy setting and decision making. The proposed model incorporates a refined nonparametric preference function into the EXPROM II method to alleviate cognitive burden on decision makers while improving the flexibility of preference articulation. Meanwhile, linear discriminant analysis, a supervised machine learning-based dimensionality reduction algorithm, is embedded to reduce data dimensionality and project features into axes that maximize class separability, which simplifies the data structure, filters noise, and emphasizes informative attributes. This transformation improves the clustering performance of K-means clustering, which yields clearer and more actionable patterns in high-dimensional or noisy datasets. A case study on transport safety engineering across the Asia-Pacific Economic Cooperation countries demonstrates the reliability, scalability, and effectiveness of the model in guiding resource allocation and strategic prioritization. The proposed framework offers a practical, interpretable, and intelligent support tool to manage complex decision tasks with enhanced stability and reliability.

Reliable prediction of strata-pressure evolution is essential for intelligent longwall mining, but steeply inclined panels show spatially heterogeneous support loading that challenges short-term warning. Here we analyse two months of hydraulic-support data from Panel II1013 in the Huaibei mining area and develop a local prediction workflow for support-pressure states. Missing and zero values were repaired, random measurement noise was reduced using a one-dimensional Kalman filter, and normalized sliding-window samples were used to compare CNN, LSTM, CNN-LSTM, Transformer and CNN-LSTM-Attention models against persistence and BP neural network baselines. Data analysis revealed a persistent high-pressure concentration from the middle to upper face. Under matched data partitioning, CNN-LSTM-Attention achieved the lowest test RMSE and highest R2 (RMSE 0.8632 ± 0.0615; MAE 0.4233 ± 0.0902; MAPE 2.5194 ± 0.4429%; R2 0.9891 ± 0.0016), reducing RMSE by 1.80% relative to BP and 9.60% relative to persistence. In a held-out 1000-sample window, the model achieved a relative regression accuracy of 98.17% (1-MAPE). A bounded multi-step validation on four representative supports, using a 24-point input window (approximately 2h), yielded valid forecast horizons of 5-10h when both horizon-level and farthest-step MAPE were ≤ 10%. These results support CNN-LSTM-Attention as a local single-support prediction module for graded warning assistance. Broader deployment will require multi-support spatiotemporal modelling and field validation of closed-loop support-control decisions.

An Integrated CLSC Optimization Model for Inventory and Green Investment Decisions under Stochastic Demand and Collected Product Quality

Structured expert elicitation (SEE) has become increasingly important in health technology assessment and economic evaluations. Complementing previous work, we aimed to synthesize recent developments in published SEE applications within health economics over the past 8 years. A systematic literature search was conducted in Medline and Embase databases from April 2017 to February 2026, supplemented with snowball sampling, to identify applications of SEE as part of economic evaluations. Data extraction and synthesis focused on expert selection, elicitation methods, and analytical techniques to identify commonalities and gaps. In total, 28 studies met the inclusion criteria. SEE applications covered diverse health interventions, from rare diseases treatments to diagnostic accuracy assessments. The number of experts recruited through purposive sampling varied from 1 to 18 clinicians per study. SEE processes remain bespoke and diverse, spanning from paper-based to software-assisted remote techniques. The studies used mainly variable and fixed interval methods (29% versus 67%) for encoding. Aggregation methods were mainly mathematical, with some studies using consensus approaches. Most studies (75%) directly incorporated pooled expert distributions into decision models. While SEE methods vary considerably across applications, suggesting that optimal approaches have yet to emerge, there is growing recognition of their potential for informing healthcare decision-making where empirical data are scarce, particularly in rare diseases and early-stage technology assessment. Future research should prioritize standardizing best practices, validating expert predictions against subsequently available empirical data, and developing enhanced bias mitigation strategies to improve the credibility of expert-informed health economic evaluations.

ABSTRACT Complex product development networks (CPDNs) are typically characterised by co-opetition relationships and restricted communication structures between manufacturers and suppliers. Traditional investment decision models often fail to capture these two inherent features simultaneously. To bridge this gap, this study proposes a novel biform game model that integrates the Position value from the graph cooperative game framework. In the non-cooperative stage, firms determine their investment levels. In the cooperative stage, collaborative benefits are allocated based on the Position value, which effectively reflects firms’ marginal contributions and their brokerage roles within the network topology. Furthermore, we explore the impact of key parameters on investment decisions and extend the model to networks with n suppliers to verify the robustness of our conclusions. The finding reveals that the optimal investments derived from the biform game model are higher than those from a pure non-cooperative model, underscoring the role of cooperation in incentivizing investment. This research presents a novel theoretical framework for analyzing strategic investment under co-opetition and communication constraints, offering practical insights to enhance investment efficiency and collaboration in CPDNs.

ABSTRACT Rural transport systems typically separate passengers, freight, and parcels, leading to low vehicle utilization and high costs under sparse demand. Existing studies overlook how parcel handling at village stops affects passenger travel time and service coordination in rural demand-responsive transport (DRT). This study models the trade-off between operational cost and passenger inconvenience in rural passenger-freight integration, focusing on the interaction between efficiency and service performance. We formulate an operational decision model for coordinated passenger-parcel activities and apply a heuristic solution to assess service configurations. A case study in Yiyuan County, China, shows that collaborative service design reduces fleet requirements and total operating costs, while the increase in passenger travel time remains moderate. The results also identify conditions under which collaborative services offer greater performance advantages in rural transport systems.

