This Week in Critical Care — Sep 15, 2026

Welcome to This Week in Critical Care. This week we're covering 10 notable papers spanning resuscitation and mechanical circulatory support, the shifting evidence around sepsis targets and phenotype-guided therapy, and a cluster of studies on mechanical ventilation, organ support and long-term recovery. Let's dive in.

We start with resuscitation, where one of the most-debated procedural questions of the last two years now has a pooled answer. In Resuscitation, Holmberg and colleagues report an individual participant data meta-analysis of the randomised trials comparing an intraosseous-first with an intravenous-first access strategy in out-of-hospital cardiac arrest [1]. Two trials contributed data on just over seven and a half thousand patients, analysed with adjustment for prognostic factors and supplemented by prespecified Bayesian modelling. Thirty-day survival was essentially identical between strategies, with moderate certainty evidence, and favourable neurological outcome likewise showed no significant difference. For sustained return of spontaneous circulation the point estimate leaned slightly toward intravenous access, but that difference was not statistically significant and rested on low-certainty evidence. The Bayesian analyses suggested that whatever difference exists between the two routes is small. The practical reading is that neither route should be treated as mandatory doctrine — the operator who can reliably establish access fastest in the hands available is still the right answer, and drug delivery route is not where the remaining survival gain in out-of-hospital arrest is hiding.

Staying with arrest but moving to the cath lab and cardiac intensive care unit, also in Resuscitation, Arai and colleagues analysed the Japanese percutaneous ventricular assist device registry, covering more than eight and a half thousand patients in cardiogenic shock supported with Impella [10]. They split patients by arrest presentation, and the gradient is striking. Thirty-day mortality was about 29 percent in those with no arrest, roughly 42 percent after out-of-hospital arrest with return of spontaneous circulation, about 48 percent in those who arrested in hospital before device implantation, and just over 60 percent in out-of-hospital arrest without return of circulation. After adjustment, in-hospital arrest before implantation carried roughly a fifty percent higher hazard of death than no arrest at all, along with more major bleeding, device-related bleeding, limb ischaemia, stroke and sepsis. Within that in-hospital arrest group, older age, lactate above four, creatinine above one point five, longer low-flow time and the addition of veno-arterial extracorporeal support all marked worse outcomes, while myocarditis as the aetiology and a defibrillated rhythm marked better ones. The message is that arresting on the ward or in the unit before mechanical support is not a milder version of out-of-hospital arrest — it defines a distinct, very high-risk shock phenotype that deserves earlier escalation decisions and franker conversations about ceilings of care.

The second theme is sepsis and shock, where three papers converge on the same uncomfortable point: our targets and our therapies are only as good as the patient we apply them to. In Critical Care Medicine, Mendes and colleagues pooled seven randomised trials and just over twenty-four hundred patients with septic shock comparing peripheral perfusion-guided resuscitation — essentially capillary refill time — against usual care or lactate-guided care [4]. In the primary frequentist analysis, twenty-eight-day mortality was about thirteen percent lower in relative terms, but the confidence bounds crossed no effect, so the result was not statistically significant, and the two largest and lowest-risk trials were individually negative. Heterogeneity was zero, and the prespecified Bayesian model was directionally concordant, but the authors are explicit that the magnitude of any benefit remains uncertain and adequately powered trials are still needed. So capillary refill remains a reasonable, free, bedside adjunct — not yet an evidence-based replacement for lactate. Alongside that, in the Journal of Critical Care, Zarama and colleagues used a nationally representative survey of United States emergency department visits from 2012 to 2022, covering some 820 million weighted visits [6]. Coded sepsis identification nearly tripled over the decade, rising to about two and a half percent of adult visits, and delivery of the full four-component bundle — fluids, antibiotics, blood cultures and lactate — climbed from under three percent to a peak of about twenty percent, with a sharp step up around 2016 when national reporting requirements landed. But twenty percent is still the ceiling, lactate remained the least often delivered element despite rising fastest, and delivery was far more likely in patients who looked severely ill at triage and markedly less likely when the source was intra-abdominal. The authors themselves question whether an all-or-nothing composite is the right quality metric, arguing that individual evidence-based components may serve patients better.

That theme of choosing the right patient runs straight into a narrative review in Shock, where García-Ramos and colleagues synthesise a decade of contradictory evidence on polymyxin B haemoadsorption in septic shock and propose a phenotype-guided pathway [3]. They argue the responder phenotype converges on three domains — detectable endotoxaemia on the endotoxin activity assay, intermediate organ dysfunction with a sequential organ failure assessment score in the seven to thirteen range, and an intra-abdominal source — with early initiation as a fourth operational criterion. They cite the Tigris Bayesian phase three trial's reported absolute reduction in ninety-day mortality of around fifteen percentage points, and cohort data suggesting the survival signal concentrates in intra-abdominal infection, where the adjusted hazard of death was roughly halved. Two cautions matter here. This is a narrative synthesis with a five-step algorithm derived by the authors, not a guideline, and without formal external consensus or validation — which they acknowledge as the necessary next step. It is worth noticing the tension across journals this week: the same abdominal source that identifies a possible haemoadsorption responder is also the presentation least likely to get a sepsis bundle in the emergency department.

