IQR indicates interquartile range.
aCHADS2 score components include congestive heart failure (1 point), hypertension (1 point), age 75 years or older (1 point), diabetes mellitus (1 point), prior stroke (2 points), CHA2DS2VASc score components include congestive heart failure (1 point), hypertension (1 point), age 75 years or older (2 points), diabetes mellitus (1 point), prior stroke (2 points), vascular disease (1 point), age 65 to 74 years (1 point), female sex (1 point).
bHigh-performing sites were defined as having achieved an adherence rate at or above the median proportion of adherence in this cohort (74%); low-performing sites were defined as an adherence rate of less than the median proportion the adherence in this cohort.
Each bar represents a participating site. The height of each bar represents the proportion of patients who had adhered to their medication regimen at that site. The dark bar represents the median proportion of the 4863 participant patients (74%) who were adherent per site.
Adherence to dabigatran was defined as the proportion of days covered as being at least 80%. The dotted line represents no effect. All observations to the right of the vertical line signify a positive association with dabigatran adherence. The error bars represent 95% CIs. Variables included in the model were age, sex, race (white), congestive heart failure, hypertension, diabetes, stroke or transient ischemic attack, chronic kidney disease, bleeding, myocardial infarction, liver disease, depression, alcohol, drug abuse, region, patients per hospital, median income in hospital’s county, proportion of urban to rural patients at hospital, education, selection, and follow-up.
eMethods. Supplemental methods
eTable 1. Components of the qualitative tool-kit
eTable 2. Indications and contra-indications for use of dabigatran released by the Veterans Health Administration Pharmacy Benefits Management
eTable 3. Baseline characteristics of study population and study sites stratified by site performance
eFigure. Pictorial representation of proportion of days covered calculation
Customize your JAMA Network experience by selecting one or more topics from the list below.
Shore S, Ho PM, Lambert-Kerzner A, et al. Site-Level Variation in and Practices Associated With Dabigatran Adherence. JAMA. 2015;313(14):1443–1450. doi:10.1001/jama.2015.2761
Unlike warfarin, which requires routine laboratory testing and dose adjustment, target-specific oral anticoagulants like dabigatran do not. However, optimal follow-up infrastructure and modifiable site-level factors associated with improved adherence to dabigatran are unknown.
To assess site-level variation in dabigatran adherence and to identify site-level practices associated with higher dabigatran adherence.
Design, Setting, and Participants
Mixed-methods study involving retrospective quantitative and cross-sectional qualitative data. A total of 67 Veterans Health Administration sites with 20 or more patients filling dabigatran prescriptions between 2010 and 2012 for nonvalvular atrial fibrillation were sampled (4863 total patients; median, 51 patients per site). Forty-seven pharmacists from 41 eligible sites participated in the qualitative inquiry.
Site-level practices identified included appropriate patient selection, pharmacist-driven patient education, and pharmacist-led adverse event and adherence monitoring.
Main Outcomes and Measures
Dabigatran adherence (intensity of drug use during therapy) defined by proportion of days covered (ratio of days supplied by prescription to follow-up duration) of 80% or more.
The median proportion of patients adherent to dabigatran was 74% (interquartile range [IQR], 66%-80%). After multivariable adjustment, dabigatran adherence across sites varied by a median odds ratio of 1.57. Review of practices across participating sites showed that appropriate patient selection was performed at 31 sites, pharmacist-led education was provided at 30 sites, and pharmacist-led monitoring at 28 sites. The proportion of adherent patients was higher at sites performing appropriate selection (75% vs 69%), education (76% vs 66%), and monitoring (77% vs 65%). Following multivariable adjustment, association between pharmacist-led education and dabigatran adherence was not statistically significant (relative risk [RR], 0.94; 95% CI, 0.83-1.06). Appropriate patient selection (RR, 1.14; 95% CI, 1.05-1.25), and provision of pharmacist-led monitoring (RR, 1.25; 95% CI, 1.11–1.41) were associated with better patient adherence. Additionally, longer duration of monitoring and providing more intensive care to nonadherent patients in collaboration with the clinician improved adherence.
