Establishment And Verification Of Risk Prediction Model Of Bronchiectasis Hemoptysis Based On Quantitative CT

ObjectiveTo establish the prediction model of hemoptysis risk in patients with bronchiectasis based on general data and quantitative CT parameters,and compare the prediction efficacy of different models for helping to guide the early intervention of patients with bronchiectasis with high risk of hemoptysis,formulate reasonable treatment plan,improve the quality of life of patients,and prevent the occurrence of adverse events.This study is divided into three chapters:Chapter one: To obtain the corresponding quantitative image indexes based on the quantitative analysis platform of bronchiectasis,and to analyze the significant differences of general data and quantitative imaging indexes between bronchiectasis patients with hemoptysis and without hemoptysis,and to evaluate the predictive efficacy of each single factor indicator.Chapter two: To established a training set and a verification set in the samples,and to screen independent risk factors for predicting the occurrence of bronchiectasis hemoptysis in the training set.To establish logistic regression models of general data,quantitative image indexes,and general data combined quantitative imaging indexes,and to evaluate and compare the predictive efficacy of each model.Chapter three: To establish the nomogram prediction model and artificial neural network(ANN)model of general data combined quantitative imaging indexes based on the above results.To evaluate and verify the predictive efficacy,and to explore the clinical practical value of each model in the training set and validation setChapter one1.Materials and methodsGeneral data and quantitative imaging indexes of 233 patients with bronchiectasis in our hospital were retrospectively collected and divided into two groups(hemoptysis group and non-hemoptysis group)according to the occurrence of hemoptysis during2-year follow-up.The independent sample t test,the Mann-Whitney U test,and the chi-square test were used to analyze significant differences in indexes between the two groups,and the receiver operating characteristic(ROC)curve were used to analyze the prediction efficiency in the single factor index.P<0.05 was considered statistically significant.2.ResultsThere were statistically significant differences in smoking history,the total number involved bronchial branches,the inner cross-sectional area,the inner diameter and the score of the maximum dilated bronchial lumen between the patients in the bronchiectasis group and the non-hemoptysis group(P<0.05).There were no significant differences in age,sex,the number of lobes involved in bronchiectasis,the left/right lung,the upper lobe/non-upper lobe where the largest bronchiectasis was located,the number and the area of vessels around the maximum dilated bronchial lumen(P>0.05).ROC curve analysis showed that the area under curve(AUC)of the score,the inner cross-sectional area,the inner diameter of the maximum dilated bronchial lumen,the total number involved bronchial branches,and smoking history were 0.898,0.855,0.760,0.852 and0.666,respectively.3.Brief summaryBased on the quantitative imaging analysis platform,this study found that the risk of hemoptysis was greater in patients with bronchiectasis who had a history of smoking,had more total number involved bronchial branches,had the largest inner cross-sectional area, had the largest inner diameter and had the largest of the maximum dilated bronchial lumen.ROC curve analysis showed that the above indexes had significant predictive efficacy on the risk of bronchiectasis hemoptysis,and the score of the maximum dilated bronchial lumen had the best efficacy.Chapter two1.Materials and methodsThe inclusion objects were same as in Chapter one.The sample size was divided into training set(n=133)and verification set(n=100)by stratified random sampling in a ratio of 6:4.In the training set,the independent influencing factors of hemoptysis in patients with bronchiectasis were screened by multi-factor stepwise Logistic regression analysis,and regression models of general data,quantitative image indexes and general data combined quantitative image indexes were established respectively.The ROC curve of each model was drawn,and the Delong test was carried out for the differences between the AUCs of different models.The net reclassification index(NRI)and integrated discrimination improvement(IDI)analysis to determine the improvement of general data combined quantitative image indexes regression model compared with general data regression model and quantitative image indexes regression model.2.ResultsMultiple Logistic regression analysis showed that smoking history,the total number involved bronchial branches,the inner cross-sectional area,and the score of the maximum dilated bronchial lumen were independent factors influencing the risk of bronchiectasis hemoptysis.Regression models of general data,quantitative image indexes and general data combined quantitative image indexes were 0.670,0.938 and0.951,respectively.The Delong test between the general data combined quantitative image indexes model and the general data model showed a significant difference in AUC(Z=7.089,P<0.001),and there was also a significant difference between the quantitative image index regression model and the general data regression model(Z=-5.701,P<0.05).There was no significant difference in AUC between general data combined quantitative image index regression model and quantitative image index regression model(Z=1.042,P>0.05).NRI and IDI analysis between general data combined quantitative image indexes model and general data model