17 articles · 3 topics
Critical care medicine6 articles
Score 0.94 · ✓ Verified
Yin-sheng Liao et al. — Critical Care · September 11, 2026
Why it’s here: This article is a systematic review and meta-analysis on using EtCO₂/PETCO₂ changes to predict fluid responsiveness in mechanically ventilated adults, highly relevant to hemodynamic management and mechanical ventilation.
Study Design
This systematic review and meta-analysis evaluated the diagnostic accuracy of dynamic EtCO₂/PETCO₂ changes for predicting fluid responsiveness in mechanically ventilated adults. Twenty studies involving 1,515 diagnostic observations were included, utilizing passive leg raising, mini-fluid challenge, or volume expansion as preload challenges. Pooled sensitivity and specificity were estimated using a bivariate random-effects model.
Key Results
The pooled sensitivity was 0.748 (95% CI, 0.680–0.806) and specificity was 0.853 (95% CI, 0.797–0.895), with a summary ROC AUC of 0.872 based on 2×2 data. Across all 20 studies, the random-effects pooled reported AUC was 0.839 (95% CI, 0.788–0.879). Exploratory analyses suggested potential benefits for PLR-based assessment and patient-level analyses, but subgroup superiority was not established.
Why It Matters
Dynamic EtCO₂/PETCO₂ changes may serve as a useful non-invasive adjunct for assessing fluid responsiveness in mechanically ventilated adults under stable conditions. However, the certainty of evidence was rated as low, and these changes should not be used as a standalone test, as the available evidence does not support a universal threshold.
Score 0.94 · ✓ Verified
Henrique Gomes Mendes et al. — Critical Care Medicine · September 11, 2026
Why it’s here: Directly addresses hemodynamic support (vasodilatory shock, resuscitation) and outcomes (mortality) in critically ill adults using RCTs and meta-analysis.
Study Design
This systematic review and dual frequentist-Bayesian meta-analysis included seven randomized controlled trials (RCTs) enrolling 2408 adult patients with septic shock. The study compared peripheral perfusion-guided resuscitation with standard or lactate-guided care, assessing 28-day mortality. Risk of bias and certainty of evidence were evaluated using established tools.
Key Results
The primary frequentist analysis found a pooled risk ratio (RR) for 28-day mortality of 0.87 (95% CI, 0.76-1.01) with peripheral perfusion-guided resuscitation, which did not reach statistical significance (p = 0.06). The Bayesian analysis estimated a 97.2% posterior probability that the RR is less than 1, indicating a consistent direction of effect.
Why It Matters
Peripheral perfusion-guided resuscitation showed a trend towards lower 28-day mortality in adults with septic shock, though the frequentist analysis did not achieve statistical significance. While the Bayesian analysis suggests a high probability of benefit, the magnitude of the effect remains uncertain, highlighting the need for adequately powered trials to confirm clinical meaningfulness.
Score 0.82 · ✓ Verified
Sandra L. Kane‐Gill et al. — Critical Care Medicine · September 11, 2026
Why it’s here: This article is a systematic evidence map of biomarker use in critically ill patients for acute kidney injury, directly relevant to critical care medicine and AKI management.
Study Design
This study created systematic evidence maps of novel kidney injury biomarkers in critically ill patients by screening 6,805 records and including 1,116 studies. The majority of included studies (78.6%) focused on adult populations, with 93.3% employing a cohort design to investigate acute kidney injury (AKI) biomarkers.
Key Results
Systematic evidence maps were synthesized for biomarker studies to predict AKI (n = 944), prognosticate clinical outcomes (n = 647), diagnose AKI etiology (n = 109), enrich clinical trials (n = 6), and manage AKI (n = 12). Mixed critically ill populations, cardiac surgery, and sepsis were the most frequently studied clinical contexts, accounting for 69.5% of the studies.
Why It Matters
While substantial clinical evidence exists for biomarker accuracy in AKI diagnosis, practical trials for clinical use are fewer, indicating a gap between predictive accuracy and clinical decision-making. Guidance for implementation can be gleaned from evaluations, particularly in surgical and nephrotoxin-exposed populations where evidence is available, highlighting priorities for future management and enrichment trials.
● Preprint · Score 0.82 · ✓ Verified
Khdir, S. A., Ahmed, O. H. — medRxiv Preprint · health informatics · September 11, 2026
Why it’s here: This article presents a deep learning model for simultaneous early detection of sepsis and heart failure in adult intensive care units.
