Prediction of therapy in primary endoscopic retrograde cholangiopancreatography.

BACKGROUND AND STUDY AIMS: It seems rational to perform endoscopic retrograde cholangiopancreatography (ERCP) if the probability of endoscopic therapy is high, but to carry out magnetic resonance cholangiopancreatography (MRCP) or endoscopic ultrasound (EUS) first if this probability is moderate or low. The aim of the present study was to develop a model describing the probability of endoscopic therapy in patients without previous biliary imaging.

PATIENTS AND METHODS: The development of the model was based on stepwise multiple logistic regression applied to 2470 prospectively registered first-time ERCP procedures. The model was evaluated by application to 442 prospectively registered first-time ERCP procedures entered in the database in the following 2 years.

RESULTS: Predictors selected were: age, gender, p-amylase >/= 400 U/l, ln(s-bilirubin), ln(s-alkaline phosphatase), common bile duct (CBD) stone seen on transabdominal ultrasonography, gallbladder stone seen on transabdominal ultrasonography, interaction of dilated bile ducts seen on transabdominal ultrasonography with ln(s-bilirubin), and interaction between age and male gender. The area under the receiver operating characteristic (ROC) curve was 0.875 and there was good fit of the model. A test with a probability cutoff value of 80 % had a positive predictive value (PPV) of 92.8 %. Specificity was 87.1 % and, using this test, 52.4 % of patients would have been selected for primary ERCP. In the application cohort, the frequency of therapy was higher than in the development cohort. The area under the ROC curve was 78.7 %. When used in the evaluation cohort, with a cutoff probability of 80 %, the test had sensitivity 84.0 %, specificity 49.5 %, negative predictive value (NPV) 46.6 % and PPV 85.6 %. Of the patients, 76.7 % would have been selected for ERCP. This would have identified 85.5 % of individuals needing therapeutic ERCP without use first of MRCP or EUS. Test-positive cases constituted 90.3 % of stent insertions and 86.3 % of stone extractions.

CONCLUSIONS: The model is useful for selection of patients for ERCP at our center.