乳腺癌患者新辅助化疗后腋窝淋巴结转移预测模型的建立

Establishment of a predictive model of axillary lymph node metastasis after neoadjuvant chemotherapy in breast cancer patients

目的通过分析影响乳腺癌患者新辅助化疗(NCT)后腋窝淋巴结转移(ALNM)的相关因素,建立预测ALNM的模型,为筛选出NCT后ALNM的患者提供一定的依据。方法回顾性分析2014年7月至2018年4月在西南医科大学附属医院接受NCT的196例乳腺癌患者资料。以患者年龄、月经状态、NCT后肿瘤分期、术前腋窝淋巴结状态、瘤体位置、化疗方案、化疗次数、ER状态、HER-2状态和Ki67表达作为评价指标,采用t检验、χ~2检验或Fisher精确概率检验分析各临床病理特征与ALNM之间的关系,将以上因素作为输入变量进行方程拟合,以Akaike信息准则值最小时的因素纳入多因素分析,构建预测模型,并画出列线图。根据患者临床病理特征在列线图中量化出ALNM的风险,采用受试者工作特征(ROC)曲线评估此模型的预测效果,最后采用十折交叉验证法将患者分为10组,轮流将其中9组用于建模,1组用于验证此预测模型的可靠性。结果 NCT后腋窝淋巴结未转移组(n=49)与转移组(n=147)比较,NCT后肿瘤分期、术前腋窝淋巴结状态、肿瘤位置、ER和HER-2状态的差异均有统计学意义(χ~2=20.876,P<0.001;χ~2=57.342,P<0.001;χ~2=13.800,P=0.008;χ~2=15.041,P<0.001;χ~2=5.770,P=0.016)。多因素分析结果显示:术前ALNM、ER阳性及Ki67低表达是NCT后ALNM的独立危险因素(OR=30.27,95%CI:10.57～108.28,P<0.001;OR=0.28,95%CI:0.11～0.69,P=0.007;OR=0.96,95%CI:0.93～1.00,P=0.032)。该预测模型的ROC曲线下面积为0.89(95%CI:0.84～0.94,P<0.001);十折交叉验证预测模型的平均准确率及Kappa系数分别为84.9%和0.611;当列线图预测模型预测ALNM可能性为0.5时,敏感度、特异度、准确率、阳性预测值、阴性预测值、阳性似然比、阴性似然比和约登指数分别为91.2%(134/147)、73.5%(36/49)、86.7%(170/196)、91.2%(134/147)、73.5%(36/49)、3.44、0.12和0.65。结论此模型能较好地预测NCT后ALNM的风险,能够为特定患者提供更加准确的决策依据。

Objective To analyze the factors affecting the axillary lymph node metastasis (ALNM) after neoadjuvant chemotherapy(NCT) for breast cancer and establish a model for predicting ALNM in order to provide a reference for screening the patients with ALNM after NCT. Methods We retrospectively analyzed the clinical data of 196 patients who received NCT in the Affiliated Hospital of Southwest Medical University between July 2014 and April 2018. Age, menstrual status, tumor stage after NCT, preoperative lymph node status, tumor location, chemotherapy regimen, times of chemotherapy, ER status, HER-2 status and Ki67 expression were used as evaluation parameters. The relationship between the clinicopathological characteristics and ALNM was analyzed by t test,χ^2 test and Fisher exact probability test. The above-mentioned factors were used as input variables to fit the equation, and the factors which met the minimum Akaike information criterion were included in the multivariate analysis. The predictive model was established and the nomogram was drawn. According to the clinicopathological characteristic of patients, the risk of ALNM was quantified in the nomogram. The predictive effect of the model was evaluated by the receiver operating characteristic (ROC) curve. Finally, the research data were divided into 10 groups by the ten-fold cross-validation method: nine groups for modeling and one group for verifying the reliability of this predictive model. Results The following clinicopathological characteristics presented a significant difference between non-ALNM group and ALNM group after NCT: tumor staging after NCT, preoperative axillary lymph node staging, tumor location, ER status, and HER-2 status (χ^2 =20.876, P<0.001;χ^2 =57.342, P<0.001;χ^2 =13.800, P=0.008;χ^2 =15.041, P<0.001;χ^2 =5.770, P=0.016). Multivariate analysis results showed that preoperative ALNM, ER positive and low expression of Ki67 (OR=30.27, 95%CI: 10.57-108.28, P<0.001;OR=0.28, 95%CI: 0.11-0.69, P=0.007;OR=0.96, 95%CI: 0.93-1.00, P=0.032) were independent risk factors for ALNM after NCT. The area under the ROC curve of this model was 0.89 (95%CI: 0.84-0.94, P<0.001). The average accuracy and Kappa value of the ten-fold cross-validation predictive model were 84.9% and 0.611, respectively. If the probability of ALNM was 0.5 according to the nomogram predictive model, the sensitivity, specificity, accuracy, positive predictive value, negative predictive value, positive likelihood ratio, negative likelihood ratio and Yoden index were 91.2%(134/147), 73.5%(36/49), 86.7%(170/196), 91.2%(134/147), 73.5%(36/49), 3.44, 0.12 and 0.65, respectively. Conclusion This predictive model can predict the risk of ALNM after NCT, which provides more accurate decision-making basis for breast cancer patients.