Predictive Model of Internal Bleeding in Elderly Aspirin Users Using XGBoost Machine Learning.

Objective: This study aimed to develop a predictive model for assessing internal bleeding risk in elderly aspirin users using machine learning. Methods: A total of 26,030 elderly aspirin users (aged over 65) were retrospective included in the study. Data on patient demographics, clinical features, underlying diseases, medical history, and laboratory examinations were collected from Affiliated Dongyang Hospital of Wenzhou Medical University. Patients were randomly divided into two groups, with a 7:3 ratio, for model development and internal validation, respectively. Least absolute shrinkage and selection operator (LASSO) regression, extreme gradient boosting (XGBoost), and multivariate logistic regression were employed to develop prediction models. Model performance was evaluated using area under the curve (AUC), calibration curves, decision curve analysis (DCA), clinical impact curve (CIC), and net reduction curve (NRC). Results: The XGBoost model exhibited the highest AUC among all models. It consisted of six clinical variables: HGB, PLT, previous bleeding, gastric ulcer, cerebral infarction, and tumor. A visual nomogram was developed based on these six variables. In the training dataset, the model achieved an AUC of 0.842 (95% CI: 0.829-0.855), while in the test dataset, it achieved an AUC of 0.820 (95% CI: 0.800-0.840), demonstrating good discriminatory performance. The calibration curve analysis revealed that the nomogram model closely approximated the ideal curve. Additionally, the DCA curve, CIC, and NRC demonstrated favorable clinical net benefit for the nomogram model. Conclusion: This study successfully developed a predictive model to estimate the risk of bleeding in elderly aspirin users. This model can serve as a potential useful tool for clinicians to estimate the risk of bleeding in elderly aspirin users and make informed decisions regarding their treatment and management.