Switching warfarin to direct oral anticoagulants in atrial fibrillation: Insights from the NCDR PINNACLE registry

Abstract Background Previous studies examining the use of direct oral anticoagulants (DOACs) in atrial fibrillation (AF) have largely focused on patients newly initiating therapy. Little is known about the prevalence/patterns of switching to DOACs among AF patients initially treated with warfarin. Hypothesis To examine patterns of anticoagulation among patients chronically managed with warfarin upon the availability of DOACs and identify patient/practice‐level factors associated with switching from chronic warfarin therapy to a DOAC. Methods Prospective cohort study of AF patients in the NCDR PINNACLE registry prescribed warfarin between May 1, 2008 and May 1, 2015. Patients were followed at least 1 year (median length of follow‐up 375 days, IQR 154‐375) through May 1, 2016 and stratified as follows: continued warfarin, switched to DOAC, or discontinued anticoagulation. To identify significant predictors of switching, a three‐level multivariable hierarchical regression was developed. Results Among 383 008 AF patients initially prescribed warfarin, 16.3% (n = 62 620) switched to DOACs, 68.8% (n = 263 609) continued warfarin, and 14.8% (n = 56 779) discontinued anticoagulation. Among those switched, 37.6% received dabigatran, 37.0% rivaroxaban, 24.4% apixaban, and 1.0% edoxaban. Switched patients were more likely to be younger, women, white, and have private insurance (all P < .001). Switching was less likely with increased stroke risk (OR, 0.92; 95%CI, 0.91‐0.93 per 1‐point increase CHA2DS2‐VASc), but more likely with increased bleeding risk (OR, 1.12; 95%CI, 1.10‐1.13 per 1‐point increase HAS‐BLED). There was substantial variation at the practice‐level (MOR, 2.33; 95%CI, 2.12‐2.58) and among providers within the same practice (MOR, 1.46; 95%CI, 1.43‐1.49). Conclusions Among AF patients treated with warfarin between October 1, 2010 and May 1, 2016, one in six were switched to DOACs, with differences across sociodemographic/clinical characteristics and substantial practice‐level variation. In the context of current guidelines which favor DOACs over warfarin, these findings help benchmark performance and identify areas of improvement.

discontinued anticoagulation. To identify significant predictors of switching, a threelevel multivariable hierarchical regression was developed.
Conclusions: Among AF patients treated with warfarin between October 1, 2010 and May 1, 2016, one in six were switched to DOACs, with differences across sociodemographic/clinical characteristics and substantial practice-level variation. In the context of current guidelines which favor DOACs over warfarin, these findings help benchmark performance and identify areas of improvement. have noted disparities in DOAC use based on sex, race, and insurance status. [5][6][7][8][9] However, these analyses have focused exclusively on patients newly initiating anticoagulation for AF. There is limited data evaluating how commonly patients on warfarin are switched to DOACs and why they are switched. 7 The NCDR PINNACLE registry longitudinally follows patients with AF, providing insight into patterns of switching oral anticoagulant therapy and identifying patient and practice-level factors associated with switching. 10 Accordingly, we sought to examine patterns of anticoagulation among patients with AF treated with warfarin, determine predictors of switching from warfarin to DOACs, and identify potential disparities in care based on sociodemographic and clinical factors. We hypothesized that there would be significant disparities in switching patterns and substantial patient/practice-level variations.

| Data source
The NCDR PINNACLE registry is a prospective outpatient quality improvement registry that was created by the ACC in 2008. 10 In this registry, both academic and private practices collect longitudinal data pertaining to the care of patients with AF, coronary artery disease, hypertension, and heart failure. This includes demographics, medical comorbidities, medications, and practice/provider data. These data are collected from the patients' charts using a standardized collection tool to obtain and transmit data. De-identified data was extracted from an electronic medical record under a quality improvement model, with approval from an institutional review board and informed consent waived. Rigorous data definitions and periodic data quality audits are conducted to maintain NCDR data quality assurance. 11 Analysis of the data was performed at the Baim Institute for Clinical Research.

| Study population
The study cohort included all patients with nonvalvular AF, in the PIN-NACLE dataset with at least one warfarin prescription between

| Statistical methods
The primary outcome of this study was the proportion of patients with AF treated with warfarin that were switched to a DOAC and the proportion switched to each specific agent. We calculated the proportion of patients with AF treated with warfarin that were switched to a DOAC, and then calculated the proportion switched to each specific agent (dabigatran, rivaroxaban, apixaban, and edoxaban). Among the patients who were switched, we calculated the proportion who were subsequently switched back to warfarin, to a different DOAC, or had all anticoagulation discontinued. The mean/median number of visits in the PINNACLE dataset and the mean/median time of follow-up were also obtained. Additionally, patterns were determined after stratifying patients based on CHA 2 DS 2 -VASc score (groups: score 0-1, score 2-3, and score 4 or more). For the population of patients who were switched, demographic, clinical, laboratory, and medication information was compared across groups.
Categorical variables are reported as numbers (percentages) and continuous variables are reported as medians (25th and 75th percentiles) or means (SD). The significance of observed differences was tested using a Wilcoxon rank sum test or t-test for continuous vari- The distribution of the proportion of patients receiving warfarin who were switched to a DOAC was determined across practices. After the distribution was examined, additional practice-level analyses were conducted by stratifying practices into tertiles based on the proportion of patients who were switched to a DOAC (tertiles were arranged from lowest switching rates to highest).
To measure practice variation, the median odds ratio (MOR) was used. The MOR is able to provide cluster-level variance when a multilevel regression analysis is performed and can assess variation in a mixed effects model. An MOR >1.2 suggests clinically significant variation in the outcome among a unit after adjustment for fixed effects. In this analysis, an MOR >1.2 suggests clinically significant variation among practices in switching from warfarin to a DOAC. 12 3 | RESULTS

