基于机器学习算法对HR阴性/HER-2不同表达乳腺癌的临床病理特征建模分析筛选  

Analysis of the clinicopathological characteristics of breast cancer with negative HR and different HER-2 expression via modelling based on machine learning algorithm

作　　者：何杨 韩书文 陈传荣 周珏[3] HE Yang;HAN Shuwen;CHEN Chuanrong;ZHOU Jue(Department of Oncology,The First Affiliated Hospital of Wannan Medical College,Wuhu 241001,Anhui,China)

机构地区：[1]皖南医学院第一附属医院肿瘤内科,芜湖241001 [2]湖州市中心医院肿瘤内科,浙江湖州313000 [3]皖南医学院病理学教研室,安徽芜湖241002

出　　处：《皖南医学院学报》2025年第2期137-141,共5页Journal of Wannan Medical College

基　　金：国家自然科学基金面上项目(82373294);安徽省自然科学基金项目(2208085MH243);安徽省高等学校自然科学重点项目(2022AH051237)。

摘　　要：目的:以机器学习算法为基础分析激素受体(HR)阴性/人表皮生长因子受体2(HER-2)不同表达的乳腺癌临床病理特征差异。方法:采用免疫组化技术及原位杂交技术检测乳腺癌标本中雌激素受体(ER)、孕激素受体(PR)及HER-2的表达,并构建机器学习算法模型分析其与乳腺癌临床病理特征的关系。结果:①共筛选出183例ER和PR均为阴性乳腺癌,其中HER-2阳性96例,HER-2阴性87例(HER-2零表达39例,HER-2低表达48例)。②采用机器学习算法选择准确率最高的随机森林模型对183例HR-乳腺癌临床病理特征进行筛选,发现三阴型最具重要性的特征为淋巴细胞/单核细胞比值(LMR)(变量重要性=0.243),其次为病理分级、血小板/淋巴细胞比值(PLR)、Ki-67表达等。③机器学习算法选择准确率最高的随机森林模型筛选出HER-2零表达组最重要的特征为Ki-67表达(变量重要性=0.222),其次为BMI、LMR、PLR等。结论:与HR阴性/HER-2阳性型乳腺癌相比,三阴型乳腺癌的恶性程度更高,预后更差。三阴型乳腺癌中HER-2零表达免疫微环境中免疫炎性细胞更为活跃,可能从免疫检查点抑制剂治疗后获益更多;而HER-2低表达可优先考虑抗体药物偶联物治疗。Objective:To analyze the differences of clinicopathological characteristics of breast cancer with hormone receptor(HR)-negative(HR-)and different expression of human epidermal growth factor receptor 2(HER-2)based on machine learning algorithm.Methods:Immunohistochemistry and in situ hybridization were used to detect the expression of estrogen receptor(ER),progesterone receptor(PR)and HER-2 in the surgical specimens of breast cancer,and then machine learning algorithm model was established to analyze the relationship between the clinicopathological features of breast cancer and the gene expression aforementioned.Results:①A total of 183 cases of breast cancer with negative ER and PR were screened,in which 96 cases were HER-2 positive and 87 were HER-2 negative(zero expression of HER-2 in 39 cases,and low HER-2 expression in 48 cases);②The random forest model with the highest accuracy selected by machine learning algorithm was used to screen the clinicopathological characteristics of 183 cases of HR-breast cancer,the results revealed that the most important feature of triple negative type was lymphocyte to monocyte ratio(LMR)(variable importance=0.243),followed by pathological grade,platelet-lymphocyte ratio(PLR)and Ki67 expression;③The random forest model with the highest accuracy selected by the machine learning algorithm demonstrated that the most important property was associated with Ki-67 expression in the samples with zero HER-2 expression(variable importance=0.222),followed by BMI,LMR and PLR.Conclusion:Triple-negative breast cancer can be highly malignant and poorer in prognosis compared with HR-/HER-2 positive neoplasms.Immunoinflammatory cells in the immune microenvironment may be more active in triple-negative breast cancers with HER-2 zero expression,suggesting that such patients can benefit more from treatment with immune checkpoint inhibitors,yet antibody-drug conjugates may be prioritized in patients with low HER-2 expression.

关 键 词：机器学习算法 HR HER-2 乳腺癌 

分 类 号：R737.9[医药卫生—肿瘤] R365[医药卫生—临床医学]