An Immunohistochemical Algorithm for Ovarian Carcinoma Typing

Abstract

There are 5 major histotypes of ovarian carcinomas. Diagnostic typing criteria have evolved over time, and past cohorts may be misclassified by current standards. Our objective was to reclassify the recently assembled Canadian Ovarian Experimental Unified Resource and the Alberta Ovarian Tumor Type cohorts using immunohistochemical (IHC) biomarkers and to develop an IHC algorithm for ovarian carcinoma histotyping. A total of 1626 ovarian carcinoma samples from the Canadian Ovarian Experimental Unified Resource and the Alberta Ovarian Tumor Type were subjected to a reclassification by comparing the original with the predicted histotype. Histotype prediction was derived from a nominal logistic regression modeling using a previously reclassified cohort (N=784) with the binary input of 8 IHC markers. Cases with discordant original or predicted histotypes were subjected to arbitration. After reclassification, 1762 cases from all cohorts were subjected to prediction models (χ Automatic Interaction Detection, recursive partitioning, and nominal logistic regression) with a variable IHC marker input. The histologic type was confirmed in 1521/1626 (93.5%) cases of the Canadian Ovarian Experimental Unified Resource and the Alberta Ovarian Tumor Type cohorts. The highest misclassification occurred in the endometrioid type, where most of the changes involved reclassification from endometrioid to high-grade serous carcinoma, which was additionally supported by mutational data and outcome. Using the reclassified histotype as the endpoint, a 4-marker prediction model correctly classified 88%, a 6-marker 91%, and an 8-marker 93% of the 1762 cases. This study provides statistically validated, inexpensive IHC algorithms, which have versatile applications in research, clinical practice, and clinical trials.