Abstract
BACKGROUND AND OBJECTIVES: Repeated EUS-guided fine-needle aspiration/biopsy (rEUS-FNA/B) is recommended in solid pancreatic lesions (SPLs) with inconclusive initial results. However, even rEUS-FNA/B cannot completely eliminate false negatives in challenging cases. This study aims to identify risk factors of false-negative results in rEUS-FNA/B and develop a prediction model. METHODS: Data of patients who had an initial inconclusive diagnosis and underwent rEUS-FNA/B for SPLs across 8 Chinese medical centers from January 2013 to June 2024 were retrospectively reviewed. Logistic regression was performed to identify the risk factors of false-negative results in rEUS-FNA/B. A clinical prediction model using a random forest algorithm was developed with leave-one-hospital-out cross-validation. RESULTS: A total of 230 patients were enrolled in this study. Among them,159 patients (69.1%) were diagnosed with malignancies, while 71 patients (30.8%) had benign lesions. The diagnostic accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of rEUS-FNA/B was 78.26%, 68.55%, 100%, 100%, and 58.68%. At multivariate analysis, lesion size 20-40 mm (odds ratio [OR] 0.26), lesion size ≥40 mm (OR 0.20), ≤2 needle passes (OR 3.1), and no liquid-based cytology (LBC) examination (OR 3.95) were found to be independently associated with false-negative results. A final random forest model with 4 features (lesion size, rEUS-FNA/B needle type, LBC, and smear cytology) showed moderate ability to distinguish false-negative results in both training (area under the receiver operating characteristic curve = 0.78) and validation sets (area under the receiver operating characteristic curve = 0.68). CONCLUSION: To minimize false-negative outcomes during rEUS-FNA/B, LBC is strongly recommended. Additionally, SPLs with a maximum diameter of ≤20 mm exhibit a higher probability for producing false-negative results during rEUS-FNA/B procedures.