Research On Prediction Of Short-Term Recurrence Risk Of Bronchial Asthma After Acute Attack Based On Artificial Neural Network

ObjectiveAs a non-infectious chronic airway disease,bronchial asthma has the characteristics of being treatable,controllable,incurable,and recurrence.We aims to construct a Logistic regression model and an artificial neural network model to predict the short-term recurrence risk of asthma after acute attack,and compare the effects of the two models by drawing the ROC curve of the two models.Objective to explore the important factors affecting the short-term recurrence of bronchial asthma after acute attack,which conductive to clinical nursing staff to formulate targeted nursing strategies for patients with bronchial asthma,and provide direction for continuous nursing of patients with bronchial asthma after discharge.MethodsIn this study,105 patients with acute attack of bronchial asthma from the Department of respiratory and critical care medicine of a 3A general hospital in Qingdao from March 2018to may 2020 were selected.The self-made general information questionnaire was used to collect the information of patients with acute attack of bronchial asthma.Three months after discharge,the patients were prospectively followed up to see if they had asthma attack again,and were divided into asthma recurrence group and non recurrence group according to the follow-up results,which used to construct risk prediction model.In addition,34 patients who were hospitalized with cute attack of bronchial asthma from June 2020 to December2020 were selected to collect the information by the same method.The prospective follow-up of patients after discharge whether have asthma symptoms was conducted to verify the predictive effectiveness of the two models.We screen out the related factors influencing the recurrence of patients with acute attack of bronchial asthma within 3 months after discharge by Univariate analysis,and screen out the more statistically significant related factors by binary logistic multivariate regression analysis,by stepwise regression method,and construct the traditional logistic regression prediction model.The related factors of P<0.05 in the results of univariate analysis were included as predictive variables in SPSS Modeler18.0 to construct the artificial neural network prediction model.The area under the ROC curve AUC was used to evaluate the predictive effect of the two models,and the data of 34 patients with acute attack of bronchial asthma were substituted into the two prediction models to verify the predictive effect of the two models.Results1.Reliability and validity test results of the evaluation of the correctness of the use of modified inhalation device:The item content validity index I-CVI was 0.86～1.00,and the scale content validity index S-CVI was 0.943,the coefficient of content consistency Cronbach’sαis 0.824,and the scale can be used.2.Univariate analysis results of recurrence related factors within 3 months after the acute attack of asthma:Three months after discharge,in the recurrence group and non-recurrence group,there were differences in Age（P=0.022）,WBC（P=0.006）,N%（P=0.010）,NLR（P=0.008）,Pa O₂（P=0.006）,FEV1（P<0.001）,FEV1%（P<0.001）,ACT（P=0.001）,Admission mode（P=0.001）,Inhaled corticosteroids in the past year（P=0.016）,Asthma attacks in the past year（P=0.004）,Excessive use of SABA（P=0.003）,Body mass index classification（P=0.017）,Asthma severity classification（P<0.001）,Classification of acute asthma attacks（P=0.002）,Hospitalization expenses（P=0.047）and Length of stay（P=0.026）,and the difference between the two groups was statistically significant（P<0.05）.3.Results of logistic multivariate analysis and regression prediction model for recurrence within 3 months after acute attack of bronchial asthma:The independent risk factors for recurrence of asthma within 3 months after discharge include Admission mode and Excessive use of SABA,FEV1%is a protective factor for the recurrence of asthma within 3 months after the acute onset of asthma,and the logistic regression model was P=1.568*Admission mode+1.811*Excessive use of SABA-0.057*FEV1%.4.Results of artificial neural network risk prediction model for recurrence within 3months after acute attack of bronchial asthma:The independent variables（P<0.05）of Univariate analysis results were included into SPSS Modeler18.0.The system default multi-layer perceptron model,default 1 hidden layer,1 input layer,1 output layer,according to the training partition and test partition 7:3,the importance of predictive variables was ranked from large to small after the artificial neural network was run.The top five importance were FEV1%,ACT,NLR,WBC,Pa O₂.We can get the importance of standardization,FEV1%100%,ACT 42.85%,NLR 32.14%,WBC 25%,Pa O₂21.43%by comparing the importance of each prediction variable with the importance of the maximum index value.The accuracy of the training set was 93.24%,AUC 0.852,the accuracy of the test set was 83.87%,AUC 0.885,and the overall accuracy percentage of the model was 93.2%.5.Accuracy evaluation of prediction model for recurrence risk within 3 months after acute attack of bronchial asthma:Logistic regression model AUC（95%CI）for recurrence within 3 months after acute asthma attack was 0.825（95%CI:0.715,0.936）,sensitivity 91%,specificity 67%,the artificial neural network prediction model AUC（95%CI）was 0.942（95%CI:0.881,1.000）,sensitivity 95%,specificity 81%.The verification results of the two prediction models in the verification population showed that the overall accuracy,sensitivity and specificity of logistic regression model were 88.23%,93.10%and 60.00%,the artificial neural network prediction model were 96.2%,93.55%and 100%.ConclusionsThe first,the recent recurrence of bronchial asthma after acute attack was related to Age,WBC,N%,NLR,Pa O₂,FEV1,FEV1%,ACT,Admission mode,Inhaled corticosteroids in the past year,Asthma attack in the past year,Excessive use of SABA,Body mass index classification,Asthma severity classification,Classification of acute asthma attacks,Hospitalization expenses and Length of stay,and Admission mode,Excessive use of SABAwere independent risk factors for short-term recurrence of bronchial asthma,and FEV1%,ACT,NLR,WBC and Pa O₂are important predictors of short-term recurrence of bronchial asthma;The second,the prediction efficiency of the artificial neural network model for the recurrence of asthma after acute attack is better than that of logistic regression model.We should pay more attention to the patients with poor lung function,poor asthma control,high NLR,admitted by emergency department,excessive use of SABA,low Pa O₂and complicated infection in clinical nursing.We also should pay attention to the training and guidance of patients’self-management ability to prevent the recurrence of asthma within 3 months after discharge.