METHODS: Data relating to 104 patients undergoing NACT were collected. Variables derived from clinical information and pathological characteristics of patients were screened through logistic regression, random forest, and Xgboost methods to formulate predictive models. The validation and comparative assessment of these models were conducted to identify the optimal model, which was then visualized and tested.
RESULTS: Following the screening of variables and the establishment of multiple models based on these variables, comparative analyses were conducted using receiver operating characteristic (ROC) curves, calibration curves, as well as net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Model 2 emerged as the most optimal, incorporating variables such as HER-2, ER, T-stage, Reg IV, and Treatment, among others. The area under the ROC curve (AUC) for Model 2 in the training dataset and test dataset was 0.837 (0.734-0.941) and 0.897 (0.775-1.00), respectively. Decision curve analysis (DCA) and clinical impact curve (CIC) further underscored the potential applications of the model in guiding clinical interventions for patients.
CONCLUSIONS: The prediction of NACT pCR efficacy based on the expression of Reg IV in breast cancer tissue appears feasible; however, it requires further validation.
方法:收集104例NACT患者的相关数据。通过logistic回归分析筛选来自患者临床资料和病理特征的变量,随机森林,和Xgboost方法来制定预测模型。对这些模型进行了验证和比较评估,以确定最优模型,然后可视化和测试。
结果:在筛选变量并根据这些变量建立多个模型之后,使用受试者工作特征(ROC)曲线进行比较分析,校正曲线,以及净重新分类改进(NRI)和综合歧视改进(IDI)。模型2成为最优的,合并变量,如HER-2,ER,T-stage,RegIV,和治疗,在其他人中。训练数据集和测试数据集中模型2的ROC曲线下面积(AUC)分别为0.837(0.734-0.941)和0.897(0.775-1.00)。分别。决策曲线分析(DCA)和临床影响曲线(CIC)进一步强调了该模型在指导患者临床干预方面的潜在应用。
结论:根据乳腺癌组织中RegIV的表达预测NACTpCR疗效似乎是可行的;然而,它需要进一步验证。