18 articles · 3 topics
Critical care medicine6 articles
Score 0.94 · ✓ Verified
Kaylee T. Stebbins et al. — Critical Care Medicine · August 31, 2026
Why it’s here: This systematic review and meta-analysis directly addresses long-term outcomes of ECMO in critically ill adults, a core topic of interest.
Study Design
This systematic review and meta-analysis evaluated long-term mortality and functional outcomes in critically ill adults treated with ECMO. Data were extracted from 163 studies including 78,053 adults, with mortality pooled using random-effects meta-analysis, while functional outcomes were narratively synthesized due to heterogeneity.
Key Results
At 1 year, pooled mortality was 37.2% for venovenous ECMO, 55.2% for venoarterial ECMO, and 74.4% for extracorporeal cardiopulmonary resuscitation (ECPR). Functional outcomes were reported in 23.3% of studies for 7,876 survivors, assessed by various instruments.
Why It Matters
Long-term mortality after ECMO varies significantly by modality, highlighting the need for better understanding of functional recovery. Inconsistently reported functional outcomes limit the ability to inform patient selection and identify factors for improved recovery.
Score 0.90 · ✓ Verified
José RAMÍREZ-ESTEBAN et al. — Critical Care · August 31, 2026
Why it’s here: This systematic review and meta-analysis examines the effect of PEEP on ICP and CPP in acute brain injury, highly relevant to mechanical ventilation and hemodynamic support in critical care.
Study Design
This systematic review and dose-response meta-analysis investigated the relationship between positive end-expiratory pressure (PEEP) and intracranial pressure (ICP), cerebral perfusion pressure (CPP), and brain-tissue oxygenation (PbtO₂) in adults with acute brain injury. The primary analysis focused on within-patient PEEP-titration studies, with between-patient and observational studies forming a secondary layer. Study-clustered multilevel meta-regression was used to estimate the slope of each outcome per cmH₂O of PEEP.
Key Results
ICP rose by +0.118 mmHg per cmH₂O of PEEP (p = 0.006), a change well below the minimal important difference (MID) of 2 mmHg, with no departure from linearity up to 21 cmH₂O. CPP fell by -0.177 mmHg per cmH₂O (p = 0.038), less than one fifth of the 5 mmHg MID. PbtO₂ was not estimable, and evidence above 15 cmH₂O PEEP was limited.
Why It Matters
The findings suggest that increasing PEEP is associated with a small, linear increase in ICP and a corresponding small change in CPP, generally below clinically important thresholds. Exploratory analyses hypothesize larger effects in patients with non-recruitable lungs, a stiff chest wall, or reduced intracranial compliance, warranting prospective, phenotype-stratified testing. The certainty of evidence for both ICP and CPP was rated as low.
Score 0.82 · ✓ Verified
Christian Karagiannidis et al. — Critical Care · September 3, 2026
Why it’s here: This randomized controlled trial investigates extracorporeal CO2 elimination for acute exacerbation of COPD requiring mechanical ventilation, directly matching the researcher's interests.
Study Design
This randomized controlled trial evaluated extracorporeal CO2 removal (ECCO2R) for adults with acute hypercapnic respiratory failure due to AE-COPD requiring invasive mechanical ventilation (IMV) who failed or were ineligible for extubation within 24 hours. Patients were randomized to ECCO2R or standard IMV, with the primary endpoint being death or severe disability at day 60.
Key Results
The trial was terminated early with only 18 patients randomized, showing 0/8 ECCO2R-treated patients experiencing the primary composite endpoint versus 3/9 IMV-treated patients (33%). IMV duration was significantly shorter in the ECCO2R group (7.1 ± 2.0 days vs. 24.3 ± 21.4 days; p = 0.043), with a trend towards more device-support-free days at day 29 (17 ± 4 vs. 8 ± 7 days; p = 0.011).
