Added Value of Serum Hormone Measurements in Risk Prediction Models for Breast Cancer for Women Not Using Exogenous Hormones: Results from the EPIC Cohort

<b>Purpose:</b> Circulating hormone concentrations are associated with breast cancer risk, with well-established associations for postmenopausal women. Biomarkers may represent minimally invasive measures to improve risk prediction models.<b>Experimental Design:</b> We evaluated improvements in discrimination gained by adding serum biomarker concentrations to risk estimates derived from risk prediction models developed by Gail and colleagues and Pfeiffer and colleagues using a nested case-control study within the EPIC cohort, including 1,217 breast cancer cases and 1,976 matched controls. Participants were pre- or postmenopausal at blood collection. Circulating sex steroids, prolactin, insulin-like growth factor (IGF) I, IGF-binding protein 3, and sex hormone-binding globulin (SHBG) were evaluated using backward elimination separately in women pre- and postmenopausal at blood collection. Improvement in discrimination was evaluated as the change in concordance statistic (C-statistic) from a modified Gail or Pfeiffer risk score alone versus models, including the biomarkers and risk score. Internal validation with bootstrapping (1,000-fold) was used to adjust for overfitting.<b>Results:</b> Among women postmenopausal at blood collection, estradiol, testosterone, and SHBG were selected into the prediction models. For breast cancer overall, model discrimination after including biomarkers was 5.3 percentage points higher than the modified Gail model alone, and 3.4 percentage points higher than the Pfeiffer model alone, after accounting for overfitting. Discrimination was more markedly improved for estrogen receptor-positive disease (percentage point change in C-statistic: 7.2, Gail; 4.8, Pfeiffer). We observed no improvement in discrimination among women premenopausal at blood collection.<b>Conclusions:</b> Integration of hormone measurements in clinical risk prediction models may represent a strategy to improve breast cancer risk stratification. <i>Clin Cancer Res; 23(15); 4181-9. ©2017 AACR</i>.