18 articles · 3 topics
Critical care medicine6 articles
★ Flagship · Score 1.00 · ✓ Verified
Intermittent or Continuous Renal-Replacement Therapy in Acute Kidney Injury
Stèphane Gaudry et al. — New England Journal of Medicine · October 14, 2026
Why it’s here: This RCT directly addresses renal-replacement therapy in critically ill adults with AKI, a key area of interest.
Study Design
This multicenter, open-label, noninferiority trial randomly assigned critically ill adults with acute kidney injury requiring renal-replacement therapy to either intermittent hemodialysis or continuous renal-replacement therapy as initial treatment. The primary outcome assessed was a major adverse kidney event by day 90, defined as death, dependence on renal-replacement therapy, or a serum creatinine concentration more than 25% above baseline.
Key Results
The trial was stopped early due to a sustained signal of higher mortality with intermittent hemodialysis. By day 90, major adverse kidney events occurred in 78.1% of the intermittent group versus 71.6% of the continuous group, and death occurred in 63.1% of the intermittent group compared to 53.5% in the continuous group.
Why It Matters
Intermittent hemodialysis was not shown to be noninferior to continuous renal-replacement therapy for major adverse kidney events in critically ill patients with acute kidney injury and was associated with higher mortality. The study was funded by the French Ministry of Health and is registered with ClinicalTrials.gov number NCT06032884.
Score 0.94 · ⚠ Check
Nicolò Sella et al. — Critical Care · October 14, 2026
Why it’s here: This article directly addresses mechanical ventilation modes for weaning in adult ICU patients and is a systematic review of RCTs.
⚠ a section is not grounded in the abstract (The 'Key Results' section states that ASV reduced severe asynchrony, but the abstract only states that ASV significantly shortened weaning duration and reduced severe asynchrony. The 'Why It Matters' section states that no significant differences were found in reintubation or mortality rates, but the abstract states that no significant differences were observed in reintubation, ICU mortality or 28-day mortality. The abstract also mentions that NAVA may have improved weaning success, reduced asynchrony and hospital mortality, shortened IMV duration, increased ventilator-free days, compared with PSV, which is not fully captured in the 'Why It Matters' section.)
Study Design
This systematic review and network meta-analysis compared advanced ventilatory modes (NAVA, PAV, ASV, SmartCare/PS) against pressure support ventilation (PSV) in adult ICU patients undergoing invasive mechanical ventilation weaning. The analysis included 23 randomized controlled trials with 3,755 patients, searching databases from inception to March 27, 2026.
Key Results
Neurally adjusted ventilatory assist (NAVA) significantly increased weaning success (RR 1.13, 95% CI 1.00–1.27) and reduced patient-ventilator asynchrony index (AI) (MD − 7.77%) compared to PSV. Adaptive support ventilation (ASV) also significantly shortened weaning duration (MD − 0.65 days, 95% CI − 1.06 to − 0.24) and reduced severe asynchrony.
Why It Matters
These findings suggest NAVA, and to a lesser extent ASV, may be preferred weaning modes, particularly for patients at high risk of asynchrony or prolonged ventilation, as they demonstrated improved weaning success and reduced asynchrony. The certainty of evidence was assessed using CINeMA, and no significant differences were found in reintubation or mortality rates.
★ Flagship · Score 0.93 · ✓ Verified
Effect of Vasopressin Initiation Threshold on Septic Shock Mortality
Ithan D. Peltan et al. — JAMA · October 14, 2026
Why it’s here: This article directly addresses vasopressor support and septic shock, key interests for critical care medicine.
Study Design
This cluster-randomized, cluster-crossover, pragmatic clinical strategy trial involved adult patients receiving vasopressors for septic shock across 13 hospitals in Idaho and Utah. Hospitals were randomized to implement either a lower-threshold strategy (vasopressin initiation at 0.1 µg/kg/min norepinephrine equivalent) or a higher-threshold strategy (vasopressin initiation at 0.4 µg/kg/min norepinephrine equivalent), with monthly crossovers.