Why It Matters
Although prematurely terminated with a very small sample size, the trial suggests ECCO2R-facilitated early extubation may shorten IMV duration and numerically favor other secondary outcomes. However, interpretation is limited, and adequately powered multicenter trials are warranted to confirm these findings.
Score 0.78 · ✓ Verified
Physiological limits of capillary refill time: a human endotoxemia study
H. GUSTAVSSON et al. — Critical Care · August 31, 2026
Why it’s here: This human endotoxemia study investigates capillary refill time as a marker of perfusion, directly relevant to hemodynamic monitoring in critically ill patients.
Study Design
This double-blind, randomized, crossover study used quantitative capillary refill time (qCRT) measured by polarized reflectance imaging in 25 healthy volunteers. Participants received either lipopolysaccharide or saline placebo, with qCRT measured at multiple time points throughout the study.
Key Results
Finger qCRT followed a biphasic course, prolonging from a baseline near 3.5 s to 8.31 s at 1.5 h, with 14 of 25 reaching the 10-s ceiling, then shortening to 1.27 s at 5 h. These changes crossed the 3-s clinical threshold in both directions within a single inflammatory episode, with significant contrasts against placebo at both time points.
Why It Matters
Finger qCRT tracks acute systemic inflammation but crosses the 3-s threshold variably, suggesting a single fixed threshold is unreliable. Resting values are highly variable and strongly dependent on skin temperature, indicating that change over time within a participant is more informative. These findings in healthy participants require confirmation in patient studies before clinical application and are not directly generalizable to septic shock.
Score 0.76 · ✓ Verified
K. Lindup et al. — Critical Care · September 1, 2026
Why it’s here: This study focuses on non-invasive assessment of inspiratory effort during mechanical ventilation in critically ill patients with acute hypoxemic respiratory failure.
Study Design
This prospective physiological validation study evaluated the accuracy of the Mixed Integer Multi-Effort (MIME) method for non-invasively estimating inspiratory muscle pressure (Pmus) during pressure support ventilation (PSV). The study included adult patients with acute hypoxemic respiratory failure (AHRF) within 48 hours of transitioning from controlled mechanical ventilation to PSV, analyzing standard ventilator signals offline.
Key Results
MIME-derived Pmus estimates showed a mean difference of 0.4 cmH₂O with limits of agreement ranging from -3.4 to 4.2 cmH₂O compared to esophageal pressure-derived Pmus on a breath-by-breath basis. Averaged values over multiple breaths narrowed the limits of agreement to -2.2 to 3.0 cmH₂O. MIME demonstrated good discriminative performance for detecting low inspiratory effort (AUROC 0.88) and high inspiratory effort (AUROC 0.83), performing similarly to end-expiratory occlusion pressure (ΔPocc) for low effort detection, while ΔPocc performed better for high effort detection.
Why It Matters
The MIME method offers a non-invasive approach to estimate inspiratory effort during PSV, potentially complementing tools like ΔPocc for longitudinal assessment. However, further validation in broader populations and physiological settings, along with assessment of convergence and computational performance, is needed before real-time bedside implementation.
✦ Something Different · Score 0.58 · ✓ Verified
Nicola Gabriela Dymek et al. — BMC Health Services Research · September 5, 2026
✦ A change of pace: This study explores implementation science in neonatal intensive care units, offering a novel methodological perspective distinct from the clinical focus of other articles.
Study Design
This study introduces and illustrates a six-step methodology combining System Support Mapping (SSM) and the Matrixed Multiple Case Study (MMCS) approach. Semi-structured interviews with key persons from five German neonatal intensive care units (NICUs) during the early intervention phase served as the empirical basis.
Key Results
The methodology generated individual System Support Maps and a sortable cross-site matrix, revealing context-dependent perceptions of determinants where resources could be facilitating or inhibiting. The cross-site analysis highlighted homogeneous and heterogeneous determinant patterns, indicating different contextual manifestations across sites.
