Atrial fibrillation (AF) is the most common cardiac arrhythmia, with around 5 million new cases each year [1]. Anticoagulation significantly reduces the incidence of clinical stroke and mortality among patients with AF at moderate to high risk of thromboembolic events (CHA₂DS₂-VASc ≥2) [2]. Based on the results of multiple randomized controlled trials (RCTs), the direct oral anticoagulants (DOACs) are now frequently used to reduce this risk [3‐5]. Additionally, they have demonstrated favorable risk-benefit profiles compared to warfarin with significant reductions in stroke, intracranial hemorrhage, and mortality [6].

While decades of knowledge have demonstrated the safety and efficacy of warfarin, less is known about how the DOACs have been adopted into routine clinical care. Furthermore, concerns remain about DOACs because of their shorter half-lives and the potential for reductions in effectiveness with poor adherence. Because of the lack of laboratory monitoring to assess therapeutic levels, it is important to assess adherence to DOACs and the extent to which it varies by patient characteristics and different DOACs currently available. The VA health care system, being the largest integrated healthcare system with an anticoagulation infrastructure and standard copayments, offers a unique opportunity to examine real-world adherence under ideal conditions.

Accordingly, we evaluated adherence patterns to the three approved DOACs (as of January 2015) and the association between adherence with outcomes among patients with AF and prescribed DOACs in the VA Healthcare System. Specifically, we sought to characterize adherence using portion of days covered (PDC) to dabigatran, rivoraxaban, and apixaban separately. Second, we assessed the patient level factors associated with nonadherence. Finally, we assessed the association between lower adherence and outcomes of stroke and mortality for each of the DOACs.

Over the study period, a total of 10,279 patients were started on DOACs. After excluding patients with less than 1-year follow-up period (n = 3111), those who were prescribed more than 1 DOAC during the study period (n = 872), those who were prescribed warfarin after DOAC initiation (n = 1446) and those with CHA₂DS₂-VASc score < 2 (n = 1968), a total of 2882 patients were included in the analysis (Fig. 1).

Baseline patient characteristics are described in Table 1. The mean age of the cohort was 67.4 ± 9.5, most whom were white (82.7%) and male (96.9%). Co-morbidities included hypertension (88.7%), diabetes mellitus (48.5%), congestive heart failure (29.6%), cerebrovascular disease (11.1%), and prior myocardial infarction (13.6%). The mean and median CHA₂DS₂-VASc score were 2.9 ± 1.1 and 3 (IQR 2–3), respectively. Most patients in the cohort were prescribed dabigatran (72.7%), compared with rivaroxaban (19.8%) or apixaban (7.5%). Differences in patient level characteristics between DOACs were largely similar with significant differences seen in age (mean age among patients prescribed dabigatran 66.9 vs. 73.1 for apixaban), rate of hypertension (89.4% for dabigatran vs. 83.3% for apixaban), and CHA₂DS₂-VASc score (2.89 for dabigatran vs. 3.2 for apixaban).

The mean PDC was 0.84 ± 0.20 for dabigatran, 0.86 ± 0.18 for rivaroxaban, and 0.89 ± 0.14 for apixaban (p < 0.01) (Table 1). The proportion of non-adherent patients, defined as PDC <0.80, was 28.8%, 25.0%, 22.8% respectively (p = 0.05). Adherence using PDC stratified by patients who had previously been on warfarin and those who were started de novo are described in Table 2. There was a statistically significant small difference in PDC between drug types and history of prior warfarin use (P < 0.01). Based on PDC <80% for nonadherence, there was no statistically significant difference between drug types and history of prior warfarin use (P = 0.44).

Table 3 describe factors associated with nonadherence, where an odds ratio of >1 is associated with greater nonadherence and an odds ratio < 1 is associated with greater adherence. Age in years (OR, 0.98; 95% CI, 0.96–0.99; p < 0.01), hypertension (OR, 0.69; 95% CI, 0.49–0.99; p = 0.04), diabetes (OR, 0.57; 95% CI, 0.41–0.79; p < 0.01), and stroke (OR, 0.36; 95% CI, 0.2–0.68; p < 0.01) were associated with greater adherence.

