Introduction Medications cannot exert their effect if not taken as prescribed by patients. Our objective was to summarise the observational evidence on adherence to oral anticoagulants (OACs) among patients with atrial fibrillation (AF).
Methods In March 2019, we systematically searched PubMed/Medline, Embase, CINAHL and PsycINFO (from inception) for observational studies measuring adherence, its determinants and impacts in patients with AF. Mean adherence measures and corresponding proportions of adherent patients were pooled using random effects models. Factors shown to be independently associated with adherence were extracted as well as the clinical and economic outcomes of adherence.
Results We included 30 studies. Pooled mean adherence scores of over half a million patients with AF 6 months and 1 year after therapy initiation were 77 (95% CI: 74–79) and 74 (68–79) out of 100, respectively. Drug-specific pooled mean adherence score at 6 months and 1 year were as follows: rivaroxaban: 78 (73–84) and 77 (69–86); apixaban: 77 (75–79) and 82 (74–89); dabigatran: 74 (69–79) and 75 (68–82), respectively. There was inadequate information on warfarin for inclusion in meta-analysis.
Factors associated with increased adherence included: older age, higher stroke risk, once-daily regimen, history of hypertension, diabetes or stroke, concomitant cardiovascular medications, living in rural areas and being an experienced OAC user. Non-adherent patients were more likely to experience stroke and death, and incurred higher medical costs compared with patients with poor adherence.
Conclusions Our findings show that up to 30% of patients with AF are non-adherent, suggesting an important therapeutic challenge in this patient population.
- stroke medicine
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Strengths and limitations of this study
This is a timely systematic review that synthesises the evidence on extent of poor adherence to oral anticoagulants, its determinants and clinical and economic outcomes, among patients with atrial fibrillation.
We focused on observational studies (retrospective and prospective) to synthesise the evidence on patients’ real-world medication taking behaviour.
We considered all oral anticoagulants, including the newer drugs (apixaban, rivaroxaban, dabigatran and edoxaban) and aimed to generate pooled adherence at the individual drug level.
Drug utilisation consists of three interconnected but distinct phases (initiation, implementation and discontinuation) and the focus of this study was confined to the implementation phase.
Atrial fibrillation (AF)—the most common chronic arrhythmia—is an epidemic affecting more than 33 million people worldwide.1 AF increases stroke risk by up to fivefold and is responsible for a third of strokes in people over 60.2–5 Strokes secondary to AF are far more debilitating and carry three times the risk of death than strokes due to other causes.6–8
Oral anticoagulants (OACs), which include vitamin K antagonists (VKA) and direct oral anticoagulants (DOACs), are the only effective agents thus far in preventing stroke in patient with AF, showing approximately 66% relative risk reduction in clinical trials.9–13 When used outside the controlled environment of clinical trials, however, the effectiveness of these drugs is impacted by patients’ adherence.14 15 The clinical consequences of non-adherence can potentially be more significant for DOACs, given their short half-lives.14–18
Studies have previously attempted to summarise the medication taking behaviour of patients with AF. These reviews, however, focus on discontinuation of therapy (not implementation or execution of dosing), or when looking at implementation, only focus on DOACs, summarise evidence from randomised controlled trials (which do not reflect the day to day behaviours of patients) and provide a narrative summary of results with no meta-analysis.19–21 Further, no studies have summarised the evidence on determinants of adherence in this patient population and the association between adherence and outcomes (clinical or economical). The objective of this systematic review and meta-analysis was to summarise the evidence from observational studies on the extent, determinants and impacts of adherence to all OACs among patients with AF.
We conducted a systematic review and meta-analysis following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses and the Meta-analysis Of Observational Studies in Epidemiology guidelines (online supplementary file 1).22 23
In March 2019, we systematically searched PubMed/Medline, Embase, CINAHL and PsycINFO (from inception) using the relevant keywords and MeSH terms (online supplementary file 2). The search strategy was designed with the help of a medical librarian and aimed to identify peer-reviewed published manuscripts that reported on extent, determinants and impacts of non-adherence to any OAC. A manual search was also performed on Google Scholar and the bibliography of included studies.
