Pain management in intensive care usually involves complex trade-offs between therapeutic goals and patient safety, since both inadequate and excessive treatment may induce serious sequelae. Reinforcement learning can help address this challenge by learning medication dosing policies from retrospective data. However, prior work on sedation and analgesia has optimized for objectives that do not value patient survival while relying on algorithms unsuitable for imperfect information settings. We investigated the risks of these design choices by implementing a deep reinforcement learning framework to suggest hourly medication doses under partial observability. Using data from 47,144 ICU stays in the MIMIC-IV database, we trained policies to prescribe opioids, propofol, benzodiazepines, and dexmedetomidine according to two goals: reduce pain or jointly reduce pain and mortality. We found that, although the two policies were associated with lower pain, actions from the first policy were positively correlated with mortality, while those proposed by the second policy were negatively correlated. This suggests that valuing long-term outcomes could be critical for safer treatment policies, even if a short-term goal remains the primary objective.
Medical image restoration is essential for improving the usability of noisy, incomplete, and artifact-corrupted clinical scans, yet existing methods often rely on task-specific retraining and offer limited control over the trade-off between faithful reconstruction and prior-driven enhancement. This lack of controllability is especially problematic in clinical settings, where overly aggressive restoration may introduce hallucinated details or alter diagnostically important structures. In this work, we propose CARE, a training-free controllable restoration framework for real-world medical images that explicitly balances structure preservation and prior-guided refinement during inference. CARE uses a dual-latent restoration strategy, in which one branch enforces data fidelity and anatomical consistency while the other leverages a generative prior to recover missing or degraded information. A risk-aware adaptive controller dynamically adjusts the contribution of each branch based on restoration uncertainty and local structural reliability, enabling conservative or enhancement-focused restoration modes without additional model training. We evaluate CARE on noisy and incomplete medical imaging scenarios and show that it achieves strong restoration quality while better preserving clinically relevant structures and reducing the risk of implausible reconstructions and show that it achieves strong restoration quality while better preserving clinically relevant structures and reducing the risk of implausible reconstructions. The proposed approach offers a practical step toward safer, more controllable, and more deployment-ready medical image restoration.
Large visual language models (VLMs) have shown strong multi-modal medical reasoning ability, but most operate as end-to-end black boxes, diverging from clinicians' evidence-based, staged workflows and hindering clinical accountability. Complementarily, expert visual grounding models can accurately localize regions of interest (ROIs), providing explicit, reliable evidence that improves both reasoning accuracy and trust. In this paper, we introduce CARE, advancing Clinical Accountability in multi-modal medical Reasoning with an Evidence-grounded agentic framework. Unlike existing approaches that couple grounding and reasoning within a single generalist model, CARE decomposes the task into coordinated sub-modules to reduce shortcut learning and hallucination: a compact VLM proposes relevant medical entities; an expert entity-referring segmentation model produces pixel-level ROI evidence; and a grounded VLM reasons over the full image augmented by ROI hints. The VLMs are optimized with reinforcement learning with verifiable rewards to align answers with supporting evidence. Furthermore, a VLM coordinator plans tool invocation and reviews evidence-answer consistency, providing agentic control and final verification. Evaluated on standard medical VQA benchmarks, our CARE-Flow (coordinator-free) improves average accuracy by 10.9% over the same size (10B) state-of-the-art (SOTA). With dynamic planning and answer review, our CARE-Coord yields a further gain, outperforming the heavily pre-trained SOTA by 5.2%. Our experiments demonstrate that an agentic framework that emulates clinical workflows, incorporating decoupled specialized models and explicit evidence, yields more accurate and accountable medical AI. Project page: https://xypb.github.io/CARE-Project-Page/
D.V. Dmytriiev, P.A. Borozenets, Yu.V. Oleshko
et al.
