自动乳腺超声联合免疫组化早期预测乳腺癌新辅助化疗病理完全缓解的临床价值

Clinical value of automated breast ultrasound combined with immunohistochemistry in early prediction of pathological complete response to neoadjuvant chemotherapy in breast cancer

目的探讨自动乳腺超声(ABUS)联合免疫组化早期预测乳腺癌患者新辅助化疗(NAC)后获得病理完全缓解(pCR)的临床价值。方法收集2019年3月至2022年6月于空军军医大学第一附属医院(西京医院)乳腺外科收治的420例女性乳腺癌患者,所有患者均经穿刺病理证实为浸润性乳腺癌,患者在行NAC前均行ABUS检查,根据术后病理结果将乳腺癌患者分为pCR组153例和非病理完全缓解(npCR)组267例。采用多因素Logistic回归分析乳腺癌pCR的独立预测因素,构建3种预测模型分别为ABUS特征预测模型、免疫组化预测模型、ABUS联合免疫组化的联合预测模型。绘制3种模型的ROC曲线评估模型效能,通过DeLong检验比较模型的ROC曲线下面积,选择最优模型构建列线图。通过Hosmer-Lemeshow检验分析模型拟合度,采用Bootstrap法对模型进行内部验证,绘制校准度曲线对模型性能进行评估,使用决策曲线评估预测模型的临床获益。结果乳腺癌NAC前pCR组与npCR组肿瘤的雌激素受体(ER)状态、孕激素受体(PR)状态、人表皮生长因子受体-2(HER-2)状态、Ki67状态和分子分型差异有统计学意义(P均<0.05);NAC前pCR组与npCR组肿瘤的皮肤侵犯、形态、后方回声、冠状面汇聚征、冠状面白墙征差异有统计学意义(P均<0.05)。多因素Logistic回归分析显示,HER-2状态(OR值:4.396,95%CI:2.089~9.250)、分子分型(HER-2过表达型OR值:4.124,95%CI:2.069~8.219;三阴性型OR值:4.775,95%CI:2.188~10.422)、皮肤侵犯(OR值:0.110,95%CI:0.038~0.318)、冠状面汇聚征(OR值:0.416,95%CI:0.220~0.787)、冠状面白墙征(OR值:11.192,95%CI:5.316~23.561)是pCR的独立预测因子(P均<0.05)。DeLong检验表明基于ABUS模型和免疫组化特征构建的联合模型的ROC曲线下面积(AUC=0.855)显著高于免疫组化模型(AUC=0.758)和ABUS模型(AUC=0.774),差异均有统计学意义(Z=5.11,P<0.001;Z=5.49,P<0.001);ABUS模型的ROC曲线下面积(AUC=0.774)略高于免疫组化模型(AUC=0.758),但差异无统计学意义(Z=0.55,P=0.58)。Hosmer-Lemeshow检验表明联合模型拟合度良好(χ^(2)=6.175,P=0.519);内部验证结果显示,C-指数为0.856(95%CI:0.810~0.886),表明模型稳定性良好;校准曲线表明联合模型的预测概率与实际概率一致性良好;决策曲线表明在广泛阈值范围内,联合模型具有较好的临床获益。结论基于NAC前肿瘤的ABUS特征联合免疫组化构建联合模型可以早期准确预测乳腺癌NAC后pCR,可为乳腺癌患者临床治疗方案的制定提供依据。

Objective To evaluate the clinical value of automated breast ultrasound(ABUS)combined with immunohistochemistry in early prediction of pathological complete response(pCR)to neoadjuvant chemotherapy(NAC)in breast cancer patients.Methods A total of 420 female breast cancer patients admitted to the Department of Breast Surgery,the First Affiliated Hospital of Air Force Military Medical University(Xijing Hospital)from March 2019 to June 2022 were included.All patients were confirmed to have invasive breast cancer by puncture pathology.All patients underwent ABUS examination before NAC.According to the postoperative pathological results,the patients were divided into either a pCR group(n=153)or a non-pathological complete response(npCR)group(n=267).Multivariate Logistic regression was used to identify the independent predictors of pCR in breast cancer.Three prediction models were constructed:ABUS feature prediction model,immunohistochemical prediction model,and ABUS combined with immunohistochemical prediction model.The receiver operating characteristic(ROC)curves of the three models were plotted to evaluate their performance.The area under the ROC curve(AUC)of the models was compared by the Delong test,and the optimal model was selected to construct a nomogram.The fitness of the model was analyzed by the Hosmer-Lemeshow test,and the Bootstrap method was used to verify the model internally.The calibration curve was drawn to evaluate the performance of the model,and the decision curve was used to evaluate the clinical benefit of the prediction model.Results There were significant differences in estrogen receptor(ER)status,progesterone receptor(PR)status,human epidermal growth factor receptor(HER-2)status,Ki67 status,and molecular typing between the pCR group and npCR group before NAC(P<0.05 for all).There were also significant differences in skin invasion,morphology,posterior echo,retraction phenomenon on coronal images,and white wall sign on coronal images between the two groups before NAC(P<0.05 for all).Multivariate Logistic regression analysis showed that HER-2 status(odds ratio[OR]=4.396,95%confidence interval[CI]:2.089-9.25),molecular typing(HER-2 overexpression type:OR=4.124,95%CI:2.069-8.219);triple negative type(OR=4.775,95%CI:2.188-10.422),skin invasion(OR=0.110,95%CI:0.038-0.318),retraction phenomenon on coronal images(OR=0.416,95%CI:0.220-0.787),and white wall sign on coronal images(OR=11.192,95%CI:5.316-23.561)were independent predictors of pCR(P<0.05 for all).The Delong test showed that the AUC(0.855)of the combined model(nomogram)based on the ABUS model and immunohistochemical features was significantly higher than that of the immunohistochemical model(AUC=0.758;Z=5.11,P<0.001)and the ABUS model(AUC=0.774;Z=5.49,P<0.001);the AUC(0.774)of the ABUS model was slightly higher than that of the immunohistochemical model(AUC=0.758),but the difference was not statistically significant(Z=0.55,P=0.58).The Hosmer-Lemeshow test showed that the model fit well(χ^(2)=6.175,P=0.519).The results of internal validation showed that the average C-index was 0.856(95%CI:0.810-0.886),indicating that the model had good stability.Calibration curve analysis showed that the prediction probability of the combined model was in good agreement with the actual probability.Decision curve analysis showed that the combined model had better clinical benefits in a wide range of thresholds.Conclusion The combined model based on ABUS features of pre-NAC tumors combined with immunohistochemistry can accurately predict pCR to NAC in breast cancer,which can provide a basis for the formulation of clinical treatment plans for breast cancer patients.