Early detection of micrometastases, or metastases traditionally too small to be discovered, may improve risk stratification and administration of adjuvant therapies in patients with early-stage breast cancer. Cancer cells, along with their microenvironment and the systemic immune response, “can yield a metabolomic ‘signature’ of cancer presence, detectable in blood or other samples,” said Dr Christopher Hart of Prato, Italy. In fact, it was shown in a previous study that postoperative serum metabolomic profiles can be predictive of future disease relapse in a single-center cohort of estrogen receptor–negative early breast cancer patients. Here, Dr Hart and colleagues explored whether preoperative serum metabolomic profiles could be used as a prognostic indicator of disease recurrence, independent of traditional clinicopathologic risk factors, in a multicenter cohort of estrogen receptor–positive (ER+), premenopausal women with early breast cancer.
To develop this model, samples from a large international phase 2 trial were first analyzed by proton nuclear magnetic resonance spectroscopy. In total, 590 early breast cancer samples (319 of which had relapse or ≥6 years of clinical follow-up) and 109 metastatic breast cancer samples were collected. Using a training set of 85 early (nonrelapsed) and all 109 of the metastatic breast cancer samples, a Random Forest classification model was built that was able to discriminate between early and metastatic breast cancer with an 84.9% accuracy rate. This model was then applied to a test set of 234 early breast cancer samples, and a risk of recurrence score was generated based on the likelihood of the sample to be misclassified as metastatic.
In the test set, the Random Forest risk of recurrence score was well correlated with relapse, having an area under the curve of 0.747 in the receiver-operating characteristic analysis. Accuracy of this model was ultimately maximized at 71.3%, with a sensitivity of 70.8% and a specificity of 71.4%. Notably, the model functioned independently of typical clinicopathologic risk factors, including age, tumor size, grade, human epidermal growth factor receptor 2 status, nodal status, and Adjuvant! Online risk of relapse score.
These data confirm that, independent of traditional risk factors, preoperative serum metabolomic profiles can be predictive of relapse in premenopausal women with ER+ early breast cancer. “Metabolomics has the potential to improve risk stratification in early breast cancer by detecting the presence of micrometastatic disease,” said Dr Hart. Further work is ongoing to make this a practical and transferable tool.
Hart C, et al. ESMO 2016. Abstract 152PD.