A validated model to predict low recurrence risk distinguished by 21-gene recurrence score in hormone receptor-positive invasive breast cancer patients.

Abstract

532 Background: In the prospective TAILORx and RxPONDER trials, the 21-gene Recurrence Score (RS) showed endocrine therapy alone was non-inferior to chemo-endocrine therapy in the analysis of invasive disease-free survival in postmenopausal hormone-receptor (HR)-positive breast cancer patients with RS < = 25. They also indicated chemotherapy was associated with benefit for women 50 years or younger with RS 11 to 25. However, in Japan, the test is not conventionally available because of non-coverage by national insurance. We aimed to develop and validate a model to predict RS using clinicopathological factors that identify patients who would have low risk shown by testing the 21-gene RS and can avoid chemotherapy. Methods: Four hundred patients, including 187 N0/1 postmenopausal, and 213 N0 premenopausal women who underwent surgery and had the RS from St. Luke’s International Hospital, Tokyo, Japan, were included in derivation cohort. Derivation cohort was divided into 2 groups by RS 25; patients with RS of 0 to 25 (n = 321) and with RS over 26 (n = 79). Multivariate logistic regression analysis was performed using candidate factors for all patients and pre- or postmenopausal patients. The prediction model was validated using an external cohort of 70 patients from Showa University School of Medicine, Tokyo, Japan. Results: Nuclear grade (NG) (adjusted OR, 5.28, 95% CI, 2.47–11.30), high Progesterone receptor (PgR) expression (Allred score 7-8) (adjusted OR, 10.62, 95% CI, 5.34–21.13) and low Ki67 level ( < = 20%) (adjusted OR, 5.29, 95% CI, 2.33-12.01) were significant independent predictors of RS of 0 to 25. With these factors could predict RS of 0 to 25 (AUC of 0.848, 95% CI, 0.803-0.893) with the highest probability of low-RS for 100%. The prediction model of the validation cohort had same discriminatory ability having an AUC of 0.812 (95% CI, 0.701-0.923). In postmenopausal patients, NG (adjusted OR, 4.81, 95% CI, 1.72–13.42), high PgR expression (adjusted OR, 10.62, 95% CI, 4.52–37.72), and low Ki67 level (adjusted OR, 4.94, 95%CI, 1.87-13.04) were significantly associated with RS of 0 to 25 in multivariate analysis. A regression model with these 4 factors could predict RS of 0 to 25 (AUC of 0.842, 95%CI, 0.782-0.902). In premenopausal patients, NG (adjusted OR, 8.76, 95% CI, 1.14–67.40), high PgR expression (adjusted OR, 3.22, 95% CI, 1.61–6.43), and low Ki67 level (adjusted OR, 2.87, 95% CI, 1.20–6.87) were significantly associated with RS of 0 to 10 in multivariate analysis. These factors could predict RS of 0 to 10 (AUC of 0.811, 95% CI, 0.731-0.891). However, the highest probability of low-RS provided this model for premenopausal women was 46.8%. Conclusions: Our validated model could provide useful information to distinguish low-RS especially for postmenopausal patients with high reproducibility. However, for premenopausal women, the 21-gene RS is warranted.