Ki-67 proliferation index to further stratify invasive breast cancer molecular subtypes: Northern African comparative cohort-study with external TCGA-BRCA and METABRIC validation

Abstract

Introduction: breast cancer (BC) is a malignancy with very high incidence and mortality in Africa, especially in Western Africa, where more than 25 thousand deaths are registered every year. Not all BC have the same prognosis, and being able to personalize treatment and predict aggressiveness is of crucial importance. The purpose of our study is to explore further subdivisions associated with prognosis, beyond breast cancer molecular classification that is routinely established in pathology departments.

Methods: we conducted a 5-year retrospective cohort study on 1266 invasive BC of Moroccan patients, collected at the Pathology Department of Ibn-Rochd University Hospital in Casablanca, and followed at King Mohammed VI National Centre for the Treatment of Cancers. We elaborated an Estimation-Maximization Clustering, based on the main BC biomarkers: Ki-67, HER2, estrogen and progesterone receptors, evaluated by immunohistochemistry. Two independent datasets (TCGA-BRCA and METABRIC) were also analyzed to assess the external reproducibility of the results.

Results: each molecular subgroup could be partitioned into two further subdivisions: Cluster1, with average Ki-67 of 16.26% (±11.9) across all molecular subgroups and higher frequency within luminal BC, and Cluster2, with average Ki-67 of 68.8%(±18) across all molecular subgroups and higher frequency in HER2 as well as in triple-negative BC. Overall survival of the two Clusters was significantly different, with 5-year rates of 52 and 37 months for Custer1 and Cluster2, respectively (p=0.000001). Moreover, mortality rates within the same molecular subgroup, especially in luminal B HER2-, varied remarkably depending on Cluster membership (6% for C1 and 18% for C2 after 1 year of follow-up). Two different algorithms to evaluate the prognostic importance, variable selection using random forests (VSURF) and Minimal depth, ranked the subdivision proposed as one of the 4 most influential features being able to predict patient survival better than several histoprognostic features, both in the Moroccan and in the external datasets.

Conclusion: our results highlight a new refinement of the BC molecular classification and provide a simple and improved way to classify tumors that could be applied in low to middle-income countries. This is the first study of its kind addressed in an African context.