Predicting The Subgroups Of ER Low Positive Breast Cancer Based On Naive Bayes And R Language

Part Ⅰ Predicting the Subgroups of ER Low Positive Breast Cancer Based on Naive BayesObjective: To establish a model for predicting the expression state of ER based on the Naive Bayes classification algorithm,by using artificial intelligence to extract the morphological characteristics of infiltrating cancer cells and combining the clinicopathological characteristics of patients.And the model will be used to identify patients with ER low positive whose clinicopathological characteristics are similar to ER-negative patients.It will provide a reference for the treatment decisions for these patients.Methods: A retrospective study of breast cancer patients who had undergone surgery at the Fourth Hospital of Hebei Medical University.The training cohort included 139(30.89%)ER-negative(<1%)patients and 311(69.11%)ER-positive(>10%)patients in 2012.The H&E stained section of tumor tissue was passed through the Unic digital scanner to obtain whole slide images.Then use image processing technology to extract the morphological characteristics of each cell,and combine the clinicopathological characteristics of the patient to construct a model that predicts the expression state of ER based on the Naive Bayes classification algorithm.The predictive performance of the model is verified by a 5-fold cross-validation.And the subgroups prediction of 260 ER low positive patients from 2012 to 2018 was made.Results: The predictive model has a good degree of discrimination for the expression state of ER.By drawing the ROC curve,its AUC is 0.91(95%CI±0.03).The subgroups prediction were performed on 260 ER low positive patients,of which 206(79.23%)patients were predicted to be negative,and 54(20.77%)patients were predicted to be positive.By comparing the two groups,it was found that in ER low positive tumors,patients with high histological grade and Ki67 high expression are more likely to have negative predicted results,and they have lower expression of ESR1 mRNA,cannot benefit from endocrine therapy and have a poor prognosis.Conclusions: Based on the Naive Bayes classification algorithm,we use artificial intelligence to extract the morphological characteristics of infiltrating cancer cells and combine the clinicopathological characteristics of the patient to develop a model that predicts the ER expression state.It can identify the clinicopathological characteristics of patients with ER low positive who are similar to ER-negative patients,it will help guide individualized and precise treatment of patients with ER low positive.Part Ⅱ Establishment and Verification of Nomogram for Predicting the Subgroups of ER Low Positive Breast CancerObjective: To establish a nomogram to predict ER expression status based on the clinicopathological characteristics of breast cancer patients,and used to identify patients with ER low positive whose clinicopathological characteristics are similar to ER-negative patients.It will provide a reference for the treatment decisions for these patients.Methods: A retrospective study of breast cancer patients had undergone surgical treatment in the Fourth Hospital of Hebei Medical University in 2012 was conducted.A total of 450 patients with invasive breast cancer were enrolled.Among them,139(30.89%)patients were ER-negative(ER<1%)and 311 patients(69.11%)were ER-positive(ER>10%),and the two groups of patients were randomly divided into training cohort and validation cohort at a ratio of 7:3.The data of the training cohort was used to establish a nomogram to predict the expression state of ER by using R language based on logistic regression analysis,and the data from the validation cohort was used to verify the predictive model.The agreement between the predicted probability and the actual probability is evaluated by calculating the Harrell C index and drawing calibration charts,and the effectiveness of the predictive model is evaluated by calculating the area under the ROC curve.The Yorden index is used to determine the best cut-off value of the predictive model for distinguishing the expression state of ER.Finally,the established predictive model is used to predict the subgroups of 260 ER low positive patients from 2012 to 2018.Results: Patients with significant nuclear pleomorphism,mitotic figures>20/10 HPF,tumor infiltrating lymphocytes>40%,and visible necrosis tend to have negative ER expression.The Harrell C index of the predictive model in the training cohort and validation cohort are 0.80(95% CI 0.75-0.86)and 0.83(95% CI 0.75-0.90),respectively.The calibration chart shows that there is a good agreement between the predicted probability and the actual probability.Using ROC curve to evaluate the predictive power of the model,the AUC of the training cohort and validation cohort were 0.804(95% CI 0.750-0.858)and 0.828(95% CI 0.752-0.903),respectively.By calculating the Yorden index,the best cut-off value for predicting ER expression status is 0.59.The subgroups prediction were performed on 260 ER low positive patients,of which 164(63.08%)patients had a negative predicted result(predicted value<0.59),and 96(36.92%)patients had a positive predicted result(predicted value>0.59).Among patients with ER low positive,those with a negative predicted result have lower expression levels of ESR1 mRNA,cannot benefit from endocrine therapy and have a poor prognosis.Conclusions: Based on the clinicopathological characteristics of the patient,we have developed and verified a nomogram that predicts the ER expression status of patients with invasive breast cancer,which can identify patients with ER low positive whose clinicopathological characteristics are similar to ER-negative patients,which will help guide individualized and precise treatment of patients with ER low positive.Part Ⅲ Study on the Consistency of RT-qPCR and IHC in Detection of Breast Cancer Molecular TypingObjective: To explore the consistency of RT-qPCR and IHC detection results and the impact on the molecular typing of breast cancer,and to research whether RT-qPCR can be used as a more accurate molecular typing method to help diagnose breast cancer.Methods: 398 patients with invasive ductal carcinoma of the breast who had undergone surgical treatment at the Fourth Hospital of Hebei Medical University from 2014 to 2015 were enrolled.The expressions of ER,PR,HER2 and Ki67 were detected by IHC and RT-qPCR,and molecular typing was performed.Compare and analyze the consistency of the results of the two methods and the impact on molecular typing.Results: There is a high degree of agreement between the results of ER,PR,Ki67 and HER2 detected by IHC and RT-qPCR(r = 0.893,0.884,0.815,0.824,P<0.01).But in ER low positive tumors,the consistency of the results of the two methods is poor.For the overall research population,the expressions of ER,PR,Ki67 and HER2 were significantly different in T stage and histological grade(P<0.05).In addition,after analyzing the impact of these two detection methods on the molecular typing of breast cancer,we found that some breast cancers that were typed as Luminal B by IHC were classified as Luminal A by RT-qPCR(P<0.01).Through survival analysis,it is found that patients who are classified as Luminal A by RT-qPCR have a good prognosis,and the RT-qPCR detection method is more helpful to identify these patients.Conclusions: This study proved that the results of RT-qPCR and IHC have a high degree of consistency.But for tumors with ER low positive,the results of these two methods have poor consistency.For the molecular typing of breast cancer in the overall research population,the prognosis of Luminal A breast cancer detected by RT-qPCR is significantly better.