Surgeon volume beats centre volume in mitral repair, and two more from JTCVS

1. Surgeon versus centre mitral repair volume

Journal · Published: JTCVS, 29 August 2026 — genuinely new, 2 days old Caruso, He, Clark, Willekes, Bolling, Ailawadi et al.

Michigan Society of Thoracic and Cardiovascular Surgeons Quality Collaborative, March 2011 – March 2025. Adults undergoing isolated mitral valve surgery for primary mitral regurgitation. High-volume centre defined as >50 repairs annually; high-volume surgeon as >25 repairs over two consecutive years. Primary outcome: repair versus replacement.

• 6,649 patients, overall repair rate 80%
• Only 3 of 33 centres (9%) and 6 of 146 surgeons (3.8%) met high-volume criteria
• High-volume centres: higher crude repair rate (87% vs 73%) — but not significant after adjustment: OR 1.23 (95% CI 0.61–2.45), p=0.56
• High-volume surgeons: adjusted repair rate 93% vs 70% — OR 1.45 (1.1–1.9), p=0.003; shorter median cross-clamp time (75 vs 91 min, p=0.001); lower permanent pacemaker implantation (1.0% vs 3.0%)

The deposited abstract truncates at the p-value for the pacemaker comparison.

Interpretation. This is the most directly actionable paper of the day, because it separates two things that referral policy usually conflates. Centre volume washed out after adjustment; surgeon volume did not. For degenerative mitral disease — where repair rather than replacement is the whole point, and a replacement is a permanent downgrade in a patient who could have kept their valve — that distinction has consequences for how patients are referred and how heart teams are constructed.

The scarcity finding deserves attention on its own: 3.8% of surgeons in a well-run statewide collaborative met the high-volume threshold. Guideline recommendations that repair be performed in high-volume settings are, on these numbers, describing a very small part of actual practice.

Two caveats. Volume is entangled with case selection — a surgeon known for repair attracts repairable valves — and the 70% repair rate among lower-volume surgeons is not obviously a failure so much as a different case mix. And this is one state, with a mature quality collaborative, which likely makes it a best case rather than a representative one.

Read the paper · PMID 42668111

2. Prior cardiac surgery and acute type A aortic dissection repair

Journal · Published: JTCVS, 28 August 2026 — genuinely new, 3 days old Gregg, Lau, Ingason, Rahouma, Krieger et al. (high-volume aortic centre)

Retrospective analysis of all acute type A dissection repairs at one high-volume aortic centre, July 1997 – October 2024. Primary outcome: composite in-hospital major adverse events — operative mortality, myocardial infarction, cerebrovascular accident, tracheostomy, renal failure, re-exploration for bleeding. Secondary: all-cause mortality over follow-up.

• 413 patients, of whom 17.4% (72) had prior cardiac surgery
• The prior-surgery cohort carried a greater comorbid burden — including prior MI in 36.1% vs 11.4%
• Prior cardiac surgery was not independently associated with increased major adverse events on multivariable analysis
• The higher observed morbidity and mortality in that cohort was attributed to the comorbidity burden rather than the redo status itself

The deposited abstract truncates at the p-value for the prior-MI comparison.

Interpretation. Useful and somewhat counterintuitive. Redo sternotomy in acute dissection is one of those situations that feels categorically more dangerous, and this says the danger travels with the patient rather than with the reoperation. That has a practical consequence: a redo aortic dissection should not be triaged as inoperable on the basis of the previous sternotomy alone.

Read the setting carefully before generalising, though. This is a single high-volume aortic centre with 27 years of accumulated technique, so “not independently associated” is a statement about what is achievable there. The comorbidity finding is also the more reliable half of the result — a 413-patient series with 72 exposed patients has limited power to exclude a moderate independent effect.

Read the paper · PMID 42665038

3. Hourly deep-learning prediction of mortality and CRRT after cardiac surgery

Journal · Published: JTCVS, 29 August 2026 — genuinely new, 2 days old El Moheb, Bitar, Putman, Mazzeffi et al.

Recurrent neural network models built on the MIMIC database, ICU admissions after CABG alone or combined with valve surgery, 2008–2019, split 70:20:10. Fixed variables (demographics, comorbidities, procedure) plus hourly dynamic ICU data — vitals, ventilation, vasopressors, labs. Risk updated at each timestep, with a self-attention layer marking influential time points.

• 7,402 patients; 1.3% died in hospital, 1.8% required CRRT
• Using full ICU sequences: AUPRC 0.877 (mortality) and 0.906 (CRRT); F1 0.808 and 0.757
• Simulating real deployment with hourly updates, performance improved as physiologic data accumulated
• The attention layer identified the timepoints driving each prediction

Interpretation. The methodological choice worth noticing is AUPRC rather than AUROC. With outcome rates of 1.3% and 1.8%, AUROC would look spectacular and mean very little; precision-recall is the honest metric for rare events, and reporting it suggests the authors know what they are doing.

What the paper does not answer is the question that decides whether such a model is useful: at a chosen operating threshold, how many false alarms per true detection? An AUPRC of 0.877 against a 1.3% event rate is genuinely strong, but “identifying high-risk patients several days before overt deterioration” only helps if the alert burden is tolerable and if something is actually done differently in response. MIMIC is also a single-institution database, and models like this characteristically lose ground on external validation.

Read the paper · PMID 42668112

4. Three phenotypes of candidemia in critical illness

Journal · Published: Critical Care, 28 June 2026 — ~2 months old, not new Reizine, Henry, Desmedt, Camus et al.

Date note: this surfaced in a search filtered to 28–31 August, but its actual first publication was 28 June. Verify each item’s own date rather than trusting the filter — see the note added to the source file today.

Multicentre retrospective cohort, 492 ICU patients with candidemia across 16 French ICUs, 2015–2023. Factor analysis of mixed data followed by hierarchical clustering on principal components; Kaplan-Meier and Cox models for survival.

Overall 90-day mortality 62.6% (median age 64, 69.1% male). Three phenotypes:

• Phenotype 1 (n=70, 14.2%) — severe immunosuppression, mostly haematological malignancy (82.9%), high severity (SAPS II 70). Mortality 72.9%
• Phenotype 2 (n=223, 45.3%) — elderly cirrhotic patients (19.3%) with early-onset digestive candidemia. Mortality 70.4%
• Phenotype 3 (n=199, 40.5%) — younger, lower severity, catheter-related candidemia. Mortality 50.3%

Cirrhosis, age and illness severity were associated with mortality; a catheter-related source was protective.

Interpretation. A 62.6% 90-day mortality is a reminder of how lethal ICU candidemia remains despite adequate antifungals. The clinically useful part is the split between the digestive and catheter-related phenotypes: a removable focus is worth roughly 20 percentage points of mortality, which is an argument for aggressive line management rather than a statement about the fungus.

The phenotyping approach mirrors what has happened in sepsis and ARDS, and carries the same caveat — unsupervised clusters derived from one retrospective cohort describe that cohort, and need prospective validation before they guide treatment. Note also that the phenotypes track host and source rather than organism, which is where the actionable decisions actually sit.

Read the paper · PMID 42365321