Time-dependent prediction model using modified DeepSurv algorithm for dynamic risk assessment of post-operative bone metastases in breast cancer.

• DOI: 10.1200/jco.2025.43.16_suppl.e13109
• Publication Date: 2025-05-28T21:11:12Z
• Authors: Fei Ma; Hewei Ge; Jiani Wang; Xiaojia Wang; Jin Yang; Zhenchuan Song; Shusen Wang; Xinhong Wu; Zheng Lv; Yafen Zhang; Quchang Ouyang

e13109 Background: Breast cancer (BC) is the most common in women, with bones being primary site of metastases (~70%). Bone often lead to skeletal-related events (SREs), adversely impacting patients' quality life and survival. For post-operative BC, timely risk identification, screening, intervention for bone will reduce SREs improve However, formulating personized real-time screening recommendations poses a significant challenge. Our study aimed develop model dynamically predict BC goal providing an individual strategy. Methods: We retrospectively analyzed patients aged 20 70 years who were firstly diagnosed between 2010 2023 received surgery from 9 medical centers China (NCT06544668). Muti-center real-world data collected divided into 3 groups: Group A (recurrence metastases), B without C (no recurrence). Univariable, multivariable correlation analyses used identify factors related metastases. Then Cox regression, machine learning (Random Forest, Support Vector Machine), deep (DeepSurv) models modified version incorporating longitudinal disease trajectories constructed respectively. Tenfold cross-validation was hyperparameter tuning. hold-out test evaluate performance concordance index (c-index) as evaluation metric. Results: total 3,970 enrolled (Group A, 2,078 [52.3%]; B, 817 [20.6%]; C, 1,075 [27.1%]). The median time first recurrence 44.1 months. In level ALP gradually increased within year prior metastases, whereas no increase observed other 2 groups. Univariate analysis showed that significantly associated post-surgery baseline characteristics (tumor stage, node pathological grade, nerve invasion, neoadjuvant therapy, Ki-67) dynamitic (ALP lung metastases). Among base prediction models, DeepSurv demonstrated favorable distinguishing dataset c-index 0.67, 3, 5 AUC 0.66 0.69. After data, 0.80, values at 0.81 0.79. Conclusions: This identified new dynamic BC. By trajectory could enable personalized screening. Validation using external databases currently underway.