Explainability of deep learning methods is imperative to facilitate their\nclinical adoption in digital pathology. However, popular deep learning methods\nand explainability techniques (explainers) based on pixel-wise processing\ndisregard biological entities' notion, thus complicating comprehension by\npathologists. In this work, we address this by adopting biological entity-based\ngraph processing and graph explainers enabling explanations accessible to\npathologists. In this context, a major challenge becomes to discern meaningful\nexplainers, particularly in a standardized and quantifiable fashion. To this\nend, we propose herein a set of novel quantitative metrics based on statistics\nof class separability using pathologically measurable concepts to characterize\ngraph explainers. We employ the proposed metrics to evaluate three types of\ngraph explainers, namely the layer-wise relevance propagation, gradient-based\nsaliency, and graph pruning approaches, to explain Cell-Graph representations\nfor Breast Cancer Subtyping. The proposed metrics are also applicable in other\ndomains by using domain-specific intuitive concepts. We validate the\nqualitative and quantitative findings on the BRACS dataset, a large cohort of\nbreast cancer RoIs, by expert pathologists.\n