Prediction of Thromboembolism Risk and Comparative Effectiveness of Warfarin Treatment Strategies in a Hemodialysis Population With Atrial Fibrillation

Hemodialysis patients with atrial fibrillation (AF) are at increased risk of thromboembolism compared to patients with normal kidney function. CHA2DS2-VASc, commonly used risk score to determine treatment to prevent thromboembolism among hemodialysis patients, was developed in the general population with AF. Therefore, we developed and validated a claims-based prediction model for 1-year risk of thromboembolism in hemodialysis patients with AF and compared it to the CHA2DS2-VASc score. Warfarin has been the mainstay of anticoagulation treatment in dialysis patients given the limitations of renal clearance with newer oral anticoagulants. Few studies have examined the association between quality of anticoagulation achieved using international normalized ratio (INR) and clinical outcomes in hemodialysis patients with AF. We estimated the effect of following different INR regimens on clinical outcomes among hemodialysis patients initiating warfarin.For aim 1, we identified 84,531 anti-coagulant naïve patients with AF on dialysis from 2007-2017 using the United States Renal Data System (USRDS). We developed and internally validated two new prediction models for 1-year risk of thromboembolism; a logistic CHA2DS2-VASc predictor model with re-estimated parameters (henceforth called “logistic”) from our study population and a claims-based LASSO logistic model. Both were compared with the CHA2DS2-VASc base model using published parameter estimates from literature (henceforth called “base model”). The LASSO logistic model had higher discrimination than the logistic model and the base model. Both the LASSO logistic and the logistic model showed excellent calibration, while the base model systematically overestimated the predicted risk. As a prognostic model, the LASSO logistic model and the logistic model with re-estimated parameters from this study showed excellent calibration. Despite a 4.5% improvement in AUC, overall discrimination was sub-optimal (AUC<0.7) for all models.For aim 2, we identified 2,355 dialysis patients with AF newly initiating warfarin from 2007-2014 using the linked USRDS and clinical database. Following an INR regimen that was allowed to vary but never exceeded 3.0 and natural course INR was found to have lower risk for all three outcomes of thromboembolism, bleeding and all-cause mortality compared to tight control in-range INR (2.0-3.0), while following an INR regimen that was ≥2.0 was associated with higher risk for all three outcomes. However, the risk difference for thromboembolism did not reach statistical significance as seen from overlapping confidence intervals. These results demonstrate that appropriate INR interventions can help reduce the risk of bleeding and all-cause mortality in dialysis patients with AF on warfarin