Establishment and Verification of a Scoring Model for the Differential Diagnosis of Pancreatic Cancer and Chronic Pancreatitis

Objectives The aims of this study were to establish a scoring model for the differential diagnosis of pancreatic cancer (PC) and chronic pancreatitis (CP) and to evaluate its diagnostic efficacy. Methods The data of 502 patients with PC and 210 patients with CP at the Peking Union Medical College Hospital from January 1999 to December 2013 were retrospectively analyzed. Binary logistic regression was applied to establish the prediction model for the differential diagnosis. The model was verified using the method of leave-one-out cross-validation. Results The scoring system was established with 5 variables including age, carbohydrate antigen 19-9 level, splenic vein invasion, irregular dilatation of the pancreatic duct, and nontruncated pancreatic duct stenosis. The score range was from -2 to 3. The area under the receiver operating characteristic curve of the objects was 0.779 (95% confidence interval, 0.744-0.814) (P < 0.01), indicating that the scoring system is good at differentiation of PC with CP. With a score of 1 as the diagnostic cut-off value, the diagnostic sensitivity, specificity, positive predictive value, negative predictive value, and accuracy rate were 71.3%, 69.0%, 70.0%, 71.4%, and 70.2%, respectively. Conclusions The scoring model may improve the differential diagnosis of PC and CP and be useful in clinical practice.