Dementia Risk of Direct Oral Anticoagulants Versus Warfarin for Atrial Fibrillation

Background Direct-acting oral anticoagulants (DOACs) have demonstrated superior efficacy in preventing stroke and death compared with warfarin in patients with atrial fibrillation (AF), but their influence on dementia risk remains unclear. Objectives The purpose of this study was to evaluate the relative risks of dementia in DOAC vs warfarin in patients with AF. Methods An electronic literature search was conducted to retrieve studies reporting comparisons of dementia incidence between patients treated with DOACs and warfarin for AF. HRs and 95% CI were pooled in a random-effects meta-analysis. Meta-regression was performed to identify prognostic baseline variables. Network meta-analysis was performed to determine dementia risk between individual DOACs and warfarin. Results Ten studies (n = 342,624) were retrieved. DOAC was associated with a significantly lower risk of developing dementia compared with warfarin (HR: 0.88; 95% CI: 0.80-0.98; P = 0.017; I2 = 75%); significance was also seen in Asian patients (HR: 0.81; 95% CI: 0.68-0.86) but not non-Asian patients. Subgroup analyses of propensity score–matched studies and patients aged 65-75 years showed similar significance, but not for patients aged ≥75 years. Meta-regression found that a lower mean age corresponded to significantly greater favoring of DOAC over warfarin. Network meta-analysis found significant reductions in dementia risk over warfarin for rivaroxaban (HR: 0.854; 95% CI: 0.763-0.955), apixaban (HR: 0.881; 95% CI: 0.778-0.997), and dabigatran (HR: 0.871; 95% CI: 0.770-0.987); the highest-ranked treatment based on P scores was edoxaban. Conclusions The use of DOAC in AF significantly reduces dementia risk compared with warfarin, particularly in Asian patients. The possible reversal of this effect with increasing age merits further randomized trials with long-term follow-up. (Dementia Risk of Direct Oral Anticoagulants Versus Warfarin for Atrial Fibrillation: A Systematic Review and Meta-Analysis; CRD42022365634)

A trial fibrillation (AF) is a prevalent condition among older people, with a lifetime risk ranging from 1 in 3 to 1 in 5, 1 and is associated with a 5-fold risk of stroke. 2 Increased risks of cognitive impairment and dementia have also been demonstrated. 3Although stroke is a known contributory factor to dementia, particularly vascular dementia, 4 the association between AF and Alzheimer's dementia has been suggested to be independent of the occurrence of stroke. 5Yet, it remains unknown whether AF is a direct causal factor for cognitive decline, or is simply a marker of global vascular disease burden.This has led to increasing interest in the "heart-brain axis," with a call for larger longitudinal studies to tease apart the multifactorial relationships between AF and cognitive dysfunction. 6][10] Maintaining a proper international normalized ratio (INR) is also crucial when using warfarin because lower time in therapeutic range (TTR) has been associated with higher risk of dementia. 11 Nonetheless, studies comparing the risk of dementia in DOAC vs warfarin have yielded conflicting results, with some showing reduced risk with DOAC and others finding no significant difference. 12ucially, previous studies with small sample sizes were not sufficiently powered to detect any differences in dementia risk, which can only be detected in large cohorts.
Given the uncertainty in the field, this systematic review and meta-analysis aims to compare the relative risks of dementia in patients with AF taking DOAC vs warfarin, and examine the impact of baseline demographics on these risks.

