Comparing Warfarin and 4 Direct Oral Anticoagulants for the Risk of Dementia in Patients With Atrial Fibrillation

Introduction

See related article, p 3469

Atrial fibrillation (AF) is the most common arrhythmia, and the prevalence of AF increases with advanced age.¹ AF was associated with increased risks of stroke, heart failure, and death, as well as cognitive decline and dementia.^2–4 The higher risk of cognitive impairment and dementia in patients with AF did not depend on a history of previous stroke.⁵ Silent brain infarcts are frequent in patients with AF, and this could be a possible mechanism of AF-associated cognitive decline and dementia.^6,7 Several observational studies have suggested that oral anticoagulation (OAC) in patients with AF was linked with a lower risk of incident dementia.^8–11

Recently, the direct oral anticoagulants (DOACs) have been introduced in clinical practice.^12,13 Compared with warfarin, DOAC showed at least comparable efficacy for the prevention of thromboembolic events and a significant lower risk of intracranial hemorrhage in patients with AF.¹⁴ For the risk of incident dementia, DOACs were also associated with a lower risk of dementia.⁸ However, conflicting results were reported on the comparative risk of incident dementia between DOACs and warfarin.^8,11,15–17 Although each DOAC showed different effectiveness and safety profiles compared with other DOACs,^18–20 there has been a lack of evidence whether the risk of dementia could vary between the different DOACs.

In this study, our aim was to compare the risk of incident dementia between warfarin and DOACs in patients with AF. Second, we evaluated whether the risk of dementia would be different in patients treated with the four DOACs, including rivaroxaban, dabigatran, apixaban, and edoxaban, compared with those treated with warfarin.

Methods

Data Source and Study Population

This study was a retrospective observational cohort study using a nationwide claims database of South Korea. The Korean National Health Insurance System is a single-payer organization, managed by the Korean government, which includes all residents in South Korea.^21,22 The Korean National Health Insurance System has adopted a fee-for-service system to pay health care providers; therefore, the Korean National Health Insurance System obtains all medical expenses claim data of the entire Korean population. It contains a qualification database including enrollees’ age and sex, a claim database including diagnoses defined by the International Classification of Diseases, TenthRevision, prescription records and procedures for inpatient and outpatient services, and death information.^21,22

Using this database, we identified 154 407 patients with AF who had ≥1 pharmacy claim for any OAC, including warfarin or DOACs (rivaroxaban, dabigatran, apixaban, or edoxaban) from January 1, 2014, to December 31, 2017. To include only new users of index OAC, we excluded patients who received any OAC treatment before January 1, 2014. A total of 89 547 OAC naive nonvalvular AF patients aged ≥40 years were identified (Figure 1). We excluded the (1) patients who were diagnosed with valvular AF, (2) those who had a history of pulmonary embolism or deep vein thrombosis, (3) those who received joint replacement surgery, and (4) those who were diagnosed with end-stage renal disease. Patients with prevalent dementia were also excluded. Finally, the study population consisted of 72 846 patients (Figure 1).

• Download figure
• Download PowerPoint

The Institutional Review Board of the Seoul National University Hospital (E-1811-010-982) approved the present study, and the informed consent was waived because anonymous and deidentified information was used for this analysis. All data have been made publicly available at the National Health Insurance Sharing Service and can be assessed at http://nhiss.nhis.or/kr/bd/ab/bada000eng.do.

The Reporting of Studies Conducted Using Observational Routinely-Collected Health Data was used to report the findings of this study.²³

Covariates and Definitions

Age, sex, and comorbidities, including hypertension, diabetes, dyslipidemia, heart failure, prior stroke, prior myocardial infarction, peripheral arterial disease, chronic obstructive pulmonary disease, and cancer, were ascertained from the Korean National Health Insurance System database.²¹ Table I in the Data Supplement summarized the definitions of comorbidities.^1,13,20 The CHA₂DS₂-VASc score for assessing stroke risk was calculated, combining covariate information.^1,13,20

Study Outcome and Follow-Up

The primary outcome was incident dementia, which was further classified as Alzheimer dementia and vascular dementia.²⁴ The definitions were based on the relevant diagnosis codes and the prescription for dementia. Detailed definitions are presented in Table I in the Data Supplement. Patients were censored at the occurrence of outcome events, discontinuation of index treatment, or the end of the study period (December 31, 2018). The discontinuation of index treatment was defined as a 30-day gap from the last prescription of index treatment. The discontinuation of index treatment included both OAC discontinuation or a switch from index treatment to other types of OAC.

