Ovarian Carcinoma Histotype: Strengths and Limitations of Integrating Morphology With Immunohistochemical Predictions

Abstract

Ovarian carcinoma histotypes are critical for research and patient management and currently assigned by a combination of histomorphology +/- ancillary immunohistochemistry (IHC). We aimed to validate the previously described IHC algorithm (Calculator of Ovarian carcinoma Subtype/histotype Probability version 3, COSPv3) in an independent population-based cohort, and to identify problem areas for IHC predictions. Histotype was abstracted from cancer registries for eligible ovarian carcinoma cases diagnosed from 2002 to 2011 in Alberta and British Columbia, Canada. Slides were reviewed according to World Health Organization 2014 criteria, tissue microarrays were stained with and scored for the 8 COSPv3 IHC markers, and COSPv3 histotype predictions were calculated. Discordant cases for review and COSPv3 prediction were arbitrated by integrating morphology with IHC results. The integrated histotype (N=880) was then used to identify areas of inferior COSPv3 performance. Review histotype and integrated histotype demonstrated 93% agreement suggesting that IHC information revises expert review in up to 7% of cases. There was also 93% agreement between COSPv3 prediction and integrated histotype. COSPv3 errors predominated in 4 areas: endometrioid carcinoma (EC) versus clear cell (N=23), EC versus low-grade serous (N=15), EC versus high-grade serous (N=11), and high-grade versus low-grade serous (N=6). Most problems were related to Napsin A-negative clear cell, WT1-positive EC, and p53 IHC wild-type high-grade serous carcinomas. Although 93% of COSPv3 prediction accuracy was validated, some histotyping required integration of morphology with ancillary test results. Awareness of these limitations will avoid overreliance on IHC and misclassification of histotypes for research and clinical management.

Figure 1

Sankey diagram showing changes in histotype from cancer registry (left) via review (middle) to integrated histotype (right). The cancer registry serous carcinoma split into high-grade serous (HGSC) and low-grade serous carcinoma (LGSC). The majority of carcinomas not otherwise specified (NOS) and some endometrioid carcinomas (EC) were classified as HGSC on review. In contrast, most reclassification from review to integrated histotype were reclassifications of HGSC to EC and MC to EC. CCC – clear cell carcinoma, MC – mucinous carcinoma.

Figure 2

COSPv3 prediction errors: clear cell versus endometrioid carcinoma. A, B, Clear cell carcinomas showing typical clear cell morphology with hyaline stromal changes. COSPv3 falsely predicted EC because of absence of Napsin A expression and diffuse Vimentin expression. C, D, Endometrioid carcinomas with focal secretory changes (C) and sex cord like architecture (D). COSPv3 falsely predicted CCC despite absence of Napsin A but based on absence of ARID1A and PR.

Figure 3

COSPv3 prediction errors: endometrioid versus low-grade serous carcinoma. A, B, C, Endometrioid carcinomas showing sex cord like (A), glandular (B, C) morphology with secretory changes in C. COSPv3 falsely predicted LGSC owing to WT1 expression in these cases. D, Low-grade serous carcinoma showing micropapillary architecture with some cytoplasmic mucin. COSPv3 falsely predicted EC owing to diffuse expression of the mucinous markers TFF3.

Figure 4

COSPv3 prediction errors: endometrioid versus high-grade serous carcinoma. A, B, C, Endometrioid carcinomas showing glandular (A, B) morphology with mucinous differentiation or solid almost squamoid morphology in C. COSPv3 falsely predicted HGSC owing to the combined presence of WT1 expression and abnormal p53 in these cases. D, High-grade serous carcinoma showing small papillae within a cystic space. Note, high-grade nuclear atypia. COSPv3 falsely predicted EC owing to lack of WT1 expression.

Figure 5

COSPv3 prediction errors: high-grade serous versus low-grade serous carcinoma. A, B, C, D High-grade serous carcinomas showing slit like spaces (A, B, C) or broad papillary architecture with microcystic spaces (D). COSPv3 falsely predicted LGSC owing to the normal p53 expression. We speculate that normal p53 expression maybe due to truncating or splice site TP53 mutation in A and B or due to antigen degradation leading to misinterpretation in C (perhaps overexpression) and D (perhaps cytoplasmic expression).