Background and Aims: Endoscopic ultrasound (EUS)-guided drainage of pancreatic fluid collections (PFCs) is effective. However, therapeutic escalation is often required and predictors are lacking. We aimed to develop a machine learning (ML)-based risk stratification to predict step-up therapy at the index procedure.
Methods: Multicenter derivation and independent validation study of patients with PFCs undergoing EUS-guided drainage at three high-volume tertiary centers (2016-2024). The primary endpoint was step-up therapy, defined as any additional intervention required to achieve clinical success (endoscopic necrosectomy, multigate drainage, percutaneous drainage, or laparoscopic/open surgery). Model performance was benchmarked against the quadrant-necrosis- infection (QNI) classification system.