The combined end-point of mortality and/or stroke based on adherence patterns are described in Table 4. The total number of follow-up days was 1,922,857, with mean (SD) and median (IQR) of 667.2 (432.2) and 582 (388, 933), respectively. The overall rate of death and stroke were 17.4% (502) and 1.7% (49), respectively.

Among patients on dabigatran, there was a significant association between lower adherence (per 0.1 decline in PDC measured over the initial 6 months after initiation of therapy) and higher risk of death or stroke; (HR, 1.07; 95% CI, 1.03–1.12; per 0.1 drop in the PDC; p < 0.01; n = 277 total events; death = 253, stroke = 24). Among patients on rivaroxaban, there was a trend for an association between lower adherence at 6 months (per 0.1 decline in PDC) and higher risk of mortality and stroke (HR, 1.07; 95% CI, 0.89–1.28; p = 0.46; n = 25 total events; death = 24, stroke = 1]. In secondary analysis, nonadherence (PDC <80%) at 6 months to dabigatran was associated with increased risk of death or stroke (HR, 1.54; 95% CI, 1.20–1.97; p < 0.01). There was a similar trend for rivaroxaban but it was not statistically significant (HR, 1.74; 95% CI, 0.77–3.94; p = 0.18). Outcomes analyses were not conducted for apixaban due to the small number of events (n = 5).

Using VHA data from June 2006 to January 2015, we characterized adherence to DOACs and assessed the association between nonadherence and outcomes among patients in the VA Healthcare System. There were three main findings. First, 1 in 4 patients had sub-optimal adherence, which varied slightly based on the DOAC. Second, several patient factors were associated with greater medication adherence, including older age, diabetes, and stroke. Third, nonadherence was associated with adverse outcomes including mortality and stroke for dabigatran, with a similar trend for rivoraxaban.

Medication nonadherence is a prevalent and growing concern among healthcare providers. Previous investigations have described adherence patterns to dabigatran. Using the Danish National Prescription Registry, Gorst-Rasmussen evaluated 2960 patients started on dabigatran for AF [9]. The mean one-year PDC was 0.84 and about one-quarter of patients were nonadherent (PDC <80%). Among 17,000 patients on dabigatran, PDC among de novo starters were 0.67 and 0.71 for those previously on warfarin [10]. Schulmann interviewed a small cohort of patients (103) and found that 88% of patients were adherent (PDC >80%) to dabigatran [11]. The VA Healthcare System offers a unique opportunity to examine real-world adherence under ideal conditions. It is the largest integrated healthcare system and has specific processes of care by which anticoagulation is safely administered (including anticoagulation clinics and a clinical pharmacist infrastructure). Additionally, copayments are standard across oral anticoagulation agents which does not lead to a differential copayment burden. A previous investigation in the VA population demonstrated that 27.8% of patients on dabigatran for AF were nonadherent, which was also associated with an increased risk for all-cause mortality and stroke [8]. Our findings further this knowledge by demonstrating that nonadherence is common, even among the newer DOACs. Furthermore, these findings are in direct discordance with adherence patterns seen in randomized controlled trials which are generally higher. Differences in adherence patterns between clinical trials and routine clinical practice are multifactorial and include a potential Hawthorne effect for those enrolled in clinical trials and lack of close follow up, high incidence of comorbidities and polypharmacy, and/or financial constraints for those in routine clinical care. Additional research is needed to determine which factors can be targeted to maximize patient outcomes.