Statistical Analysis

Continuous variables are presented as mean±SD, and categorical variables are presented as percentages. For the comparison, we used the inverse probability of treatment weights (IPTW) of the propensity score (PS) method to balance baseline covariates between the comparator and reference groups.^25–27 First, we conducted a comparison of the risk of incident dementia between warfarin and pooled DOAC. Between the two groups, the PS of being in each treatment group was calculated using an ordinary logistic regression based on all baseline covariates.²⁵ We used the absolute standardized difference of ≤0.1 (10%) as a negligible difference in potential confounders between the two treatment groups.²⁸ Second, we performed a comparison between warfarin as the reference group and the four DOAC-treated groups (rivaroxaban, dabigatran, apixaban, and edoxaban) using the IPTW method. The maximum absolute standardized difference of ≤0.1 (10%) in all covariates indicates that all groups were well balanced.²⁸

Crude incidence rates (IRs) were calculated using crude event numbers during the follow-up period divided by 100 person-years at risk for new-onset dementia. After the IPTW, the risk of incident dementia in the pooled DOAC versus warfarin (reference) and each DOAC versus warfarin (reference) was compared using Kaplan-Meier curves (log-rank test) and weighted Cox proportional hazard regression models with IPTW. For all statistical analyses, statistical significance was defined as P of <0.05. Statistical analyses were performed using SAS 9.3 (SAS Institute, Inc, Cary, NC).

Sensitivity Analyses

We conducted 3 sensitivity analyses for the comparison between pooled DOAC and warfarin and between each DOAC and warfarin. First, the main analysis was conducted as an on-treatment analysis, which censored the index treatment discontinuation. We additionally performed an intention-to-treat analysis that did not censor the index treatment discontinuation as a sensitivity analysis. Second, considering that the development of incident dementia is a chronic process, to confirm the long-term effect of OAC treatment on the risk of incident dementia, we performed a sensitivity analysis including patients who received index treatment at least ≥1 year with a 1-year lag period to exclude the events during the first year of follow-up. Third, incident stroke during follow-up might affect the development of new-onset cognitive dysfunction and dementia, so a sensitivity analysis was performed with censoring incident stroke during the follow-up period.

Complementary Analyses

Smoking is a known risk factor of dementia.²⁹ Approximately 70% of the total study population had the national health screening examination data.^21,22 We performed a complementary analysis to adjust for smoking status in a subgroup available for the national health screening examination data. In this subgroup, multivariable Cox analyses were conducted: (1) model 1 adjusted the same variables used in the PS calculation, including age, sex, CHA₂DS₂-VASc score, hypertension, diabetes, dyslipidemia, heart failure, prior stroke, prior MI, PAD, COPD, and cancer; (2) model 2 additionally adjusted smoking status (nonsmoker, ex-smoker, and current smoker) in model 1; and (3) model 3 excluded the COPD and additionally adjusted smoking status in model 1.

Subgroup Analyses

The analyses of comparisons between warfarin and pooled DOAC were supplemented by subgroup analyses according to the following subgroups: age strata (<65, 65 to <75, and ≥75 years), baseline CHA₂DS₂-VASc score (<3, 3–5, and ≥6), and history of stroke. For subgroup analyses, we used multivariable Cox proportional hazards regression models using all variables included in the PS calculation (age, sex, hypertension, diabetes, dyslipidemia, heart failure, prior stroke, prior myocardial infarction, peripheral arterial disease, chronic obstructive pulmonary disease, and cancer). The statistical significance of the interaction between treatment was defined as P interaction of <0.1.