Conclusions and Relevance
Among nonvalvular atrial fibrillation patients treated with dabigatran, there was variability in patient medication adherence across Veterans Health Administration sites. Specific pharmacist-based activities were associated with greater patient adherence to dabigatran.
Atrial fibrillation is the most common cardiac arrhythmia, affecting more than 3 million patients and necessitating treatment with oral anticoagulation in moderate- to high-risk patients to reduce stroke risk.1,2 Warfarin was the only treatment available until the recent introduction of target-specific oral anticoagulants (TSOACs). Dabigatran is the first of several in this class approved in United States for prevention of stroke and systemic embolism in atrial fibrillation.3 Unlike warfarin, for which periodic laboratory testing is required, TSOACs do not require routine testing to evaluate anticoagulation effect.
Prior studies with warfarin have demonstrated site-level variation in medication safety and effectiveness.4 Particular components of higher performing sites that have been found to improve warfarin adherence include frequent patient contact and identifying delinquent patients through involvement of anticoagulation clinics.5 Although TSOACs do not require monitoring, drug adherence in observational studies has been poorer than observed in clinical trials.6,7 A previous study reported that suboptimal adherence to dabigatran was associated with increased risk of stroke and death.6 Since TSOACs do not need dose titration or level monitoring, whether patient follow-up infrastructure similar to warfarin is needed remains unknown.
Accordingly, we aimed to explore site-level variation in patient adherence to dabigatran and modifiable site-level practices associated with improved adherence in the Veterans Health Administration (VHA). We first evaluated the extent of variation in patient adherence across VHA sites. Next, we conducted qualitative inquiries to identify variation in site-level practices for management of patients taking dabigatran. Finally, we evaluated the association between site-level practices and patient adherence after adjustment for site and patient-level characteristics.
We performed a mixed-methods study qualitatively examining site practices followed by quantitative estimation of site-level adherence and the association between site practices and adherence. We included VHA sites with 20 or more patients filling a dabigatran prescription for at least 30 days at a VHA pharmacy between October 1, 2010, and September 30, 2012, with at least 30 days of follow-up (Figure 1). Outpatient and anticoagulation clinic supervisors were initially contacted by email to explain the goals and procedures of this study and request participation. Clinic supervisors provided contact information for persons in the hospital deemed most appropriate for inquiry. In brief, we used semistructured telephone interviews composed of open-ended questions with specific probes asking about all procedures involved in managing patients who were precribed dabigatran (eTable 1 in the Supplement). We then used an iterative, inductive, and deductive toolkit of analytical strategies drawing primarily on content analysis methodology to identify site-level strategies aimed at improving TSOAC adherence.8 These strategies were subsequently analyzed quantitatively to examine if they were statistically associated with improved adherence. The Colorado Multiple Institutional Review Board approved this study and waiver of informed patient consent was granted. Further details regarding site selection, methods for qualitative and quantitative data collection, and analyses are detailed in the eMethods section of the Supplement.
Primary exposure variables were site-level strategies identified during our qualitative inquiries. These were chosen since they are amenable to easy change and included
Patient selection: Once dabigatran was ordered by a clinician, a detailed review of patient charts was carried out by dedicated pharmacists to ensure appropriate indication and to rule out contraindications consistent with VHA recommendations (eTable 2 in the Supplement). This included assessment of adherence to warfarin and other twice-daily medications. The primary rationale behind such selection of patients adherent to other medications was concern for worse adherence in the absence of regular monitoring such as that provided for warfarin.