Due to active hostilities, the number of patients experiencing phantom pain has sharply increased nowadays. In case of limb amputation, 50 % of patients experience phantom pain, and about 70 % report phantom sensations. This issue is extremely relevant and insufficiently studied in modern medicine. Only a few medical institutions provide adequate management of chronic pain syndrome (including phantom pain). It is also worth noting that untreated phantom pain makes the use of prosthesis impossible, which, in turn, nullifies the potential for complete socialization and adaptation of the patient, thereby increasing the burden not only on the medical system but also on social services. The use of neuraxial analgesia methods has proven to be an effective treatment for this pathology; however, the short duration of effect encourages further exploration and research. This case report highlights the combination of neuraxial methods with the use of botulinum toxin type A for the treatment of phantom pain in patients with traumatic limb amputations. Given the limited number of relevant studies and the small sample size regarding the use of botulinum toxin type A, we would like to present our own clinical case with a positive outcome.
Medical emergencies. Critical care. Intensive care. First aid
Boris Penev, Hallmon Hughes, Katherine Scarpino
et al.
Introduction: The trigeminocardiac reflex (TCR), a physiologic response to irritation of the branches of the trigeminal nerve, was first described in humans in 1870. Gastric hypermotility, hypotension, bradycardia, and even asystole have been reported in response to surgical manipulation of the trigeminal nerve and its branches, but literature is limited in patients not undergoing surgery. Although effects are generally transient and benign, TCR can present a significant diagnostic and therapeutic challenge in patients undergoing surgical manipulation of the trigeminal nerve and its branches. Case Report: We describe a case of severe bradycardia secondary to facial trauma causing hemodynamic compromise and diagnostic uncertainty. Conclusion: This case highlights a possible case of TCR, as well as therapeutic considerations, in a patient presenting to the emergency department with severe facial trauma.
Medical emergencies. Critical care. Intensive care. First aid
Katja Naasunnguaq Jagd, Rachael DeVries, Ole Winther
Domain-specific foundation models for healthcare have expanded rapidly in recent years, yet foundation models for critical care time series remain relatively underexplored due to the limited size and availability of datasets. In this work, we introduce an early-stage pre-trained foundation model for critical care time-series based on the Bi-Axial Transformer (BAT), trained on pooled electronic health record datasets. We demonstrate effective transfer learning by fine-tuning the model on a dataset distinct from the training sources for mortality prediction, where it outperforms supervised baselines, particularly for small datasets ($<5,000$). These contributions highlight the potential of self-supervised foundation models for critical care times series to support generalizable and robust clinical applications in resource-limited settings.
In critical care settings such as the Intensive Care Unit, clinicians face the complex challenge of balancing conflicting objectives, primarily maximizing patient survival while minimizing resource utilization (e.g., length of stay). Single-objective Reinforcement Learning approaches typically address this by optimizing a fixed scalarized reward function, resulting in rigid policies that fail to adapt to varying clinical priorities. Multi-objective Reinforcement Learning (MORL) offers a solution by learning a set of optimal policies along the Pareto Frontier, allowing for dynamic preference selection at test time. However, applying MORL in healthcare necessitates strict offline learning from historical data. In this paper, we benchmark three offline MORL algorithms, Conditioned Conservative Pareto Q-Learning (CPQL), Adaptive CPQL, and a modified Pareto Efficient Decision Agent (PEDA) Decision Transformer (PEDA DT), against three scalarized single-objective baselines (BC, CQL, and DDQN) on the MIMIC-IV dataset. Using Off-Policy Evaluation (OPE) metrics, we demonstrate that PEDA DT algorithm offers superior flexibility compared to static scalarized baselines. Notably, our results extend previous findings on single-objective Decision Transformers in healthcare, confirming that sequence modeling architectures remain robust and effective when scaled to multi-objective conditioned generation. These findings suggest that offline MORL is a promising framework for enabling personalized, adjustable decision-making in critical care without the need for retraining.
Emerson B. Nairon, BSA, Jeslin Joseph, BS, Abdulkadir Kamal, BSN, RN
et al.
