CRRT on ECMO answers the question ADQI asked three days ago, a ventilator benchmark that flags the wrong hospitals, a decade of sepsis bundle delivery, and a preprint on sedation for LAA occlusion

A thin weekend in the tracked journals — the 12–14 September sweep across 22 named journals returned three hits, two of them from Saturday’s Annals of Thoracic Surgery. The index is healthy (1,378 records for 12 September, 626 for the 13th, 300 so far for today), so this is a real weekend lull rather than a lag. The topic-wide sweep across all journals rescued the day, and in particular a strong Journal of Critical Care issue dated 13 September.

1. CRRT during ECMO — 1,358 patients, and the answer ADQI asked for three days ago

Journal · Published: Journal of Critical Care, 13 September 2026 — yesterday Hirata, Honsell, Takeuchi, Rehman, Pranto, Gonzalez Flores, Shampa, Tangchitthavorngul, Wille, Tolwani, Kashani M, Singh, Kashani KB, Neyra

Retrospective multicentre cohort, 2009–2023, adults receiving ECMO, categorised by CRRT exposure and its timing relative to cannulation: CRRT-before (≤7 days before), CRRT-early (<48 h after), CRRT-late (48 h to 14 days after), or ECMO-only. Primary outcome in-hospital mortality, with stratification by ECMO configuration.

1,358 ECMO patients; 512 (37.7%) received CRRT — 82 before, 234 early, 196 late. 61.5% VA-ECMO, 36.6% VV, 1.9% both.

• In-hospital mortality: ECMO-only 35.2%; CRRT-before 74.4%; CRRT-early 65.0%; CRRT-late 66.3%
• Adjusted, versus ECMO-only: CRRT-before aOR 4.95 (2.62–9.35), CRRT-early aOR 3.10 (2.14–4.50), CRRT-late aOR 4.08 (2.82–5.89)
• Independent predictors of mortality: older age, sepsis, and VA-ECMO
• Among survivors, RRT dependence at discharge was 71.2% after VA-ECMO + CRRT versus 46.2% after VV-ECMO + CRRT

Interpretation. Three days ago this archive covered the ADQI XXXVI / ELSO consensus on AKI and CRRT during extracorporeal support — the epidemiology and risk-factor workgroup, which reported AKI in 40–80% of ECMO patients and CRRT in 40–50%, and which closed by calling for “standardised definitions, prospective multicentre trials” and better data. This is a multicentre cohort landing in the same week, and its CRRT incidence — 37.7% — sits just below the consensus range.

The headline numbers are stark and almost entirely confounded, and the paper is honest about which is which. An adjusted odds ratio of 3–5 for death is not a claim that CRRT kills ECMO patients. It is a claim that needing CRRT identifies a much sicker population, and the timing gradient supports that reading: CRRT-before carries the highest odds (4.95) — those patients arrived at cannulation already in established renal failure. The conclusion states it carefully: CRRT was “independently associated with” mortality regardless of configuration and timing, which is a statement about prognosis, not about the therapy.

The genuinely new and less confounded finding is the renal recovery split: 71.2% versus 46.2% RRT dependence at discharge, VA versus VV. Both groups survived and both received CRRT, so the comparison is between two surviving cohorts rather than between the sick and the less sick. That maps onto exactly what ADQI named as mechanistic drivers — non-pulsatile flow, venous congestion, haemolysis — all of which differ between venoarterial and venovenous support. VA-ECMO imposes retrograde non-pulsatile aortic flow and is used in cardiogenic shock with high venous pressures; VV does neither. Two-thirds of VA-ECMO survivors leaving hospital dialysis-dependent is a number worth carrying into a consent conversation.

Caveats: retrospective, fourteen years of changing ECMO practice, and no information in the abstract on CRRT indication — fluid overload versus metabolic versus solute clearance are different exposures with different prognoses, and pooling them limits what the timing categories can mean. It also, notably, does not report whether earlier CRRT initiation helped, which is the actual clinical question; the timing categories describe when it happened, not a randomised strategy.

Read the paper · PMID 42732731

2. Benchmarking ventilator liberation — the conventional metric flags the wrong hospitals, and reverses the direction

Journal · Published: Journal of Critical Care, 13 September 2026 — yesterday Msemakweli, Onesphore, Okpanachi, Nyambod

The problem, stated with unusual clarity:

Duration of invasive mechanical ventilation is used to benchmark intensive care units, but published metrics model duration directly and do not treat death as a competing event, so a patient who dies on day two counts as a short ventilation.

Retrospective cohort, 43,880 adults starting invasive ventilation at 114 US hospitals in the eICU Collaborative Research Database. The rebuilt estimand: hospital-specific, risk- and reliability-adjusted cumulative incidence of liberation by day 7, with death as a competing event. Overdispersion-adjusted funnel plots, compared against a conventional observed-to-expected ventilation-duration metric at matched control limits.