Results

A total of 72 846 patients were finally included in this analysis (mean age, 71.8±10.5 years; 58% men; and mean CHA₂DS₂-VASc, 4.0±1.8; Figure 1; Table II in the Data Supplement). Among these, 25 948 were treated with warfarin, and 46 898 were treated with DOACs (17 193 with rivaroxaban, 9882 with dabigatran, 11 992 with apixaban, and 7831 with edoxaban). Baseline characteristics are presented in Table II in the Data Supplement. Compared with patients treated with warfarin, those treated with DOACs were older and had a higher CHA₂DS₂-VASc score.

During 911 171 person-years of follow-up, 4437 patients were diagnosed with dementia (1097 was vascular dementia and 3009 was Alzheimer dementia). The crude incidence of dementia was 4.87 per 100 person-years, vascular dementia was 1.20 per 100 person-years, and Alzheimer dementia was 3.30 per 100 person-years. Median follow-up duration of the different OAC groups is presented in Table III in the Data Supplement.

Comparison Between Warfarin and Pooled DOAC

Crude event numbers, crude IRs, unadjusted and multivariable adjusted hazard ratios (HRs), and 95% CIs are presented in Figure 2 and in Table II in the Data Supplement. After IPTW, the baseline characteristics of 2 groups were well balanced (Table 1). Kaplan-Meier curves of warfarin and pooled DOAC groups before and after IPTW are presented in Figure I in the Data Supplement. Compared with warfarin, the pooled DOAC-treated group showed comparable risks for dementia (HR, 0.929 [95% CI, 0.858–1.007]), vascular dementia (HR, 0.921 [95% CI, 0.814–1.043]), and Alzheimer dementia (HR, 1.049 [95% CI, 0.969–1.136]; Figure 2).

• Download figure
• Download PowerPoint

Sensitivity Analyses and Complementary Analyses

The results of 3 sensitivity analyses were consistent with the main analysis (Table IV in the Data Supplement). DOAC use showed comparable risks for incident dementia, vascular dementia, and Alzheimer dementia in all 3 sensitivity analyses: intention-to-treat analysis, the analysis including patients with at least 1 year or longer follow-up, and the analysis censoring incident stroke during follow-up (Table IV in the Data Supplement).

The results of the health screening examination were available in 70% of the total study population. The smoking status of this subgroup is presented in Table V in the Data Supplement. Although there were some differences in the ranges of CIs compared with the main analysis, overall directionality of HRs was consistent with the main results in the three multivariable models (Table VI in the Data Supplement).

Subgroup Analyses

Table VII in the Data Supplement shows crude event numbers and IRs of various subgroups. Crude IRs of dementia, vascular dementia, and Alzheimer dementia tended to be higher in older age groups, those with higher CHA₂DS₂-VASc score, and those with a history of stroke.

Age (<65, 65 to <75, and ≥75 Years)

DOAC use was associated with a significantly lower risk of dementia than warfarin in patients aged 65 to <75 years but comparable with warfarin in other age groups (P for interaction, 0.009; Figure 3). Similar trends were observed in vascular dementia but not for Alzheimer dementia.

• Download figure
• Download PowerPoint

CHA₂DS₂-VASc Score (<3, 3–5, and ≥6)

In all subgroups categorized by different CHA₂DS₂-VASc scores, DOAC use showed comparable risks for dementia, vascular dementia, and Alzheimer dementia, and there was no significant interaction among subgroups (Figure 3).

Presence of History of Stroke

Among the total study population, 18 717 (25.7%) patients had a history of stroke. While DOAC and warfarin groups showed comparable risks of dementia in patients without stroke history, DOAC use was associated with a lower risk of dementia than warfarin in patients with stroke history (P for interaction, 0.005; Figure 3).

Comparisons Between Warfarin and Each DOAC

Before IPTW, baseline characteristics among warfarin and 4 DOAC groups were different (Table VIII in the Data Supplement). After IPTW, the maximum absolute standardized difference for all covariates was <0.1 (Table 2). Crude event numbers and IRs are presented in Figure 4.