IMPORTANCE:. Patients admitted with cerebral hemorrhage or cerebral edema often undergo external ventricular drain (EVD) placement to monitor and manage intracranial pressure (ICP). A strain gauge transducer accompanies the EVD to convert a pressure signal to an electrical waveform and assign a numeric value to the ICP.
OBJECTIVES:. This study explored ICP accuracy in the presence of blood and other viscous fluid contaminates in the transducer.
DESIGN:. Preclinical comparative design study.
SETTING:. Laboratory setting using two Natus EVDs, two strain gauge transducers, and a sealed pressure chamber.
PARTICIPANTS:. No human subjects or animal models were used.
INTERVENTIONS:. A control transducer primed with saline was compared with an investigational transducer primed with blood or with saline/glycerol mixtures in mass:mass ratios of 25%, 50%, 75%, and 100% glycerol. Volume in a sealed chamber was manipulated to reflect changes in ICP to explore the impact of contaminates on pressure measurement.
MEASUREMENTS AND MAIN RESULTS:. From 90 paired observations, ICP readings were statistically significantly different between the control (saline) and experimental (glycerol or blood) transducers. The time to a stable pressure reading was significantly different for saline vs. 25% glycerol (< 0.0005), 50% glycerol (< 0.005), 75% glycerol (< 0.0001), 100% glycerol (< 0.0005), and blood (< 0.0005). A difference in resting stable pressure was observed for saline vs. blood primed transducers (0.041).
CONCLUSIONS AND RELEVANCE:. There are statistically significant and clinically relevant differences in time to a stable pressure reading when contaminates are introduced into a closed drainage system. Changing a transducer based on the presence of blood contaminate should be considered to improve accuracy but must be weighed against the risk of introducing infection.
Medical emergencies. Critical care. Intensive care. First aid
Jeffrey N. Clark, Matthew Wragg, Emily Nielsen
et al.
There is a growing need to understand how digital systems can support clinical decision-making, particularly as artificial intelligence (AI) models become increasingly complex and less human-interpretable. This complexity raises concerns about trustworthiness, impacting safe and effective adoption of such technologies. Improved understanding of decision-making processes and requirements for explanations coming from decision support tools is a vital component in providing effective explainable solutions. This is particularly relevant in the data-intensive, fast-paced environments of intensive care units (ICUs). To explore these issues, group interviews were conducted with seven ICU clinicians, representing various roles and experience levels. Thematic analysis revealed three core themes: (T1) ICU decision-making relies on a wide range of factors, (T2) the complexity of patient state is challenging for shared decision-making, and (T3) requirements and capabilities of AI decision support systems. We include design recommendations from clinical input, providing insights to inform future AI systems for intensive care.
Medical image segmentation has been significantly advanced with the rapid development of deep learning (DL) techniques. Existing DL-based segmentation models are typically discriminative; i.e., they aim to learn a mapping from the input image to segmentation masks. However, these discriminative methods neglect the underlying data distribution and intrinsic class characteristics, suffering from unstable feature space. In this work, we propose to complement discriminative segmentation methods with the knowledge of underlying data distribution from generative models. To that end, we propose a novel hybrid diffusion framework for medical image segmentation, termed HiDiff, which can synergize the strengths of existing discriminative segmentation models and new generative diffusion models. HiDiff comprises two key components: discriminative segmentor and diffusion refiner. First, we utilize any conventional trained segmentation models as discriminative segmentor, which can provide a segmentation mask prior for diffusion refiner. Second, we propose a novel binary Bernoulli diffusion model (BBDM) as the diffusion refiner, which can effectively, efficiently, and interactively refine the segmentation mask by modeling the underlying data distribution. Third, we train the segmentor and BBDM in an alternate-collaborative manner to mutually boost each other. Extensive experimental results on abdomen organ, brain tumor, polyps, and retinal vessels segmentation datasets, covering four widely-used modalities, demonstrate the superior performance of HiDiff over existing medical segmentation algorithms, including the state-of-the-art transformer- and diffusion-based ones. In addition, HiDiff excels at segmenting small objects and generalizing to new datasets. Source codes are made available at https://github.com/takimailto/HiDiff.
