Predictors of adverse events after endoscopic ultrasound-guided through-the-needle biopsy of pancreatic cysts: a recursive partitioning analysis.

BACKGROUND AND STUDY AIMS : Endoscopic ultrasound-guided through-the-needle biopsy (TTNB) of pancreatic cystic lesions (PCLs) is associated with a non-negligible risk for adverse events (AEs). We aimed to identify the hierarchic interaction among independent predictors for TTNB-related AEs and to generate a prognostic model using recursive partitioning analysis (RPA). PATIENTS AND METHODS : Multicenter retrospective analysis of 506 patients with PCLs who underwent TTNB. RPA of predictors for AEs was performed and the model was validated by means of bootstrap resampling. RESULTS : Mean cysts size was 36.7 mm. Most common diagnoses were intraductal papillary mucinous neoplasm (IPMN, 45 %), serous cystadenoma (18.8 %), and mucinous cystadenoma (12.8 %). Fifty-eight (11.5 %) AEs were observed. At multivariate analysis, age (odds ratio [OR] 1.32, 1.09-2.14; p = 0.05), number of TTNB passes (OR from 2.17, 1.32-4.34 to OR 3.16, 2.03-6.34 with the increase of the number of passes), complete aspiration of the cyst (OR 0.56, 0.31-0.95; p = 0.02), and diagnosis of IPMN (OR 4.16, 2.27-7.69; p < 0.001) were found to be independent predictors of AEs, as confirmed by logistic regression and random forest analyses. RPA identified three risk classes: high-risk (IPMN sampled with multiple microforceps passes, 28 % AEs rate), low-risk (1.4 % AE rate, including patients < 64 years with other-than-IPMN diagnosis sampled with ≤ 2 microforceps passes and with complete aspiration of the cyst) and middle-risk class (6.1 % AEs rate, including the remaining patients). CONCLUSION : TTNB should be selectively used in the evaluation of patients with IPMN. The present model could be applied during patient selection as to optimize the benefit/risk of TTNB.