Studies that compared patients using VKAs and DOACs (activated factor Xa inhibitors and factor IIa inhibitors) with ≥ 18 years and diagnosed with non-valvular AF were eligible. Studies that included valvular AF, or did not report their exclusion, were excluded. DOACs are contraindicated for patients with moderate to severe mitral stenosis or mechanical prostatic heart valve (valvular AF); in these cases, only VKAs are recommended.^{2 , 4}

Medline, Embase, LILACS, SciELO, CINAHL, and Cochrane Central Register of Controlled Trials databases were searched until June 9, 2021. The bibliography of selected studies was reviewed for retrieving further references.

The complete search strategies are presented in Chart S1. Outcome-related terms were not included to increase search sensitivity. No language or publication date restrictions were applied.

After removing duplicates, four peer reviewers independently analyzed titles and abstracts and identified eligible studies. The discrepancies were resolved by a third reviewer.

Data extraction was performed independently by three peer reviewers. The discrepancies were resolved by a fourth reviewer.

The following information was extracted: publication data, design, location, study period, study population and demographics of participants, sample size by exposure group, anticoagulants evaluated, the method used to measure outcomes, time of follow-up, and outcome measures results. Missing information was requested from the authors by email.

The evaluated outcomes were adherence and QoL. Eighty percent (80%) adhesion was considered the conventional threshold for ‘good adherence.’^{3 , 16} When studies reported crude and adjusted results, adjusted results were extracted.

We presented adherence as a percentage or mean of the adherence score with the standard deviation and QoL as the mean, median, or mean difference of the QoL score in absolute number or percentage, with the respective standard deviation or confidence interval (CI).

Two reviewers independently assessed the methodological quality of the studies included using the Joanna Briggs Institute (JBI) critical appraisal tools.¹⁷ There are four possible answers for each criterion: ‘yes,’ ‘no,’ ‘not clear,’ or ‘not applicable.’ Discrepancies were resolved by consensus.

Considering the checklists do not propose scores to classify the methodological quality, we assigned one point to the items evaluated with 'yes' and zero to the items evaluated as ‘no,’ ‘it is unclear’ and ‘not applicable.’

Descriptive statistics were used to present the data by the outcome. We synthesized evidence narratively. Meta-analyses could not be performed due to the high methodological heterogeneity (especially interventions, settings, study designs, and outcome measures).

A total of 18,271 abstracts were retrieved, of which 71 were inspected for eligibility criteria through the full article, with 24 studies included in the review. Figure 1 shows the selection process. The excluded studies and reasons are presented in Table S1.

Figure 1

Treatment adherence was assessed in 16 studies, including 122,458 patients with non-valvular AF (DOACs 56,749, VKAs 65,709). QoL was measured in 11 studies, including 5,687 (DOACs 3,329, VKAs 2,358). Three out 24 studies evaluated both outcomes simultaneously.¹⁸

More than half of adherence^{18 , 19 , 21} and QoL^{18 , 19 , 28} studies were carried out in Europe. All studies were published between 2012 and 2021, of which 79.2% (n = 19) were published in the last five years. All the studies included were published in English, except three, which were published in Spanish,¹⁹ Portuguese²⁵ and French.²⁶

The detailed risk of bias assessment is presented in Tables S2-S4. For cross-sectional studies that measured adherence (n = 8), four^{25 , 32} had low methodological quality. The main criteria that contributed to reducing the quality were lack of clarity in the definition of the sample inclusion criteria, lack of identification and strategies to deal with confounding, and adequate statistical analysis.

Five of the cohort studies (n = 8) had moderate¹⁹ quality. The factors that contributed the most to the reduction in the methodological quality were differences between groups, incomplete follow-up, and strategies for dealing with incomplete follow-up.

All cohort studies that assessed QoL had moderate methodological quality (n = 4).^{19 , 20 , 36 , 37} Of the cross-sectional studies (n = 6), the majority (n = 4) had low quality.^{18 , 29 , 38 , 39} The criteria that contributed to the reduction in the methodological quality were the absence of identification and strategies to deal with confounding and inadequate statistical analysis.

The randomized clinical trial (RCT) was of moderate quality,²⁸ as it was not clear about the allocation concealment and blinding.