• By day 7: 72.9% liberated, 11.4% died, 15.7% still ventilated
• Risk-standardised 7-day liberation ranged from 26.7% to 85.8% (median 70.7%)
• Case mix explained 3.7% of between-hospital variance; hospital structure and measurable processes a further 31.6% combined
• The two metrics ranked hospitals only moderately alike (Spearman ρ 0.79) and agreed on no outlier at matched limits (weighted κ 0.00)
• The conventional metric called 26 hospitals better than expected and 11 worse. The competing-risk metric reversed it: 3 better, 26 worse.
• Sixteen hospitals (14%) with significantly worse liberation went unflagged by the conventional metric — and had higher risk-standardised ICU mortality, 18.2% versus 13.6%, P < 0.001

Interpretation. This is the sharpest methodological paper this archive has carried, and its finding is not subtle: a widely used quality metric was rewarding the units where patients died faster.

A weighted kappa of 0.00 on outlier identification is the number to hold onto. Not poor agreement — no agreement. Two metrics applied to the same 43,880 patients at matched control limits identified entirely different hospitals as outliers, and inverted the direction: 26 hospitals that looked good became 26 that looked bad. That is not a refinement; it is a different answer.

The mechanism is simple once stated, and the paper states it in one sentence: a patient who dies on day two counts as a short ventilation. A unit with high mortality accumulates short ventilation episodes and scores well. The confirmation is in the sixteen missed hospitals — significantly worse at liberating patients, invisible to the conventional metric, and with mortality 4.6 percentage points higher.

Where this sits in the archive. The WEAN SAFE failed-weaning phenotypes sent on 10 September found 78% ICU mortality among patients who never weaned, and raised the problem of a failed first separation attempt shaping withdrawal decisions. This is the institutional version of the same confound: at the patient level, death and liberation compete; at the hospital level, ignoring that competition inverts your league table. Both papers are about the same statistical fact seen at different scales.

The 3.7% figure deserves its own note. Case mix explained under 4% of between-hospital variance while structure and process explained 31.6% — nearly nine times as much. The usual defence against benchmarking (“our patients are sicker”) is, on this data, mostly wrong.

Limits: a single database, US hospitals contributing to eICU are not a random sample, and “measurable processes” in an administrative dataset are crude. And the 26.7%-to-85.8% spread in risk-standardised liberation is so wide that residual confounding must be contributing something. But none of that rescues the conventional metric, which is the point.

Read the paper · PMID 42732730

3. Sepsis identification and bundle delivery in US emergency departments, 2012–2022

Journal · Published: Journal of Critical Care, 13 September 2026 — yesterday Zarama, Espinola, Camargo

Serial cross-sectional analysis of the National Hospital Ambulatory Medical Care Survey, a representative sample of US ED visits, 2012–2022. Sepsis identified as explicit (sepsis-specific code) or implicit+ (infection plus organ-dysfunction codes); composite analysed. Outcome: documented four-component bundle — IV fluids, antibiotics, blood cultures, lactate.

820.1 million weighted adult ED visits; 18.3 million (2.2%) met the sepsis definition.

• Sepsis identification rose from 0.9% to 2.5% — OR 1.14 per year (99% CI 1.10–1.17)
• Bundle delivery rose from 2.4% to a peak of 20% — OR 1.15 per year (1.07–1.23), with a 2016 level shift, OR 4.38
• Lactate rose most steeply yet remained the least delivered component
• Strongest associations: severe presentation at triage OR 2.84 (1.73–4.65); abdominal infection source OR 0.30 (0.12–0.72)
• Conclusion: delivery “increased modestly but remained incomplete, even among explicitly recognized sepsis patients,” and because the all-or-nothing composite may not capture partially delivered care and its benefit is contested, individual components and “measures of personalized resuscitation” may be better quality indicators

Interpretation. A peak of 20% is the number, and it is the story. After a decade of regulatory attention, national quality reporting and three revisions of the sepsis definition, four in five patients meeting a sepsis definition did not have all four bundle components documented. The 2016 level shift (OR 4.38) is SEP-1 becoming a reportable measure, and it is a real discontinuity — but it moved delivery to a fifth, not to a majority.

The abdominal-source finding is the clinically interesting one and is easy to misread. OR 0.30 for bundle delivery in abdominal infection is not obviously a failure. Abdominal sepsis frequently presents as pain with a surgical question attached, and the patient goes to CT and theatre rather than through a bundle checklist. A quality measure that penalises the pathway toward source control is measuring the wrong thing — and source control is the intervention with the largest effect size in abdominal sepsis.

