Abstract P1-01-06: Digital image analysis outperforms manual scoring for breast cancer subclassification and prognostication

Abstract

Abstract Introduction Categorization according to the four gene expression-based 'intrinsic' subtypes "Luminal A", "Luminal B", "HER2-enriched" and "Basal-like" is the method of choice for practical prognostic and predictive value in the heterogeneous spectrum of Breast Cancers. Gene expression tests are however not yet universally available, which has created an opportunity for routine immunohistochemical stains to act as surrogate markers (biomarkers) for the gene expression-based subtypes. As recommended by international expert consensus, the expressions of Estrogen receptor α (ER), Progesterone receptor (PR), Human Epidermal Growth factor Receptor 2 (HER2) and the proliferation-associated protein Ki67 are scored during the routine pathological work-up of breast cancer specimens. Thus, congruence of these biomarker tests to the gene expression tests are of utmost importance as discrepancies in classification induces dissimilar treatment decisions. In this study, we compare a novel system for Digital Image Analysis (DIA) with the manual scoring of biomarkers used in current clinicopathological routine and suggest methods to improve their congruence to gene expression assays, provide more robust prognoses for survival as well as reduce time consumption for pathologists. Methods 1 tissue micro array (TMA) cohort and 2 cohorts of primary breast cancer specimens (total n = 436) with >20 years survival data, were sectioned into physical glass slides and digitally scanned at ×20 and then reviewed for manual vs. DIA test congruence to PAM50 gene expression assays in terms of classification into the four intrinsic subtypes, and their prognostic power. This included the evaluation of 6 different methods for DIA biomarker testing. The DIA system used was the Visiopharm Integrator System (VIS) by Visiopharm A/S, Hoersholm, Denmark. In short, this system facilitates ER, PR and Ki67-testing by a 'sandwich' technology, in which each biomarker slide is aligned with an adjacent 3 μm slide stained with a pancytokeratin marker. Thus, non-tumor tissue is to a large extent automatically excluded from analysis and only cells that express cytokeratin are eligible for automatic detection of positivity or negativity of the respective biomarker. Results 60,8 % of the cases in the TMA cohort was classified in concordance with PAM50 (κ = 0,46) using DIA. Classification with regard to HER2-enriched, Basal-like and Luminal tumors without dichotomization of A and B subtype (thereby avoiding the impact of Ki67-scoring) was 80,9 % (κ = 0,63). In the whole slide cohorts, DIA performed with 95 % of the cases classified in concordance with PAM50 (κ = 0,90), compared to 67 % (κ = 0,59) for the manual method recommended by international consensus. In addition to this DIA produced better prognostication vs. the manual method in terms of hazard of all-cause death for each subtype. Conclusion The system for DIA evaluated here outperforms manual scoring in both predictive and prognostic value. It also has the potential to reduce time consumption for pathologists, as many of the steps in the workflow is either automatic or feasible to manage without pathological expertise. Based on the findings in this study however, TMA cannot be recommended for DIA scoring of Ki67. Citation Format: Stålhammar G, Rosin G, Kis L, Lippert M, Moelholm I, Grunkin M, Bergh J, Hartman J. Digital image analysis outperforms manual scoring for breast cancer subclassification and prognostication. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P1-01-06.