Key Results
Initiating vasopressin at a lower threshold (0.1 µg/kg/min norepinephrine equivalent) compared to a higher threshold (0.4 µg/kg/min norepinephrine equivalent) did not significantly reduce 28-day mortality, with rates of 31.8% versus 33.0% respectively (adjusted marginal odds ratio, 0.98; P = .62). Kidney replacement therapy-free days through day 28 were also similar between the strategies, with a median of 29 days in both groups.
Why It Matters
In this trial, a hospital-level strategy recommending earlier initiation of vasopressin did not improve 28-day mortality in patients with septic shock compared to a strategy recommending later initiation. These findings suggest that the specific dose threshold for initiating a second vasopressor may not be the primary determinant of outcomes in septic shock management.
● Preprint · Score 0.76 · ✓ Verified
Rapid Sepsis Diagnosis in Critically Ill Adults with Remoscope Label-Free Whole-Blood Imaging
Bratvold, N. et al. — medRxiv Preprint · intensive care and critical care medicine · October 11, 2026
Why it’s here: This article directly addresses rapid sepsis diagnosis in critically ill adults using a novel imaging technique, aligning perfectly with the researcher's interest in sepsis and critical care diagnostics.
Study Design
This prospective study utilized label-free imaging cytometry with Remoscope on 10 µL whole-blood samples from 29 critically ill adults and 22 healthy controls. A multiple instance learning (MIL) model was employed to analyze thousands of blood images per patient, distinguishing sepsis from non-infectious critical illness and healthy states.
Key Results
The three-class MIL model, optimized for rule-out, achieved 100% sensitivity and 85.3% specificity in detecting sepsis. Analysis revealed abnormal leukocyte and erythrocyte morphology, including immature neutrophil-like cells, as key hematologic signatures.
Why It Matters
This label-free whole-blood imaging approach with MIL offers a rapid diagnostic method for sepsis-associated hematologic signatures without requiring stains or cold chain. It establishes a foundation for developing rapid diagnostics in critical care settings.
Score 0.72 · ✓ Verified
Yongpo Jiang, Yinghe Xu, 金胜威 — Critical Care · October 12, 2026
Why it’s here: This study investigates lymphatic dysfunction in sepsis using a novel bedside technique, directly relevant to sepsis management in the ICU.
Study Design
This prospective observational cohort study enrolled adult patients meeting Sepsis-3 criteria and non-septic critically-ill controls. Indocyanine green (ICG) was injected subcutaneously into intact upper extremities within 24 hours of diagnosis, and lymphatic flow velocity was measured using a portable laser lymphatic imager.
Key Results
Lymphatic flow velocity was significantly different between groups (P = 0.014), with non-septic patients at 0.24 cm/s, sepsis survivors at 0.08 cm/s, and sepsis non-survivors at 0.02 cm/s. Among septic patients, 9 had zero lymphatic flow velocity, and 55.6% of these died within 28 days.
Why It Matters
Bedside ICG lymphography provides a safe, rapid, and visual assessment of peripheral lymphatic drainage in septic adults, with reduced flow velocity potentially serving as a prognostic biomarker. The complete examination was finished within 10 minutes with no adverse events reported, though these findings require validation in larger cohorts.
✦ Something Different · Score 0.53 · ✓ Verified
Felix Hieber et al. — Critical Care · October 17, 2026
✦ A change of pace: This article shifts focus from direct patient physiology to the human factors and well-being of the physicians themselves, offering a novel perspective on critical care practice.
Study Design
This real-time observational study shadowed five intensive care physicians over 38 twelve-hour shifts in a German university hospital's surgical ICU. Clinical decisions were classified by severity using a study-specific Decision Severity Score, and psychometric assessments of fatigue, sleepiness, and mood were administered at multiple time points. Temporal workload structure was quantified using the burstiness metric, and shifts were categorized into four types based on decision density and burstiness.