Why It Matters
The combined SSM and MMCS approach offers a structured methodology for identifying and analyzing implementation determinants, accounting for individual perspectives and enabling systematic comparison. This distinction between determinant patterns can inform the development of differentiated cross-site and site-specific implementation strategies.
Healthcare machine learning6 articles
★ Flagship · Score 0.87 · ⚠ Check
Michael Ko et al. — The Lancet Digital Health · September 1, 2026
Why it’s here: This article develops and validates a foundation model using contrastive learning on ECGs for predicting cardiovascular diseases and outcomes.
⚠ a section is not grounded in the abstract (The 'Why It Matters' section contains the sentence 'The model's ability to leverage single ECG leads, even those typically considered less informative, further enhances its clinical utility.' While the abstract mentions 'The model has good performance in the detection of anterior and inferior acute myocardial infarction when using single ECG leads, outperforming supervised baseline models when using leads typically considered less informative,' the 'Why It Matters' section generalizes this to all clinical utility, which is not directly and unambiguously supported by the abstract.)
Study Design
This retrospective study developed and validated ECG-CLIP, a foundation model for ECG analysis, using a two-stage pretraining framework combining ECG data with textual annotations. The model was pretrained on over 1.7 million ECGs and paired clinician-overread ECG reports from the Scripps Health GE MUSE system, collected between January 15, 2008, and January 15, 2019. External validation was performed on the MIMIC-IV dataset for various cardiovascular disease and adverse health outcome prediction tasks.
Key Results
ECG-CLIP demonstrated superior performance across all detection and prediction tasks compared to supervised and non-ECG foundation models, achieving high label efficiency. For instance, it reached the same AUC as the best comparator model with 90.8% less training data in cardiovascular disease detection. In low-label settings with only ten positive labels, ECG-CLIP significantly outperformed the runner-up on acute myocardial infarction detection (AUC 0.910 vs 0.884) and atrial fibrillation prediction (AUC 0.777 vs 0.765).
Why It Matters
ECG-CLIP represents a generalizable foundation model capable of learning clinically meaningful ECG representations for diverse diagnostic and prediction tasks, enabling accurate and data-efficient risk predictions. This approach supports scalable deployment in real-world and resource-limited settings by facilitating interpretable risk predictions across various clinical tasks. The model's ability to leverage single ECG leads, even those typically considered less informative, further enhances its clinical utility.
Score 0.84 · ✓ Verified
Artificial intelligence tools in sepsis prediction: a systematic review and meta-analysis
Aadith Ashok et al. — npj Digital Medicine · August 31, 2026
Why it’s here: This systematic review and meta-analysis evaluates the accuracy of machine learning models for sepsis prediction, directly aligning with the researcher's interest.
Study Design
This systematic review and network meta-analysis evaluated the diagnostic accuracy of machine learning (ML) models for sepsis prediction in adults. The study searched five major databases for studies comparing ML models against comparators, synthesizing data using random effects models and meta-regression. The PROBAST-AI tool was used to assess risk of bias across 53 studies encompassing over 7 million patient encounters.
Key Results
The best-performing ML models achieved a pooled area under the receiver operator curve of 0.88 (95% CI [0.86, 0.90]), significantly outperforming traditional comparators by a mean difference of 0.119 (p < 0.001). Pooled sensitivity was 77.2% and pooled specificity was 84.7%, with decision tree and ensemble models showing the highest performance in the network meta-analysis.
Why It Matters
While ML models demonstrate statistically superior performance for sepsis prediction, with decision trees performing best, substantial heterogeneity (I²>95%) and a high risk of bias in most studies are critical barriers to translation. Wide prediction intervals and the high risk of bias highlight the need for future research to prioritize standardized validation and prospective trials to establish real-world impact.
Score 0.84 · ✓ Verified
AI models for medication adherence prediction: closing the gap to clinical readiness
Alys Cowdy et al. — npj Digital Medicine · September 2, 2026
Why it’s here: This article reviews AI models for medication adherence prediction and discusses their readiness for clinical application, directly aligning with the researcher's interest.