We categorized the studies according to the measurement methods and characteristics (Table S5). The following indirect methods were found: questionnaire (n = 8); administrative database (n = 5); reporting of missed doses (n = 2); and electronic monitoring of drug administration (MEMS) (n = 1). All studies that used the administrative dispensing method assessed the Proportion of Days Covered (PDC) formula to calculate adherence, and established a threshold of 80% for a compliant patient. The proportion of adherent participants in studies that used PDC²² to assess adherence ranged from 42-81% in the DOACs group and 51-78% in the VKAs.

The Morisky Medication Adherence Scale (MMAS-8) was used in four studies,^{18 , 20 , 26 , 34} followed by the Morisky-Green Scale (n = 2).^{19 , 27} The short-form Adherence to Refills and Medications Scale (ARMS) was applied in one study.⁴¹ Others adopted both the Brief Medication Questionnaire (BMQ) and the Measurement of Treatment Adherence (MAT).²⁵

For questionnaires, six studies^{18 , 19 , 25} reported the outcome as the proportion of adherents, with the percentage of adherents of 76.3% (1103/1446) in the DOACs group as compared to 74.3% (823/1107) in the VKAs group.

Two studies assessed adherence by the number of doses missed.^{32 , 33} One of the studies³² found a less mean number of missed doses in the warfarin group (0.2) than dabigatran users (0.4) (p = 0.007). The other study did not find any difference between groups.³³

The only study that used MEMS found a more significant difference in the proportion of adherents in patients using VKAs (97.9%) as compared to DOACs users (95.8%) (p = 0.003).²¹

The 12 studies that measured the proportion of adherent patients^{18 , 19 , 21} showed an overall adherence rate of 66.8%. In the DOACs group, the rate was 71.6%, and in the VKAs, 62.8%. (a percentage obtained through the ratio between the number of adherent patients in each anticoagulant class and the total number of patients included in the studies). Adherence results varied between anticoagulants, with a significant difference in five studies,^{19 , 21 , 25 , 32 , 40} with three favorable to VKAs.^{21 , 25 , 32} Among studies that showed better adherence in the VKAs group, different measurement methods were used (MEMS,²¹ missed doses³² and questionnaires²⁵ ).

Table S6 presents the characteristics and results of the studies analyzed. Three studies adopted specific^{19 , 20 , 38} questionnaires, while eight^{18 , 28} used generic questionnaires.

Of the studies with specific instruments, two^{20 , 38} used the Duke Anticoagulation Satisfaction Scale (DASS). In both studies, total mean scores in the three domains evaluated (limitations, inconvenience and psychological impact) were lower for users of DOACs, indicating better QoL than VKAs. However, only Cabbar et al. demonstrated a significant difference between groups (DOACs 75.19±18.52 vs VKAs 90.12±17.28; p < 0.001).³⁸

One study used the AF QoL instrument (AF-QoL 18) and demonstrated better QoL in patients using dabigatran as compared to VKAs (50.6 vs. 42.8; p < 0.001).²⁰

Modifications of the Medical Outcomes Study Form 36 (SF-36) were used in four studies.²⁸ Ng et al .³⁹ demonstrated better measures in the eight domains evaluated for patients using VKAs, without statistical significance.

Two studies^{29 , 30} showed better health conditions for DOACs when compared to VKAs. Apsite et al. ²⁹ demonstrated a favorable difference to DOACs in the physical functioning (p = 0.012) and social (p = 0.026) domains. At the same time, Cabbar et al. found a relevant difference between groups in all domains (p < 0.05).³⁸ van Miert et al. ²⁸ showed no significant difference between groups in the SF-36 domains, comparing the mean difference in the score at baseline and after one year of follow-up.

Four studies^{18 , 31 , 36 , 37} evaluated the outcome by the EuroQol 5-dimension questionnaire (EQ-5D). Two studies^{18 , 31} used the EQ-5D questionnaire with three response levels (EQ5D-3L), showing better health status for patients using DOACs, without statistical significance (p = 0.29).

For measurement of the utility score, two studies adopted EQ5D comparing dabigatran to warfarin. Ho et al. ³⁶ found a utility score of 0.77±0.17 for the dabigatran group and 0.74±0.16 for VKAs. Monz et al. ³⁷ found a utility score of 0.77 (IC 95% 0.75- 0.80) for dabigatran and 0.78 (IC 95% 0.76- 0.81) for VKAs. Both studies did not find statistical significance.

Three studies^{18 , 36 , 37} reported data from the visual analog scale (VAS); the mean score reported by patients in the VKAs group ranged from 65.0 to 73.9, and from 67.1 to 72.8 in the DOACs group. The difference between groups was not significant in any of the studies.