Safety and Effectiveness of Oral Anticoagulants in Patients with Atrial Fibrillation and Cancer

Abstract

Background: The use of oral anticoagulants (OACs) in patients with atrial fibrillation (AFib) and cancer remains suboptimal due to insufficient evidence in clinical guidelines. In this dissertation, our goals are to (1) screen for clinically relevant drug-drug interactions (DDIs) between direct oral anticoagulants (DOACs) and antineoplastic agents (Aim 1); (2) examine benefits and risks of OAC initiation strategies according to stroke risk and compare the effectiveness and safety between DOACs and warfarin (Aim 2); and (3) develop and validate machine learning (ML) algorithms to predict 1-year risk of stroke and bleeding among patients with AFib and cancer (Aim 3). Methods: In Aim 1, we conducted a retrospective cross-sectional analysis of adverse events (AE) reports from the 2004-2021 US Food and Drug Adverse Events Reporting System (FAERS) database. Disproportionality analysis, logistic regression, and multi-item Gamma-Poisson shrinker were used to calculate the safety signals in bleeding or stroke of DOAC-antineoplastic agents combination. For Aims 2 and 3, population-based retrospective cohort studies including patients with newly diagnosed AFib and a record of breast, lung, or prostate cancer were conducted using the 2011-2019 Surveillance, Epidemiology, and End Results (SEER)-Medicare database. In Aim 2, we emulated target trials to compare risk of stroke and bleeding among patients who initiated OACs at CHA2DS2-VASc score ≥1, ≥2, ≥4, ≥6, and never initiated (Aim 2.1) and between DOAC and warfarin initiators (Aim 2.2). Pooled logistic regressions with inverse probability weighting were used to estimate the treatment effect, adjusting for baseline and time-varying covariates. In Aim 3, we fitted elastic net, Random Forest (RF), extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and Neural Network on the training dataset and validated ML algorithms on the testing dataset. Area under the curve (AUC), sensitivity, specificity, F2 score, and Brier score were compared across the models. Results: No significant risk signal of DDIs between DOACs and antineoplastic agents on therapeutic class level was detected. For individual antineoplastic drugs, only the concomitant DOAC-neratinib exhibited a risk signal [EBGM (EB05-EB95) = 2.71 (2.03-3.54)]. Next, we found that OAC initiation at higher level of CHA2DS2-VASc score (≥6) reduced risk of stroke compared to non-initiators, while all OAC initiation strategies at different level of CHA2DS2-VASc score reduced risk of bleeding. However, OAC initiation at low CHA2DS2-VASc scores were not beneficial or even harmful among patients with lung cancer or advanced cancer stages. Further, DOACs had a similar risk of stroke and major bleeding compared with warfarin in patients with AFib and cancer. Lastly, in prediction of ischemic stroke, RF outperformed other ML models [AUC 0.916, sensitivity 0.868, specificity 0.801, F2 score 0.375, Brier score=0.035]. However, the performance of ML algorithms in prediction of major bleeding was poor. Conclusions: Most in vitro DDIs between DOACs and antineoplastic agents may not be clinically relevant. In addition, among cancer patients with new AFib diagnosis, OAC initiation at higher risk of stroke (CHA2DS2-VASc score ≥6) may be more beneficial in preventing ischemic stroke and bleeding. Patients with advanced cancer status or low life-expectancy should be given OACs when CHA2DS2-VASc score≥6. Regarding treatment choice, DOACs are safe and effective alternatives to warfarin in patients with AFib and cancer. For risk assessment, we demonstrated a promising application of ML in prediction of stroke among patients with AFib and cancer. This tool may be leveraged in assisting clinicians to identify patients at high risk of stroke and optimize treatment decisions.