Background: The MONARCH 2 trial (NCT02107703) showed the efficacy of abemaciclib, a cyclin-dependent kinase 4 & 6 inhibitor (CDK4/6i), in combination with fulvestrant for hormone receptor-positive, HER2-negative metastatic breast cancer (MBC). The aim of this analysis was to explore the prediction of circulating tumor cells (CTCs) stratification using machine learning for hypothesis generation of biomarker-driven clinical trials. Patients and Methods: Predicted CTCs were computed in the MONARCH 2 trial through a K nearest neighbor (KNN) classifier trained on a dataset comprising 2436 patients with MBC. Patients were categorized into predicted Stage IVaggressive (pStage IVaggressive, ≥5 predicted CTCs) or predicted Stage IVindolent (pStage IVindolent, <5 predicted CTCs). Prognosis was tested in terms of progression-free-survival (PFS) and overall survival (OS) through Cox regression. Results: Patients classified as predicted pStage IVaggressive and predicted pStage Stage IVindolent were, respectively, 183 (28%) and 461 (72%). After multivariable Cox regression, predicted CTCs were confirmed as independently associated with prognosis in terms of OS, together with ECOG performance status, liver involvement, bone-only disease, and treatment arm. Patients in the pStage Stage IVindolent subgroup treated with abemaciclib experienced the best prognosis both in terms of PFS and OS. The treatment effect of abemaciclib on OS was then explored through subgroup analysis, showing a consistent benefit across all subgroups. Conclusion: This study is the first analysis of CTCs modeling for stage IV disease stratification. These results show the need to expand biomarker profiling in combination with CTCs stratification for improved biomarker-driven drug development.