Most of the evidences for beneficial effects of beta-blockers in patients with acute myocardial infarction (AMI) were from the clinical studies published in the pre-reperfusion era when anti-platelet drugs, statins or inhibitors of renin-angiotensin-aldosterone system which are known to reduce cardiovascular mortality of patients with AMI were not introduced. In the reperfusion era, beta-blockers’ benefit has not been clearly shown except in patients with reduced ejection fraction (EF; ≤40%). In the era of the early reperfusion therapy for AMI, a number of patients with mildly reduced EF (>40%, <50%) or preserved EF (≥50%) become increasing. However, because no randomized clinical trials are available until now, the benefit and the optimal duration of oral treatment with beta-blockers in patients with mildly reduced or preserved EF are questionable. Registry data have not showed the association of oral beta-blocker therapy with decreased mortality in survivors without heart failure or left ventricular systolic dysfunction after AMI. In the Korea Acute Myocardial Infarction Registry-National Institute of Health of in-hospital survivors after AMI, the benefit of beta-blocker therapy at discharge was shown in patients with reduced or mildly reduced EF, but not in those with preserved EF, which provides new information about beta-blocker therapy in patients without reduced EF. However, clinical practice can be changed when the results of appropriate randomized clinical trials are available. Ongoing clinical trials may help to answer the unresolved issues of beta-blocker therapy in patients with AMI.
Medical emergencies. Critical care. Intensive care. First aid
Nicolás Riera-López, Francisco Aranda-Aguilar, Montse Gorchs-Molist
et al.
Abstract Introduction Stroke is the most common time-dependent pathology that pre-hospital emergency medical services (EMS) are confronted with. Prioritisation of ambulance dispatch, initial actions and early pre-notification have a major impact on mortality and disability. The COVID-19 pandemic has led to disruptions in the operation of EMS due to the implementation of self-protection measures and increased demand for care. It is crucial to evaluate what has happened to draw the necessary conclusions and propose changes to improve the system’s strength for the future. The study aims to compare prehospital time and neuroprotective care metrics for acute stroke patients during the first wave of COVID-19 and the same periods in the years before and after. Methods Analytical, observational, multicentre study conducted in the autonomous communities of Andalusia, Catalonia, Galicia, and Madrid in the pre-COVID-19 (2019), “first wave” of COVID-19 (2020) and post-COVID-19 (2021) periods. Consecutive non-randomized sampling. Descriptive statistical analysis and hypothesis testing to compare the three time periods, with two by two post-hoc comparisons, and multivariate analysis. Results A total of 1,709 patients were analysed. During 2020 there was a significant increase in attendance time of 1.8 min compared to 2019, which was not recovered in 2021. The time of symptom onset was recorded in 82.8% of cases, and 83.3% of patients were referred to specialized stroke centres. Neuroprotective measures (airway, blood glucose, temperature, and blood pressure) were performed in 43.6% of patients. Conclusion During the first wave of COVID-19, the on-scene times of pre-hospital emergency teams increased while keeping the same levels of neuroprotection measures as in the previous and subsequent years. It shows the resilience of EMS under challenging circumstances such as those experienced during the pandemic.
Special situations and conditions, Medical emergencies. Critical care. Intensive care. First aid
Bourke W. Tillmann, Julie Hallet, Rinku Sutradhar
et al.
