Vascular Surgery

ML models outperform RCRI for postoperative cardiac risk but lack external validation

Systematic review of 13 studies (54 models; 927,113 patients) found machine learning generally exceeded the Revised Cardiac Risk Index for postoperative cardiac events. Automated machine learning ranked highest (SUCRA 96.6) and gradient boosting showed consistent discrimination gains. Only six models had low risk of bias and none reported external validation, limiting immediate clinical adoption. 1

Five‑year GREAT registry: GORE EXCLUDER IBE shows low aortic-related mortality

In 94 real‑world patients at five years, aortic‑related mortality was 1.1% while all‑cause mortality was 14.3% (cumulative 23.4%). Cumulative reintervention through five years was 16.0%, with device‑related reintervention 15.0% and endoleaks in 12.8%. Device migration, fracture, and aortic rupture were rare, supporting acceptable midterm performance in routine practice. 2

European multisociety consensus: standards for high‑risk endovascular stroke device evidence

Eighteen experts used Delphi methods to define methodological standards for regulatory studies of high‑risk endovascular stroke devices. They endorse prospective, multicenter observational designs and single‑arm studies with benchmarks for many device approvals. Core outcomes recommended include functional independence at 90 days, successful reperfusion, and symptomatic intracranial hemorrhage. 3