METHODS
LITERATURE SEARCH.This systematic review and meta-analysis was performed in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Guidelines and registered with PROSPERO (CRD42022365634).
An electronic literature search from inception to June 23, 2022, was conducted by 2 independent investigators (Khi Fong and Vern Tan) on PubMed, EMBASE, and Web of Science for relevant articles, using the concepts of dementia, atrial fibrillation, DOAC, and anticoagulants (Supplemental Table 1).No language restrictions were applied.Bibliographies of included studies were screened, and a search on Google Scholar using the first and last author of each included study was conducted, to ensure inclusion of all relevant studies.
Retrieved abstracts and full texts were reviewed by 2 independent investigators, with conflicts being resolved via group consensus among all authors.
Prospective or retrospective studies reporting comparisons of the outcome of dementia incidence between patients treated with DOAC vs patients treated with warfarin for AF were included.Case reports, case series, reviews, and conference abstracts were excluded.
A standardized data collection template with predefined data fields including study characteristics, patient demographics and outcomes was used for data extraction by 2 independent investigators.
Studies were assessed for risk of bias using the Newcastle-Ottawa Scale.
META-ANALYSIS.The primary outcome in this metaanalysis was the incidence of dementia during followup.This outcome was analyzed in several ways.First, HRs for development of dementia and their 95% CIs were pooled in a random-effects meta-analysis.
Where studies provided both raw and corrected HR estimates, the corrected HR was used.This forest plot was also stratified according to the region in which the study was conducted.Next, subgroup analysis of HR was performed for: 1) studies reporting stratified outcomes for patients $75 years of age and 65-75 years of age; and 2) studies that were propensity score-matched (PSM).Finally, numbers of dementia diagnoses and person-years of follow-up were pooled to determine incidence rate ratios.
Where PSM was used in the study and HRs were clearly provided for the matched group, these were preferentially used instead of HRs for the unmatched cohort. 13If studies did not specify summary HRs across all subgroups (eg, different age groups) and instead reported them separately, they were considered as distinct studies and pooled separately in the meta-analysis.If there was any suspected overlap in these separately reported groups due to PSM, the group with the largest number of analyzed participants was used for analysis.Random-effects Mantel-Haenszel or Inverse-Variance models were used for all analyses due to heterogeneity in definitions of dementia and study region, due to lack of specification of DOAC dosages and durations in some studies, and due to support for generalization inferences beyond the included studies.Heterogeneity was values <40%, 40%-75%, and >75%, respectively. 14Funnel plot symmetry was visually assessed for publication bias.Certainty of evidence was assessed using the Grading of Recommendations Assessment, Development and Evaluation approach. 15ta-regression was performed to identify variables that had an influence on the results of the meta-analysis.Baseline variables that were reported in $10 included studies were used for meta-regression against the HR of respective studies.Bubble plots were generated for any significant associations.
Prior to meta-regression, missing means from medians, ranges, and IQRs were derived via imputation. 16,17TWORK META-ANALYSIS.If $2 studies reported HRs of dementia risk for DOAC vs warfarin stratified by component DOACs, a network meta-analysis HRs were pooled in a random-effects, 2-stage Frequentist NMA, using warfarin as the common comparator.Despite overlap in the PSM warfarin cohorts for some studies, HRs were considered as separate studies due to different numbers of matched warfarin subjects to each DOAC.I 2 and Cochran Q were used to test for heterogeneity and inconsistency.A network graph, league table, and forest plot were also generated for this analysis.P scores were used to numerically rank the treatment strategies in the overall cohort as well as within subgroups, with higher P scores corresponding to greater efficacy.
P scores are based solely on HRs and standard errors of the frequentist NMA estimates under the normality assumption, providing an intuitive way to appraise treatments and inform medical decision-making. 18l analyses were performed in RStudio using R-4.1.2and the packages "meta," "netmeta," and "metafor," with P < 0.05 regarded to indicate statistical significance.There was no funding source for this study.This article made use of publicly available data from published studies, hence ethics approval was not required.
Only 1 study 21 was available for the American region.
In the NMA, 3 studies 21,25,26 provided stratified data for individual DOAC agents vs warfarin.Rivaroxaban, dabigatran, and apixaban demonstrated significant reduction in dementia risk over warfarin, with the upper bound of all 3 CIs <1 (Figure 4).The reduction in dementia risk for edoxaban was not significant (pooled HR: 0.830; 95% CI: 0.665-1.036),but this was likely due to its use in only 1 study.
Moderate heterogeneity was observed (I 2 ¼ 62.4%; Q ¼ 23.9; P ¼ 0.004).P scores ranked edoxaban as the therapy with the numerically highest reduction in dementia risk compared with warfarin, followed by rivaroxaban, dabigatran, and apixaban, but no significant differences between DOACs were seen in the league table (Table 3).The network plot is shown in Supplemental Figure 10.