| Predictors of switching
Baseline demographic, clinical, and practice characteristics of the overall population with AF treated with warfarin (Table S1), patients who were and were not switched (Table S2), and within specific medication groups (Table S3)

| Practice variation in switching
There was substantial variation in switching patterns at the practice-  Figure 4.
The 355 practices were divided into tertiles (arranged from lowest switching rates to highest) based on the proportion of individuals switched from warfarin to DOACs (Table S4), with tertile 1 comprised practices with the lowest switching rates. Practices with higher switching rates tended to have higher clinical volume, more electrophysiologists, and were more likely to be located in the West region.

| Timing of switching
As time progressed, the proportion of individuals switched from warfarin to DOACs increased (Table S5)  The updated ACC/AHA/HRS guidelines recommend DOAC therapy over warfarin given data demonstrating comparable efficacy with less bleeding. A meta-analysis of randomized controlled trials demonstrated that DOACs provide significant reduction in stroke, intracranial hemorrhage, and mortality with similar major bleeding compared to warfarin. 13 Against this backdrop, previous analyses from the PIN-NACLE registry have shown higher rates of DOAC initiation in AF than the switching we observed, with approximately 60% of individuals newly initiated on anticoagulation receiving a DOAC. [5][6][7][8][9] Potential explanations for the relatively low rate of switching include therapeutic inertia, where providers may not choose to make a change to anticoagulation therapy if a patient has been stable on warfarin without adverse events. Additionally, when DOACs were new, some providers may have waited for accumulating evidence of safety and benefit prior to changing therapy. This may explain why more switching occurred later in the study. Providers may have also chosen to not switch patients they believed had adequate Time in Therapeutic Range (TTR) on warfarin. European guidelines suggest that providers can consider continuing warfarin if patients are able to maintain a TTR on warfarin ≥70% of the time. 14,15 However, studies suggest that less than 50% of patients on warfarin are able to maintain a TTR ≥70% at 6 months and a smaller proportion of these patients are able to maintain a TTR ≥70% at 12 months. 16 Finally, cost is a significant barrier to DOAC use for many patients in the United States, but is less of an issue in Europe where there is widespread prescription drug coverage. This is further emphasized by a study from a Danish cohort that showed significantly higher rates of switching than we observed in our analysis. 17 Broadly speaking, the substantial practice level variation suggests that there are multiple mechanisms underlying these findings. Both at the practice level and among providers within the same practice, there was substantial variation in switching patterns, emphasizing a lack of a Another key finding from our analysis is that switching was less prevalent among patients with increasing age, non-white race, those without private insurance, and men. There have been numerous disparities in many aspects of AF management, including initiation of DOACs and ablation. [5][6][7][8][9] Although predictors of switching have not been extensively examined in the literature, disparities in DOAC initiation have been observed based on age, race, insurance status, geographic region, female sex, lower household income, higher stroke risk, and higher bleeding risk. [18][19] In a Danish study that specifically focused on switching from warfarin to DOACs, it was similarly observed that younger age, female sex, history of stroke, and history of bleeding were predictors of switching. 17 Our analysis adds to this literature and further heightens the need for quality improvement efforts to address and close these disparities.
It is also notable that switching was more prevalent with increased bleeding risk (based on HAS-BLED scores) but less prevalent with increased stroke risk (based on CHA 2 DS 2 -VASc scores). We know from other studies that bleeding avoidance is an important driver of decision-making 20 4 Increased cost and other barriers to switching must be addressed, including patient perspectives and preferences.
The large degree of practice variation seen in this study highlights the lack of a standardized approach for switching patients.

| Limitations
There were a few important limitations of this analysis. First, since the PINNACLE registry contains data abstracted from practices participating in a quality improvement registry, it is possible that the data may not be representative of other practices in the United States. This program provides additional support to facilitate guideline recommended treatments that may not be available in most practices. Information such as accessibility to INR testing (ie, patients living far away from testing centers), household income, and drug benefit plans would be helpful to better define the role of cost in limiting switching. Additionally, as with any clinical management decision, there are often relevant factors that fall outside of information that can be collected in a registry, such as patient preference and unmeasured practice variation.
Other factors such as nonadherence, or TTR with warfarin may have affected anticoagulation patterns and switching to DOACs, but were not collected in the PINNACLE registry. Another limitation is the relatively sparse data on renal function available for our analysis. Renal dysfunction is an important consideration in choosing anticoagulation for AF patients and previously was a limitation for DOAC use, although current guidelines now suggest it is reasonable to anticoagulate these patients with apixaban. 4 With a relatively low proportion of patients with CKD, we were not able to systematically examine the association of renal function on switching.

| CONCLUSIONS
In a contemporary cohort of patients with AF treated with warfarin, switching from warfarin to DOACs was relatively uncommon.
Switching was less likely with rising stroke risk, but more likely with higher bleeding risk. Additionally, there was significant practice variation in switching. In the context of the 2019 update to the ACC/AHA/ HRS guidelines, which favor DOACs over warfarin, these findings may help benchmark performance and identify areas of improvement.