Key Results
A higher mean decision rate was associated with greater end-of-shift decision fatigue, with an increase of 0.5 decisions per 5-minute interval corresponding to a 2.1-point higher fatigue score. Conversely, higher burstiness was linked to less mood disturbance, where a 0.10 increase in burstiness correlated with a 6.3-point lower mood score. Response certainty in a post-shift task decreased with more task interruptions, and a trend towards more conservative responses was seen in busy-clustered shifts.
Why It Matters
These findings suggest that reducing decision volume alone may not improve physician wellbeing, as quieter, uniformly paced shifts showed the least favorable psychometric values. Mitigation strategies might need to focus on protecting concentrated clinical activity and minimizing hindrance stressors like interruptions, though these exploratory findings require replication. Future multicenter studies are needed to determine reproducibility and potential links to objective decision quality or patient safety outcomes.
Healthcare machine learning6 articles
★ Flagship · Score 0.83 · ✓ Verified
Human–artificial intelligence collaboration for comanaging epilepsy in real time
Zack Goldblum et al. — Nature Medicine · October 13, 2026
Why it’s here: This article presents a human-AI collaboration platform for epilepsy management, demonstrating real-time clinical care application and patient interaction with AI.
Study Design
This study presents a device platform that links electroencephalography (EEG) to a large language model-based artificial intelligence (AI) for real-time epilepsy management. The inpatient prototype communicates bidirectionally with patients via a smartphone app, processing biomarkers and notifying them of clinical events. The platform was validated in 13 patients undergoing video-EEG monitoring for epilepsy.
Key Results
In a pseudoprospective simulation, seizure detectors improved precision solely from patient feedback, without expert intervention. Patient feedback reduced seizure detector false alarms by approximately 80% in a small epilepsy monitoring study. This AI-driven system allows patients to converse with their EEG monitoring in real time.
Why It Matters
This human-AI collaboration offers a scalable, safeguarded platform for medical devices to adapt to patient and provider needs, potentially improving epilepsy management. The system enables devices to learn from clinical events and teach patients about their health behaviors. The validation involved 13 patients, demonstrating the AI's ability to refine seizure detection through patient input.
Score 0.78 · ✓ Verified
Fardin Afdideh et al. — npj Digital Medicine · October 13, 2026
Why it’s here: This article presents an infrastructure for post-market governance of adaptive medical AI, focusing on regulatory frameworks and operational monitoring, which is crucial for the real-world deployment and safety of AI in healthcare.
Study Design
This study presents AEGIS, an internal governance and post-market-monitoring infrastructure designed to support predetermined governance pathways for adaptive medical AI. AEGIS comprises dataset assimilation and retraining, model monitoring, and conditional decision-making modules, utilizing a four-category taxonomy and an independent ALARM signal.
Key Results
Instantiating AEGIS for sepsis prediction and brain-tumor segmentation exercised all decision categories, successfully routing failures to the critical REJECT + ALARM state. On real cross-site data, drift detection ROC-AUC rose from 0.82 to 1.00, reliably triggering near 500-patient batches with zero false alarms.
Why It Matters
AEGIS provides operational tooling for post-market governance of medical AI under US and EU regulatory frameworks, addressing a gap in current approaches. This proof-of-concept utilizes illustrative clinical models, and drift scoring quantifies covariate shift only, leaving concept drift for future work.
Score 0.76 · ✓ Verified
Maria Mahbub et al. — npj Digital Medicine · October 16, 2026
Why it’s here: This article directly addresses the use of large language models for clinical information extraction and proposes a validation framework for trustworthy deployment, aligning perfectly with the researcher's interests.
Study Design
This study proposes a multi-stage validation framework for large language model (LLM)-based clinical information extraction, designed for rigorous assessment under weak supervision. The framework integrates prompt calibration, rule-based filtering, semantic grounding, judge-LLM evaluation, expert review, and predictive validity analysis to quantify uncertainty without exhaustive manual annotation.
