Development and Validation of a Preoperative Prediction Model for Neoplastic Gallbladder Polyps.

Abstract

Aims/Background The primary goal in evaluating gallbladder polypoid lesions (GPLs) is to identify neoplastic polyps (NP). Numerous studies have investigated risk factors for NP. This study aimed to develop a practical preoperative prediction model for NP using simple and easily accessible clinical variables. Methods We retrospectively analyzed clinical data from patients with GPLs who underwent cholecystectomy at Lanzhou University Second Hospital between January 2018 and September 2022. A total of 621 cases were included and randomly assigned into a training set (70%) and an internal validation set (30%). An external validation set was established using data from 117 patients treated at other centers between January and December 2023. Univariate logistic analyses were performed, followed by backward stepwise multivariate logistic regression analysis for variables with p< 0.2 to identify significant variables associated with NP. These predictors were included in the final logistic regression model and visualized as a nomogram model. The discrimination, calibration, and clinical utility of the model were evaluated. Results Age (odd ratio (OR) = 1.06, 95% CI = 1.03-1.09, p= 0.0001), polyp size (OR = 19.01, 95% CI = 6.48-55.79, p < 0.0001), polyp number (OR = 0.25, 95% CI = 0.12-0.56, p = 0.0006), gallbladder wall thickness (OR = 1.57, 95% CI = 1.02-2.41, p= 0.0385), and polyp echo characteristics (OR = 0.41, 95% CI = 0.19-0.85, p = 0.0169) were identified as independent influencing factors for NP. The area under the curve (AUC) of the nomogram model in the training, internal validation, and external validation sets were 0.886 (95% CI, 0.841-0.930), 0.836 (95% CI, 0.753-0.919), and 0.867 (95% CI, 0.743-0.978), respectively. Calibration curves for the three datasets showed Brier scores of 0.079, 0.092, and 0.070, all below 0.25, indicating good calibration. Decision curve analysis (DCA) and clinical impact curve (CIC) analysis suggested that a threshold probability of 0.6 provided the most significant clinical benefit. Conclusion This prediction model, incorporating easily accessible variables, demonstrated excellent performance in the identification of NP and contributed to clinical decision-making in GPL management.