Patient education: The provision of mandatory pharmacist-led education prior to dispensing dabigatran for the first time with subsequent reenforcements at each patient-pharmacist contact. Education included, at minimum, appropriate storage, therapy indication, review of adverse-effects (ie, gastrointestinal discomfort, bleeding, or strokelike symptoms); importance of adherence, management of missed doses, risks of falls; and review of drug interactions and appropriate daily dosing.
Patient monitoring: The provision of dedicated pharmacist-led adverse event and adherence monitoring. This included assessment of gastrointestinal adverse effects, bleeding events, or strokelike symptoms. In addition, immediately before and after a procedure, the pharmacist in collaboration with a clinician would guide patients on using TSOACs to minimize bleeding events. Adherence monitoring included determining how medication was taken and stored, frequency of missed doses with timely laboratory testing (including renal and liver function, complete blood counts). Four domains were identified in this component: (a) Anticoagulation Clinic Involvement: We included this based on the significant heterogeneity observed in responses to our qualitative inquires regarding an ideal follow-up coordinator for patients prescribed TSOAC. Additionally, previous research has demonstrated that patients who had previously taken warfarin had a higher adherence to dabigatran than patients who had not taken warfarin. This association could have potentially resulted from these patients receiving treatment from anticoagulation clinics; (b) Mode of Patient Contact: The mode of contacting patients was characterized as telephone clinics, face-to-face clinics, or contact strategies tailored to patient convenience; (c) Monitoring Stratification: The duration of monitoring was stratified as 3, 6, or 12 or more months; and (d) Tailoring the Monitoring Process to Nonadherent Patients and Collaborative Care: Pharmacists providing more intensive monitoring to patients who were nonadherent in collaboration with the patients’ physicians to improve adherence.
The primary outcome was patient adherence to dabigatran, measured as a dichotomous variable for the proportion of days covered (PDC) of at least 80%. This PDC cut-off is consistent with published research9,10 including the analysis of patients taking treatment that was reported in the Randomized Evaluation of Long-term Anticoagulation Therapy (RE-LY) trial.11 Consistent with prior literature, we defined PDCs as the total number of nonhospitalized days in which dabigatran was supplied divided by observation time (eFigure in the Supplement). This method has been detailed and validated, showing that it correlates with other direct adherence measures such as blood-drug levels.6,9,12,13 The primary rationale for assessing adherence was the association between a low PDC and increased hazard for all-cause mortality and stroke that has been previously demonstrated. For every 10% decrease in dabigatran PDC, the hazard for all-cause mortality and stroke increased by 13% (hazard ratio [HR], 1.13; 95% CI, 1.07-1.19).6
Further details regarding our proportion of days estimates and other covariates are in the eMethods section of the Supplement.
The comparisons of patient and site characteristics between participating and nonparticipating sites and low- and high-performing sites were made using the χ2 test for categorical variables and the t test for continuous variables. We decided a priori that sites participating in the qualitative inquiry would be classified as high-performing if the site-level PDC was at or higher than the median PDC of the entire cohort. Remaining sites were classified as low-performing. To assess site-level variation in dabigatran PDC, we used a mixed-effects logistic regression model with a random intercept for the site adjusted for other covariates listed in the eMethods of the Supplement. Next, we used median odds ratios (ORs) to summarize the magnitude of site-level variation on a similar scale that covariate effects are usually expressed, as an OR.14 For 2 patients with same covariates at different sites, the ratio of the odds of patient adherence to nonadherence when moving from a higher-performing site to a lower-performing site was first calculated for all possible pairs of patients. The median value from this distribution was selected as the median OR.
Modified Poisson models were generated using a generalized estimate equation to account for clustering of patients within sites. This analysis was limited to sites participating in the interview process. The models used key site-level strategies identified during the interviews as exposure variables and PDC as the outcome and were adjusted for selected covariates. Secondary models looked separately at characteristics of monitoring including duration of follow-up, contact method, clinician, and follow-up of patients who were nonadherent. Modified Poisson models were used in place of log binomial models due to issues with model convergence15 and were generated from the geeglm function within the geepack package in R 3.1.1 (R Core Team, 2014).16-18 Furthermore, we explored interactions between education and monitoring as well as education plus monitoring and selection using the same modified Poisson approach. For all analyses, significance testing was 2-sided and P values <.05 were considered statistically significant.