Study Design
This systematic review assessed 41 studies on AI models for medication adherence prediction. The review utilized the PROBAST + AI framework to evaluate model development and data across participants, predictors, outcomes, and analyses.
Key Results
The majority of models (71%) exhibited significant concerns regarding development quality, and 80% showed a high risk of bias in their evaluation. These issues were frequently linked to poorly defined adherence outcomes, insufficient handling of missing data, and limited validation.
Why It Matters
Methodological rigor, not algorithm complexity, appears to be the primary obstacle to translating these models into clinical practice, as discrimination did not consistently improve with more complex algorithms. The findings highlight the need for improved robustness, interpretability, and clinical actionability in future adherence prediction models.
Score 0.82 · ✓ Verified
Dajun Jiang et al. — npj Digital Medicine · September 1, 2026
Why it’s here: This article develops a multi-modal framework using CT and EHR data for predicting conversion to arthroplasty in hip fractures, demonstrating clinical prediction.
Study Design
This study introduces MMHIP, a multi-modal framework for predicting early conversion to arthroplasty in patients with intracapsular hip fractures. The framework integrates preoperative pelvic CT scans with electronic health record (EHR) data, trained on a large multi-center development cohort (n=951) and externally validated (n=240). This represents the largest multi-center cohort to integrate 3D CT and long-term follow-up scans for this prediction task.
Key Results
MMHIP achieved an AUROC of 0.920 in development and 0.873 on external validation, significantly outperforming image-only and EHR-only models. Incorporating CT data improved discrimination over EHR-based models, with absolute AUROC gains of 0.150–0.252 in development and 0.151–0.221 externally. At a specific operating threshold, MMHIP achieved a positive predictive value (PPV) of 0.901 and maintained a PPV of 0.900 on external validation.
Why It Matters
These findings support MMHIP as an interpretable preoperative risk-stratification model that provides clinically relevant prognostic information. This may aid in individualized, clinician-led treatment decisions for patients with intracapsular hip fractures. The model's ability to integrate CT data with EHRs advances beyond conventional approaches for this patient population.
Score 0.76 · ✓ Verified
Runhuang Yang et al. — npj Digital Medicine · September 1, 2026
Why it’s here: This article uses deep learning on retinal images to predict obesity subtypes associated with cardiometabolic risk, demonstrating real-world deployment.
Study Design
This study developed ROSA (Retinal-based Obesity Subtyping Algorithm), a deep learning framework utilizing knowledge distillation from multimodal cardiometabolic data and retinal images. The framework was applied to 63,693 participants across the UK Biobank and a Beijing Health Management Cohort external validation set.
Key Results
ROSA classified participants into three obesity subtypes: baseline concordant, discordant inflammatory, and discordant hyperglycaemic, achieving a macro-averaged AUROC of 0.66 in the external validation cohort. In the independent external BHMC-EV cohort, class-specific external sensitivity and PPV were 0.84 and 0.87 for baseline concordant (BC), 0.02 and 0.73 for discordant inflammatory (DIS), and 0.51 and 0.34 for discordant hyperglycaemic (DHG), respectively.
Why It Matters
ROSA demonstrates that retinal imaging can capture cardiometabolic heterogeneity within obesity, with the hyperglycaemic subtype associated with increased type 2 diabetes risk and accelerated cardiometabolic disease progression. However, the limited sensitivity of ROSA indicates that further calibration and validation are needed before clinical deployment.
✦ Something Different · Score 0.58 · ✓ Verified
Differences in tone of AI and care team responses to patient messages by patient demographics
Angela Mastrianni et al. — npj Digital Medicine · September 3, 2026
✦ A change of pace: This article offers a unique perspective on AI in healthcare by examining the nuances of human-AI interaction and communication, which is a departure from purely technical or diagnostic applications.
Study Design
This retrospective observational study evaluated demographic differences in the tone of AI-generated draft replies (AI-GDRs) and care team responses to patient portal messages. The study included 12,202 message triads from three internal and family medicine practices in New York City.