Inclusion criteria and study selection
Studies were included if they used a prospective or retrospective observational study design, and quantitatively measured secondary adherence (also known as the ‘implementation’ phase), which looks at medication dose omissions, additions or delays and does not involve those who did not initiate their therapy.15 Studies published in English, French, Spanish, Persian, Finnish, Cantonese or Korean were included.24 No limitations were imposed on setting, country, publication date or quality.
While we were primarily interested in OAC adherence in patients with non-valvular AF, we included studies that did not specifically restrict inclusion to this population, with notation in quality assessment. Studies of self-reported adherence were excluded (including those using validated scales such as Morisky Medication Adherence Scale©) as they are prone to overestimation of adherence (social desirability bias).24 Cross-sectional and interventional studies, editorials, conference proceedings and studies that evaluated or validated adherence measurement methods were also excluded.
Two authors independently screened titles and abstracts of the retrieved studies followed by full text review of candidate studies. Disagreements about inclusion were resolved by discussion with a third author.
Data extraction and synthesis
The primary adherence measure extracted was the mean and SD of patients’ adherence over 6 months or 12 months post index date (after therapy initiation). The secondary adherence measure was proportions of adherent patients, that is proportion of patients reported in each study to have mean adherence score more than 80 (this could be > or ≥ depending on how the study defined ‘adherent’). The 80% adherence is the conventional threshold for ‘good adherence’.25 26 Six or twelve months were chosen as these were the most common follow-up times. If a study had variable follow-up time (eg, from initiation to permanent discontinuation or death), the median follow-up time was used. For studies that reported the proportion of non-adherent participants, data were transformed to proportion adherent to allow pooling. When both unadjusted and adjusted outcomes were reported, we extracted and analysed the adjusted results. When unmatched and propensity score matched results were reported, we extracted the matched results as they were expected to be more accurate estimates. When a study reported adherence to both index OAC and current OAC (allowing for switching), adherence to index OAC was analysed to minimise heterogeneity since studies defined switching differently. Adherence results with switching allowed were still reported.
We extracted information on the determinants or factors shown in the included studies to be independently associated with adherence in multivariable regression analyses. We classified the identified determinants under the WHO’s five dimensions of medication adherence to identify areas in need of more research.27 Finally, we extracted information on the clinical and economic consequences of poor adherence.
Meta-analyses were carried out using DerSimonian and Laird random-effects models to determine the pooled mean adherence and the corresponding pooled proportion of adherent patients over 6 months and 1 year of observation. If a study reported adherence scores for multiple cohorts, all were included in the meta-analysis (multiple entries per study). In anticipation of heterogeneity, subgroup analysis was performed for each adherence measure, by presence of potential conflict of interest and study quality. Additional meta-analyses were also performed focusing only on studies that reported comparative adherence between different OACs in the same cohort, to calculate the pooled OR of adherence for each comparison.
I2 statistics was used to quantify heterogeneity between studies.28 Leave-one-out analysis was also performed for outliers to explore and potentially reduce heterogeneity.29 Forest plots and funnel plots were constructed using OpenMeta-Analyst (Microsoft Corporation, Redmond, Washington, USA) or RevMan5 (V. 5.3, Copenhagen, Denmark) software to illustrate the results and assess publication bias using funnel plots where relevant, that is, where studies reported measures of association (eg, OR).30 31 Clinical and economic impacts of poor adherence were summarised narratively as meta-analysis was not possible.