Key Results
Applied to extract substance use disorder diagnoses from 919,783 clinical notes, the framework's rule-based filtering and semantic grounding removed 14.59% of unsupported or implausible extractions. A judge LLM showed substantial agreement with subject matter experts for high-uncertainty cases, enabling the primary LLM to achieve an F1 score of 0.80.
Why It Matters
The LLM-extracted SUD diagnoses demonstrated greater accuracy in predicting subsequent engagement in SUD specialty care compared to structured-data baselines. This work presents a generalizable human-in-the-loop framework for establishing the trustworthiness of LLM-based clinical information extraction at scale while minimizing the need for exhaustive manual annotation.
Score 0.76 · ✓ Verified
Sihang Zeng et al. — npj Digital Medicine · October 15, 2026
Why it’s here: This article presents a multi-agent LLM framework for temporal reasoning over EHR data for multi-cancer early detection, which is a core interest in predictive modeling and AI in clinical care.
Study Design
This study introduces TrajOnco, a training-free, multi-agent large language model (LLM) framework designed for temporal reasoning over longitudinal electronic health records (EHRs) for multi-cancer early detection. The framework utilizes a chain-of-agents architecture with long-term memory to generate patient-level summaries, evidence-linked rationales, and predicted risk scores.
Key Results
In a zero-shot evaluation on de-identified Truveta EHR data across 15 cancer types, TrajOnco achieved AUROCs of 0.64–0.80 for predicting cancer diagnosis risk at 1 year. It performed comparably to supervised machine learning in a lung cancer benchmark and demonstrated better temporal reasoning than single-agent LLMs.
Why It Matters
TrajOnco demonstrates the potential of multi-agent LLMs for interpretable temporal reasoning over longitudinal EHRs, advancing scalable multi-cancer early detection and clinical insight generation. Its interpretable outputs can reveal population-level risk patterns aligning with established clinical knowledge, and the multi-agent design enabled effective temporal reasoning even with smaller-capacity models.
Score 0.72 · ✓ Verified
Cathy Shyr et al. — npj Digital Medicine · October 15, 2026
Why it’s here: This article develops an AI framework using LLMs for rare disease phenotyping from clinical notes, directly applying AI to clinical data for diagnostic support.
Study Design
This study introduces RARE-PHENIX, an end-to-end artificial intelligence framework for rare disease phenotyping. The framework integrates large language model-based phenotype extraction, standardization to Human Phenotype Ontology (HPO) terms, and supervised ranking of HPO terms. It was trained on data from 2671 patients across 11 Undiagnosed Diseases Network clinical sites and validated on 16,357 clinical notes from 143 patients.
Key Results
RARE-PHENIX generates structured, prioritized HPO terms that are highly concordant with clinician curation. The framework demonstrates potential downstream disease-prioritization signal. External validation was performed on real-world clinical notes.
Why It Matters
This AI framework may support human-in-the-loop rare disease phenotyping in real-world settings by automating labor-intensive manual curation. By operationalizing the full workflow from feature extraction to HPO term prioritization, RARE-PHENIX addresses limitations of existing AI approaches. The findings suggest the framework's utility in improving the phenotyping process for rare diseases.
✦ Something Different · Score 0.66 · ✓ Verified
Liang-Wei Tseng et al. — npj Digital Medicine · October 13, 2026
✦ A change of pace: This article explores the challenges of deploying medical LLMs in a specific, less common domain (traditional Chinese medicine), offering a critical perspective on prompt engineering that differs from the more direct application-focused articles.
Study Design
This study evaluated 16 large language models (LLMs) using seven prompt strategies on 475 questions from the Taiwan national TCM licensing examination. The evaluation aimed to understand LLM performance on traditional Chinese medicine (TCM) knowledge and the factors influencing performance variation.
