A large-scale retrospective study in metastatic breast cancer patients using circulating tumor DNA and machine learning to predict treatment outcome and progression-free survival

Abstract

PurposeMonitoring levels of circulating tumor-derived DNA (ctDNA) represents a non-invasive snapshot of tumor burden and potentially clonal evolution. Here we describe how a novel statistical model that uses serial ctDNA measurements from shallow whole genome sequencing (sWGS) in metastatic breast cancer patients produces a rapid and inexpensive assessment that is predictive of treatment response and progression-free survival. Patients and MethodsA cohort of 188 metastatic breast cancer patients had DNA extracted from serial plasma samples (total 1098, median=4, mean=5.87). Plasma DNA was assessed using sWGS and the tumor fraction in total cell free DNA estimated using ichorCNA. This approach was compared with ctDNA targeted sequencing and serial CA 15-3 measurements. The longitudinal ichorCNA values were used to develop a Bayesian learning model to predict subsequent treatment response. ResultsWe identified a transition point of 7% estimated tumor fraction to stratify patients into different categories of progression risk using ichorCNA estimates and a time-dependent Cox model, validated across different breast cancer subtypes and treatments, outperforming the alternative methods. We then developed a Bayesian learning model to predict subsequent treatment response with a sensitivity of 0.75 and a specificity of 0.66. ConclusionIn patients with metastatic breast cancer, sWGS of ctDNA and ichorCNA provide prognostic and predictive real-time valuable information on treatment response across subtypes and therapies. A prospective large-scale clinical trial to evaluate clinical benefit of early treatment changes based on ctDNA levels is now warranted.