We have previously shown that serum metabolomic spectra can be used to predict relapse in a single-centre cohort of ER- early breast cancer (EBC) patients. Here, we investigated this further using serum from a larger cohort of premenopausal, ER+ EBC patients enrolled in a multicentre phase III trial investigating the effect of timing of adjuvant surgical oophorectomy in the menstrual cycle. We also explore the ability of the metabolomic risk score to improve prognostication in patients at different levels of risk by traditional methods.
Methods: Proton NMR spectra were generated for 590 serum samples obtained preoperatively in the adjuvant trial, and 109 serum samples from women with metastatic breast cancer. In a training set, a model was built to discriminate EBC from MBC based on the NMR profiles, using Random Forest classification. A recurrence risk score for EBC patients was generated, based on the likelihood of an EBC sample being misclassified as MBC. It was then applied to a test set of 234 EBC patients with relapse or follow-up greater than 6 years, to predict relapse. A cut off for the risk score was identified using ROC analysis. Exploration of outcome by recurrence score was performed using Kaplan Meier. Impact of individual metabolites is assessed.
Results: The RF model separated EBC from MBC with discrimination accuracy of 84.9% in the training set. In the test set, the RF recurrence risk score correlated with relapse, with an area under the curve of 0.747 in ROC analysis. Accuracy was maximised at 71.3% (sensitivity 70.8%, specificity 71.4%). The model performed independently of age, tumor size, grade, HER2 status and nodal status. When stratified by Adjuvant Online score, the RF risk score improved prognostication in the lower two tertiles.
Conclusions: In a multicentre group of ER+ EBC patients, a model based on preoperative serum metabolomic profiles was prognostic for disease recurrence. This was independent of traditional clinicopathological risk factors, although not discriminatory in the very high risk subgroup by Adjuvant Online.