The third theme is mechanical ventilation, and two papers reframe how we think about failure to liberate. In Intensive Care Medicine, Caldecott and colleagues report a secondary analysis of the WEAN SAFE cohort, more than five and a half thousand patients [2]. About a fifth of the cohort never underwent a separation attempt at all; among those who did, around fifteen percent had still failed to wean by day ninety. Those failed-wean patients had the longest ventilation and the longest unit stay, underwent more separation attempts but fewer extubation attempts, and more reintubations and tracheostomies. Unit mortality in that group was 78 percent, against 2 percent in those who weaned successfully. Latent analysis identified three distinct phenotypes distinguished by the number of separation attempts and the timing of decisions to withdraw or withhold life-sustaining therapy — and a failed first separation attempt was strongly associated with subsequent withdrawal decisions. That association deserves reflection, because it raises the question of how much weaning failure is physiology and how much is a prognostic judgement set in motion by a single early failed trial. Complementing that, in the Journal of Critical Care, Msemakweli and colleagues rebuilt the ventilation benchmarking metric on a competing-risk footing across nearly forty-four thousand patients at 114 United States hospitals [9]. The point is conceptually simple: if you benchmark ventilation duration directly, a patient who dies on day two counts as a short, successful ventilation. By day seven, about 73 percent of patients were liberated, roughly eleven percent had died, and sixteen percent remained ventilated. Case mix explained under four percent of between-hospital variance. The conventional metric and the competing-risk metric agreed on not a single outlier, and they reversed each other's conclusions — the conventional approach flagged twenty-six hospitals as better than expected, the competing-risk approach flagged twenty-six as worse. Sixteen hospitals with genuinely poor liberation went entirely unflagged by the conventional metric, and those units had higher risk-standardised mortality. If your unit is being benchmarked on ventilator days, ask how death is handled in the model.

Finally, three papers on organ support and recovery. In the Journal of Critical Care, Hirata and colleagues studied continuous renal replacement therapy in 1,358 adults on extracorporeal membrane oxygenation across multiple centres [7]. Just under four in ten received renal replacement, and mortality was far higher in every renal replacement group than in the extracorporeal-only group — roughly 35 percent without renal replacement versus around 65 to 74 percent with it — with adjusted odds of death three to five times higher regardless of timing relative to cannulation. This is observational, and the obvious confounder is that needing renal replacement marks sicker physiology; the timing analysis gives no support to the idea that starting earlier rescues these patients. Among survivors, dependence on renal replacement at discharge was much commoner after veno-arterial than veno-venous support, at about 71 percent versus 46 percent. Also in Critical Care Medicine, Kane-Gill and colleagues mapped the entire kidney biomarker literature — more than eleven hundred studies screened in from nearly seven thousand records [8]. The distribution is telling: the overwhelming majority were observational cohorts, over nine hundred studies addressed prediction of acute kidney injury, but only six used biomarkers to enrich a clinical trial and only twelve addressed management. The accuracy question has been answered many times over; the question of what to do differently at the bedside has barely been asked. And in Intensive Care Medicine, Schaller and colleagues offer a narrative review of post-intensive care syndrome and rehabilitation across the recovery continuum [5]. They recommend a two-step assessment across the physical, cognitive and mental health domains, they reaffirm the ABCDEF bundle with light or no sedation, delirium prevention, early mobilisation and family participation, and they note that post-intensive care syndrome in family members affects somewhere between a fifth and three fifths of relatives, with nurse-led family interventions looking promising. Their honest conclusion is that the evidence base remains too heterogeneous for firm recommendations on the dose, timing or model of follow-up care.

If you only have time for one paper this week, make it the individual participant data meta-analysis of intraosseous versus intravenous access in cardiac arrest in Resuscitation [1]. It settles a question that has driven protocol rewrites and equipment purchasing across emergency medical services systems, and it settles it with pooled patient-level data rather than another single trial.

Here are the key takeaways from this week in Critical Care. First, route of vascular access in out-of-hospital arrest does not meaningfully change survival — speed and reliability in the hands you have still decide. Second, peripheral perfusion-guided resuscitation in septic shock did not significantly reduce mortality in pooled randomised data, so treat capillary refill as an adjunct rather than a validated target. Third, arrest occurring in hospital before Impella implantation marks a distinct, very high-risk cardiogenic shock phenotype with nearly half of these patients dead at thirty days. Fourth, failure to wean carries a unit mortality approaching eighty percent and is tightly bound to decisions made after the first failed separation attempt — be deliberate about what that first failure means. And fifth, any ventilation benchmark that ignores death as a competing risk can invert your unit's ranking and hide the units that are genuinely failing to liberate patients.

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.

If this weekly briefing is useful, follow the show in your podcast app so new episodes arrive automatically. And a quick rating on Apple Podcasts or Spotify helps other physicians discover the show.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.