Integrating pharmacovigilance and machine learning: bleeding signal detection and risk prediction for oral anticoagulants in cancer patients.

BACKGROUND: Oral anticoagulants (OACs) have been widely used for the prevention and treatment of thrombosis in cancer patients, but the bleeding signals associated with different OACs in this population have not been fully studied. OBJECTIVES: This study aimed to compare bleeding signals, identify risk factors, and establish a prediction model for OACs in cancer patients. METHODS: This study was based on the Food and Drug Administration Adverse Event Reporting System database and used the reporting odds ratio to evaluate the bleeding event signals associated with OACs in cancer patients from Q1 2014 to Q1 2024. Additionally, this study used a retrospective cohort design to assess the impact of OACs on bleeding risk in cancer patients. RESULTS: This study analyzed 29,897 adverse event reports of OACs in cancer patients, which included 4084 bleeding events. We found that bleeding events were mainly concentrated within the first 2 months of medication use. Compared with apixaban, the other 4 OACs (warfarin, edoxaban, rivaroxaban, and dabigatran) showed increased risks of bleeding, especially in the gastrointestinal system. Male, cardiovascular diseases, cerebral infarction, renal insufficiency, gastrointestinal cancer, and elevated international normalized ratio are linked to an increased risk of bleeding from OACs in cancer patients. The XGBoost model was the top-performing model among the 4 bleeding prediction models (training set area under the curve = 0.788; test set area under the curve = 0.783). CONCLUSION: Bleeding adverse events related to OACs primarily occur early in the treatment. Apixaban shows weaker bleeding signals. Male sex, cardiovascular diseases, cerebral infarction, renal insufficiency, gastrointestinal cancer, and an elevated international normalized ratio may be potential risk factors for bleeding events. The prediction model for OACs-related bleeding developed in this study demonstrated acceptable performance in internal validation.