Abstract PS3-07-16: Validation of Late Distant Recurrence Prediction Model in Premenopausal Women with ER-Positive/HER2-Negative Breast Cancer

Abstract Background: The Clinical Treatment Score post-5 years (CTS5) model was developed to predict late recurrence in estrogen receptor (ER)-positive/human epidermal growth factor receptor 2 (HER2)-negative breast cancer (BC) patients who have completed five years of endocrine therapy (ET). However, CTS5 has shown lower predictive accuracy and risk underestimation in premenopausal women, and the model may not fully reflect the currently improved prognosis as it was developed based on trials conducted decades ago. To address these limitations, we previously developed a machine learning-based model to predict the probability of late distant metastasis (DM) using a multi-institutional cohort of premenopausal women. In this study, we aimed to externally validate our model and compare its performance with that of CTS5. Methods: We retrospective reviewed the patients who underwent primary breast cancer surgery between 2000 and 2013 at Asan Medical Center, Severance Hospital, Boramae Medical Center, and those enrolled in the ASTRRA trial. Eligible patients were premenopausal women aged ≤45 years with ER-positive/HER2-negative BC who received ET with or without ovarian function suppression (OFS). Patients who discontinued ET before 4.5 years, received neoadjuvant chemotherapy, and had distant metastasis within five years were excluded. Our previously developed machine learning-based prediction model utilized eight clinicopathologic features: age, tumor size, number of positive lymph nodes, nuclear grade, histologic grade, progesterone receptor status, use of OFS, and chemotherapy. Patients were stratified into high- or low-risk groups based on a model output probability of 0.480. The primary endpoint was distant metastasis-free survival (DMFS) between five and ten years after surgery. Results: A total of 1,465 patients were included. The median age at the time of operation was 41.0 years (IQR, 38.0-43.0 years). T1 stage and N0 stage tumors were present in 920 (62.8%) and 897 (61.2%) patients, respectively. ET duration was extended in 283 patients (19.3%), and OFS was administered in 537 (36.7%). Among 1,182 patients who did not undergo ET extension, the 10-year DMFS rate was 95.3% during a median follow-up of 157.7 months (IQR 137.6-185.6). The high-risk group had significantly worse DMFS than the low-risk group (p0.001, HR 2.71; 95% CI, 1.60-4.58). When including those who extended ET, high-risk patients appeared to benefit from ET extension (p=0.008, HR 4.36; 95% CI, 1.33-14.29), whereas low-risk patients did not (p=0.126). In comparison with CTS5, our model achieved a slightly better AUC (0.701 vs. 0.654). According to the CTS5 classification, high-risk group demonstrated significantly worse DMFS (p0.001, HR, 3.39; 95% CI, 1.91-6.00), but the intermediate-risk group did not show a different survival outcome (p=0.885) compared to the low-risk group. Notably, 45 among 698 patients classified as low-risk by CTS5 were identified as high-risk by our model, and had significantly worse DMFS (p=0.023, HR, 3.26; 95% CI, 1.11-9.57). Finally, we developed our model using the combined development and validation datasets (n=2,395). The updated model demonstrated robust performance, with an AUC of 0.744, accuracy of 0.718, sensitivity of 0.615, specificity of 0.725, positive predictive value of 0.128, and negative predictive value of 0.966. Conclusions: Our machine learning-based model provides valuable prognostic information on the risk of late distant recurrence between five and ten years after surgery in premenopausal, ER-positive/HER2-negative BC patients. This model would assist clinicians in personalizing decisions regarding the duration of ET and the need for OFS in patients who remain recurrence-free at five years postoperatively. Citation Format: J. Cheun, D. Shin, D. Noh, J. Ahn, Y. Lee, E. Kang, J. Lee, J. Lee, S. Kwon, H. Lee, J. Ryu, S. Ahn. Validation of Late Distant Recurrence Prediction Model in Premenopausal Women with ER-Positive/HER2-Negative Breast Cancer abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-07-16.