基于超声图像特征的列线图模型预测乳腺叶状肿瘤Ki-67表达水平的临床价值

Clinical value of nomogram model based on ultrasonic characteristic for predicting Ki-67 expression level in phyllodes tumor of breast

目的基于超声图像特征构建列线图模型,探讨其预测乳腺叶状肿瘤(PTB)Ki-67表达水平的临床价值。方法选取我院经术后病理证实的PTB患者119例,根据Ki-67表达水平分为Ki-67低表达组(Ki-67≤10%)82例和Ki-67高表达组(Ki-67>10%)37例。比较两组超声图像特征的差异,包括最大径、纵横比、形态、边缘、实质回声、囊变、钙化、后方回声、血流信号、阻力指数、收缩期峰值流速;应用多因素Logistic回归分析筛选预测Ki-67高表达水平的独立预测因子,基于独立预测因子构建列线图模型。采用Bootstrap自助抽样法进行内部验证,绘制受试者工作特征(ROC)曲线分析模型的区分度;采用Hosmer-Lemeshow检验并绘制校准曲线评估模型的校准度;临床决策曲线分析模型的临床适用性。结果两组最大径、囊变、血流信号、收缩期峰值流速比较差异均有统计学意义(均P<0.05);两组纵横比、形态、边缘、实质回声、钙化、后方回声、阻力指数比较差异均无统计学意义。多因素Logistic回归分析显示,最大径、囊变和收缩期峰值流速均为预测Ki-67高表达水平的独立影响因子(OR=1.164、0.078、1.404,均P<0.05),根据上述独立预测因子构建列线图模型。ROC曲线分析显示,列线图模型预测PTB患者Ki-67高表达水平的曲线下面积(AUC)为0.812(95%可信区间:0.704~0.920);Hosmer-Lemeshow检验显示,列线图模型预测概率与实际概率比较差异无统计学意义(P=0.330);校准曲线显示,预测曲线与校准曲线的校准度较高,C-index为0.814(95%可信区间:0.710~0.913);临床决策曲线分析显示,当阈值为0.10~0.48时,该模型的临床获益较高。结论基于超声图像特征的列线图模型可用于预测PTB患者Ki-67表达水平,该模型具有较高的准确性和可靠性。

Objective To establish the nomogram model based on ultrasonic characteristics,and to explore the clinical value of nomogram for predicting Ki-67 expression level in phyllodes tumor of breast(PTB).Methods A total of 119 patients with PTB confirmed by pathology after surgery in our hospital were selected,and they were divided into low expression group(Ki-67≤10%)consisting of 82 cases and high expression group(Ki-67>10%)consisting of 37 cases according to the Ki-67 expression level.The differences in ultrasonic characteristics between the two groups were compared,including maximum diameter,aspect ratio,morphology,margins,parenchymal echo,cystic degeneration,calcification,posterior echo,blood flow signal,resistance index and peak systolic flow velocity.The multivariate Logistic regression analysis was applied to screen the independent predictors for predicting the high expression level of Ki-67,and a nomogram model was established based on the independent predictors.The Bootstrap method was used for internal validation,and receiver operating characteristic(ROC)curve was drawn to assess the discrimination of the model.The calibration of the model was evaluated by Hosmer-Lemeshow test and calibration curve.The clinical decision curve was used to assess the model’s clinical utility.Results The differences in maximum diameter,cystic degeneration,blood flow signal and peak systolic flow velocity between the two groups were statistically significant(all P<0.05).The differences in aspect ratio,morphology,margins,parenchymal echo,calcification,posterior echo and resistance index between the two groups were not statistically significant.Multivariate Logistic regression analysis showed that the maximum diameter of lesion,cystic degeneration and peak systolic flow velocity were independent predictors of the high expression level of Ki-67(OR=1.164,0.078,1.404,all P<0.05).Subsequently,nomogram model was established based on above independent predictors.ROC curve analysis showed that the area under the curve(AUC)of the nomogram model for predicting the high expression level of Ki-67 in PTB patients was 0.812(95%confidence interval:0.704~0.920).The Hosmer-Lemeshow test showed no statistically significant difference between the predicted probability and the observed probability(P=0.330).The calibration curve showed a strong correlation between the prediction curve and the calibration curve,with a high C-index value of 0.814(95%confidence interval:0.710~0.913).The clinical decision curve showed that the model had a higher clinical benefit within the threshold range of 0.10~0.48.Conclusion The nomogram model based on ultrasonic characteristics can be used for the prediction of Ki-67 expression levels in PTB patients,exhibiting a high degree of accuracy and reliability.