Study Design
This study introduces DualStream-MTCA, a hybrid deep-learning model designed for the simultaneous early detection of sepsis and heart failure in adult intensive care units. The model integrates dual Bidirectional LSTM streams for vital signs and laboratory data, incorporating multi-head cross-attention and XGBoost leaf embeddings. It was trained on 53,229 ICU stays from MIMIC-IV v3.1.
Key Results
On a held-out test set, DualStream-MTCA achieved an AUROC of 0.867 for sepsis and 0.899 for heart failure, demonstrating excellent probability calibration with an expected calibration error below 0.015. External validation on 102,695 eICU-CRD stays showed strong generalizability for heart failure, with only a 0.036 AUROC drop.
Why It Matters
The developed model addresses the challenge of simultaneous early detection for sepsis and heart failure, which share physiological warning signs but require distinct treatments. While external validation shows strong generalizability for heart failure, sepsis performance drops were attributed to data sparsity in the external dataset. The model also demonstrated positive net clinical benefit.
Score 0.70 · ✓ Verified
Chloe Apps et al. — Critical Care Medicine · September 11, 2026
Why it’s here: This article directly addresses driving ability in adults recovering from critical illness, including those who received mechanical ventilation, aligning with ICU outcomes and recovery.
Study Design
This observational cohort study was conducted at a certified driving assessment center in the United Kingdom. It evaluated adult ICU survivors who received mechanical ventilation or ECMO, assessing their physical and cognitive abilities, as well as practical driving skills using simulators and real-world driving tests.
Key Results
Of 33 participants who completed the assessment around 12.5 weeks post-discharge, 26 (79%) were deemed capable of driving a standard vehicle, though only 15 (45%) had resumed driving. Seven participants (21%) were unable to drive due to physical, psychological, or cognitive impairments, but four of these were able to resume driving after interventions within the 12-month period.
Why It Matters
Approximately three-quarters of this cohort were deemed able to drive a standard vehicle about 3 months after hospital discharge, yet less than half had resumed driving, highlighting a gap between capability and resumption. The study also found that most individuals with driving impairments could resume driving after interventions, suggesting the utility of comprehensive assessments and subsequent support.
● Preprint · ✦ Something Different · Score 0.58 · ✓ Verified
Jansen, N. J. et al. — medRxiv Preprint · intensive care and critical care medicine · September 11, 2026
✦ A change of pace: This article offers a perspective on building critical care capacity in a low-resource setting, which is a significantly different focus from the highly technical, data-driven, and often Western-centric research typically found in critical care journals.
Study Design
This mixed-methods descriptive study details the establishment of a six-bed ICU at a rural hospital in Uganda. The study collected quantitative data on patient demographics, diagnoses, and interventions, alongside qualitative data from group discussions to identify challenges and lessons learned.
Key Results
During the first three months, 45 patients were admitted to the new ICU. Key challenges encountered included an unstable power supply, procurement delays for medications, limited critical care knowledge, cultural communication barriers, and financial constraints.
Why It Matters
Establishing a functional ICU in a low-resource setting is feasible with extensive planning, continuous training, and robust infrastructure, as demonstrated by this study. Simple, affordable interventions like continuous monitoring and reliable oxygen supply significantly impacted patient safety, offering practical insights for similar initiatives in LMICs.
Healthcare machine learning6 articles
★ Flagship · Score 0.81 · ✓ Verified
Arsela Prelaj et al. — Nature Medicine · September 1, 2026
Why it’s here: This article discusses explainable AI, decision support tools, multimodal models, real-world deployment, and clinical usability in cancer treatment.
Study Design
The I 3 LUNG study is the largest international, real-world, multimodal, AI-based study, enrolling 2,396 patients with non-small cell lung cancer (NSCLC). It integrated real-world clinical and blood (CB) data, CT images, digital pathology, and genomics into machine learning early fusion (MLEF) and deep learning intermediate fusion (DLIF) models. ML and DL CB-only models were also developed and evaluated.
Key Results
AI models significantly surpassed PD-L1, ECOG PS, NLR, LDH, and LIPI scores in the independent TEST set, with CB-only models achieving an AUC up to 0.77. While multimodal integration with MLEF showed higher performance, its incremental benefit was uncertain and did not translate in TEST and EXVAL. Performance dropped in external validation (EXVAL) to an AUC range of 0.55–0.72, likely reflecting population differences.
Why It Matters
The study demonstrates the clinical usefulness of AI tools in NSCLC treatment selection, with an explainable AI (XAI) ML CB-only tool improving physician predictions. Although multimodal integration showed promise, its added value requires further investigation. A prospective validation of the decision support system is currently underway in over 2,000 patients.