Predictors of greater adherence included older age and a history of hypertension, diabetes, and stroke. Factors previously known to be associated with nonadherence include male gender, homelessness, and psychiatric disorders, particularly depression while those associated with greater adherence include high level of education, stability of family background, and affordability to therapy [12‐15]. Additionally, treatment of asymptomatic disease is associated with poor adherence [15]. Importantly, cardiovascular disease, such as in some cases of atrial fibrillation, are generally asymptomatic chronic diseases where the perceived benefits of daily medical therapy may not be apparent to patients. Among these patients, physicians should have a heightened awareness of the possibility of poor adherence and consider directly asking patients about their adherence.

Adherence to DOACs was a significant predictor of outcomes including all-cause mortality and stroke. A similar association of risk of death or stroke was seen among the 27.8% of non-adherent patients on dabigatran, with a hazard ratio of 1.13 (95% CI 1.07–1.19) [8]. Furthermore, prior studies have shown similar associations of risk and nonadherence among patients with coronary artery disease, heart failure, hypertension, and hyperlipidemia [16‐19]. For example, among 10,000 patients with diabetes, 21% were non-adherent which was associated with an increased risk for all-cause hospitalization and mortality [16]. These findings suggest that medication nonadherence is common among many chronic illnesses and highly impacts outcomes.

There are a few implications of these findings. First, while adherence was significantly different stratified by DOAC, the absolute difference was very small. Similarly, differences in demographic and co-morbidities associated with nonadherence were small and determining nonadherence to DOACs based on these clinical characteristics may be difficult. This suggests that other mechanisms to detect patients at risk for nonadherence need to be identified, including directly asking patients about adherence or utilizing data systems in which prescription refills, or lack thereof, are recorded and provided in real-time. Further evaluation of such data systems should be developed to determine if they can successfully be implemented and improve patient adherence. Additionally, it is often thought that adherence to medical therapy improves with simplification; however, we found that adherence to the once daily rivaroxaban did not have better adherence rate. Further studies are needed to evaluate this perception.

Certain factors should be considered in the interpretation of this study. First, our analysis was confined to patients in the VA Healthcare System with the clear majority of patients being white males. Thus, our results may not apply to the broader population of patients being prescribed DOACs. However, the VA provides an opportunity to examine adherence in a closed pharmacy system with highly reliable data sources because refill data for these medications are well captured given the smaller copay in the VA. Second, while we utilized pharmacy databases to capture medication dispensing, there is a lack of distinction between dispensing and consumption, making this type of analysis less reliable at the individual level. However, refill adherence has previously been shown to be an accurate marker of patients’ adherence in the VA system [20]. Additionally, direct methods of assessing adherence (i.e. directly observed therapy) have limitations and are not practical for routine clinical use. Third, there were relatively fewer patients on rivaroxaban and apixaban as these are newer to the VA system and their use is dependent on contraindications or previous intolerance to dabigatran. Therefore, broader examination of outcomes is limited due to lack of power given the relatively smaller sample size. Furthermore, longer term outcomes are limited given the exclusion of a large majority of patients who had less than one year follow up. Additionally, the impact of longer duration of follow up on adherence or outcomes was not addressed but should be explored in future analysis. Fourth, direct comparisons between classes of DOACs were not performed as we are unable to account for unmeasured confounders and selection bias for the determination of which DOAC was selected. Fifth, we did not evaluate the association between adherence and bleeding given the low rate of bleeding events. Sixth, a large portion of patients (1446) were excluded after having been prescribed warfarin after DOAC initiation. The reasons for such were unclear from this study and further evaluation is needed to address reasons for a change in anticoagulation strategies. Finally, our results are likely an overestimation of adherence based on the exclusions. For example, by excluding patients who stopped or did not tolerate a DOAC, we selected a group of patients more likely to adhere to medical therapy and thus increasing the adherence rate beyond what is seen in practice.

This study characterizes adherence patterns and outcomes among patients in the VA Healthcare System using DOACs for AF. First, more than one quarter of patients had sub-optimal adherence with DOACs. While there were differences in adherence between DOAC and patient characteristics, these were not clinically significant. Second, outcomes including all-cause mortality and stroke were associated with medication adherence. Efforts towards identifying non-adherent patients and targeting adherence interventions are needed.