We critically appraised the quality of adherence measurement in the included studies by adapting a condensed version of the checklist designed by the International Society of Pharmaco-economics and Outcomes Research (ISPOR) Group, designed specifically for medication adherence studies, to establish standards for data sources, operational definitions, measurement of medication adherence and reporting of results, previously used in a systematic reviews of adherence to gout medication.32 We also critically appraised individual study reporting quality using Strengthening the Reporting of Observational Studies in Epidemiology.33 Studies received a point for each checklist item they met and a 0 score if not met. A quality score was computed for each study (number of items satisfactorily met/the total number of applicable items) and reported as a percentage. Items deemed not applicable were excluded from the denominator of the study's score. Studies were categorised as low, moderate or high quality if they scored ≤50%, 51%–80% or >80%, respectively (arbitrary thresholds defined by authors).
Following Cochrane’s commercial sponsorship policy as a guide, potential conflicts of interest were deemed present if any of the following were met: (1) provision of study funding by the for-profit manufacturer or marketer of any of the OACs included in the corresponding study or (2) disclosure of potential conflict of interest with a for-profit manufacturer or marketer of any of the OACs included in the corresponding study.34
Patient and public involvement
Patients or the public were not involved in the design, or conduct, or reporting, or dissemination of our research.
Initial search led to 1122 studies, all of which were in English (figure 1). A total of 30 studies were included in this systematic review35–64 involving 593 683 participants (DOAC: 437 610, VKA: 156 073). Most studies were published after 2015 (n=22, 73% of total included), conducted in North America (n=19, 63%) and retrospective (n=29, 97%) (table 1). Adherence measurement was assessed to be of high quality in 59% of the included studies and moderate in 38%, according to the ISPOR checklist (online supplementary file 3). The most frequently reported adherence measures were proportion days covered (PDC) (n=21, 70% of the included studies) and medication possession ratio (MPR) (n=9, 20%) over 6 months or 1 year post index date (table 2). The majority of the included studies focused on adherence to DOACs with only four observational studies measuring and reporting adherence to warfarin. There were no data on adherence to edoxaban, betrixaban, phenprocoumon, acenocoumarol or fluindione.
The range of reported adherence results was quite wide. Reported mean adherence ranged between 67 (out of 100)58 61 64 and 8655 over 6 months and 5758 and 8641 over 1 year post index date, with corresponding reported proportion of adherent patients ranging between 47%59 and 82%56 over 6 months and 41%58 and 95%45 over 1 year. A wide range of adherence results were observed even at the individual OAC level (table 2).
Pooled mean adherence scores over 6 month and 1 year post medication initiation were 77 (95% CI: 74–79) and 74 (68–79), with the corresponding pooled proportion of adherent patients as 63% (58%–68%) and 70% (65%–76%), respectively. Adherence was similar between DOACs, although adherence to apixaban and rivaroxaban was slightly higher than dabigatran (table 3). No meta-analysis could be conducted for mean adherence to warfarin since this was not reported by the included studies. Pooled estimates of proportion of adherent patients for warfarin resulted from meta-analysis of two studies only (as illustrated in tables 2 and 3). Due to the limited data in warfarin, no drug class comparison could be made. Figure 2 illustrates the forest plots for patients’ mean adherence score over 6 months and 1 year. The remaining forests plots, including forest plots of proportion adherent, adherence to individual OACs, subgroup analyses (by adherence measure (PDC and MPR), study quality and potential for conflict of interest) can be found in online supplementary file 4.
Between-study variance (represented as I2) was high and not reduced by the leave-one out analysis or subgroup analysis. Exclusion of studies with potential conflicts of interest led to lower adherence scores for all OACs but did not change the rank-order of OACs (adherence to dabigatran remained lower than the others). Excluding studies of low and moderate quality or stratifying the analysis by adherence measure (PDC vs MPR) or country (USA vs others) had only minor impacts on pooled adherence results and the detected heterogeneity (online supplementary file 4).
