This Week in Critical Care — May 21, 2026

Welcome to This Week in Critical Care. This week we're covering 10 notable papers spanning advanced bedside diagnostics, mechanical ventilation, novel therapies for lung disease, and systems-level innovations in resuscitation and artificial intelligence. Let's dive in.

Bedside Diagnostics for Pulmonary Embolism When a patient with suspected pulmonary embolism is too hemodynamically unstable for a trip to the CT scanner, what are our options at the bedside? Two papers in *Critical Care* this week explore advanced, non-invasive diagnostic tools for this exact scenario. First, a review article summarizes the latest American Heart Association guidelines, reinforcing the central role of point-of-care ultrasound, or POCUS [1]. The authors highlight that a multi-organ assessment—integrating cardiac, pulmonary, and venous ultrasound—can rapidly identify signs of right ventricular dysfunction or even thrombus in transit. This approach provides clinically actionable information to support timely decisions, like initiating thrombolysis, when definitive imaging is not immediately feasible.

Building on this theme, a second paper in *Critical Care* evaluates the accuracy of Electrical Impedance Tomography, or EIT, for detecting perfusion defects in PE [4].

Methods Investigators developed a novel “wasted-ventilation index” derived from EIT-based ventilation-perfusion maps. They first validated this in a piglet model of PE, comparing it against dynamic contrast-enhanced CT and CTPA. They then tested the index in a cohort of 66 patients with acute respiratory failure and 10 with chronic thromboembolic disease.

Results In animal models, the wasted-ventilation index was highly accurate, with an AUC of 0.99, demonstrating 96% sensitivity and 94% specificity for detecting pulmonary artery occlusions. In the human cohorts, the index also performed well, with an AUC of 0.92, corresponding to a sensitivity of 81% and a specificity of 94%. Notably, the index consistently decreased after patients received thrombolysis, suggesting it can track response to therapy. These papers together suggest that for the unstable PE patient, both multi-organ ultrasound and EIT are powerful bedside aids that can guide critical decisions when the gold standard is out of reach.

Refining Mechanical Ventilation Next, we turn to mechanical ventilation, with two papers that quantify the risks of ventilator support and explore new ways to monitor patient-ventilator interaction. A systematic review and meta-analysis in *Critical Care Medicine* investigates the association between mechanical power and mortality in invasively ventilated adults [7].

The Study Researchers analyzed 34 studies, pooling data to compare mechanical power between survivors and non-survivors. Mechanical power is an index that combines variables like tidal volume, respiratory rate, driving pressure, and PEEP into a single measure of energy delivered to the respiratory system per minute.

Results The analysis found that mechanical power was consistently and significantly higher in non-survivors compared to survivors, with a mean difference of about 1.9 Joules per minute. This association held even after normalizing for body weight or respiratory system compliance. Each 1 Joule per minute increase in mechanical power was associated with a 4% increase in the odds of death. The authors also found that a mechanical power threshold above 17 Joules per minute was associated with a 60% higher odds of mortality. While this highlights mechanical power as a marker of ergotrauma, the authors caution that prospective trials are needed to see if actively lowering it improves outcomes.

So, if high mechanical power is a concern, how can we better monitor its components, such as patient effort? A study also in *Critical Care Medicine* details a novel artificial intelligence algorithm designed to do just that [3]. Current methods to estimate inspiratory muscle pressure are either invasive, requiring an esophageal catheter, or intermittent, requiring occlusion maneuvers. This AI algorithm was designed to provide a continuous, non-invasive estimate in real-time from ventilator waveform data. In a prospective study of 48 ICU patients, the AI-estimated muscle pressure showed good agreement with the gold-standard esophageal manometry. The algorithm could also reliably detect high or low patient effort and identify patient-ventilator dyssynchronies like ineffective efforts or autotriggering with a sensitivity of over 86%. This tool represents a potential step forward, moving from intermittent snapshots of patient effort to continuous, real-time monitoring at the bedside, which could help clinicians better manage the contributors to mechanical power.

New Biologics for Chronic Lung Disease This week brings several important updates on biologic therapies for severe pulmonary diseases. First, for chronic obstructive pulmonary disease, or COPD, we have contrasting results from two different agents. A pooled analysis from the BOREAS and NOTUS trials, published in the *American Journal of Respiratory and Critical Care Medicine*, evaluated dupilumab in COPD patients with type 2 inflammation, defined as a blood eosinophil count of 300 or more [8]. The results were compelling. Over 52 weeks, dupilumab reduced emergency department visits or hospital admissions by 38% compared to placebo. It also reduced the risk of a first such event by 45%. Furthermore, for patients who did have an exacerbation, those on dupilumab required significantly fewer systemic corticosteroids. This provides strong evidence for dupilumab in this specific patient subgroup.

In contrast, a pair of trials in *The Lancet* report more mixed results for astegolimab, an antibody targeting the ST2 receptor in the IL-33 pathway [10]. Unlike the dupilumab trials, the ALIENTO and ARNASA trials enrolled COPD patients with a history of frequent exacerbations regardless of their baseline eosinophil count. In the ALIENTO trial, astegolimab given every two weeks was associated with a 15% reduction in the annual rate of moderate or severe exacerbations, a result that was statistically significant. However, the every-four-week dose showed no significant benefit. In the second trial, ARNASA, neither dosing regimen met statistical significance for the primary endpoint, though the every-four-week dose showed a trend towards benefit. Taken together, these findings suggest a potential, albeit modest, role for targeting the ST2 pathway in a broad COPD population, but the benefit is not as clear-cut as that seen with dupilumab in a more targeted, eosinophilic phenotype.

