Prediction study of bleeding risk in coronary artery disease patients based on XGBoost algorithm combined with photoplethysmography and clinical characteristic variables

ZHANG Li-yue; DONG Shi-yong; SHI Jun-shan; MIHELAYI·Adile; WANG Rong

doi:10.3969/j.issn.1672-5301.2024.01.009

Abstract：Objective To develop a predictive model for bleeding events during antithrombotic therapy in patients with coronary artery disease(CAD)based on machine learning combining photoplethysmography(PPG)and clinical characteristic variables.Methods PPG and clinical characteristic data from the online database of antithrombotic therapy for patients with coronary artery disease(CAD)at the General Hospital of the People's Liberation Army in China,diagnosed by coronary angiography from January 2018 to October 2019,with at least 1 reported bleeding event,were collected.The PPG-clinical characteristic dataset were randomly divide into the training and validation sets in an 80∶20 ratio.The training set was used to construct CAD-bleeding event prediction models by random forest,support vector machine and XGBoost algorithms.SHAP(SHapley Additive exPlanations)was utilized to explain the models and select the clinical variables included in the best machine learning predictive model.Finally,the selected predictive model on the validation set in terms of sensitivity,specificity and area under the receiver operating characteristic curve(AUC)was evaluated.Results A total of 155 CAD patients'clinical data and PPG data were included in the study.The XGBoost model demonstrated the best predictive performance in the training set(AUC=0.927).Following the screening of clinical feature variables,12 predictive factors for bleeding events during antiplatelet therapy in CAD patients were identified,including systolic blood pressure,history of diabetes and use of glucose-lowering medications.Subsequently,the predictive models constructed using the PPG-clinical data set features and PPG features alone were compared using validation set data.The model constructed with PPG-clinical data set features exhibited superior predictive performance(AUC= 0.892)compared to the model utilizing PPG features alone.Additionally,it demonstrated high sensitivity and specificity.Conclusion The XGBoost algorithm combined with PPG and clinical feature variables demonstrated superior predictive performance in forecasting bleeding events in patients with coronary artery disease(CAD).Building upon this,the application of portable wearable PPG devices holds promise for further enabling accurate and dynamic at-home monitoring of bleeding risk in CAD patients undergoing antiplatelet therapy,thereby improving their long-term clinical outcomes.