Between October 1, 2010, and September 30, 2012, a total of 5376 patients from 125 sites filled a dabigatran prescription (≥30 days at a VHA pharmacy with at least 30 days of follow-up). Of those, 3882 patients (72.2%) were adherent (PDC ≥80%) to taking dabigatran (median, PDC of 94%; interquartile range [IQR], 76%-100%; mean, 84%; SD, 22%). Furthermore, 1247 patients (23%) had a medication gap of more than 30 days. Details have been previously published.6 Overall, 58 sites with fewer than 20 patients prescribed dabigatran during the study period were excluded. The final cohort comprised 4863 patients who were prescribed dabigatran at 67 VHA sites (Figure 1). At total of 47 pharmacists at 41 sites were interviewed. Baseline patient and site characteristics of the study cohort stratified by their participation are listed in eTable 3 in the Supplement.
Across 67 eligible sites, a wide variation in proportion of patients who were adherent ranging from 42% to 93% (Figure 2) was observed. The overall mean proportion of patients who were adherent was 72% (SD, 12%; median 74%; IQR 66%-80%). This variation persisted after adjustment for measured confounders with a median OR of 1.57. Sites participating in the qualitative inquiry were classified as high performing if the site-level PDC was at least a median proportion of days of 74%. Remaining sites were classified as low performing.
Out of the 41 eligible sites participating in the qualitative inquiry, 40 sites (98%) reported review of indications and contraindications by dedicated pharmacists to ensure appropriate use of dabigatran after it was ordered by the clinician. However, 10 sites (24%) reported not examining adherence to warfarin or other medications as a criterion for selecting patients. Overall, 30 sites (73%) reported providing mandatory pharmacist-led educational initiative prior to dabigatran initiation. Additionally, 28 (68.3%) sites reported that all patients received pharmacist-led adverse event and adherence monitoring, and 13 (31.7%) sites reported that such monitoring was the initiating physician’s responsibility.
Among 28 sites providing dedicated pharmacist-led monitoring, 12 sites reported monitoring for 3 months, 6 sites for 6 months, and 10 sites monitored patients indefinitely for as long as dabigatran was continued. At all these sites, this duration was determined largely by local policies established by consensus of cardiologists, hematologists, anticoagulation clinic supervisors, and administrators. At 22 sites, monitoring was performed by anticoagulation clinics; at the remaining 6 sites, by outpatient pharmacists not associated with anticoagulation clinics because the burden was too large for anticoagulation clinics to provide such follow-up or their involvement was thought to be unnecessary. Patients were contacted by telephone only at 18 sites, 4 sites required all patients to attend in-person anticoagulation clinics, and the remaining 6 sites tailored the contact mode to patient preference and location with patients residing in remote sites receiving telephone follow-up. Additionally, 20 sites tailored their monitoring process to patient adherence and collaborated with clinicians in managing patients who were nonadherent. At all sites, patients were given contact information for a clinic telephone line in case of nonemergency questions. Between 2010 and 2012, management practices across all sites were constant. In addition, all sites reported having the same practices for patients who did not receive care at the VHA but only used VHA pharmacies for obtaining dabigatran.
The Table shows patient and site characteristics stratified by their performance (low performing defined as site PDCs being less than the national median of 74% observed in this data set). Of the participating sites, 18 (43.9%) were low performing and 23 (56.1%) were high performing. Compared with low-performing sites, high-performing sites were from the Midwest and had a higher proportion of white patients and had a lower proportion of patients with chronic kidney disease. There were no other clinically meaningful differences in patient and site characteristics. However, site-level practices differed significantly between low- and high-performing sites as highlighted in the Table. Overall, only 3 sites belonging to a single geographic area reported using software to track patient adherence to dabigatran. They also reported a regional performance measure of the proportion of the patients who were adherent to taking dabigatran. All of these sites were high performing. However, due to the small number of such sites, use of tracking software to identify nonadherent patients was not included in the multivariable model to assess its association with proportion of days.