showed that the prediction efficiency of the former model was significantly improved compared with that of the latter model(NRI=1.599,P<0.001;IDI=0.465,P<0.001).In addition,NRI and IDI analysis between general data combined quantitative image indexes model and quantitative image indexes model also found that the prediction efficiency of the former was significantly.improved compared with that of the latter(NRI=0.681,<0.001;IDI=0.040,P< 0.001).3.Brief summaryThis study found that smoking history,the total number involved bronchial branches,the inner cross-sectional area,and the score of the maximum dilated bronchial lumen were independent factors affecting the occurrence of bronchiectasis hemoptysis.Based on this,Logistic regression models of general data,quantitative image indexes and general data combined quantitative image index were established.The AUC,sensitivity and specificity of multi-factor Logistic regression model of general data combined quantitative image indexes were 0.951,0.925 and 0.925,respectively,which was the highest among the three models,and had the best prediction efficiency.Compared with the other two models,the prediction efficiency is significantly improved after adding indexes.Chapter three1.Materials and methodsBased on the results of Chapter two,a Nomogram prediction model of general data combined quantitative image indexes was established,and the calibration curve was used to evaluate the good fit of the model in the training set and validation set,the distinguishing ability of the model was evaluated by ROC curve,and the decision curve analysis(DCA)was used to evaluate the clinical application value of the model.In addition,hemoptysis or not as output,all general and quantitative image indexes as input,and a random sample size at 70% as the training sample,30% sample as the validation sample,artificial neural network(ANN)model is established.Exploring the influence extent of each index on the risk of hemoptysis,and using ROC curve to evaluate the predictive ability of the model.2.ResultBased on the results of Chapter two,a Nomogram prediction model was established,and the calibration curve showed that the model had a good consistency between the predicted risk and the actual risk in the training set and the verification set.In the training set,AUC=0.951,sensitivity and specificity were 92.5% and 92.5%,respectively,while in the verification set,AUC=0.956,sensitivity and specificity were 86.0% and 94.7%,respectively.The DCA showed that when the prediction probability threshold is within the range of 0.00 to 0.84 and 0.00 to 0.89,the Nomogram model has a good clinical net benefit in predicting the risk of bronchiectasis hemoptysis in both the training set and the validation set.In the ANN model that included all general data and quantitative imaging indexes,the top 5 indexes of influence extent were the score of the maximum dilated bronchial lumen,the total number involved bronchial branches,the inner cross-sectional area,the inner diameter of the maximum dilated bronchial lumen and smoking history.The AUC of ANN model was 0.949,and the accuracy of model prediction was 87.6%and 87.5% respectively in the training samples and the verification samples.3.Brief summaryIn this study,the Nomogram prediction model were established based on the above independent factors,and it showed good calibration and differentiation ability in both the training set and the verification set.DCA showed that the nomogram model has significant clinical practical value.However,among all general data and quantitative imaging indexes included in the ANN model,the top 5 indexes of influence extent were the score of the maximum dilated bronchial lumen,the total number involved bronchial branches,the inner cross-sectional area,the inner diameter of the maximum dilated bronchial lumen and smoking history.Moreover,the ANN model also showed good predictive efficacy in the training set and the verification set.ConclusionsBased on the quantitative imaging analysis platform,this study obtained the relevant quantitative parameters of the airway of patients with bronchiectasis,and found that smoking history,the total number involved bronchial branches,the inner cross-sectional area,the inner diameter and the score of the maximum dilated bronchial lumen all had significant predictive efficacy on the risk of bronchiectasis hemoptysis.The score of the maximum dilated bronchial lumen has the best efficacy.This study found that smoking history,the total number involved bronchial branches,the inner cross-sectional area,and the score of the maximum dilated bronchial lumen were independent factors affecting the occurrence of bronchiectasis hemoptysis.The AUC,sensitivity and specificity of the general data combined quantitative image indexes Logistic regression model established on this basis were 0.951,0.925 and 0.925,respectively,which was the best among the three models,and the prediction efficiency was significantly improved compared with other models.The Nomogram model established in this study showed good calibration and differentiation ability in both training and verification sets,and the DCA showed that it has significant clinical value.In the ANN model,the top 5 indexes of importance for the influence extent on the prediction results were the score of the maximum dilated bronchial lumen,the total number involved bronchial branches,the inner cross-sectional area,the inner diameter of the maximum dilated bronchial lumen and smoking history,which also showed good predictive efficacy.The above models could provide help for clinicians to evaluate the risk of hemoptysis and early intervention in patients with bronchiectasis.