Finally, for pulmonary arterial hypertension, or PAH, a study in *Science Translational Medicine* provides a fascinating look at the mechanism of sotatercept [6]. Sotatercept is a promising new therapy for PAH, thought to work by rebalancing BMPR-II signaling. This study used a transcriptomic biomarker panel in peripheral blood to measure target engagement in nine patients treated with the drug. Counterintuitively, the analysis suggested that sotatercept did not increase BMPR-II pathway signaling. Instead, it led to a reduction, possibly because it was depleting circulating BMP9 and BMP10, which are ligands for the receptor. This work underscores the complexity of these signaling pathways and the critical need for biomarkers to understand how our therapies actually work in vivo.

Systems and Strategies in Critical Care Our next theme covers large-scale systems, one for delivering advanced resuscitation and another for building reliable AI models. From the journal *Resuscitation*, the FLIGHT-to-ECPR study evaluated a helicopter-based pathway to expand access to extracorporeal cardiopulmonary resuscitation, or ECPR, for patients in rural areas with refractory out-of-hospital cardiac arrest [2]. The hypothesis was that using a helicopter could shorten transport times enough to keep patients within the critical 60-minute window of low-flow time. In this prospective study, 45 patients were activated for the helicopter pathway, with 27 ultimately receiving ECPR. When compared to a matched cohort of urban patients transported by ground, the low-flow times were similar. Most importantly, the primary outcome—survival to discharge with a favorable neurologic status—was identical between the helicopter and ground transport groups, at just under 26% for cannulated patients. This study demonstrates that a helicopter-facilitated, hospital-based ECPR strategy is feasible and can safely expand access to this advanced resuscitation for rural populations without compromising outcomes.

Shifting from logistical systems to data systems, a study in *Critical Care* addresses a major challenge for artificial intelligence in the ICU: generalizability [9]. Most AI models for predicting outcomes like mortality or length of stay are trained on data from a single hospital or database, and their performance often drops significantly when applied elsewhere. Investigators used four large, international ICU databases to compare different training strategies. They found that models trained on a single source performed poorly on external validation. The most effective strategy was data pooling—combining data from multiple databases to train a single, robust model. This approach yielded the best performance for both generalization to new sites and specialization for specific sites. The authors advocate for a “Data-Centric AI” approach, emphasizing that the key to building reliable clinical AI is not just a better algorithm, but better, more diverse, and openly shared data.

From the Bench: Rethinking a 'Toxic' Metabolite Finally, we have a fundamental science paper from *Nature* that challenges a long-held view of a specific metabolite, with potential long-term implications for our understanding of critical illness [5]. The metabolite is L-2-hydroxyglutarate, or L-2-HG. It’s primarily known for accumulating in a rare genetic disease, causing severe neurologic problems, and has therefore been classified as a toxic byproduct. This study provides compelling evidence that L-2-HG is, in fact, a physiological signaling molecule. Its levels are actively regulated by the mitochondrial NADH-to-NAD ratio, and it acts on specific enzymes to regulate gene expression. To prove its importance, the researchers overexpressed the enzyme that breaks down L-2-HG in mice, systemically reducing its levels. These mice had impaired postnatal growth, kidney damage, and increased mortality. This work establishes that a metabolite previously regarded as purely toxic has a crucial physiological function, a paradigm that may apply to other metabolic derangements we see in the ICU.

Editor's Pick If you only have time for one paper this week, make it the meta-analysis on mechanical power and mortality in *Critical Care Medicine* [7]. This paper synthesizes data from 34 studies to confirm a strong, consistent association between higher energy delivery and worse outcomes. It provides a useful, albeit preliminary, benchmark of 17 Joules per minute that can help clinicians frame the risk of ergotrauma at the bedside.

Clinical Bottom Line Here are the key takeaways from this week in Critical Care. First, for unstable patients with suspected pulmonary embolism, bedside multi-organ ultrasound and Electrical Impedance Tomography are emerging as valuable diagnostic tools when CT imaging is not a safe option. Second, higher mechanical power is consistently associated with increased mortality in ventilated patients. While not yet a proven therapeutic target, a value over 17 Joules per minute should raise concern for ergotrauma. Third, in patients with COPD and type 2 inflammation, defined by an eosinophil count of 300 or more, dupilumab significantly reduces severe exacerbations and the need for systemic corticosteroids. Biologics targeting other pathways in broader COPD populations, like astegolimab, have shown more modest or mixed results. Fourth, helicopter-based transport for ECPR can successfully expand access to advanced resuscitation for rural out-of-hospital cardiac arrest patients, achieving neurologic outcomes comparable to urban systems. And finally, for artificial intelligence models to be reliable in clinical practice, they must be trained on diverse, multi-center data. Models developed at a single institution are unlikely to perform well when deployed externally.

That's your roundup for This Week in Critical Care. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. See you next week.