Key Results
Without prompt engineering, LLM accuracy varied widely from 35.4% to 85.9%, with roughly half of the models passing. Prompt optimization significantly altered performance for individual models, reversing failing scores in one model (49.7% to 87.4%) while decreasing accuracy in others, and no single prompt strategy consistently benefited all models.
Why It Matters
These findings highlight that reliable deployment of LLMs in TCM necessitates empirical validation of both model choice and prompt strategy, as model selection dictates baseline capability and prompt-model interactions can substantially change performance. A key limitation is that comparisons between model families were descriptive due to the evaluation of only 16 individual models.
Hospital logistics and operations6 articles
★ Flagship · Score 0.83 · ✓ Verified
Convexity of Inverse Blocking Curves in Loss Systems
Benjamin Legros — Operations Research · October 12, 2026
Why it’s here: Directly addresses capacity, scale, and resource pooling in loss systems, highly relevant to hospital capacity and operations research.
Study Design
This study investigates economies of scale in loss systems by analyzing the relationship between system capacity and supportable demand. The key method involves proving that the maximum supportable demand is a function of the number of servers.
Key Results
The paper establishes that the maximum supportable demand is an increasing and globally convex function of the number of servers. This means that as a system grows larger, each additional server can support progressively more additional demand while maintaining the same blocking target.
Why It Matters
This result provides a rigorous mathematical foundation for resource pooling in capacity-constrained services, with applications in telecommunications, healthcare, and emergency response. The findings apply to the classical Erlang-B loss model and can be extended to systems with finite waiting rooms.
★ Flagship · Score 0.77 · ✓ Verified
Dialysis Clinic Patient Flow and Work Design
Vincent W. Slaugh, André A. Ciré — Manufacturing & Service Operations Management · October 16, 2026
Why it’s here: This article directly addresses patient flow and work design in a healthcare setting, using operations research methods to improve efficiency.
Study Design
This study analyzes patient flow and work design in US outpatient dialysis clinics, which serve 433,000 patients annually. The research compares the prevailing "push" approach with two proposed alternatives: an "interlaced" scheme and a "tandem" scheme for scheduling patients and patient care technicians (PCTs).
Key Results
Under constant treatment durations, interlaced schedules achieve the shortest completion time, while tandem configurations can reduce labor per patient. Numerical analysis incorporating heterogeneous treatment durations and stochastic factors confirms the robustness of these insights. The study shows that interlacing finishes earliest with homogeneous treatment durations.
Why It Matters
These findings highlight minimizing unoccupied station time as the primary driver of labor efficiency, contrasting with the push-focused approach. Transitioning to interlaced or tandem systems, potentially with increased standard task times (e.g., from 15/15 to 25/20 minutes), can alleviate PCT stress and promote patient safety without raising labor costs.
Score 0.76 · ✓ Verified
İbrahim H. Kayral, Dilek DÜLGER, Seda Taşır Yılmaz — BMC Health Services Research · October 13, 2026
Why it’s here: This article systematically reviews digital and organizational strategies for preventing healthcare-associated infections in adult hospitals, directly addressing hospital operations and implementation.
Study Design
This systematic review synthesizes evidence on digital, organizational, and hybrid strategies for preventing healthcare-associated infections (HAIs) in adult inpatient hospital settings. A systematic search of PubMed/MEDLINE, Scopus, and Web of Science was conducted for studies published between January 2010 and December 2025, with 47 studies ultimately included. A structured narrative synthesis was performed due to methodological heterogeneity.
Key Results
Digital interventions primarily supported surveillance and feedback processes, while organizational strategies commonly supported adherence and standardization. Hybrid approaches frequently reported improvements in both process and infection-related outcomes, though direct category-level comparisons were absent. The reported effects varied according to hospital setting, organizational capacity, and implementation context.
Why It Matters
Successful implementation was commonly associated with leadership support, workflow integration, and organizational readiness. The findings suggest that digital technologies may better support sustained HAI prevention when embedded within organizational workflows and quality improvement structures rather than implemented as standalone solutions. Gaps in real-world implementation and organizational integration continue to limit the effectiveness of infection prevention efforts.