The authors’ closing argument is the same one this archive has been tracking all month, and they make it explicitly: the all-or-nothing composite is a poor instrument, its benefit is contested, and “measures of personalized resuscitation” may serve better. That is the SEP-1 critique the IDSA multisociety position paper made on 29 August, arriving now with national denominators behind it. And “personalised resuscitation” is not an empty phrase this month: yesterday’s peripheral perfusion-guided meta-analysis put capillary-refill-guided resuscitation at a 97.2% Bayesian posterior probability of reduced 28-day mortality. The alternative to counting bundle components is beginning to have evidence behind it.

Limits: documentation, not delivery — NHAMCS records what was recorded, and under-capture is certain. The implicit+ definition using codes in any position will pull in patients no clinician considered septic. And 2.2% of ED visits meeting a sepsis definition is higher than most clinicians would guess, which itself suggests the definition is broad.

Read the paper · PMID 42732729

4. Conscious sedation versus general anaesthesia for LAA occlusion — preprint

PREPRINT — not peer reviewed. F1000Research, 12 September 2026 · DOI 10.12688/f1000research.188329.1 Nazir, Hassan, Amir, Jahangir, Rasul, Raza, Malik, Ajmal, Rauf, Maaz, Gulzar, Burki, Ejaz, Hasan

Included because it speaks directly to a question this archive raised six days ago, and labelled as a preprint throughout. It has not been peer reviewed and the numbers may change.

Systematic search to September 2024, studies comparing conscious sedation with general anaesthesia during left atrial appendage occlusion. Three studies, 1,395 patients.

• Device-related thrombus: OR 0.17 (0.03–0.98), P = 0.05, favouring conscious sedation
• Procedure time: −7.17 minutes (−9.34 to −5.01), P < 0.0001, favouring conscious sedation
• No significant difference in pericardial tamponade, bleeding, shock, vascular haematoma, stroke, cardiovascular or all-cause mortality
• No difference in technical success, procedural success, peri-device leak <5 mm, or fluoroscopy time

Interpretation. Read the resonance first and the evidence second, because they point different ways.

In the 8 September guidelines watch this archive covered the ELAACC expert consensus on multimodality imaging for LAA closure, and I wrote that its prioritisation of 3D transoesophageal echo and fusion imaging pushes toward a general anaesthetic with a probe for the duration, against the ICE-based lightly sedated workflow several groups have been moving toward; and that if your centre had been reducing anaesthetic involvement on ICE grounds, the consensus did not support that. Six days later, here is a meta-analysis arguing the opposite — that conscious sedation is safe, faster, and possibly associated with less device thrombus.

But the evidence here is weak, and I would not change practice on it. Three studies. OR 0.17 with a confidence interval of 0.03 to 0.98 and P = 0.05 — an interval spanning a 33-fold range, touching the null, on what must be a handful of events. That is the same shape as the sepsis estimate I declined to quote from the mitral endocarditis meta-analysis yesterday, and it deserves the same treatment: the point estimate is not usable. A mechanism by which lighter sedation prevents device-related thrombus months later is also not obvious.

Seven minutes of procedure time is real and small. The genuinely reassuring finding is the long list of nulls — tamponade, stroke, leak, technical success all equivalent — which is consistent with conscious sedation being a reasonable option in selected patients. That is what the abstract concludes, and it is appropriately hedged.

How to hold the two documents together: the ELAACC consensus is about imaging quality and standardisation and says what it needs to see; this preprint is about anaesthetic technique and procedural outcomes and says the lighter option does not appear to harm. Neither addresses the other’s question. The honest position is that the imaging requirements drive the anaesthetic decision, and a centre with the ICE skill to meet those requirements under sedation can reasonably do so.

Read the preprint

---

Notes for the next run

• Index state on 14 September: 12 Sep 1,378 records, 13 Sep 626, 14 Sep 300. Healthy — the thin tracked-journal yield was a genuine weekend lull, not a lag. Keep the trailing window.
• Worth a slot on a quiet day, seen 12–14 September and not pursued: Heart & Lung on core components of transitional care for ECMO survivors after ICU discharge (three-round modified Delphi, 12 September) — pairs with item 1 above and with the DESTINE 2.0 and ADQI documents; Journal of Burn Care & Research on predicting unplanned CRRT interruptions in burn patients with stacked ensemble machine learning; European Journal of Nuclear Medicine on practices and barriers in PET/CT use for critically ill patients, an international survey; and Catheterization and Cardiovascular Interventions on balloon-expandable TAVR in patients with prior endovascular aortic stent grafts — niche but squarely cardiothoracic.
• Still blocked: the Annals of Thoracic Surgery pulsatility commentary (fourth attempt, no abstract) and the EHJ–Cardiovascular Imaging bicuspid valvulo-aortopathy piece. Both remain on the outstanding list; neither will be characterised from its title.