DISCUSSION
This meta-analysis presents a comprehensive summary of the existing literature, demonstrating that DOAC use in AF is indeed associated with a reduced risk of dementia compared with warfarin (Central Illustration).This effect was seen even when data was restricted to PSMs only, and on NMA of individual DOACs vs warfarin.Significance was seen for the subset of Asian patients but not non-Asian patients.
Included studies had large cohorts and were sufficiently powered to detect differences in dementia risk.
Previous meta-analyses demonstrated a similar reduction in dementia risk with DOAC use as opposed to warfarin. 29,30However, they were limited by the inclusion of data from the pivotal randomized trials comparing DOAC vs warfarin.Information on dementia diagnoses were only posted on the corresponding trial websites of the National Library of Medicine and not analyzed directly in the final manuscripts, the numbers of dementia cases were too low due to studies being underpowered to detect differences in this outcome, and time-to-event analysis in the form of HR was not provided.These trials were excluded from the present analysis to reduce the risk of bias and heterogeneity.
Several mechanisms linking AF to an increased risk of dementia have been proposed.Stroke has been proposed as a significant contributory factor to the pathogenesis of dementia, with studies showing that new-onset dementia and cognitive impairment are the sequelae of stroke. 31,32Nonetheless, other evidence supports a stroke-independent risk of dementia. 33,342][43] The prevalence of renal disease ranged from 1.7%-30.4% among included studies, although insufficient studies were present for metaregression.Thus, further studies are required to investigate this correlation.
A significantly lower risk of dementia was seen in DOACs compared with warfarin for the subgroup of Asian patients, but not European patients.A similar discordance has been previously reported for the setting of venous thromboembolism prophylaxis. 44ysiologically, Asian patients are likely to be HRs for dementia risk of each DOAC vs warfarin in all studies reporting this stratification were pooled in a random-effects network analysis.
Abbreviations as in Figure 1.
Fong et al shown to increase in AF. 5,34 Nevertheless, given the steep increase in incidence rates of all types of dementia beginning at approximately 70-75 years, 48,49 it may be possible that the AF-independent increase in dementia diluted the influence of anticoagulation in the elderly population.This could explain why the correlation between age and HR for dementia risk between DOAC and warfarin had a high heterogeneity (I 2 ¼ 70%) despite being statistically significant.To  The table should be read from left to right.HRs and 95% CIs for development of dementia for each comparison are in the cell in common between the column-defining and row-defining treatment.A HR of <1 favors column-defining treatment (lower risk of dementia).Significant comparisons are highlighted in italics.P scores for each treatment are provided in the same cell.Nonetheless, the impact on dementia is likely to be multifactorial, beyond anticoagulation (or DOACs) alone, hence the move toward a more holistic or integrated care approach to AF management. 54deed, there is increasing literature on lifestyle factors and risk factor management impacting on incident dementia in patients with AF, 55,56 which is not accounted for in this analysis.An integrated care approach to AF management has been associated with improved clinical outcomes 57 and is now recommended in international guidelines. 58Of note, adherence to such an integrated care approach is associated with a lower risk of incident Alzheimer's and vascular dementia. 59

CONCLUSIONS
Dementia Risk of DOACs vs Warfarin considered low, moderate, or considerable for I 2

FIGURE 1
FIGURE 1 PRISMA Flowchart of Included Studies Dementia Risk of DOACs vs Warfarin largest number of patients (rivaroxaban vs warfarin) was selected for subsequent analysis.Five studies provided HRs for patients $75 years of age, of which one, Friberg et al 22 provided separate HRs for patients 75-85 years and $85 years; 3 studies provided HRs for patients 65-75 years of age.The mean age of participants ranged from 70.4-75.7 years, with a slight male predominance (50.6%).Follow-up time ranged from 0.3-3.7 years.The presence of heart failure or diabetes mellitus at baseline was <50%, but the majority of patients had hypertension.The presence of previous stroke ranged from 0%-27%.The dementia endpoint in all studies was determined by reviewing the incidence of new disease codes for a dementia diagnosis after anticoagulant initiation from the respective patient databases.Risk of bias was generally low (Supplemental

