XGBoost-based machine learning test improves the accuracy of hemorrhage prediction among geriatric patients with long-term administration of rivaroxaban

Abstract

Abstract Background Hemorrhage is a potential and serious adverse drug reaction, especially for geriatric patients with long-term administration of rivaroxaban. It is essential to establish an effective model for predicting bleeding events, which could improve the safety of rivaroxaban use in clinical practice. Methods The hemorrhage information of 798 geriatric patients (over the age of 70 years) who needed long-term administration of rivaroxaban for anticoagulation therapy was constantly tracked and recorded through a well-established clinical follow-up system. Relying on the 27 collected clinical indicators of these patients, conventional logistic regression analysis, random forest and XGBoost-based machine learning approaches were applied to analyze the hemorrhagic risk factors and establish the corresponding prediction models. Furthermore, the performance of the models was tested and compared by the area under curve (AUC) of the receiver operating characteristic (ROC) curve. Results A total of 112 patients (14.0%) had bleeding adverse events after treatment with rivaroxaban for more than 3 months. Among them, 96 patients had gastrointestinal and intracranial hemorrhage during treatment, which accounted for 83.18% of the total hemorrhagic events. The logistic regression, random forest and XGBoost models were established with AUCs of 0.679, 0.672 and 0.776, respectively. The XGBoost model showed the best predictive performance in terms of discrimination, accuracy and calibration among all the models. Conclusion An XGBoost-based model with good discrimination and accuracy was built to predict the hemorrhage risk of rivaroxaban, which will facilitate individualized treatment for geriatric patients.