Biliary strictures present a major diagnostic challenge as benign and malignant conditions often exhibit similar clinical and imaging findings. In cases of indeterminate strictures, obtaining a tissue sample is crucial. Notably, up to 25% of patients undergoing surgery for suspected cholangiocarcinoma (CCA) are ultimately diagnosed with benign disease, underscoring the difficulty of preoperative diagnosis. The low cellularity and desmoplastic nature of CCA further complicate cytological and histological assessments. Current ESGE guidelines recommend ERCP with tissue acquisition (TA) as the preferred diagnostic approach for perihilar strictures requiring drainage. This study aims to evaluate the diagnostic accuracy of EUS-TA for both distal and perihilar biliary strictures in a high-volume bilio-pancreatic endoscopy referral center. 

A retrospective analysis was conducted on a prospectively maintained database of patients who underwent EUS-TA at San Raffaele Hospital between April 2021 and June 2024 due to obstructive jaundice or biliary strictures identified via CT or MRI. Patients with preexisting pancreatic lesions, extrinsic biliary compression, or choledocholithiasis without biliary abnormalities were excluded. EUS-guided tissue sampling was performed using 25–22-gauge FNA or FNB needles with Rapid On-Site Evaluation (ROSE) by cytotechnicians. The final diagnosis was established based on surgical pathology, imaging, or a minimum follow-up of six months. Sensitivity, specificity, and likelihood ratios (LLR+ and LLR−) were calculated using 2×2 contingency tables, with 95% confidence intervals estimated via the Clopper-Pearson method. Additionally, a Random Forest model was developed using an 80:20 stratified split of the dataset, preserving class distribution. The model achieved an accuracy of 83.33% on the test set.

Of the 249 patients screened, 86 met the inclusion criteria. Among them, 47 (54.7%) had perihilar strictures, and 39 (45.3%) distal ones. Strictures presented as lesions in 65% of cases and as wall thickening in 35%. Malignancy was diagnosed in 76.7% of cases, while 23.2% were benign. Adequate tissue sampling was achieved in 94% of cases, with five false-negative results. EUS-TA demonstrated an overall sensitivity, specificity, and accuracy of 86.4%, 95%, and 88.4%, respectively. In perihilar strictures, sensitivity was 85.7%, specificity 100%, and accuracy 87.2%, with an NPV of 45.4% and negative LR of 0.14. Combining EUS-TA with ERCP-TA improved sensitivity to 89% and accuracy to 92%.The Random Forest model showed high diagnostic performance, with a precision of 0.88, recall of 0.94, and an F1-score of 0.91 for malignant lesions. Multivariate analysis identified male sex, jaundice, mass-like presentation, and vascular invasion as independent predictors of malignancy. No clinical or EUS factors were significantly associated with diagnostic accuracy prediction.

This study confirms the strong diagnostic performance of EUS-TA for biliary strictures, including perihilar cases. The combination of EUS-TA and ERCP-TA significantly improves diagnostic yield, particularly for perihilar strictures. These findings support the use of EUS-TA as a first-line diagnostic approach in high-expertise centers, with a same-session EUS-ERCP strategy optimizing procedural efficiency and reducing risks.