Abstract Background Older adults are at high-risk for a post-operative intensive care unit (ICU) admission, yet little is known about the impact of these admissions on quality of life. The objective of this study was to evaluate the impact of an unexpected post-operative ICU admission on the burden of cancer symptoms among older adults who underwent high-intensity cancer surgery and survived to hospital discharge. Methods We performed a population-based cohort study of older adults (age ≥ 70) who underwent high-intensity cancer surgery and survived to hospital discharge in Ontario, Canada (2007–2017). Using the Edmonton Symptom Assessment System (ESAS), a standardized tool that quantifies patient-reported physical, mental, and emotional symptoms, we described the burden of cancer symptoms during the year after surgery. Total symptom scores ≥ 40 indicated a moderate-to-severe symptom burden. Modified log-Poisson analysis was used to estimate the impact of an unexpected post-operative ICU admission (admission not related to routine monitoring) on the likelihood of experiencing a moderate-to-severe symptom burden during the year after surgery, accounting for potential confounders. We then used multivariable generalized linear mixed models to model symptom trajectories among patients with two or more ESAS assessments. A 10-point difference in total symptom scores was considered clinically significant. Results Among 16,560 patients (mean age 76.5 years; 43.4% female), 1,503 (9.1%) had an unexpected ICU admission. After accounting for baseline characteristics, patients with an unexcepted ICU admission were more likely to experience a moderate-to-severe symptom burden relative to those without an unexpected ICU admission (RR 1.64, 95% CI 1.31–2.05). Specifically, among patients with an unexcepted ICU admission the average probability of experiencing moderate-to-severe symptoms ranged from 6.9% (95 CI 5.8–8.3%) during the first month after surgery to 3.2% (95% CI 0.9–11.7%) at the end of the year. Among the 11,229 (67.8%) patients with multiple ESAS assessments, adjusted differences in total scores between patients with and without an unexpected ICU admission ranged from 2.0 to 5.7-points throughout the year (p < 0.001). Conclusion While unexpected ICU admissions are associated with a small increase in the likelihood of experiencing a moderate-to-severe symptom burden, most patients do not experience a high overall symptom burden during the year after surgery. These findings support the role of aggressive therapy among older adults after major surgery.
Medical emergencies. Critical care. Intensive care. First aid
Federated learning (FL) enables multiple client medical institutes collaboratively train a deep learning (DL) model with privacy protection. However, the performance of FL can be constrained by the limited availability of labeled data in small institutes and the heterogeneous (i.e., non-i.i.d.) data distribution across institutes. Though data augmentation has been a proven technique to boost the generalization capabilities of conventional centralized DL as a "free lunch", its application in FL is largely underexplored. Notably, constrained by costly labeling, 3D medical segmentation generally relies on data augmentation. In this work, we aim to develop a vicinal feature-level data augmentation (VFDA) scheme to efficiently alleviate the local feature shift and facilitate collaborative training for privacy-aware FL segmentation. We take both the inner- and inter-institute divergence into consideration, without the need for cross-institute transfer of raw data or their mixup. Specifically, we exploit the batch-wise feature statistics (e.g., mean and standard deviation) in each institute to abstractly represent the discrepancy of data, and model each feature statistic probabilistically via a Gaussian prototype, with the mean corresponding to the original statistic and the variance quantifying the augmentation scope. From the vicinal risk minimization perspective, novel feature statistics can be drawn from the Gaussian distribution to fulfill augmentation. The variance is explicitly derived by the data bias in each individual institute and the underlying feature statistics characterized by all participating institutes. The added-on VFDA consistently yielded marked improvements over six advanced FL methods on both 3D brain tumor and cardiac segmentation.
A hospital readmission is when a patient who was discharged from the hospital is admitted again for the same or related care within a certain period. Hospital readmissions are a significant problem in the healthcare domain, as they lead to increased hospitalization costs, decreased patient satisfaction, and increased risk of adverse outcomes such as infections, medication errors, and even death. The problem of hospital readmissions is particularly acute in intensive care units (ICUs), due to the severity of the patients' conditions, and the substantial risk of complications. Predicting Unplanned Readmissions in ICUs is a challenging task, as it involves analyzing different data modalities, such as static data, unstructured free text, sequences of diagnoses and procedures, and multivariate time-series. Here, we investigate the effectiveness of each data modality separately, then alongside with others, using state-of-the-art machine learning approaches in time-series analysis and natural language processing. Using our evaluation process, we are able to determine the contribution of each data modality, and for the first time in the context of readmission, establish a hierarchy of their predictive value. Additionally, we demonstrate the impact of Temporal Abstractions in enhancing the performance of time-series approaches to readmission prediction. Due to conflicting definitions in the literature, we also provide a clear definition of the term Unplanned Readmission to enhance reproducibility and consistency of future research and to prevent any potential misunderstandings that could result from diverse interpretations of the term. Our experimental results on a large benchmark clinical data set show that Discharge Notes written by physicians, have better capabilities for readmission prediction than all other modalities.