Score 0.72 · ✓ Verified
Solving operating room scheduling problems using large language model based multi-agent architecture
Chu Tang et al. — npj Digital Medicine · October 15, 2026
Why it’s here: Focuses on solving operating room scheduling problems using LLM-based multi-agent systems, directly relevant to OR scheduling and optimization.
Study Design
This study evaluates a novel iterative multi-agent (IMA) framework for generating and refining optimization code for operating room (OR) scheduling. The framework leverages iterative collaboration among specialized large language model (LLM)-based agents to address the complexity of OR operations. Experiments were conducted across 10 LLMs and 35 scenarios, including synthetic and real OR datasets.
Key Results
The IMA framework consistently outperformed single-agent and traditional multi-agent baselines, achieving an average feasible schedule rate of 82.91% compared to 65.59% and 52.53% respectively. Furthermore, IMA-generated code produced solutions comparable to expert-written mixed-integer programming models. A case study demonstrated improvements in OR utilization and overtime reduction.
Why It Matters
These findings suggest that LLM-based multi-agent systems offer a practical and effective approach for OR scheduling optimization, a critical task given the substantial share of hospital costs associated with OR operations. The IMA framework's ability to generate and refine optimization code addresses the diversity of surgical types, high patient demand, and limited hospital resources inherent in this complex problem.
● Preprint · Score 0.62 · ✓ Verified
Paganini, M. et al. — medRxiv Preprint · emergency medicine · October 13, 2026
Why it’s here: This study validates a tool for forecasting heatwave-related emergency department demand, directly relevant to capacity and demand management in hospitals.
Study Design
This retrospective, ecological, time-series study validated a heatwave-related emergency department (ED) demand forecasting tool across four clusters in Ticino Canton, Switzerland. The model used daily mean temperature and ED presentations, trained on summers from 2016-2024 and tested on summer 2025, defining heatwaves as daily mean temperature >25°C for ≥3 consecutive days.
Key Results
A non-linear temperature-demand association was observed, with ED visits increasing at higher temperatures, and model fit varied across clusters (R² 0.389-0.672). The forecasting tool's prediction errors remained within acceptable ranges during two heatwaves in 2025, outperforming naive baselines.
Why It Matters
The findings suggest that heat-health action plans should utilize locally calibrated demand forecasts for emergency care preparedness, rather than standardized activation protocols. Baseline demand, heatwave occurrence, and model performance demonstrated substantial variation within the canton.
★ Flagship · ✦ Something Different · Score 0.57 · ✓ Verified
Shipment Monitoring, Allocation, and the Impact on Food Waste
Tao Lu — Management Science · October 15, 2026
✦ A change of pace: This article shifts the focus from direct patient care logistics to the broader operational challenges of shipment monitoring and allocation, introducing a novel perspective on resource management and waste reduction within a supply chain context.
Study Design
This paper considers a firm that purchases and transports a fresh produce product for sale in two destination markets. Using sensor technologies to monitor product condition, the firm decides on a purchase quantity and then allocates units between markets based on their condition and transportation distance. The analysis characterizes optimal allocation policies to efficiently determine the optimal purchase quantity.
Key Results
The study shows that monitoring technology may inadvertently increase total waste by inducing larger purchase quantities, though this is often accompanied by a lower waste rate and improved distribution efficiency. Improvements in postharvest processes can strengthen the waste-reduction benefits of monitoring technology.
Why It Matters
As firms adopt monitoring technology for shipment allocation, additional policy and managerial attention may be required to mitigate potential increases in total waste. Firms aiming for sustainability goals can capture profit gains from monitoring by carefully moderating increases in purchase quantities.
Curated from OpenAlex, Crossref & medRxiv. Summaries are machine-generated and grounded in each article’s abstract or open-access text; ⚠ flags a summary that needs a second look, and ✦ marks a deliberate change-of-pace pick.
Leave a Reply