Research On The Prediction Of Short-term Thrombosis Events In Patients With Coronary Heart Disease Based On The Technology Of Photoplethysmography

Purpose:Explore the application value of prediction for short-term thrombosis events in patients with coronary heart disease(CHD)based on the technology of Photoplethysmography(PPG)and machine learning,and provide new ideas for establishing the short-term(within 3 months)thrombosis prediction model.Methods:Analysis outpatients and inpatients CHD Patients in PLA general Hospital during June 2017-January 2018.PPG data are collected when patients participate in the Clinical trials.Endpoints are confirmed by the reception of re-visiting patients,consulting inpatient medical record and phone calls during the follow-up period of 3 months.According to whether thrombosis events or not,the patients are divided into thrombosis group and non-thrombosis group.The patients' baseline data and PPG data are collected through HIS system and network database.The baseline information and PPG signal parameters of both patients' group are compared by t test and chi-square test,use logistic regression find the independent risk factors,use collinearity diagnostics choose variable of predict model.The receiver operating characteristic curve(ROC)is established by three machine learning algorithm.The diagnosis efficiency of short-term thrombosis prediction model is compared after three algorithms.Results:the PPG data of 91patients who meet the inclusion criteria are collected.9 patients(9.890%)are in the thrombosis group,and 82 patients(90.110%)are in non-thrombosis group.There is statistical difference in coronary artery lesion number(P=0.024)and heart rate(p<0.001).The statistical difference is not existed in age,gender,height,weight,BMI,smoke history,hypertension,diabetes,hyperlipemia,myocardial infarction history,left ventricular ejection function,NYHA class ?/?,aortic valve insufficiency,left main disease,stroke,carotid artery stenosis,peripheral vascular disease,chronic lung disease,serum creatinine,abnormal liver function,use of nitrates.The areas under the ROC curve(AUC)by Logistic regression,Logistic regression(Bootstrap)and XGboost is 0.9119,0.9010,and 0.9837.Conclusion:The PPG feature parameters SI,RI,T,Tbd,Pab are significant for the establishment of prediction model of thrombosis in patients with CHD.The combination of PPG technology and machine learning,a new idea for the prediction model establishment in future,is of higher prediction value for short-term thrombosis events in patients with CHD.PPG is characterizes as widely used in clinical,simple and repeatable,independent and efficient.It is worth to be popularized.