Studies comparing adherence between different OACs in the same cohort
Nineteen studies reported comparative adherence between different OACs in the same cohort (table 4).35–37 39–45 49 50 52 55–58 60 62 Odds of being adherent was significantly higher for apixaban compared with dabigatran over both 6 months (OR:1.24, 95% CI: 1.07–1.45) and 1 year post index date (OR:1.76, 95% CI: 1.35–2.29). Odds of adherence was significantly higher for rivaroxaban compared with dabigatran over 6 months (OR:1.39, 95% CI: 1.15–1.67), but not 1 year (OR:1.17, 95% CI: 0.38–3.60). Odds of adherence did not differ between apixaban and rivaroxaban over 6 months (OR:0.80, 95% CI: 0.51–1.24) or 1 year (OR:1.02, 95% CI: 0.79–1.33).
Studies reporting adherence among several cohorts with different characteristics
Three studies compared adherence between new versus experienced users.37 50 56 McHorney et al reported greater mean PDC score for both rivaroxaban and apixaban (0.90 and 0.88, respectively) among prior OAC users compared with naïve users (0.87 and 0.86, respectively).56 Borne et al reported a higher mean PDC score for apixaban users with prior warfarin experience compared with naïve users (0.89±0.14 vs naïve: 0.87±0.15, p<0.01).37 Confirming these results, Manzoor et al reported higher mean PDC for experienced users compared with naïve users over 6 months (83.3±24.6 vs 72.3±31.3; p<0.05), 9 months (81.2±26.4 vs 67.3±33.8); p<0.05) and 1 year (79.9±27.6 vs 63.7±35.2; p<0.05).50
One study, Eapen et al, compared adherence among those prescribed OAC at discharge versus after discharge and reported that patients prescribed warfarin at discharge had significantly higher prescription fill rates compared with those prescribed after discharge at 3 months (84.5% vs 12.3%; p<0.001) and 1 year (91.6% vs 16.8%; p<0.001).44
Determinants of adherence
Many factors were identified by the included studies as significant determinants of adherence. Summarising these under WHO’s classification, the factors identified in the included studies to be significantly and positively associated with adherence were: Patient factors: history of hypertension,43 49 diabetes,37 stroke37 52; Regimen factors: once daily dosing,35 49 concomitant use of statin,43 52 ACE inhibitor or angiotensin II receptor blockers,43 52 higher risk of bleeding43; and Social/economic factors: living in rural or deprived areas.52 53 Factors found to be significantly and negatively associated with adherence to OAC were: being a naïve OAC user,50 56 twice daily dosing35 49 and impaired cognitive or functional ability.56 No healthcare system and condition factors related predictors of adherence were identified.
Conflicting results were reported for female sex,47 48 53 age,37 43 47–50 52 53 risk of stroke,43 47 53 presence of multiple comorbidities43 50 51 56 and higher number of concomitant medications.50 51 These factors were found to be predictors of high and low OAC adherence in different studies.
Impacts of adherence
Four studies assessed the clinical impact of adherence.35 37 42 59 Alberts et al reported 50% increased hazard of ischaemic stroke with DOAC non-adherence (aHR:1.50, 95% CI:1.30–1.73).35 Deshpande et al reported non-adherent patients to be 1.82 times (aHR:1.82, 95% CI: 1.24–2.67; p=0.002) and 2.08 times (aHR:2.08, 95% CI: 1.11–3.89; p=0.02) more likely to experience an ischaemic stroke compared with adherent patients, over 6 and 12 months, respectively.42 Similarly, Borne et al reported a higher risk of death or stroke per 0.1 drop in the PDC among dabigatran users (HR:1.07, 95% CI: 1.03–1.12; p<0.01).37 Shore et al reported a 13% increase in risk of combined all-cause mortality and stroke with lower adherence (aHR:1.13, 95% CI: 1.07–1.19 per 10% decrease in PDC) but found no association between adherence and non-fatal bleeding events (aHR:1.04 per 10% increase in PDC, 95% CI: 0.94–1.14) or myocardial infarction (aHR:0.97 per 10% increase in PDC, 95% CI: 0.78–1.21).59
Two studies measured the economic impacts of adherence.38 43 Casciano et al reported significantly more inpatient and emergency room encounters and longer length of stay for non-adherent patients compared to adherent patients and Deshpande et al reported significantly higher annual adjusted per-patient medical cost (inpatient and outpatient) for non-adherent users compared with adherent ones (US$30 485 vs $23 544; p≤0.001).38 43
In this systematic review, we synthesised observational data of over half a million patients with AF to reveal that up to 30% are non-adherent to OACs, and that non-adherent patients are more likely to experience stroke, death and incur higher medical costs compared with adherent patients. We also found that older age, higher stroke risk, once-daily regimen, history of hypertension, diabetes or stroke, concomitant cardiovascular medications, living in rural areas and being an experienced OAC user could be associated with better adherence.