Key Results
AI-GDRs had lower odds of including polite language for Hispanic patients compared to White patients, and lower odds of conveying positive affect for Hispanic patients, those preferring non-English, those assigned female at birth, or those in lower-income areas. Care team responses also showed lower odds of conveying positive affect for Hispanic patients or those preferring a non-English language.
Why It Matters
These findings highlight the need for careful AI implementation to prevent the introduction or amplification of inequities in patient-provider communication. Some observed tone differences in AI-GDRs were also present in care team responses, suggesting potential systemic issues.
Hospital logistics and operations6 articles
Score 0.80 · ✓ Verified
Yin Yuan et al. — BMC Health Services Research · August 31, 2026
Why it’s here: This systematic review directly addresses systems engineering and design approaches applied to healthcare services and systems, aligning with operations research principles.
Study Design
This systematic review synthesized findings from 32 reviews, which collectively included 1127 studies, focusing on systems engineering and design methodologies applied to healthcare services and systems. The review excluded primary studies or those focused on specific interventions like healthcare apps, instead concentrating on secondary studies of systems-level applications. Data were extracted using a pre-developed framework to analyze approaches across micro, meso, and macro healthcare levels.
Key Results
The synthesis revealed an increasing application of systems engineering and design in healthcare, with service design, improvement, and development being the most common areas. User-centered and co-design approaches were found to enhance patient responsiveness, while systems approaches aided in managing complexity at policy and organizational levels. Methodological distinctions in structure, user engagement, and intervention level underscore the need to align choices with healthcare context and stakeholder needs.
Why It Matters
The findings highlight that user-centered and co-design methods are most effective for patient-level challenges, while systems thinking and structured stakeholder engagement are crucial for system-level transformation. A developed conceptual framework maps how these methodologies operate across micro, meso, and macro healthcare levels, guiding the matching of methodological choice to context. However, significant gaps remain in outcome standardization, implementation evaluation, and adaptation for low-resource or diverse settings.
Score 0.70 · ✓ Verified
Operating distribution center networks at scale: Efficient formulations and policies
Chenxi Li et al. — Production and Operations Management · September 4, 2026
Why it’s here: This paper develops efficient formulations and policies for operating distribution center networks at scale, directly relevant to hospital supply chain and logistics management.
Study Design
This study addresses the challenge of distributing inventory across large networks of distribution centers for retailers and sellers facing volatile demand. The authors develop distributionally robust network inventory management policies with cross-fulfillment using a scenario-based approach, incorporating feature information and demand correlation.
Key Results
The key technical development is a computationally efficient linear programming formulation for large networks, integrated into a dynamic inventory policy with look-ahead approximation. Validation using real-world data from China demonstrated that the proposed policy outperforms existing data-driven approaches, achieving operating cost savings of up to 19% compared to the company's current policy.
Why It Matters
The findings offer a scalable and efficient solution for inventory allocation in complex distribution networks, demonstrating significant cost reduction potential. The study analytically shows the value of scenario modeling for inventory problems, providing a practical framework for improving operational efficiency in logistics.
● Preprint · Score 0.68 · ✓ Verified
McHenry, R. D. et al. — medRxiv Preprint · emergency medicine · August 31, 2026
Why it’s here: This article investigates the impact of point-of-care testing on Emergency Department (ED) flow and crowding, directly addressing ED operations and patient throughput.
Study Design
This multi-centre controlled interrupted time series (CITS) study evaluated the impact of point-of-care (POC) high-sensitivity troponin testing and reduced sampling intervals in two large urban intervention Emergency Departments (EDs) compared to one untreated control ED. Analyses used a 120-day window on either side of implementation and segmented ITS models, with permutation tests against 147 pre-intervention placebo dates.
