Abstract PO5-26-05: Machine learning-based risk prediction for late distant recurrence in young women with estrogen receptor-positive/human epidermal growth factor 2-negative breast cancer
Dong Seung Shin,Janghee Lee,Jong-Ho Cheun,Jun-Hee Lee,Yewon Shin,Soong June Bae,Eunhye Kang,Sunyoung Kwon,Han-Byoel Lee,Jai Min Ryu,Sung Gwe Ahn
DOI: https://doi.org/10.1158/1538-7445.sabcs23-po5-26-05
IF: 11.2
2024-05-03
Cancer Research
Abstract:Background There is a clinical unmet need to predict the risk of late recurrence in premenopausal women with estrogen receptor (ER)-positive/human epidermal growth factor 2 (HER2)-negative breast cancer (BC) who completed endocrine treatment (ET). Previous studies suggest that the CTS5 model, developed using data from cohorts with postmenopausal women and predicts a likelihood of late recurrence after five years of ET, is not highly predictive in premenopausal women. We developed a machine learning-based model predicting the chance of late distant recurrence (DR) using multi-institutional cohorts of premenopausal women with ER-positive/HER2-negative BC who have had no DR for five years from the surgery. Methods The study conducted a retrospective review of patients who underwent primary surgery between 2000 and 2011 at Samsung Medical Center, Gangnam Severance Hospital, and Seoul National University Hospital. The study included premenopausal women with ER-positive/HER2-negative BC aged 45 years or younger, who were treated over two years of ET with or without ovarian function suppression and had not experienced distant recurrence for at least five years following the surgery. Patients who had received neoadjuvant chemotherapy were excluded from the study. The primary endpoint is the area under the curve (AUC) and sensitivity (Recall) of the model predicting late DR after five years from primary operation. A total of nine clinical features, including age, tumor size, the number of positive lymph nodes, nuclear grade (NG), histologic grade (HG), progesterone receptor (PR) status, chemotherapy, extension of ET, and addition of OFS, were utilized for supervised machine learning classification. Results A total of 2,555 patients were included in this study. The median age was 41 (range, 21-45) years. During a median follow-up duration of 130.3 (range, 60.0-257.6) months, 157 women (6.1%) had late DR after five years of breast cancer surgery. Age, NG, HG, tumor size, and the number of positive LNs were important clinicopathologic variables for late recurrence. The treatment variables included chemotherapy, extension of ET, and the addition of OFS. There were 1,951 patients (76.4%) who had adjuvant chemotherapy, 650 patients (25.4%) with extension of ET over 5 years, and 529 patients (20.7%) who underwent addition of OFS. To develop the model, we performed repeated 5-fold cross-validation using a Balanced Random Forest classifier. The AUC and sensitivity of the model with six features, including age, tumor size, the number of positive lymph nodes, NG, HG, and PR status, were 0.705 and 0.768, respectively. When chemotherapy was added to this model, the AUC and sensitivity of the model increased to 0.713 and 0.771, respectively. When two more treatment features, such as ET extension and addition of OFS, were added to the model, the AUC and sensitivity were 0.714 and 0.752, respectively. Conclusions Demographic and clinical characteristics of the patients in this study Table. Citation Format: Dong Seung Shin, Janghee Lee, Jong-Ho Cheun, Jun-Hee Lee, Yewon Shin, Soong June Bae, Eunhye Kang, Sunyoung Kwon, Han-Byoel Lee, Jai Min Ryu, Sung Gwe Ahn. Machine learning-based risk prediction for late distant recurrence in young women with estrogen receptor-positive/human epidermal growth factor 2-negative breast cancer [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl) nr PO5-26-05.
oncology