Mathematical model for predicting biliary therapeutic endoscopic retrograde pancreatography (ERCP).

Introduction: Magnetic resonance cholangiopancreatography is as sensitive as endoscopic retrograde pancreatography in the evaluation of biliary tract diseases but does not offer therapeutic options. The aim of the present study was to develop a mathematical model to predict 'therapeutic endoscopic retrograde pancreatography' using clinical variables so that patients with low probability could be more appropriately investigated by magnetic resonance cholangiopancreatography in future.

Methods: Endoscopic retrograde pancreatography cases between January 1996 to December 1997 were retrospectively reviewed (before introduction of magnetic resonance cholangiopancreatography). Clinical, biochemical and radiological variables were analysed and a model was developed using multiple logistic regression.

Results: Case notes for 573 patients were successfully reviewed. A total of 330 patients underwent therapeutic endoscopic retrograde pancreatography (sphincterotomy or stent insertion). Clinical indications of obstructive jaundice and cholangitis, ultrasonographic findings of dilated common bile duct, and raised liver function tests (two or more elevated parameters) were each found to be predictive for 'therapeutic' endoscopic retrograde pancreatography. Using these variables, the mathematical model in the present study has specificity of 77% and sensitivity of 75% at the probability level of 50% or higher. This model has been tested in a separate group of endoscopic retrograde pancreatography cases carried out in 1998 and was found to have sensitivity 77.6%, specificity 80.3%, positive predictive value 68.5% and negative predictive value 86.6%.

Conclusions: The model reported in the present study can help clinicians to identify cases for therapeutic endoscopic retrograde pancreatography and diagnostic magnetic resonance cholangiopancreatography.