The proportion of patients adherent was higher at sites performing appropriate selection (75% vs 69%), included patient education (76% vs 66%), and maintained monitoring (77% vs 65%). Following multivariable adjustment, there was no statistically significant association between the provision of pharmacist-led education prior to dabigatran initiation and adherence (adjusted relative risk [RR], 0.94; 95% CI, 0.83-1.06). However, appropriate patient selection defined as evaluation of indication, ruling out contraindications, and assessing adherence to other medications as a selection criterion prior to initiation of dabigatran was significantly associated with dabigatran adherence (RR, 1.14; 95% CI, 1.05-1.25). Similarly, pharmacist-led adverse event and adherence monitoring was associated with improved dabigatran adherence (RR, 1.25; 95% CI, 1.11-1.41). The interactions examined were not statistically significant (P > .05).
Adherence progressively increased with longer follow-up duration (Figure 3). Tailoring the monitoring process and collaborating with physicians in managing nonadherent patients was associated with better adherence (RR, 1.31; 95% CI, 1.16-1.47). There was no difference in patient adherence to dabigatran at sites with involvement of anticoagulation clinics (median PDC, 0.97; IQR, 0.83-1.00) vs sites without their involvement (median PDC, 0.94; IQR, 0.80-1.00). Association between mode of contacting patients and dabigatran adherence was not statistically significant in the multivariable model.
In this site-level analysis of patients who had atrial fibrillation and were treated with dabigatran, we found variation in the patient proportion adherent to dabigatran across VHA sites. We also found variation in their management strategies with some sites using pharmacists and others deferring management solely to clinicians. The principal finding is that appropriate patient selection was associated with better dabigatran adherence. Similarly, pharmacist-led monitoring was associated with higher adherence with a progressive increase in adherence with longer monitoring duration. Additionally, pharmacist collaboration with clinicians for patients who were nonadherent was associated with higher adherence rates. These findings suggest that such site-level practices provide modifiable targets to improve patient adherence to dabigatran as opposed to patient characteristics that frequently cannot be modified.
Our results highlight the importance of selecting patients and monitoring strategies to translate the efficacy of TSOACs in randomized trials to clinical practice. Prior studies have described variation in patient performance on warfarin across sites further highlighting the importance of management strategies in improving patient performance to anticoagulants.4,19,20 Similar relationship between pharmacist-led initiatives and medication effectiveness has been described previously as well.5,21,22 The higher adherence rates associated with provision of such dedicated monitoring even for a short time is potentially due to consistent contact made with patients. Prescribing physicians may not always be able to contact patients on a routine basis due to their high workload and limited time available during clinic visits. Regular adverse-event monitoring may help mitigate medication discontinuation due to minor adverse effects. However, due to the observational nature of our study, a cause and effect relationship cannot be estimated.
These findings may have several implications for present day health care systems because altering the management of dabigatran patients may help improve adherence. In the Post-Myocardial Infarction Free Rx Event and Economic Evaluation (MI FREEE) trial23 that assessed the effect of eliminating medication co-payments on medication adherence, adherence rates were 4% to 6% higher in the group that did not have a co-payment than the group that did. In comparison, we observed higher adherence rates among sites providing various pharmacist-based interventions. Hence, we believe our results are not clinically trivial and future randomized trials are needed to better evaluate the effect of our findings. Feasibility of these practices is demonstrated by their successful implementation at VHA sites. Although delivery of all proposed components is likely feasible in an integrated health care system, individual components such as pharmacist-led education prior to dispensing medications can readily be implemented elsewhere. With rapid adoption of TSOACs into practice, randomized trials evaluating the effect of multimodal interventions involving pharmacists on patient adherence to dabigatran and other TSOACs are needed. Evaluation of lower-cost alternatives to pharmacist-led management, such as clinician-directed automated voice messaging or electronic reminders for medication renewals, could also prove valuable.
