Real-World Comparisons of Low-Dose NOACs versus Standard-Dose NOACs or Warfarin on Efficacy and Safety in Patients with AF: A Meta-Analysis

Objective We aimed to further investigate the efficacy and safety of low-dose NOACs by performing a meta-analysis of cohort studies. Background Meta-analyses of randomized controlled trials (RCTs) have demonstrated that low-dose non-vitamin K antagonist oral anticoagulants (NOACs) showed inferior efficacy compared with standard-dose NOACs, although they are still frequently prescribed for patients with atrial fibrillation (AF) in the clinical practice. Methods Cochrane Central Register of Controlled Trials (CENTRAL), Embase, and MEDLINE were systematically searched from the inception to September 9, 2021, for cohort studies that compared the efficacy and/or safety of low-dose NOACs in patients with AF. The primary outcomes were ischemic stroke and major bleeding, and the secondary outcomes were mortality, intracranial hemorrhage (ICH), and gastrointestinal hemorrhage (GH). Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated with the random-effect model. Results Twenty-five publications involving 487856 patients with AF were included. Compared with standard-dose NOACs, low-dose NOACs had comparable risks of ischemic stroke (HR = 1.03, 95% CI 0.96 to 1.11), major bleeding (HR = 1.12, 95% CI 0.97 to 1.28), ICH (HR = 1.09, 95% CI 0.88 to 1.36), and GH (HR = 1.11, 95% CI 0.92 to 1.33), except for a higher risk of mortality (HR = 1.41, 95% CI 1.21 to 1.65). Compared with warfarin, low-dose NOACs were associated with lower risks of ischemic stroke (HR = 0.72, 95% CI .67 to 0.78), mortality (HR = 0.67, 95% CI 0.59 to 0.77), major bleeding (HR = 0.64, 95% CI 0.53 to 0.79), ICH (HR = 0.57, 95% CI 0.42 to 0.77), and GH (HR = 0.78, 95% CI 0.64 to 0.95). Conclusions Low-dose NOACs were comparable to standard-dose NOACs considering risks of ischemic stroke, major bleeding, ICH, and GH, and they were superior to warfarin. Low-dose NOACs might be prescribed effectively and safely for patients with AF. Considering limitations, further well-designed prospective studies are foreseen.


Introduction
Atrial fibrillation (AF), known as a common cardiac arrhythmia worldwide, can cause ischemic stroke and seriously jeopardize the health of global elder patients [1]. For decades, warfarin was prescribed to prevent ischemic stroke from AF by decreasing the production of several clotting proteins that rely on vitamin K [2]. However, the adherence to warfarin is severely affected by the frequent international normalized ratio (INR) monitoring, drugdrug and drug-food interactions [3]. In recent years, the approval of non-vitamin K antagonist oral anticoagulants (NOACs), which directly inhibit the critical factors of the coagulation cascade, provided new anticoagulant strategies for the patients with AF.
Meta-analyses of RCTs assessed the efficacy and safety of standard-dose NOACs, low-dose NOACs, and warfarin in patients with AF. Moreover, the results revealed that lowdose NOACs were inferior to standard-dose NOACs in the efficacy with a higher risk of ischemic stroke and had no superior efficacy and safety than warfarin [4,5]; standarddose NOACs were superior to warfarin in the efficacy and safety with less ischemic stroke, mortality, and ICH [4][5][6]. However, low-dose NOACs are still frequently prescribed for patients with AF. Low-dose NOACs were prescribed for 31%, 19%, and 29% of patients in Korea [7], France [8], and America [9], respectively. RCTs were performed under optimized conditions, strict inclusion and exclusion criteria, which might not reflect real-world conditions. Moreover, RCTs enroll a small, nonrepresentative subset of patients and overlook the important interactions between the patients and the real world, which can affect treatment outcomes [10]. Furthermore, medication adherence, the key point for treatment effectiveness, is closely monitored in RCTs, which is not always the case in clinical practice [10]. Real-world cohort studies, which enroll patients of broad-spectrum baseline characteristics, may provide a comprehensive picture of the clinical practice. erefore, we aimed to further investigate the efficacy and safety of low-dose NOACs by conducting a meta-analysis of all relevant cohort studies.