Medical treatments often involve a sequence of decisions, each informed by previous outcomes. This process closely aligns with reinforcement learning (RL), a framework for optimizing sequential decisions to maximize cumulative rewards under unknown dynamics. While RL shows promise for creating data-driven treatment plans, its application in medical contexts is challenging due to the frequent need to use sparse rewards, primarily defined based on mortality outcomes. This sparsity can reduce the stability of offline estimates, posing a significant hurdle in fully utilizing RL for medical decision-making. We introduce a deep Q-learning approach to obtain more reliable critical care policies by integrating relevant but noisy frequently measured biomarker signals into the reward specification without compromising the optimization of the main outcome. Our method prunes the action space based on all available rewards before training a final model on the sparse main reward. This approach minimizes potential distortions of the main objective while extracting valuable information from intermediate signals to guide learning. We evaluate our method in off-policy and offline settings using simulated environments and real health records from intensive care units. Our empirical results demonstrate that our method outperforms common offline RL methods such as conservative Q-learning and batch-constrained deep Q-learning. By disentangling sparse rewards and frequently measured reward proxies through action pruning, our work represents a step towards developing reliable policies that effectively harness the wealth of available information in data-intensive critical care environments.
S. Valdoleiros, Cristina Torrão, Laura S Freitas
et al.
Background Nosocomial meningitis is a medical emergency that requires early diagnosis, prompt initiation of therapy, and frequent admission to the intensive care unit (ICU). Methods A retrospective study was conducted in adult patients diagnosed with nosocomial meningitis who required admission to the ICU between April 2010 and March 2020. Meningitis/ventriculitis and intracranial infection were defined according to Centers for Disease Control and Prevention guidelines. Results An incidence of 0.75% of nosocomial meningitis was observed among 70 patients. The mean patient age was 59 years and 34% were ≥65 years. Twenty-two percent of patients were in an immunocompromised state. A clear predisposing factor for nosocomial meningitis (traumatic brain injury, basal skull fracture, brain hemorrhage, central nervous system [CNS] invasive procedure or device) was present in 93% of patients. Fever was the most frequent clinical feature. A microbiological agent was identified in 30% of cases, of which 27% were bacteria, with a predominance of Gram-negative over Gram-positive. Complications developed in 47% of cases, 24% of patients were discharged with a Glasgow coma scale <14, and 37% died. There were no clear clinical predictors of complications. Advanced age (≥65 years old) and the presence of complications were associated with higher hospital mortality. Conclusions Nosocomial meningitis in critical care has a low incidence rate but high mortality and morbidity. In critical care patients with CNS-related risk factors, a high level of suspicion for meningitis is warranted, but diagnosis can be hindered by several confounding factors.
Heena Chhanwal, Divya Kheskani, Parita Gandhi
et al.
Abstract Background Awareness under general anesthesia is an unpleasant phenomenon that usually goes unnoticed and neglected. Numerous incidences of intraoperative awareness are not reported. Reasons for awareness might be the inadequate depth of anesthesia, less effective drugs, lack of proper anesthesia monitoring equipment, and untrained medical staff. The purpose of this study is to evaluate intraoperative awareness during general anesthesia and titrate the amount of anesthetic agents according to BIS values among patients and monitor hemodynamic parameters throughout the surgery. Results The intraoperative awareness reported was 2% in the BIS group and 8% in the non-BIS group. The total propofol consumption in the BIS group was significantly less as compared to the non-BIS group (P value<0.0001). Conclusions The incidence of definite awareness with postoperative recall and propofol consumption was reduced in the BIS group as compared to the non-BIS group.
Anesthesiology, Medical emergencies. Critical care. Intensive care. First aid