Optimising the management of patients with cholecystitis and suspected common bile duct stones: an external validation of current prediction models

Background: Identifying patients with common bile duct (CBD) stones in the setting of acute calculous cholecystitis (ACC) is challenging. Three prediction models were previously developed to streamline this process. This study externally validates these models and evaluates whether a new model is needed. Methods: In this retrospective single-centre cohort study, patients who underwent early cholecystectomy for ACC between 2015 and 2021 were included. The previously published models were assessed for discriminative performance, calibration, and two pre-defined evaluation criteria. These criteria, based on an expert survey and Dutch diagnostic practice, were: 1) a positive predictive value (PPV) ≥ 75% for high-risk patients (indicating the need for direct ERCP), and 2) < 10% of patients assigned to either intermediate or high-risk (indicating the need for additional diagnostics to investigate the CBD). If either criterion was not met, the model was considered insufficient, and the development of a new model was pursued. Results: Of 1016 patients, 74 (7%) had confirmed CBD stones. Among those, 28 were diagnosed immediately on presentation via initial imaging for ACC. This left an at-risk population of 988 patients, 46 (5%) of whom were ultimately diagnosed with CBD stones. Chisholm’s model performed best (c-statistic: 0.89, 95%CI: 0.83–0.94) but failed to meet the criteria (50% PPV for high-risk patients and 14% assigned to intermediate or high-risk). The models by Khoury and Reddy showed PPVs of 56% and 34%, respectively. Consequently, a new prediction model was developed incorporating AST, ALP and CBD diameter (c-statistic: 0.90, 95%CI: 0.85–0.95) that met the criteria, achieving a 75% PPV for high-risk patients and assigning 8% of patients to intermediate or high-risk. Conclusion: Existing CBD stone prediction models demonstrate insufficient efficiency in the work-up of ACC patients. A new model using AST, ALP and CBD diameter was developed, which had superior characteristics when applied to the current data set.