Methods
is meta-analysis was prepared according to the PRISMA (Preferred Reporting Items for Systemic Reviews and Metaanalysis) and MOOSE (Meta-Analysis of Observational Studies in Epidemiology) guidelines [11,12].

Search Strategy and Study Selection. Cochrane Central
Register of Controlled Trials (CENTRAL) (from inception to September 9, 2021), MEDLINE (from inception to September 9, 2021), and Embase (from inception to September 9, 2021) were systematically searched. Details of the search strategy are illustrated in Supplementary Tables S1-S3. We developed inclusion criteria for this meta-analysis prospectively.
e criteria of studies screening were as follows: (1) the target population was patients with AF; (2) studies involved lose-dose NOACs and standard-dose NOACs or warfarin; (3) studies included efficacy (ischemic stroke and mortality) or safety outcomes (major bleeding, intracranial hemorrhage, and gastrointestinal hemorrhage); (4) the study type was the cohort. AF patients with valvular heart disease (VHD) or receiving NOACs after catheter ablation and studies published in the forms of conference abstracts, letters, or protocols were excluded. In addition, for the same data source or overlapping data reported in more than one study, only the most comprehensive data with the longest follow-up period was included. References of included studies and relevant meta-analyses were screened for additional eligible studies as well.

Data Extraction and Quality Assessment.
e primary efficacy outcome was ischemic stroke, and the secondary efficacy outcome was mortality (all-cause mortality). e primary safety outcome was major bleeding, defined as fatal bleeding or bleeding in a critical site, and the secondary safety outcomes were intracranial hemorrhage (ICH) and gastrointestinal hemorrhage (GH).
Two reviewers (Ze Li and Xiaozhen Wang) independently screened titles and abstracts of the retrieved studies to exclude those that did not explore questions of interest and then independently screened full texts of the remaining studies to identify those that met all the inclusion criteria. We manually checked the reference list of each acquired article for relevant studies. For each included study, two reviewers independently extracted the characteristics of the included studies and patients, as well as outcomes as predefined. Discrepancies were resolved by discussing with the third reviewer (Aiping Wen).
Bias risks were assessed with the Newcastle-Ottawa quality assessment scale [21]. e publication bias was quantitatively assessed by the Begg's [22] and Egger's tests [23], P < 0.05 was taken as statistically significant. Two reviewers (Ze Li and Xiaozhen Wang) assessed risks of bias independently and in duplicate. Any disagreements were resolved in consultation with the supervisor (Aiping Wen).

Data Synthesis and Statistical Analysis.
Intention to treat analysis (ITT) results were used wherever possible. If ITT results were not available, we used the data the author reported. All analyses were performed by Stata 16.0. Hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) were estimated with the random-effect model. e heterogeneity among studies was assessed by I 2 with <25%, 25-50%, and >50% indicating low, moderate, and high degree of heterogeneity, respectively. Meta-regression analyses were performed to examine possible sources of heterogeneity in the data.
To explore the influence for different regions of patients, subgroup meta-analyses were performed by stratifying patients into Asia and non-Asia. Most cohort studies used the propensity score matching (PSM) method or multivariable model (MM) to balance the confounding factors between groups and minimize the heterogeneity, so we enrolled adjusted cohort studies to perform adjusted subgroup metaanalyses. For all comparisons in this meta-analysis, P < 0.05 was taken as statistically significant.

