ABSTRACT
Objective:To establish a long-term mortality rate prediction model for patients aged 60 years and over with atrial fibrillation and coronary heart disease using the machine learning method, and identify the corresponding risk factors of mortality.Methods:In this retrospective cohort study, a total of 329(11 cases lost of follow-up)patients with 183 males(55.6%)and 146 females(44.4%), aged(77.8±7.3)years, and 142 patients aged 80 years or older(43.2%)were selected in our hospitals from January 2013 to March 2015.And their clinical data on atrial fibrillation and coronary heart disease were analyzed.They were divided into the death group(151 cases)and the survival group(167 cases)according to the survival outcome.In addition, 60 patients aged 60 years and over admitted to our hospitals from April to July 2015 with atrial fibrillation and coronary heart disease were selected as external data validation set.The clinical data included age, gender, body mass index, diagnosis, co-morbidity, laboratory indicators, electrocardiogram, echocardiogram, treatment data.These patients were followed up for at least 6 years, and the main adverse cardiovascular and cerebrovascular events(MACCE), including death, were recorded.Finally, the data of the enrolled patients were randomly divided into the training set and the test set according to the ratio of 9∶1, Different models were established to predict the long-term mortality of patients with atrial fibrillation and coronary heart disease by machine learning algorithm.The optimal model was established by substituting external data(60 cases)into the model for verification and comparison.The top 20 risk factors for mortality were determined by Shapley additive explanation(SHAP)algorithm.Results:A total of 329 hospitalized patients were included in this study, the overall median follow-up time was 77.0 months(95% CI: 54.0~84.0), 11 cases lost during follow-up(3.3%), and 151 cases died(45.9%). The analysis found that the areas under the ROC curve for a support vector machine(SVM)model, k-Nearest Neighbor(KNN)model, decision tree model, random forest model, ADABoost model, XGBoost model and logistic regression model were 0.76, 0.75, 0.75, 0.91, 0.86, 0.85 and 0.81, respectively.The random forest model had the highest prediction efficiency, with the accuracy of 0.789 and F1 value of 0.806, which was better than the logistic regression model[the Area Under Receiver Operating Characteristic Curve(AUC): 0.91 vs.0.81, P<0.05]. D-dimer, age, number of MACCE, left ventricular ejection fraction, serum albumin level, anemia, New York Heart Association(NYHA)grade, history of old myocardial infarction, estimated glomerular filtration rate(eGFR)and resting heart rate were important risk factors for predicting long-term mortality. Conclusions:The random forest model based on machine learning method can predict the long-term mortality of patients with atrial fibrillation and coronary heart disease aged 60 years and over, have a good identification ability.Its accuracy is higher than that of the traditional Logistic regression model.Reducing the long-term mortality and improving the long-term outcomes can be achieved by intervening on D-dimer levels, correcting hypoproteinemia and anemia, improving cardiac function and controlling resting ventricular rates.
ABSTRACT
Rivaroxaban is one of the new oral anticoagulants (NOAC) for preventing stroke and systemic embolism in patients with non-valvular atrial fibrillation. It has clear pharmacokinetic parameters, stable plasma concentration, less drug-drug interaction and higher compliance of patients. However, the discrepancy of pharmacokinetics between individuals and drug-induced hemorrhage events frequently occur clinically, therefore the association of gene polymorphism with drug metabolism has become a research hotspot. This article reviews the research progress on pharmacokinetic characteristics of rivaroxaban and its relationship with gene polymorphism, to provide a reference for the individualized rational use of rivaroxaban.