A Bayesian network model to predict neoplastic risk for patients with gallbladder polyps larger than 10 mm based on preoperative ultrasound features

Abstract

Background: Polyp size of 10 mm is insufficient to discriminate neoplastic and non-neoplastic risk in patients with gallbladder polyps (GPs). The aim of the study is to develop a Bayesian network (BN) prediction model to identify neoplastic polyps and create more precise criteria for surgical indications in patients with GPs lager than 10 mm based on preoperative ultrasound features.

Methods: A BN prediction model was established and validated based on the independent risk variables using data from 759 patients with GPs who underwent cholecystectomy from January 2015 to August 2022 at 11 tertiary hospitals in China. The area under receiver operating characteristic curves (AUCs) were used to evaluate the predictive ability of the BN model and current guidelines, and Delong test was used to compare the AUCs.

Results: The mean values of polyp cross-sectional area (CSA), long, and short diameter of neoplastic polyps were higher than those of non-neoplastic polyps (P < 0.0001). Independent neoplastic risk factors for GPs included single polyp, polyp CSA ≥ 85 mm ², fundus with broad base, and medium echogenicity. The accuracy of the BN model established based on the above independent variables was 81.88% and 82.35% in the training and testing sets, respectively. Delong test also showed that the AUCs of the BN model was better than that of JSHBPS, ESGAR, US-reported, and CCBS in training and testing sets, respectively (P < 0.05).

Conclusion: A Bayesian network model was accurate and practical for predicting neoplastic risk in patients with gallbladder polyps larger than 10 mm based on preoperative ultrasound features.

Keywords: Bayesian network; Gallbladder carcinoma; Gallbladder polyps; Polyp cross-sectional area; Prediction model.