Using gene expression to identify patients at high risk for late breast cancer recurrence

PROJECT SUMMARY Non-metastatic breast cancer is treated surgically and with adjuvant therapies. As these treatments have improved, recurrence-free survival has steadily increased, and late recurrence (recurrence beyond 10 years’ survival) has become a pressing issue for patients and providers. There are existing and emerging therapies that prophylactically target dormant breast tumors to prevent late recurrence. However, these therapies carry substantial toxicities and should only be used in women who are at high risk for late recurrence. No prognostic assay is recognized by the American Society of Clinical Oncology or by the National Comprehensive Cancer Network to reliably predict recurrence risk beyond 5 to 10 years. Our proposed work differs from earlier studies of late recurrence biomarkers by including large samples of premenopausal and postmenopausal ER-positive breast cancer patients who have survived 10 years recurrence-free, and by following patients for up to 20 years after initial diagnosis. Within strata of menopausal status at diagnosis, we will identify 200 cases of late distant recurrence (occuring >10 years after initial diagnosis and treatment) and match to these 200 recurrence-free controls on follow-up time as well as calendar year, menopausal status, age, and stage at diagnosis. We will identify gene expression profiles specific to late recurrence risk by comparing expression patterns in late recurrences and controls using a robust ensemble of conventional and machine learning models We will bolster the rigor and reproducibility of the identified expression panel by (a) fitting a host of viable models to identify predictive gene expression levels and combining these into a powerful, consolidated ensemble; (b) employing in-sample cross- validation to minimize the likelihood of overfitting and false-positive findings, and (c) assessing independence of disovered genes from genes used in existing early recurrence models. At project completion, we will have developed a primary breast tumor gene expression profile that can stratify ER-positive breast cancer patients with respect to their late recurrence risk. This approach will provide accurate risk stratification and a long lead time for initiating prophylactic therapy with one of several existing and emerging drugs that eradicate or stabilize dormant tumor foci. The genes contributing to our new profile and their associated biological pathways may also suggest new therapeutic targets to prevent late recurrence.