★ Flagship · Score 0.77 · ✓ Verified
On-premise medical AI agents for reliable clinical decision-making
Li Zhang et al. — Nature Medicine · September 15, 2026
Why it’s here: This article presents on-premise medical AI agents powered by LLMs for reliable clinical decision-making, directly addressing AI in clinical care and decision support.
Study Design
This study developed and evaluated a fully on-premise clinical AI agent designed for reliable decision-making in intensive care settings. The agent couples local operational control with a multi-perspective reliability framework to support selective autonomy, utilizing the MIMIC-IV database for evaluation.
Key Results
The on-premise agent achieved 90.04% accuracy on a seven-disease task and 83.8% accuracy on a four-disease task, approaching a cloud baseline. Diagnostic behavioral consistency was a strong indicator of correctness (AUC = 0.860), and at a 0.90 consistency threshold, 49.4% of cases were retained with 98.9% diagnostic accuracy.
Why It Matters
These findings propose a practical framework for institutionally governed clinical AI agents, enabling selective autonomy by using reliability signals to identify lower-risk cases for autonomous handling. This approach addresses limitations in clinical translation by ensuring institutionally governed deployment and reliable decision-time uncertainty estimation.
Score 0.72 · ⚠ Check
Tobias Zimmermann et al. — npj Digital Medicine · September 14, 2026
Why it’s here: This article details the training and prospective validation of a deep learning model for predicting myocardial infarction subtypes from ECGs, demonstrating clinical utility.
⚠ a section is not grounded in the abstract (The 'Key Results' section states that the model 'outperformed physician interpretation for ST-segment elevation MI (STEMI) with an AUROC of 0.96 versus 0.89.' The abstract states that for STEMI, the model's AUROC was 0.96 [0.94–0.98] versus physician interpretation's AUROC of 0.89 [0.85–0.93], p < 0.001. While the AUROC values are correct, the abstract does not explicitly state that the model *outperformed* physician interpretation, only that it achieved a higher AUROC. The 'Why It Matters' section states 'However, the model's performance in detecting NSTEMI-OMI cases within Occlusion MI detection was noted as a limitation.' The abstract states 'Occlusion MI detection reached AUROC 0.91 [95%-CI 0.89–0.93], though NSTEMI-OMI cases were frequently missed.' This implies a limitation but does not explicitly state it as a limitation in the same way the section does.)
Study Design
This study developed and validated a deep-learning model using convolutional neural networks to detect acute myocardial infarction (AMI) subtypes from digital 12-lead ECGs. The model was trained on 173,396 patient ECGs and underwent fine-tuning and internal validation in 7591 patients, followed by external validation in 4370 patients with suspected AMI.
Key Results
For non-ST-segment elevation MI (NSTEMI), the model achieved an AUROC of 0.81, with type 1 at 0.82 and type 2 at 0.75, and outperformed physician interpretation for ST-segment elevation MI (STEMI) with an AUROC of 0.96 versus 0.89. Occlusion MI detection reached an AUROC of 0.91, though NSTEMI-OMI cases were frequently missed in this context.
Why It Matters
This ECG-based deep learning model shows good discrimination and calibration across AMI subtypes, suggesting potential clinical utility for rapid risk stratification and early cardiology intervention. However, the model's performance in detecting NSTEMI-OMI cases within Occlusion MI detection was noted as a limitation.
Score 0.72 · ✓ Verified
Xiawei Li et al. — npj Digital Medicine · September 15, 2026
Why it’s here: This article uses deep learning (Vision Transformer) for automated pathology image segmentation to quantify a prognostic biomarker in pancreatic cancer, directly aligning with AI in clinical care and predictive models.
Study Design
This study developed and validated a Vision Transformer-based model for automated segmentation and quantification of the lymphocyte-stroma ratio (LSR) in pancreatic ductal adenocarcinoma (PDAC) whole-slide images (WSIs) from multiple centers. The model was trained and validated using WSIs to classify tissue and quantify LSR, addressing the limitations of manual assessment.
Key Results
The automated model achieved high accuracy in tissue classification and demonstrated that a higher LSR is significantly correlated with improved overall survival. Kaplan–Meier survival curves showed a significantly higher risk for the LSR-low group in both development (P = 0.002) and multicenter cohorts (P < 0.001).
Why It Matters
This study establishes a robust, automated LSR quantification system validated across multicenter PDAC cohorts, offering a scalable solution for precision oncology by linking computational biomarkers to therapeutic outcomes. The developed online platform provides visualization and quantification of tissue categories, enhancing clinical utility.