Adherence to OACs among patients with AF has been thoroughly studied in developed countries. In our study, pooled proportion of adherent patients at 6 months and 1 year were 63% and 70%, respectively, which are higher than those found for other chronic cardiovascular medications such as statins (54%) and antihypertensives (59%).65 However, our finding that up to 37% of patients with AF do not adhere to OACs is concerning considering the detrimental consequences of non-adherence in this particular clinical context. We were unable to ascertain whether the conveniences of DOACs translates into better adherence compared with warfarin due to lack of adherence data on warfarin, a likely result of warfarin dose variations complicating MPR and PDC ascertainment from administrative data. Between DOACs, however, adherence was found to be similar, although dabigatran appeared to have slightly lower adherence than apixaban and rivaroxaban.
Many patient-related-related, regimen-related and social/economic-related factors were identified by the included studies as significant determinants of adherence. It should be noted that each of these factors were reported to have a significant impact on adherence by one or two studies. The limited number of prospective observational studies on the topic restricted our ability to identify important psychosocial determinants as administrative data fall short in recording patients’ knowledge gaps, misconceptions and varying values and preferences, all of which have frequently been reported in patients with AF.66–71 Further, questions remain about the role of sex, age, risk of stroke, presence of multiple comorbidities and number of concomitant medications on adherence. One explanation for the inconsistencies we observed could be differences in how these factors were defined in our included studies. A 2019 systematic review of 34 systematic reviews on determinants of adherence to cardiovascular medications (beta blockers, calcium channel blockers, ACE inhibitors, angiotensin II receptor blockers and diuretics) also reported inconsistent results for the role of gender in adherence.72 These authors also found that the effects of concomitant medications and comorbidities seem to be drug-specific and condition-specific, which could explain some of the interstudy variability with this factor.72 A multivariate patient-level meta-regression analysis could provide more clarity to these issues with OACs in patients with AF. Nevertheless, our findings indicate potential opportunities for interventions such as education and counselling for younger or newly diagnosed patients (naïve users) and adherence support for those on twice daily dosed OACs.
Lastly, we looked at outcomes of poor adherence. Our review found evidence of association between lower adherence and strokes, mortality, healthcare utilisation and costs. Our findings confirm the results of a 2017 systematic review of 79 studies across 14 disease groups which reported that $3347–$19 472 are attributed to non-adherence per patient per year among those with cardiovascular conditions (hypertension, hypercholesterolaemia and chronic heart failure).73 Our findings in relation to clinical outcomes are in line with results of meta-analyses of a large body of research showing that poor adherence across a range of conditions was associated with a 26% increased risk of poor treatment outcomes.74 The adherence–outcome relationship is, however, very complex, and dependant on many factors, including the nature of the disease.74 This is why it was important to summarise the strength of this relationship specifically in AF. Our findings, while based on only four studies, reveal the relationship between lower adherence and poor clinical outcomes in patients with AF, and support the potential of interventions aimed at increasing adherence in patients with AF.73–79
This review was primarily limited by gaps in the available evidence. Given our interest in observational data, our evidence was narrowed to developed countries where the technology and infrastructure for systematic collection of such data is available. The high number of studies from a few developed countries introduced the possibility of duplicate patients in the analysis since many of the included studies used the same database with overlapping periods.35 38–40 50 64 Furthermore, there may be potential for publication bias or under-representation from studies from developing countries. As described in the Methods section, we attempted to assess publication bias using funnel plots but were limited with few studies reporting measures of association. Nonetheless, for these meta-analyses, findings do not suggest presence of publication bias (online supplementary file 3).
