Breast cancer is a molecularly heterogeneous disease in which treatment benefit varies by tumor subtype, disease stage and patient characteristics. Precision oncology aims to match treatment to tumor biology, and clinical trials are a gold standard for prospective evaluation of the biomarkers’ clinical utility. Biomarker trials pose distinct methodological and statistical challenges related to the study design, conduct and data analysis. To illustrate these challenges, we will discuss two biomarker-focused trials of CDK4/6 inhibitor–based therapy recently completed by the International Breast Cancer Study Group. PYTHIA was a single-arm phase II biomarker discovery trial, downstream of AURORA molecular screening program, in patients with endocrine-resistant hormone receptor (HR) -positive/HER2-negative metastatic breast cancer treated with palbociclib plus fulvestrant. In this trial we evaluated prognostic role of a novel non-invasive proliferation biomarker serum thymidine kinase activity (sTKa), RNA-Seq based signature of E2F pathway activity linked to functional RB1 loss and CDK4/6 inhibitor resistance (RBsig), and pre-treatment ctDNA detection. TOUCH was a randomized phase II trial in postmenopausal patients with early-stage, HR-positive/HER2-positive breast cancer. It compared neoadjuvant paclitaxel with a chemotherapy-free regimen of palbociclib plus letrozole, both combined with trastuzumab and pertuzumab. Its primary objective was to explore the interaction between the RBsig status (HIGH or LOW) and treatment activity, assessed by pathological complete response, defined as absence of invasive tumor cells in the breast and in the axillary lymph nodes at the time of surgery. The hypothesis was that patients with RBsig HIGH would derive benefit from chemotherapy and might be resistant to CDK4/6 inhibitors. Conversely, patients with RBsig LOW would derive little benefit from chemotherapy and may benefit from CDK4/6 inhibitors. RNA-Seq was performed on pre-treatment biopsies to derive RBsig values. Using these examples we highlight key statistical issues: distinguishing prognostic from predictive biomarkers, dichotomization of continuous markers, biases due to missing specimens.
Dr Svitlana Tyekucheva is a Principal Research Scientist and independent investigator at the Department of Data Science at the Dana-Farber Cancer Institute and Department of Biostatistics at the Harvard T.H. Chan School of Public Health. She is trained in statistics and its applications to genomics. During graduate training she was a member of the Center for Comparative Genomics and Bioinformatics at the Pennsylvania State University. Dr Tyekucheva’s research interests lie in developing new, and tailoring existing, statistical methods to efficiently process and analyze high-dimensional genomic data, including next-generation sequencing, gene expression profiling and metabolomics, to integrate diverse genomic data types using genes sets, and to normalize gene expression data across different studies. Dr Tyekucheva leads the Biostatistics and Computational Core of the Harvard Cancer Consortium Prostate SPORE and co-leads Biostatistics and Computational Biology Core of the AML Program Project and Biostatistics Core of the Leukemia SPORE. Dr. Tyekucheva extensively collaborates with the hematologic malignancies investigators at DFCI. Within this collaboration she works on design and data analysis of basic laboratory experiments, genomic data from clinical cohorts, and development of biomarkers for patient outcomes. She also serves as a primary statistician on multiple Phase I and Phase II clinical trials. Dr Tyekucheva is a member of the International Breast Cancer Study Group (IBCSG) statistical center, where she leads translational analyses for several international clinical trials.