Score 0.66 · ✓ Verified
Tongjuan Zou et al. — Critical Care · September 12, 2026
Why it’s here: This study uses AI-assisted phenotyping for gastrointestinal dysfunction assessment in critically ill patients, directly applying AI to clinical care and prospective validation.
Study Design
This multicentre prospective observational cohort study enrolled adult ICU patients with acute brain injury (ABI) from three teaching hospitals in Sichuan, China. Gastric antral ultrasound videos and ECG recordings were acquired on ICU days 1, 2, 3, and 7 to analyze AI-assisted gastric antral rhythm phenotypes and heart rate variability (HRV).
Key Results
The gastroparesis-like phenotype, accounting for 37.6% of monitoring days, was associated with a higher GIDS total score (adjusted β = 0.293) and increased odds of GIDS ≥ 1 (adjusted OR = 1.991) compared to the regular phenotype. The gastroparesis-like phenotype also showed associations with higher mean heart rate, stress index, DFA α1, and lower SampEn and SDNN after multivariable adjustment.
Why It Matters
AI-assisted gastric antral rhythm phenotypes derived from continuous ultrasound CSA curves were associated with daily GIDS and selected HRV-derived measures in patients with ABI. However, these associations do not establish causality or diagnostic equivalence to clinical gastroparesis.
● Preprint · ✦ Something Different · Score 0.52 · ✓ Verified
Nosa-Ihaza, E. A. et al. — medRxiv Preprint · health systems and quality improvement · September 11, 2026
✦ A change of pace: This article explores antimalarial pharmacotherapy gaps in Nigerian children using predictive machine learning, offering a perspective on global health and infectious diseases that is distinct from the more common focus on Western healthcare systems and chronic conditions.
Study Design
This study analyzed data from the 2023-24 Nigeria DHS to assess antimalarial pharmacotherapy gaps in 3,962 children under five who reported fever. Four predictive modeling approaches, including logistic regression, elastic-net, random forest, and gradient boosting, were compared to predict two outcomes: Gap A (access) and Gap B (quality of treatment).
Key Results
The proportion of febrile children with access to care and testing was 16.0%, and 60.2% of those treated with antimalarials received an ACT. Caregiver-reported financial barriers were significantly associated with lower odds of ACT receipt (β = -0.97, p = 0.013), a finding consistent across most models. All eight models had AUCs between 0.48 and 0.62, with 95% CIs containing 0.50, indicating discrimination no better than chance.
Why It Matters
Limited household and maternal characteristics explained variation in the malaria care cascade, suggesting interventions should target point-of-sale costs within distribution channels. The predictive models' inability to confidently discriminate beyond chance highlights limitations in using standard survey data for this purpose. The consistent association between financial barriers and reduced ACT receipt points to cost as a key intervention target.
Hospital logistics and operations5 articles
Score 0.82 · ✓ Verified
Azam Nasrollahi et al. — BMC Health Services Research · September 15, 2026
Why it’s here: Constructs a framework for climate-resilient hospitals, covering domains like sustainable infrastructure, energy management, and waste management, which are operational aspects.
Study Design
This study conducted a systematic review in 2024, with a search rerun in January 2025, to construct a framework for hospital climate adaptation. The review included 30 studies published between 2000 and 2024, assessed using the QATSDD tool, and employed thematic synthesis for data extraction.
Key Results
The review identified seven key domains and 165 indicators for hospital climate adaptation. These domains encompass sustainable infrastructure, energy management, greenhouse gas emission reduction, leadership and governance, waste management, workforce capacity building, and education and awareness.
Why It Matters
The findings aim to assist managers and policymakers in developing frameworks to enhance hospital resilience and reduce environmental impacts, aligning healthcare with global climate initiatives. This systematic review protocol was prospectively registered in PROSPERO (CRD420251001110).
★ Flagship · Score 0.71 · ✓ Verified
Zhou Chen, Yichuan Ding, Luyi Yang — Operations Research · September 11, 2026
Why it’s here: Analyzes customer queueing behavior and service system design with implications for capacity-constrained services like healthcare wait lists.
Study Design
Chen, Ding, and Yang study customers who can join multiple provider queues, choose only their preferred provider, or decline service. They compare this flexible multilisting design with single-listing and pooling strategies. The analysis focuses on implications for throughput and social welfare in horizontally differentiated services.
