A lasso-based model combining miRNA and clinical variables predicts future risk of breast and ovarian cancer

Hereditary breast and ovarian cancer syndrome (HBOC) is principally caused by germline mutations in BRCA1 and BRCA2. However, most women with HBOC are undiagnosed, and some patients meeting clinical criteria for HBOC will have no identifiable mutation after genetic testing. Here, we deploy a lasso-based model to combine serum miRNA profiles with clinical data to identify women at elevated risk for ovarian cancer among a population of 1831 individuals enrolled in an institutional biobank. The miRNA and metadata variables are mapped to two-dimensional space using lasso, after which a linear classification model is trained to estimate "BRCAness" and long-term risk of cancer. After tenfold cross-validation, the method offers a BRCA prediction AUC score of 0.98 (95% CI 0.94-1.0) and generalizes across subgroups stratified by age, cancer history, and racial/ethnic group. To demonstrate the clinical relevance of this phenotype, we use the lasso-based model to assess 5-year ovarian cancer risk among an independent cohort of 1044 subjects agnostic to genetic testing results enrolled in a randomized clinical trial. In this unselected population, the output of the lasso-based model strongly correlates to the log 5-year relative risk of ovarian cancer (R = 0.93, 95% CI 0.83-0.97, p < 0.0001). When the model was used to predict future onset of ovarian cancer directly, the AUC offered was AUC = 0.75 (95% CI 0.70-0.78). Together, these data suggest the proposed model is a predictor of future ovarian cancer risk.