PO.PS01.02 · 人群科学
韩国乳腺癌的性别特异性生存模式：来自K-CURE队列的可解释AI洞见
Sex-specific survival patterns in korean breast cancer: Explainable AI insights from K-CURE Cohort
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
背景：男性乳腺癌（MBC）罕见，但其发病率在全球范围内正在上升。关于其与女性乳腺癌（FBC）相比生存差异的证据有限。本研究旨在描述韩国乳腺癌的性别特异性生存模式，并使用可解释人工智能（XAI）方法识别关键促成因素。
方法：本研究利用了韩国临床数据研究卓越应用（K-CURE）的数据，这是一个涵盖韩国所有乳腺癌病例的全国性登记库。对2012年至2021年间确诊的患者按性别进行分析。使用多变量Cox比例风险模型评估总生存期（OS）和乳腺癌特异性生存期（BCSS）。应用包括SHAP和LIME在内的XAI技术，以识别生存的性别特异性影响因素。
结果：在200,222例患者中（846例男性，199,376例女性），男性的OS和BCSS显著低于女性（p<.001）。这一差异在校正后仍然存在（男性OS HR≈1.30-1.40；BCSS HR≈1.40-1.50）。远处分期在两性中均与最高死亡风险相关（男性HR≈9-12；女性HR≈10-24）。观察到独特的性别特异性模式。在男性中，代谢指标与生存呈相反关联：较高的血红蛋白与较低的死亡率相关（HR≈0.88），而较高的空腹血糖与死亡风险增加相关（HR≈1.04）。在女性中，激素治疗和靶向治疗与死亡率降低相关（HR≈0.30-0.55）。XGBoost模型对两项结局均达到中等预测性能（OS AUC≈0.85；BCSS AUC≈0.87）。SHAP和LIME分析一致支持这些模式，提示代谢特征对男性的风险估计贡献更大，而肿瘤分期和治疗因素对女性的影响相对更大。
结论：这些发现表明，乳腺癌生存的性别差异不仅源于总体风险特征的差异，还源于各性别特有的不同预后因素。在男性中，代谢指标与生存的关联更为密切，而在女性中，肿瘤分期和治疗因素的相关性更大。总体而言，这些结果反映了不同性别间不同的潜在机制，并凸显了在乳腺癌照护中制定量身定制的性别特异性管理策略的必要性。
查看英文原文 English abstract
Background: Male breast cancer (MBC) is rare, but its incidence is increasing worldwide. Evidence regarding survival differences compared with female breast cancer (FBC) is limited. This study aimed to characterize sex-specific survival patterns in Korean breast cancer and to identify key contributing factors using explainable artificial intelligence (XAI) methods.
Methods: This study utilized data from the Korean Clinical Data Utilization for Research Excellence (K-CURE), a nationwide registry of all breast cancer cases in Korea. Patients diagnosed between 2012 and 2021 were analyzed by sex. Overall survival (OS) and breast cancer-specific survival (BCSS) were assessed using multivariable Cox proportional hazards models. XAI techniques, including SHAP and LIME, were applied to identify sex-specific contributors to survival.
Results: Among 200,222 patients (846 males, 199,376 females), males showed significantly poorer OS and BCSS than females ( p <.001). This disparity persisted after adjustment (male OS HR ≈1.30-1.40; BCSS HR ≈1.40-1.50). Distant stage was associated with the highest mortality risk in both sexes (male HR ≈9-12; female HR ≈10-24). Distinct sex-specific patterns were observed. In males, metabolic indicators showed opposite associations with survival: higher hemoglobin was linked to lower mortality (HR ≈0.88), whereas higher fasting blood sugar was associated with increased mortality risk (HR ≈1.04). In females, hormone and targeted therapies were associated with reduced mortality (HR ≈0.30-0.55). XGBoost models achieved moderate predictive performance for both outcomes (OS AUC ≈0.85; BCSS AUC ≈0.87). SHAP and LIME analyses consistently supported these patterns, suggesting that metabolic profiles contributed more to risk estimation in males, whereas tumor stage and treatment factors were relatively more influential in females.
Conclusion: These findings indicate that sex differences in breast cancer survival arise not only from differences in overall risk profiles but also from distinct prognostic factors specific to each sex. In males, metabolic indicators were more closely associated with survival, whereas in females, tumor stage and treatment factors had greater relevance. Overall, these results reflect differing underlying mechanisms by sex and highlight the need for tailored, sex-specific management strategies in breast cancer care.
利益披露 Disclosure
S. Ahn, None..
J. Hwang, None..
S. Kong, None..
J. Park, None..
S. Jung, None..
H. Kim, None.