Another limitation of our analysis was the high heterogeneity (I2 >80%) among the studies. Possible sources of heterogeneity include differences in patient inclusion criteria (eg, OAC naïve vs experienced); methods for handling and defining medication switches, stockpiling, refill gaps and hospitalisation dates; fixed versus variable observational periods and adherence measure used (PDC vs MPR). Subgroup analyses did not affect the amount of statistical heterogeneity detected. Nonetheless, in addition to the summary measures derived from meta-analysis, we were able to detect the range of adherence measures from the included studies. Finally, drug utilisation consists of initiation, implementation and discontinuation,15 80 and the focus of this study was confined to the implementation phase. Systematic reviews of OAC initiation and discontinuation are needed to provide a complete picture of medication taking behaviour in patients with AF.
Our understanding of the comparative adherence between warfarin and DOACs among patients with AF is currently impeded by lack of observational data on warfarin. Sophisticated statistical models are needed to calculate days’ supply of warfarin, despite its varying dose, to allow measurement of MPR or PDC for this drug using administrative data. Furthermore, we lack information on patterns of non-adherence to OACs. All of the current studies have treated adherence as a static behaviour, calculating and reporting it using a single summary measure. This methodological approach does not provide a complete picture of adherence, which is a dynamic behaviour that changes over time.25 81 Characterisation of adherence patterns over time is vital in understanding the problem of poor adherence and targeting the right patients at the right time with the right interventions.82–86
There is a need for more research investigating the clinical and economic consequences of poor adherence as the current evidence is limited to findings of four studies. Moreover, a clinically meaningful OAC adherence threshold has yet to be determined in AF.35 37 42 59 While the association between taking more than 80% of medications and improved clinical outcomes has been shown in four AF studies, it remains unclear if this is the optimal threshold for AF.35 37 42 59 Clinically relevant adherence cut-off values have been shown to differ widely (from 58% to 85%) in different diseases, and even among drug classes.14 87 As with antiretroviral medications, given the detrimental consequences of OAC non-adherence, the clinically meaningful threshold for ‘good adherence’ to OACs may need to be much higher than 80%.87
Synthesis of observational data suggests that overall OAC adherence in patients with AF is below the conventional threshold of ‘adherent’ (80%). These findings, combined with evidence that lower adherence is associated with poor clinical outcomes and higher costs, suggest an important therapeutic challenge in this patient population. Our study also highlights the need for more consistent measures of adherence, and more research to characterise patterns of OAC non-adherence, identifying determinants of poor OAC adherence and investigate the clinical and economic consequences of OAC non-adherence.
Contributors Conceived the study: SS, PSL, MADV; Designed the search strategy: SS, MADV, PSL; Conducted the literature search: SS; Screened titles and abstracts: SS, RT; Screened full texts: SS, RT; Extracted data: SS, RT; Made methodological decisions (data synthesis and analysis): MADV, SS; Analysed the data: SS; Conducted quality assessment; SS, RT; Interpreted the results: SS, PSL, JA, MADV; Prepared the manuscript first draft: SS, MDV, PSL, RT; Reviewed the manuscript and provided critical feedback: JA, MADV, PSL; Revised the manuscript: SS, PSL, RT, MADV.
Funding PSL’s research is partially supported by the UBC David H MacDonald Professorship in Clinical Pharmacy. MADV holds a Canada Research Chair in Medication Adherence, Utilisation and Outcomes and is a Michael Smith Foundation for Health Research Scholar.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Patient consent for publication Not required.
Ethics approval Ethical approval for this study was not required per our institution’s policies.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement No data are available. No data are available.