Building Childhood Asthma Prediction Model With Artificial Neural Network And BRFSS Database

Objective: Childhood asthma is a respiratory disease that seriously affects children's physical and mental health.In recent years,the prevalence and mortality of asthma in children have increased,affecting 9.6-13% of children in the world,causing huge economic and social burdens.Because children's airway development is not perfect,children are vulnerable to environmental factors and are at high risk of asthma in children.80% of asthma symptoms occur before the age of 3.With the increase of age,the development of the airway and immune system is gradually improved.Some children will recover,while others will still have asthma at school age and will receive lifelong treatment,so it is important to predict whether or not you will continue to have asthma during school age.Based on the BRFSS database and the artificial neural network method,this study provides a simple and easy prediction model for childhood asthma,which is beneficial to strengthen the health management of high-risk children with asthma,alleviate the pressure of medical resources,and improve the health and quality of life of patients.Methods: Data of the asthma call-back survey was downloaded from the BRFSS database,and the data of asthma patients who participated in the call-back survey from 2011 to 2014 were selected.According to the latest asthma GINA guidelines,previous research,and consulted clinical experts,20 characteristic variables were initially selected from the database,including: parental asthma history,family income,birth month,insurance,maternal age,passive smoking,obesity,chronic obstructive pulmonary disease / history of chronic bronchitis.The outcome variable is whether or not the child has asthma at school age.Subsequently,the data is preprocessed by deleting missing values,data conversion,and data discretization,and the data is divided into training sets and test sets according to a ratio of 70% : 30%.The statistical methods are applied,and the information gain theory is combined with the Feature_Selection feature filter in the Python language to filter the predictors.Aiming at the phenomenon of gradient dispersion in traditional artificial neural networks,the model is optimized by using the stochastic gradient descent algorithm.The training set is used to continuously learn and adjust the parameter settings to construct the optimal prediction model.The testing set is used to evaluate the model and the evaluation indicators are accurate,sensitivity,specificity and receiver operating characteristic curve.Finally,the model is compared with the logistic regression,support vector machine and decision tree algorithm.Results: After data preprocessing,a total of 930 children were included,of which 587 had asthma in school age and 343 recovered,and the ratio of positive samples was 1:1.71,which was more balanced.After comparing the screening methods of the two characteristic variables,the final selected high correlation predictor set includes: history of asthma,passive smoking,age of diagnosis,correct use of inhaler,parental education,family income,and the birth month.The accuracy of the prediction model based on the optimized BP artificial neural network is 73.4%,the sensitivity is 72.7%,the specificity is 69.3%,and the area under the ROC curve is 0.731.The performance of the model is better than the logistic regression model,decision tree and support vector machine model performance.Conclusion: In this study,children with asthma were studied.The artificial neural network algorithm was used to construct the asthma prediction model with accuracy of 73.4%,sensitivity of 72.7% and specificity of 69.3%.The performance of the model is optimized by the stochastic gradient descent algorithm,and the accuracy is improved.The predictive variables applied are all behavioral risk factors,which greatly improve the compliance of children,facilitate the clinical primary and secondary prevention work,improve the patients' health and the quality of life.