Polypharmacy and potentially inappropriate medications are highly prevalent in primary care and are associated with adverse drug events, reduced adherence, and diminished quality of life. Deprescribing is a key strategy to address these challenges, but its implementation is complex, particularly when long-term preventive medications are involved and decisions are preference-sensitive. Digital tools may support shared decision making in deprescribing, yet many existing tools lack clear implementation orientation. This study describes the iterative refinement of arribaMediQuit, a digital deprescribing tool for primary care, with the aim of improving usability, ethical robustness, and implementation potential. The refinement followed an iterative, participatory development process informed by the Medical Research Council framework for complex interventions and the International Patient Decision Aid Standards. Multiple stakeholder groups were involved, including general practitioners, researchers in health services and pharmacology, members of a patient advisory board, experts in medical ethics, and the original developers of the tool. Regular expert meetings were used to review content, terminology, visual design, decision logic, and deprescribing strategies. Feedback was continuously integrated into successive versions of the tool. The study focused on qualitative refinement rather than outcome evaluation. Key refinements included the development of a dynamic medication database and a structured categorization of medications into three categories (symptomatic, intermediate, and preventive medications) with tailored decision processes. A new linear decision model was introduced for preventive medications to better reflect value-sensitive trade-offs under uncertainty. Terminology and visual elements were revised to align with everyday clinical language and to enhance patient comprehensibility. Ethical considerations, including the communication of benefits, harms, and withdrawal symptoms, were explicitly addressed. Stakeholders also identified potential future uses of the tool, such as educational applications and integration with other deprescribing or medication review tools. The iterative, theory-informed refinement of arribaMediQuit illustrates how shared decision making principles can be operationalized in a deprescribing tool designed for routine primary care. By integrating technical guidance with value-sensitive deliberation and implementation considerations, the tool shows promise for supporting ethically grounded and feasible deprescribing. Future studies will evaluate feasibility, acceptability, and use in routine practice.

ABSTRACT OBJECTIVES Prehospital routing decisions for patients with suspected stroke are complex, and consideration of actual hospital performance data for door-to-needle and door-to-puncture times for thrombolysis and endovascular thrombectomy (EVT) may impact on-scene decision making. We sought to determine whether optimal routing destinations for patients with suspected stroke in the prehospital setting differ between three strategies: 1) American Heart Association (AHA) consensus recommendations, 2) personalized decision modeling without hospital-specific performance data (base model) and 3) with inclusion of actual hospital data for door-to-needle and door-to-puncture times for thrombolysis and EVT (enhanced model). METHODS Our previously published decision-analytic model (base model) incorporates geographic, patient, and hospital data to determine optimal transport destination, using a standard distribution of thrombolysis and EVT times by hospital type (primary vs comprehensive stroke center). We enhanced the model by incorporating real hospital performance data on thrombolysis and EVT treatment times from all Northeast United States Get with the Guidelines-Stroke hospitals (enhanced model). We generated 500 probable pick-up locations with 800 patient profiles of varying age, stroke severity, and last known well time, giving 400,000 patient-location scenarios. For each scenario we determined the optimal destination based on AHA consensus recommendations and from 1000 simulations each using base and enhanced models to determine the optimal destination for each strategy. We compared destination recommendations and examined conditions under which results changed. RESULTS Of the 400,000 patient-location scenarios, 56.1% and 63.1% of routing destinations were different from the consensus recommendations in the base and enhanced models, respectively, as patients were more often directed to hospitals with faster times to reperfusion. Higher stroke severity and greater odds of large vessel occlusion were associated with concordance of recommendations. CONCLUSIONS Prehospital routing models for patients with suspected stroke may benefit from inclusion of actual hospital performance characteristics.

Artificial intelligence has increasingly been applied to orthodontic diagnosis and treatment planning. Clinical decision-making in orthodontics is complex and often varies among practitioners, particularly for decisions such as tooth extraction, skeletal classification, and selection of treatment modality. This systematic review aimed to evaluate how artificial intelligence and knowledge-based systems have been developed and assessed for orthodontic treatment planning, to compare their performance with expert clinicians, and to examine how different learning approaches influence accuracy, interpretability, and potential clinical integration. A comprehensive search of PubMed, Embase, Scopus, Web of Science, Cochrane Library, and IEEE Xplore was conducted from database inception to 15 October 2025 without language or date restrictions. Eligible studies evaluated artificial intelligence-based or knowledge-based systems using real orthodontic patient data to support diagnosis or treatment planning. Two reviewers independently screened records, extracted data, and assessed risk of bias using the ROBINS-I tool. Due to heterogeneity in study designs, decision tasks, and outcome measures, findings were synthesized narratively rather than through meta-analysis. Nineteen studies met inclusion criteria. Most investigated supervised machine-learning models for extraction decisions, skeletal classification, or multi-step treatment planning, while others evaluated rule-based, fuzzy, Bayesian, or hybrid systems. Reported accuracies frequently exceeded 80%, and in some studies surpassed 90% when compared with expert decisions. Knowledge-based systems, including rule-based, fuzzy, and Bayesian/probabilistic approaches, offered greater transparency in reasoning, whereas data-driven models often demonstrated higher discriminative performance. Overall risk of bias was predominantly moderate to serious, largely due to retrospective designs, single-center samples, limited external validation, and reliance on expert opinion as the reference standard. Artificial intelligence systems can approximate expert orthodontic decision-making for defined diagnostic and planning tasks and show promise as clinical decision-support tools. However, current evidence is limited by methodological weaknesses and lack of prospective validation. These systems should presently be considered adjunctive aids rather than autonomous planners. Robust multicenter studies with external validation are needed before routine clinical implementation can be recommended. PROSPERO 1025581. Registered prospectively.