Key Results
Multilisting attracts more customers and consistently achieves at least as much throughput as single-listing or pooling. While multilisting improves social welfare relative to pooling, the increased participation can worsen congestion and reduce welfare compared to single-listing. The analysis also shows that welfare-maximizing allocations may treat providers asymmetrically.
Why It Matters
The findings suggest that while multilisting expands access, it may not always improve social welfare due to increased congestion. Asymmetric prices can be used to guide customers toward socially efficient choices. These insights are applicable to designing capacity-constrained services like healthcare wait lists and public housing systems.
★ Flagship · Score 0.67 · ✓ Verified
Finite-Time Minimax Bounds and an Optimal Lyapunov Policy in Queueing Control
Yujie Liu, Vincent Y. F. Tan, Yunbei Xu — Operations Research · September 15, 2026
Why it’s here: Applies queueing theory and operations research to scheduling and performance guarantees, directly relevant to operations.
Study Design
This study develops a minimax framework for analyzing finite-time scheduling performance in systems with bursty, nonstationary demand. The framework quantitatively characterizes how expected total queue length scales with system capacity and variability. It provides a basis for evaluating and comparing scheduling policies.
Key Results
The authors introduce LyapOpt, a policy that accounts for both first- and second-order effects in service allocation, outperforming MaxWeight which optimizes only first-order drift. Under a specific structured condition, LyapOpt achieves optimal finite-time performance while retaining stability guarantees. In contrast, MaxWeight's finite-time performance scales suboptimally with system parameters.
Why It Matters
These results clarify the limitations of classical drift-based scheduling methods like MaxWeight. They motivate new queueing-control methods that offer rigorous finite-time guarantees. The framework provides a basis for evaluating and comparing scheduling policies in various infrastructure settings.
★ Flagship · Score 0.67 · ✓ Verified
The Dedicated Docket in U.S. Immigration Courts: An Analysis of Fairness and Efficiency Properties
Daniel Freund, Wentao Weng — Manufacturing & Service Operations Management · September 16, 2026
Why it’s here: This article uses queueing models to analyze the efficiency and fairness of a dedicated docket in immigration courts, which is relevant to queueing models and operational efficiency in a service setting.
Study Design
This study analyzes the operational implications of the U.S. immigration system's dedicated docket using a stylized queueing model. The model considers a policymaker routing asylees to either a regular or dedicated docket with a set delay target for the latter, and lawyers allocating their time to maximize successful cases.
Key Results
The dedicated docket system can Pareto-improve both speed and accuracy compared to a single docket, but only if the policymaker sets the delay target appropriately; otherwise, the system may be dominated by a single docket. Furthermore, the dedicated docket system can only satisfy three natural fairness rules if it is dominated by the single docket.
Why It Matters
The analysis reveals that the lack of fairness in the dedicated docket is a fundamental design flaw, yet the program enables surprising efficiency gains through delay differentiation. Policymakers and legal advocacy groups must recognize that this same delay differentiation, while boosting efficiency, inherently creates unfairness between dockets.
★ Flagship · Score 0.67 · ✓ Verified
Last-Mile Humanitarian Logistics Planning with Isolated Communities
Mahdi Noorizadegan et al. — Manufacturing & Service Operations Management · September 17, 2026
Why it’s here: This paper directly addresses humanitarian logistics planning, including staging areas, fleet allocation, and routing under uncertainty, highly relevant to logistics operations.
Study Design
This paper introduces a new Last-mile Humanitarian Logistics Planning problem to address aid delivery to isolated communities after disasters, motivated by hurricanes in Honduras. The problem jointly determines staging area location and capacity, heterogeneous fleet sizes (ground and aerial), mobile unit allocation, and routing to meet delivery time targets under demand and travel time uncertainty. It is formulated as a Parallel Drone–Vehicle Routing Problem with Location and Fleet-Size Decisions, modeled as a route-based mixed-integer program.
Key Results
A tailored Branch-and-Price algorithm is developed to solve the problem exactly, with a nonlinear pricing subproblem reformulated and efficiently solved by a customized dynamic programming approach. The complexity of the uncertainty model remains comparable to the deterministic counterpart, ensuring practicality and scalability. A case study and synthetic instances demonstrate the framework's versatility and practical relevance.
Why It Matters
The findings reveal critical trade-offs between robustness and investment, and highlight the effects of delivery time targets and isolation on network structure and fleet size. Analyses also show differences across uncertainty modeling approaches and the role of aerial fleet composition on operational efficiency. These insights offer actionable guidance for network configuration and fleet-sizing in humanitarian response settings with limited data and resources.
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 comment