Novel score for prediction of malignant bile duct obstruction based on biochemical and clinical markers

Summary Background Early differentiation of malignant from benign bile duct obstruction is of utmost importance. Aim To identify biochemical and clinical predictors for malignancy in patients with bile duct obstruction, and establish a predictive model by combining pre‐treatment patient characteristics. A web‐based application was developed for easy assessment of malignant bile duct probability (www.pmal-score.org). Methods One thousand hundred and thirty‐five patients [median age 66 (52–75) years, 53% male] with bile duct obstruction of various aetiologies were retrospectively evaluated at our tertiary referral centre. Multivariate logistic regression analysis identified factors as independently significant for malignant bile duct obstruction. A predictive risk score was established using ROC analysis and applied to an external validation cohort of 101 patients. Results Three hundred and ninety‐four patients had malignant bile duct obstruction proven by surgery, while in 741 patients benign obstruction was observed. Multivariate analysis identified various clinical factors to be predictive for malignancy. On the basis of eight predictors, a risk score for malignancy was developed [X = 0.025 * [age] + 1.239 * [1 if weight loss, otherwise 0] − 0.235 * [1 if pain, otherwise 0] + 0.649 * [1 if diabetes, otherwise 0] + 0.896 * [1 if jaundice, otherwise 0] + 0.109 * [bilirubin] + 0.0007 * [γ‐GT] + 0.0003 * [AP] − 4.374]: A significant correlation between the predicted malignancy and the actual malignancy was found by ROC (AUC: 0.862; 95% CI 0.838–0.886, P