FIGURE 2
FIGURE 2 Random-Effects Meta-Analysis of HR for Dementia Development

FIGURE 3
FIGURE 3 Subgroup Analysis of Dementia Development in Propensity Score-Matched Studies J A C C : A S I A , V O L . 3 , N O . 5 , 2 0 2 3 Fong et al O C T O B E R 2 0 2 3 : 7 7 6 -7 8 6 Dementia Risk of DOACs vs Warfarin OAC. 36The lower risk of intracranial bleeding and stroke observed with DOAC compared with warfarin theoretically translates to lower rates of microembolization and cerebral hypoperfusion.Moreover, given the marked influence of TTR and labile INR on dementia and intracranial bleed risk in patients on warfarin, 11 the use of DOAC may provide better control of coagulation profile and prevent the aforementioned pathways, which may lead to dementia.From a dietary standpoint, patients taking a DOAC would not require limitation of vitamin K intake, unlike those on warfarin.Low dietary intake or low blood levels of vitamin K have been associated with cognitive decline and Alzheimer's disease.37Furthermore, green leafy vegetables-a source of vitamin K-are also high in vitamin B 12 and folate, which are associated with lower incidences of cognitive decline and dementia.38Hence, the benefits of DOAC over warfarin may not be solely related to the effect of DOAC alone.Meta-regression found that higher age was associated with lower dementia risk with warfarin compared with DOAC.These findings are at odds with studies of nondementia outcomes in the literature, with DOAC demonstrating continued benefit over warfarin even in patients of advanced age, with a cohort study even deeming DOAC of greatest benefit in the very elderly.39It is uncertain whether confounders, such as the nature of AF (paroxysmal, persistent, or permanent), type and dose of DOAC, and TTR of warfarin users, may have impacted these results.These baseline variables were infrequently reported and insufficient studies reporting this outcome were available for meta-regression.Pharmacokinetic differences between OACs-DOACs with predominantly renal clearance and warfarin with predominantly hepatic clearance-could have led to altered effects of DOACs in older patients because glomerular filtration rate decreases and chronic kidney disease increases with age. 40Dose reductions in DOACs for patients of more advanced age may have led to differing effects from younger patients.Altered pharmacokinetics may have led to suboptimal dosing of DOACs and a resulting worsening of dementia risk compared with warfarin.Nonetheless, investigations into DOAC use in renal impairment have yielded mixed results contributory factors to dementia, 3-arm studies (with DOAC, warfarin, and untreated groups) are needed to analyze these age-related effects.Apart from demonstrating significant associations, it is difficult to infer cause-and-effect between AF and dementia because both are end results of pathologic processes that develop over many years.Hence, it is difficult to conclude whether the effects of DOAC or warfarin treatment on dementia risk are due to their influence on AF or other background processes that also influence AF.Despite the inclusion of different DOAC therapies in varying proportions, NMA of 3 studies showed no significant differences in dementia risk between the 4 therapies.Edoxaban, which was investigated in only 1 study, had a slight but nonsignificant edge over the other DOACs.Further head-to-head trials of various DOACs are needed to support these suggestions.STUDY LIMITATIONS.The nonrandomized nature of most studies precludes a definite conclusion on the true effect of DOACs vs warfarin in lowering dementia risk.The use of disease codes for dementia may not truly encompass the entirety of cases because this only captures patients who were admitted to the hospital during the follow-up period, but not those who developed dementia but were not hospitalized and remained undiagnosed in the community.Nonetheless, the corroboration of the PSM-only subgroup analysis with the main analysis is a strong point in favor of true significant effect.PSMs reduce the effect of confounding variables and have been shown to be empirically equivalent to RCTs in generating unbiased estimates of treatment efficacy. 50Hence, future RCTs are needed; indeed, several are currently underway (eg, GIRAF [Cognitive Impairment Related to Atrial Fibrillation Prevention Trial; NCT01994265], BRAIN-AF [Blinded Randomized Trial of Anticoagulation to Prevent Ischemic Stroke and Neurocognitive Impairment in AF; NCT02387229], and CAF [Impact of Anticoagulation Therapy on the Cognitive Decline and Dementia in Patients With Non-Valvular Atrial Fibrillation; NCT03061006]).The incorporation of standardized tests, such as the Montreal Cognitive Dementia Risk of DOACs vs WarfarinAssessment and Mini Mental State Exam, in these trials will allow detailed evaluation of cognitive function.Longer follow-up is also suggested; included studies had a follow-up of <5 years, despite a longer duration required for development of dementia suggested by other studies.51NMA accuracy is dependent on assumptions of methodologic equivalence and similarity of patient profiles.52Although methodology was grossly similar in terms of study inclusion criteria and follow-up duration, patient profiles varied considerably.The effect of rhythm control of AF on dementia risk was not analyzed due to the heterogeneity contributed by its procedural nature, although significant benefit has previously been shown.53 The use of DOACs in patients with AF significantly reduces dementia risk compared with warfarin, especially among Asian patients.Nonetheless, a suggestion of reversal of this effect with increasing CENTRAL ILLUSTRATION Meta-Analysis of Dementia Risk in DOAC Versus Warfarin for AF : benefit of DOAC over warfarin with age Network meta-analysis: no differences in dementia risk between individual DOACs dementia risk for DOAC vs warfarin Significance found in Asians, but not non-Asians 2 = 0.0263; Chi 2 = 23.36,df = 3 (P < 0.01); I 2 = 87% Heterogeneity: Tau 2 = 0.0264; Chi 2 = 13.77,df = 5 (P = 0.02); I 2 = 64% Heterogeneity: Tau 2 = 0.0170; Chi 2 = 39.52,df = 10 (P < 0.01); I 2 30.Cheng W, Liu W, Li B, Li D. Relationship of anticoagulant therapy with cognitive impairment among patients with atrial fibrillation: a meta-PERSPECTIVES COMPETENCY IN MEDICAL KNOWLEDGE: This meta-analysis of 10 large observational studies and 342,624 patients found that the use of DOACs in AF appears to significantly reduce dementia risk compared with warfarin, especially in Asian populations.However, the benefit appears to diminish with increasing age.TRANSLATIONAL OUTLOOK: This study demonstrates yet another clinical standpoint from which DOACs show benefit over warfarin, and strengthens the existing evidence base for the use of DOACs for AF.J A C C : A S I A , V O L . 3 , N O . 5

TABLE 1
Characteristics of Included Studies a Values taken from unmatched cohort.HTN ¼ hypertension; PSM ¼ propensity score-matched study.

TABLE 2
Meta-Regression of Baseline Characteristics Against HR of DOAC Vs Warfarin DM ¼ diabetes mellitus; DOAC ¼ direct-acting oral anticoagulant; HTN ¼ hypertension.FIGURE 4 Network Meta-Analysis of Dementia Risk Across Individual DOACs and Warfarin