Clinical Application Of Intelligent Predictive Model For Atrial Fibrillation In Patients With Hypertension

OBJECTIVE: Hypertension plays an important role in patients with atrial fibrillation(AF).However,little is known about how to predict AF.This study aimed to construct an intelligent predictive model to predict AF in hypertensive patients.METHODS: A total of 2065 hypertension patients admitted to the Heart Center of Zhongshan Hospital from January 2017 to September 2018 were enrolled into the study,including 716 patients with AF,and 1349 patients without AF.Patients with secondary hypertension,rheumatic heart disease and systemic congestion due to heart failure,such as interstitial pulmonary edema and cardiogenic cirrhosis,were excluded.And we collected and recorded clinical data,diagnosis and physical tests,including information,gender,height,weight,age,body mass index(BMI),hypertension(Grade 1-3),AF,coronary heart disease,myocardium disease,cardiac insufficiency,pneumonia,pulmonary embolism,asthma,chronic obstructive pulmonary disease,diabetes,gout,cirrhosis,hyperthyroid,cerebral infarction,cerebral hemorrhage,chronic kidney disease,anemia,blood routine,liver function,renal function,blood glucose,blood lipids,electrolyte,NT-pro BNP,Carcinoembryonic Antigen(CEA),Alpha-fetoprotein(AFP),Carcinoembryonic Antigen 199(CA199),creatine kinase,Cardiac Ultrasound(including IVSD,LVDD,LVPWD,AOD,LAD,RVOT,MPA,RVDD,LASD,RASD,LVEF),24-hours dynamic electrocardiogram(DCG),etc.Next,We preprocessed the collected data,including dimensionality reduction,discretization,removal of outliers and filling of vacancies.Finally we filtered out 110 attributes in the more than 200 original data.Then we formed a data set using the above 110 attributes.Finally,we constructed an intelligent predictive model using C4.5 decision tree algorithm and evaluated the accuracy of the model.RESULTS: The decision tree model involves 10 indicators,including BMI,gender,age,heart failure,coronary heart disease,gout,cardiac ultrasound(RASD,LASD,RVDD,LVDD,LAD,IVSD,AOD),apolipoprotein B,direct bilirubin,eosinophil ratio.By consulting the literature,we found that 4 indicators in the model had been clearly identified as independent risk factors for AF,3 indicators had been confirmed to be significantly related with AF,and 2 indicators had also been shown to be related with AF.However,only 1 indicator was not associated with AF.At the same time,the predictive accuracy of this model for AF is up to 79.8% using computer-generated random data.CONCLUSIONS: A model with higher predictive accuracy for AF is set up.We can use the model to evaluate patients with hypertension and predict the risk of AF.If it suggests that the risk of AF is high,clinicians should strengthen monitoring and follow-up of patients.This model will be improved by more and more clinical data and self-leaning ability of artificial intelligent model in the future.