Tracing the origin of adenocarcinomas with unknown primary using immunohistochemistry: Differential diagnosis between colonic and ovarian carcinomas as primary sites

To discriminate adenocarcinoma metastases originating from either colon or ovary, a panel of immunohistochemical markers was evaluated. For this purpose, paraffin sections from 157 primary and metastatic colonic and ovarian carcinomas were immunostained. These cases were divided into a learning group of 46 colonic and 54 ovarian carcinomas and a test group of 29 colonic and 28 ovarian carcinomas, including all metastatic tumors, among which were five with unknown primary site at the time of testing. The sections were immunostained with antibodies against carcinoembryonic antigen (CEA), cytokeratin 7 (CK7), cytokeratin 20 (CK20), CA 125, vimentin, and CA 19.9. Staining results were expressed as the product of staining intensity and percentage of positive tumor tells. Stepwise discriminant analysis was applied on the learning set to obtain a classification function for both tumors. The validity of the classification function was evaluated using the test set. There was considerable overlap in immunostaining for both tumor types, but colonic carcinomas were typically positive for CEA and CK20 and negative for CK7 and CA125. Ovarian carcinomas were typically positive for CK7 and CA125 and negative for CEA and CK20. Tn discriminant analysis, the best combination of markers appeared to be CK7 and CEA. Only one sample of the test group (2%) was misclassified. Taking learning and test groups together, 136 of the 157 samples (87%) were correctly classified with high posterior probability (PP > .8). However, from the 28 mucinous ovarian carcinomas, only 19 (68%) could correctly be classified with high PP. When excluding the nonmucinous ovarian carcinomas from the analysis, overall 87 of 103 (84.5%) of the samples were correctly classified (PP > .8) with a combination of CEA, CK7, and also vimentin. From the 28 mucinous ovarian carcinomas, only two (7%) were misclassified, and four could not be classified with sufficient certainty. In neither analysis did CK20, CA125, or CA19.9 emerge as discriminatory parameters. Based on the same data, an intuitive flow chart was constructed with which 129 of 157 cases could be classified (only one falsely) without further statistical analysis. The five metastases with an at first unknown primary could, according to the followup, all be classified correctly with high PP. Most ovarian carcinomas, including the mucinous ones, can be discriminated with high probability from colonic carcinomas using a panel of three antibodies directed against CEA, cytokeratin 7, and vimentin. Copyright (C) 1998 by W.B. Saunders Company.