Studies Identification and Characteristics.
A total of 2846 publications were identified through database search ( Figure 1). After the study screening process, twenty-five cohort studies meeting the inclusion criteria were included.
In general, there were 487856 patients in all enrolled studies. 238292 patients were included in the standard-dose group, including 115518 patients receiving NOACs and 122774 patients receiving warfarin, and 249564 patients were involved in the low-dose NOACs group. e baseline characteristics of included studies are shown in Table 1. e detailed previous medical history and group contents of included studies are illustrated in Supplementary Tables S4  and S5. Supplementary Tables S6 and S7. Overall, most cohort studies reported low risks of bias, while seven studies did not balance the confounding factors between groups, which had risks of comparability bias [8,25,30,31,[39][40][41]. ree studies did not report the length of follow-up [27,31,35], and most studies did not show the lost follow-up rate, which had risks of outcome bias. In addition, there was no publication bias for this meta-analysis by the Begg's and Egger's tests, except for the risk of ICH (P � 0.035, Egger's test) in the comparison of low-dose NOACs versus warfarin.
To minimize the heterogeneity and obtain more reliable results, adjusted subgroup meta-analyses including cohort studies with the PSM or MM method were performed. Results of all outcomes were consistent with the overall meta-analysis as well. Details of adjusted subgroup metaanalyses are illustrated in Supplementary Figures S6-S9.
For meta-regression analyses, no significant correlations were observed in most efficacy and safety outcomes. However, a significant correlation was found between major bleeding and mean age (P � 0.010), with HR increasing as the mean age of included patients ascended (Supplementary Figure S10); other significant predictors of HR were found between ICH, mean age (P � 0.046), and female (P � 0.035) as well, with HR increasing as the mean age (Supplementary Figure S11) or female percent of included patients ascended (Supplementary Figure S12). Details of meta-regression analyses are illustrated in Supplementary Table S8.
To balance the confounding factors, subgroup metaanalyses stratified by mean age (divided into older and younger groups by median) were performed, respectively. In general, all results were consistent with the overall metaanalysis. Details of subgroup meta-analyses are shown in Supplementary Figures S13 and S14.
Results of adjusted subgroup meta-analyses were consistent with the overall as well. Details of adjusted subgroup meta-analyses are illustrated in Supplementary Figures S20  and S21. For meta-regression analyses, no significant correlations were observed in efficacy and safety outcomes. Details of meta-regression analyses are illustrated in Supplementary Table S9.

Discussion
To the best of our knowledge, this is the first meta-analysis of cohort studies for low-dose NOACs versus standard-dose NOACs or warfarin in patients with AF. A few previous meta-analyses had tried to assess the efficacy and safety of    [4][5][6]. ese may not be generalizable to underrepresented patients, such as the patients with low weight, older age, or who were not yet represented in RCTs, so we performed this meta-analysis. Our meta-analysis revealed that compared with standard-dose NOACs, low-dose NOACs had comparable risks of ischemic stroke and bleeding (including major bleeding, ICH, and GH), except for a higher risk of mortality; compared with warfarin, low-dose NOACs showed lower risks of ischemic stroke, mortality, and bleeding. To assess the influence of different regions, we stratified the patients into the Asia subgroup and non-Asia subgroup. Results of subgroup meta-analyses were consistent with the overall except for the comparable risk of GH for the non-Asia subgroup in the comparison of low-dose NOACs versus warfarin.
We need to note that the baseline characteristics of cohort studies may be diverse compared to RCTs. Concerning some included studies, the mean or median ages of low-dose NOACs group were much older than standard-dose NOACs group, which led to the relatively lower CrCL, higher CHA 2 DS 2 -VASc and HAS-BLED scores [8,24,26,32,40,[42][43][44][45]. Moreover, there were some heterogeneities in the previous medical history, including hypertension, diabetes, heart failure, vascular disease, stroke/transient ischemic attack (TIA), and major bleeding. Due to the broad-spectrum baseline characteristics, most cohort studies used the PSM or MM method to adjust the data and minimize the heterogeneity. Adjusted subgroup metaanalyses including cohort studies with PSM or MM were performed as well, and the results were consistent with the overall meta-analysis.
Meta-regression analyses indicated that the mean age and female percent of included patients captured a very substantial portion of the heterogeneity in the data, so subgroup meta-analyses stratified by those were performed to balance the confounding factors. Similarly, the results were consistent with the overall as well. Nonetheless, considering the relatively few studies and ineluctable heterogeneity in this meta-analysis, further high qualified prospective studies are required to validate these results.
Most of our results were similar to the previous metaanalyses of RCTs. However, there were some conflicting results in our meta-analysis compared with RCTs, such as the comparable risk of ischemic stroke and higher risk of mortality in the comparison of standard-dose NOACs, and lower risks of ischemic stroke, major bleeding, and GH in the comparison of warfarin [4,5]. e difference in outcomes could be partially explained by several reasons: firstly, the patients' baselines of RCTs were narrow and nonrepresentative, with the approximate age of 70 years old, bodyweight of 66 kg, female percent of 40%, and CHADS 2 score of 2.0-3.0 [4,5]. ese might only address a special population of AF patients. On the contrary, cohort studies in this meta-analysis presented broad-spectrum baseline characteristics, with age ranging from 63.3 to 88.7 years old, BMI ranging from 23.1 to 31.7, CHA 2 DS 2 -VASc and HAS-BLED scores ranging from 1.9 to 5.2, 0.8 to 3.0, respectively. Compared with RCTs, cohort studies involved the individual of older age, lower body weight and CrCL, or higher CHA 2 DS 2 -VASc and HAS-BLED scores, who might be more susceptible to low-dose NOACs. Secondly, the adherence to standard-dose NOAC was about 60%, and more than one-third of patients with label NOAC prescription received a reduced low-dose [46]. As a result, some patients might be prescribed for standard-dose NOACs, whereas they take low-dose NOACs in reality. We believed that the above two reasons might contribute to the   noninferiority of low-dose NOACs versus standard-dose NOACs in the real world. irdly, the mean or median ages of low-dose NOACs were much older than standard-dose NOACs in nine studies [8,24,26,32,40,[42][43][44][45], and we considered this might explain the higher risk of mortality for low-dose NOACs. As another study showed, the older patients with AF were faced more comorbidities and death factors [47], which might eventually result in the higher risk of mortality. In addition, as it was not convenient to monitor the quality of warfarin routine usage, and many patients cannot reach the baseline requirement of time in therapeutic range (TTR) [48]; this might lead to the superiority of low-dose NOACs versus warfarin in clinical practice.
Warfarin showed some therapeutic limitations in the clinical practice, whose effect was widely affected by food and drugs, and patients need to monitor the INR frequently to supervise the efficacy and risk of major bleeding [49]. Major bleeding can seriously affect the anticoagulation treatment, such as higher risks of stroke and mortality [50], longer hospitalization [51], and more health care resource utilization [52]. At the same time, patients taking warfarin often had less time within the therapeutic range [48]. In this meta-analysis, low-dose NOACs were noninferior to standard-dose NOACs and superior to warfarin. us, considering the excellence and convenience, low-dose NOACs could be an effective and safe alternative to warfarin.

