Validation of the PREDICT breast cancer tool in a multiethnic population of U.S. women after a second primary breast cancer.

Abstract

10551 Background: PREDICT is a widely used clinical tool to estimate prognosis based on treatment after a diagnosis of early-stage breast cancer (BC). However, it has not been validated for use in women diagnosed with a second primary BC. Methods: Women diagnosed with one or two BCs between 2000-2013 were identified from the US Surveillance, Epidemiology, and End Results (SEER) program. Validation analyses were conducted separately among women with one and two BCs. For each group, absolute and relative differences in predicted (P) and observed (O) 5-year BC-specific mortality were estimated stratified by estrogen receptor (ER) status. For women with two BCs, mortality prediction was based on demographic and tumor characteristics of their second BCs. Model discrimination was evaluated by receiver-operator curve (ROC) and area under the curve (AUC). Results: 6,729 women diagnosed with a second BC and 357,204 women with only one BC were included. Among women with one ER-positive BC, the predicted (P) and observed (O) BC mortality were comparable [absolute difference (P-O): -0.22%, 95%CI: -0.29%, -0.15%]. For women diagnosed with a second ER-positive disease, the model largely underestimated the BC mortality (P-O: -6.24%, 95%CI: -6.96%, -5.49%). An underestimation of 3%-15% in absolute BC mortality was observed in subgroup analyses based on age, race/ethnicity [P-O for Non-Hispanic (NH) White, NH Black, and Hispanic were -5.63%, -9,3%, and -9.24%], chemotherapy (P-O: -9.06, 95%CI: -10.79, -7.4), radiation therapy (P-O: -5.06, 95%CI: -6.29, -3.93) and tumor characteristics. PREDICT overestimated BC mortality for women diagnosed with one ER-negative BC (P-O: 4.54%, 95%CI: 4.27%, 4.86%) but it performed well for women diagnosed with a second ER-negative BC (P-O: -1.69%, -3.99%, 0.16%). However, in subgroup analyses among women with two BCs the tool underestimated mortality for women with a second ER-negative disease if their first cancer was diagnosed before age 50 years (P-O: -3.27, 95%CI: -6.48, -0.06), was poorly differentiated (P-O: -4.08, 95%CI: -7.09, -1.2), or ER-negative (P-O: -4.68, 95%CI: -7.62, -1.39). The overall AUCs for first and second ER-positive and ER-negative BCs ranged between 0.73 and 0.82. Conclusions: The PREDICT tool largely underestimated the 5-year BC-specific mortality in women diagnosed with a second ER-positive breast cancer and in many subgroups of women diagnosed with a second ER-negative cancer. Novel prognostic tools are needed for women diagnosed with a second breast cancer.