Our study should be interpreted in light of several limitations. First, data on practice patterns were evaluated based on interview responses without direct observation. The interview pool consisted only of pharmacists. Although these pharmacists may have been considered the most appropriate respondents by the site, it is possible that important care processes not known to these individuals could have been underascertained and therefore not captured in the responses. Second, we could not obtain data on all eligible sites, highlighting potential selection bias. We also did not include sites with fewer patients due to concern for unstable PDC calculations. In an attempt to minimize selection bias, we obtained detailed data on practice patterns at both low- and high-performing sites. Third, some covariates had small numbers making it difficult to estimate their influence on patient adherence to dabigatran. However, we used robust statistical methods in an attempt to minimize this. Fourth, although PDC has been validated as an adherence measure, we cannot ascertain whether dispensed medications were actually taken, only that they were dispensed. However, direct measures of TSOAC use are currently not available, and we accounted for cancelled prescriptions VHA and non-VHA hospitalizations as detailed in the online Supplement providing accurate PDC estimates. Fifth, generalizability of our results is limited and may not apply to under-represented populations such as women. Sixth, we assumed that all patients refilled dabigatran at VHA pharmacies and did not have linked Medicare Part D data available for the observation period. However, because dabigatran is available at a nominal cost due to the fixed, nontiered co-payment system at the VHA, patients are less likely to use non-VHA sources. Nevertheless, we empirically evaluated the potential for veterans receiving dabigatran outside of the VHA system in a similar cohort (linked to Medicare Part D data),24,25 and found less than 0.1% use of Medicare for obtaining dabigatran. Still, non-VHA pharmacy use could lead to more conservative PDC estimates through our calculations. Seventh, as with all observational studies, residual confounding from unmeasured covariates persists in our study. Although we adjusted for a large number of covariates, there are other unmeasured covariates such as cognitive condition, marital status, etc, that we could not adjust for.
Among patients with atrial fibrillation who filled dabigatran prescriptions within the VHA, there was variability in patient medication adherence across sites. Specific pharmacist-based activities were associated with greater patient adherence to dabigatran.
Corresponding Author: Mintu P. Turakhia, MD, MAS, Department of Cardiology, VA Palo Alto Health Care System, 3801 Miranda Ave, Ste 111C, Palo Alto, CA 94304 (firstname.lastname@example.org).
Author Contributions: Drs Shore and Turakhia had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Shore, Ho, Turakhia.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Shore, Lambert-Kerzner, Glorioso, Turakhia.
Critical revision of the manuscript for important intellectual content: Shore, Ho, Carey, Cunningham, Longo, Jackevicius, Rose, Turakhia.
Statistical analysis: Glorioso, Carey, Turakhia.
Administrative, technical, or material support: Shore, Cunningham, Longo, Turakhia.
Study supervision: Lambert-Kerzner, Turakhia.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Turakhia reported that he is a consultant to Precision Health Economics, Medtronic Inc, and St Jude Medical Inc. No other disclosures are reported.
Funding/Support: Dr Turakhia is supported by a Career Development Award from VA Health Services Research & Development, an American Heart Association National Scientist Development Grant, and a Gilead Sciences Cardiovascular Research Scholars Program award.
Role of the Funder/Sponsor: None of the funders were involved in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, and approval of the manuscript; or decision to submit it for publication.
Disclaimer: The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States government.
Previous Presentation: This study has presented as an abstract presentation at AHA-QCOR 2014 Scientific Session in Baltimore, MD.
Additional Contributions: We thank Madeline McCarren for review of the manuscript and discussion of possible interpretations of the findings.