Limitations.
However, there were some potential limitations for our meta-analysis. Firstly, due to the limited number of included studies, we pooled all NOACs together even though rivaroxaban, apixaban, and edoxaban are the factor Xa inhibitors [53], while dabigatran is the thrombin inhibitor [54], which was conducted in another meta-analysis [55]. is may not cause significant bias, for they are all directacting oral anticoagulants inhibiting the critical factors in the coagulation cascade. Secondly, this meta-analysis might have some fundamental heterogeneity due to the nature of cohort studies, such as the mean age, CHA 2 DS 2 -VASc and HAS-BLED scores. However, most studies had used the PSM or MM method to adjust the data and minimize the heterogeneity. In addition, the results of adjusted subgroup metaanalyses including studies with PSM or MM were consistent with the overall as well. irdly, most included studies did not report the quality of TTR for warfarin. As the efficacy of warfarin was affected by the TTR, many patients cannot reach the baseline requirement of TTR [48], which might lead to the unexpected bias of low-dose NOACs versus warfarin. is limitation could be found in other meta-analyses involving warfarin [55,56]. However, the effectiveness of the treatment is ensured not only by effective and potent drugs, but also by patients' adherence to the therapy [57], and we should have various and comprehensive views of this limitation.

Conclusions
In general, for patients with AF, this meta-analysis of cohort studies demonstrated that low-dose NOACs were comparable to standard-dose NOACs considering risks of ischemic stroke, major bleeding, ICH, and GH, and they were superior to warfarin. us, low-dose NOACs might be prescribed effectively and safely for patients with AF. However, considering limitations, further well-designed prospective studies are required to validate these results.

Data Availability
All data generated or analyzed during this study are included in this published article and its supplementary information files.

Ethical Approval
is is a meta-analysis and needs no ethical committee approval.

Conflicts of Interest
e authors declare that there are no conflicts of interest.

Authors' Contributions
Ze Li was responsible for the study design, literature search, data collection, data analysis, data interpretation, drafting, and critical revision of the manuscript. Xiaozhen Wang and Dandan Li were responsible for the literature search and data collection. Aiping Wen was responsible for the study concept and design, data interpretation, critical revision of the manuscript, approval of the final submission, and took responsibility for the data's integrity and accuracy.