Immunohistochemical differentiation between primary adenocarcinomas of theovary and ovarian metastases of colonic and breast origin. Comparison between a statistical and an intuitive approach

Aim-To discriminate between adenocarcinomas that are primary to the ovary a nd metastatic to the ovary, especially of colonic and breast origin, by imm unohistochemistry, using stepwise discriminant analysis or a decision tree. Methods-312 routinely processed, formalin fixed tissue specimens were used. The tumours were divided into a learning set (n = 159), composed of primar y tumours of ovary, breast, and colon, and a test set, comprising 134 metas tases from these sites and an additional 19 primary ovarian carcinomas. The immunohistochemical panel was composed of antibodies against cytokeratin 7 (CK7) and 28 (CK20), CA125, vimentin, carcinoembryonic antigen (CEA), gros s cystic disease fluid protein- 15 (GCDFP-15), and the oestrogen receptor ( ER). The staining results of the tumours were expressed as the product of t he staining intensity and the percentage of positive tumour cells. Analyses were first performed on the learning set and then evaluated on the test se t. Results-Although the immunostaining patterns showed a considerable overlap between the three types of adenocarcinoma, the breast carcinomas were typic ally positive for GCDFP-15 and often for ER, and negative for vimentin. Ova rian carcinomas were always positive for CK7 and to a lesser extent for CR1 25. Colonic carcinomas showed prominent positivity for CEA and CK20, while no staining was seen for ER and vimentin. In discriminant analysis, six ant ibodies (alpha CK7, alpha CK20, alpha CA125, alpha CEA, alpha ER, and alpha GCDFP-15) appeared to be necessary for optimal classification: 89% of the learning set and 82% of the test set were classified correctly. In the deci sion tree, only four antibodies (aCK7, alpha CEA, alpha ER, and alpha GCDFP 15) were used to obtain ri correct classification score of 89% for the lear ning set and 84% for the test set. Conclusions-Using a semiquantitative assessment of the immunostaining resul ts by a restricted panel of six antibodies with stepwise discriminant analy sis, 80-90% of the adenocarcinomas of colon, breast, and ovary can be corre ctly classified. Discriminant analysis is computer aided and therefore an e asy method and for each case a probability value of the classification resu lt is obtained. The intuitive decision tree method provides a slightly bett er result, requires only four antibodies, and offers a more practical metho d for the surgical pathologist.