A Bayesian network prediction model for gallbladder polyps with malignant potential based on preoperative ultrasound

Abstract

Background: It is important to identify gallbladder polyps (GPs) with malignant potential and avoid unnecessary cholecystectomy by constructing prediction model. The aim of the study is to develop a Bayesian network (BN) prediction model for GPs with malignant potential in a long diameter of 8-15 mm based on preoperative ultrasound.

Methods: The independent risk factors for GPs with malignant potential were screened by χ² test and Logistic regression model. Prediction model was established and validated using data from 1296 patients with GPs who underwent cholecystectomy from January 2015 to December 2019 at 11 tertiary hospitals in China. A BN model was established based on the independent risk variables.

Results: Independent risk factors for GPs with malignant potential included age, number of polyps, polyp size (long diameter), polyp size (short diameter), and fundus. The BN prediction model identified relationships between polyp size (long diameter) and three other variables [polyp size (short diameter), fundus and number of polyps]. Each variable was assigned scores under different status and the probabilities of GPs with malignant potential were classified as [0-0.2), [0.2-0.5), [0.5-0.8) and [0.8-1] according to the total points of [- 337, - 234], [- 197, - 145], [- 123, - 108], and [- 62,500], respectively. The AUC was 77.38% and 75.13%, and the model accuracy was 75.58% and 80.47% for the BN model in the training set and testing set, respectively.

Conclusion: A BN prediction model was accurate and practical for predicting GPs with malignant potential patients in a long diameter of 8-15 mm undergoing cholecystectomy based on preoperative ultrasound.

Keywords: Bayesian network; Gallbladder carcinoma; Gallbladder polyps; Prediction model.