Abstract / Summary
Patients with unexplained pancreatobiliary duct dilatation may harbor clinically significant findings requiring intervention despite negative conventional imaging. Endoscopic ultrasonography (EUS) can detect such findings, but evidence for pre-examination risk stratification remains limited. This study aimed to evaluate the independent predictive value of common bile duct (CBD) and main pancreatic duct (MPD) diameters for EUS-detected clinically significant findings and to develop and internally validate a predictive model for patient selection. This single-center retrospective cohort study consecutively included adult patients with unexplained CBD dilatation, MPD dilatation, or combined pancreatobiliary duct dilatation on conventional imaging prior to EUS (January 2019–December 2024). The primary outcome was EUS-detected clinically significant findings. We employed restricted cubic splines, multivariable logistic regression (with prespecified variable inclusion, no stepwise selection), receiver operating characteristic (ROC) curves with DeLong testing, calibration curves with Brier score, subgroup analyses, and multiple sensitivity analyses. Multiple imputation using chained equations (20 imputed datasets) was used for missing covariate data. Internal validation was performed using bootstrap resampling with 1,000 iterations. Among 683 included patients, 151 (22.1%) had clinically significant EUS-detected findings. Both CBD diameter (adjusted odds ratio [aOR] = 1.18, 95% confidence interval [CI]: 1.10–1.27, P < 0.001) and MPD diameter (aOR = 1.34, 95% CI: 1.20–1.49, P < 0.001) were independent predictors, with no significant nonlinearity (P_nonlinear > 0.05). The combined prediction model achieved an area under the curve (AUC) of 0.867 (95% CI: 0.831–0.903), significantly superior to either diameter alone (both P < 0.001, DeLong test). The optimal cutoff values for individual duct diameters were 11.3 mm for CBD (sensitivity 66.89%, specificity 72.56%) and 3.3 mm for MPD (sensitivity 74.17%, specificity 72.56%). The optimal predicted probability cutoff of 0.247 yielded sensitivity 80.13%, specificity 79.14%, and negative predictive value 93.35%. Internal validation via bootstrap showed an optimism-corrected AUC of 0.851 and calibration slope of 0.94, indicating minimal overfitting. The model showed good calibration (Brier score = 0.135). Results remained robust across all sensitivity and subgroup analyses. CBD and MPD diameters independently predict EUS-detected clinically significant findings in patients with unexplained pancreatobiliary duct dilatation. Clinically, CBD diameter ≥ 11.3 mm or MPD diameter ≥ 3.3 mm should raise suspicion for underlying pathology. A combined model incorporating duct diameters, age, abdominal pain, cholestatic indices, and CA19-9 can aid pre-EUS risk stratification and patient selection, though external validation is needed before clinical implementation.