Effectiveness of multivariate analysis of tumor markers in diagnosis of pancreatic carcinoma: a prospective study in multiinstitutions.

Summary In 403 patients suspected of having pancreatic cancer, we prospectively studied a combination assay of various serum tumor markers: CA19-9, DUPAN2, tissue polypeptide antigen, elastase 1, γ-glutamyltranspepti-dase, lactate dehydrogenase, lipase, amylase, and alkaline phosphatase. The diagnostic value of each marker was compared with a multivariate analysis (computer-aided multivariate and pattern analysis system for pancreatic cancer examine-1: CAMPAS-PX1). Pancreatic cancer was subsequently identified in 47 patients. CAM-PAS-PX1 had higher negative in health and positive predictability than those of each marker used alone in the diagnosis of pancreatic cancer. CAMPAS-PX1 proved the most effective marker for diagnosing pancreatic cancer, but in terms of its codbenefit ration CAMPAS-PX1 was not superior to CA19-9 used alone. In this prospective trial, we experienced poor generalizability in the statistical models (CAMPAS-PX1). We believe that selection bias was present in samples used for model building. Based on this study a new model has been designed.