Key Results
The intervention produced no statistically significant change in any whole-ED flow metric (admissions, mean occupancy, maximum occupancy, or mean length of stay) when compared to the untreated control. While one intervention site initially appeared to show reductions in mean and maximum occupancy (-6.08 and -7.60 respectively), these attenuated to the null once the control was applied, with the untreated control showing similar reductions.
Why It Matters
POC cardiac biomarker testing and reduced sampling intervals did not demonstrably improve whole-ED flow, highlighting the vulnerability of uncontrolled interrupted time series designs to confounding in complex healthcare systems. Researchers must use concurrent controls and falsification tests for evaluating operational interventions, and policymakers should be aware of system-wide shocks when interpreting single-site before-and-after studies.
● Preprint · Score 0.68 · ✓ Verified
Wen, J. et al. — medRxiv Preprint · health policy · September 4, 2026
Why it’s here: This study evaluates the effect of surgical hubs on elective surgery volume, length of stay, and waiting times, directly addressing hospital capacity and operational efficiency.
Study Design
This retrospective longitudinal study used Hospital Episode Statistics from NHS acute trusts in England between April 2014 and September 2024. It compared 76 trusts, with 29 opening surgical hubs for trauma and orthopaedic surgery against 47 that did not, using a difference-in-differences approach with trust and quarter fixed effects.
Key Results
Surgical hub opening was associated with a 19.1% increase in hip and knee replacement procedures per quarter (43.75 procedures) and a 7.8% reduction in mean length of stay (0.32 days lower). However, there was no statistically significant overall change in mean waiting time from addition to the waiting list to admission.
Why It Matters
The findings suggest that surgical hubs can effectively increase the volume of elective hip and knee replacement surgeries and shorten hospital stays. While hubs increase surgical activity, this additional capacity did not translate into a statistically significant overall reduction in waiting times, indicating potential limitations in directly addressing waiting lists.
Score 0.68 · ✓ Verified
Xiangjie Zhao et al. — Production and Operations Management · September 4, 2026
Why it’s here: Directly applies queueing theory and arrival time analysis to a hospital department, relevant to patient flow and scheduling.
Study Design
This study examines customer strategic arrival times in a queueing game, building on a mixed strategy framework. Using a dynamic choice model validated with data from a Chinese hospital's endoscopy department, the research investigates transient system behavior with irrational customers and repeated game settings.
Key Results
The study proves the existence of a unique equilibrium and system convergence under fluid and M t / M /1 queueing models. Analysis of the arrival time distribution at equilibrium indicates that optimal patient arrival occurs at the start or near the end of the morning session, depending on model parameters.
Why It Matters
The findings suggest that service providers can decrease social cost by strategically modulating waiting time information shared with patients. Counter-intuitively, increasing earliness cost can also reduce social cost at equilibrium, highlighting a complex interplay of cost parameters.
✦ Something Different · Score 0.56 · ✓ Verified
Electric ambulances: will the need for charging affect response times?
Nanne A. Dieleman, Caroline Jagtenberg — Health Care Management Science · September 1, 2026
✦ A change of pace: This article explores the practical operational challenges of adopting new technology like electric ambulances, offering a novel perspective on the trade-offs between sustainability and core service delivery.
Study Design
This paper introduces a novel application of discrete-event simulation to assess the impact of electric ambulance charging times on response times. The study developed an open-source decision support system called ELASPY, built in Python, to simulate emergency response processes. ELASPY was used to predict optimal battery capacities and charger locations for electric ambulance fleets to match diesel fleet response times.
Key Results
The study's results did not reveal concerning signals for daily operations under expected operating conditions, indicating comparable response times for electric ambulances. However, the research highlights potential issues for even low busyness ambulance providers during calamities such as prolonged power outages.
Why It Matters
These findings are important for ambulance providers considering the transition to electric vehicles, as they provide a tool to assess potential impacts on response times. The authors encourage decision makers to use the publicly available ELASPY software with their own historical data to evaluate this transition in their specific regions. A key caveat is the need to consider extreme events like prolonged power outages, even for less busy operations.
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