• Download figure
• Download PowerPoint

For the risk of incident dementia, rivaroxaban, dabigatran, and apixaban showed comparable risks with warfarin, whereas edoxaban was associated with a lower risk of dementia than warfarin (HR, 0.830 [95% CI, 0.740–0.931]; Figure 4). Rivaroxaban and edoxaban were associated with a lower risk of vascular dementia compared with warfarin (Figure 4). For the risk of Alzheimer dementia, all DOACs showed comparable risks with warfarin.

In the sensitivity analyses for the risk of dementia, results were generally consistent with the main analysis (Table IX in the Data Supplement). In the intention-to-treat analysis, edoxaban was associated with a lower risk of dementia (HR, 0.852 [95% CI, 0.777–0.936]) compared with warfarin, whereas other DOACs showed comparable risks with warfarin.

When only including patients with at least 1-year follow-up, all DOACs showed a comparable risk for dementia with warfarin, except for apixaban. Apixaban was associated with a higher risk of dementia (HR, 1.195 [95% CI, 1.017–1.404]) compared with warfarin in patients with longer than the 1-year follow-up. When censoring incident stroke during follow-up, rivaroxaban (HR, 0.889 [95% CI, 0.805–0.982]) and edoxaban (HR, 0.870 [95% CI, 0.758–0.998]) were associated with lower risks of dementia compared with warfarin.

Discussion

In this large population-based cohort study, we observed that DOAC showed a comparable risk for new-onset dementia with warfarin in AF patients. These were consistent in both vascular dementia and Alzheimer dementia. Second, in patients aged 65 to <75 years and those with a history of stroke, DOAC use was associated with a lower risk of dementia compared with warfarin. Third, among the four DOACs, edoxaban use showed a significantly lower risk of dementia than warfarin. In addition, rivaroxaban and edoxaban were associated with a lower risk of vascular dementia compared with warfarin. For the risk of Alzheimer dementia, all DOACs showed comparable risks with warfarin.

It is well-recognized that AF increases the risk of dementia.^2–5 In a previous meta-analysis, for example, AF was significantly associated with increased risk of developing cognitive impairment by 40%, and the association was significant for vascular dementia (relative risk, 1.72 [95% CI, 1.27–2.32]).⁵ A large population-based study demonstrated that AF was independently associated with all types of dementia, including vascular, Alzheimer, and senile dementia.³ In another prospective cohort study, AF was associated with a higher risk of dementia (HR, 1.87 [95% CI, 1.37–2.38]).⁴ Notably, the increased risk of dementia in patients with AF was not fully explained by clinical stroke events, suggesting the independent risk of dementia irrespective of preceding clinical stroke.⁴ Subclinical cerebral ischemia and cerebral microinfarcts have been proposed as the possible mechanism of cognitive dysfunction or dementia in patients with AF.^6,7 Therefore, appropriate anticoagulation therapy would be important to lower the risk of dementia.

Several studies evaluated the benefit of OAC treatment on the risk of dementia in patients with AF.^8–11,15 In a meta-analysis, OAC treatment was associated with a risk reduction of dementia by 21% compared with no OAC treatment.¹⁰ In more recent observational studies, although the magnitude of risk reduction was slightly varied, the beneficial effect of OAC treatment on the risk of dementia was consistently observed both in non-Asian and Asian populations and even observed in patients with lower risk stroke.^9,11,15 In addition, taking OAC has been associated with a lower risk of Alzheimer dementia and vascular dementia.⁹

While OAC treatment generally lowers the risk of dementia by ≈10% to 40% in patients with AF, there have been limited and conflicting data for the comparisons between warfarin and DOACs on the risk of dementia. Since DOAC use was comparable or even superior to warfarin in stroke prevention,^14,18–20 DOACs would be expected to be superior to warfarin for dementia prevention as well. One recent large-scale observational study showed no difference between DOACs and warfarin in dementia risk (HR, 0.89 [95% CI, 0.70–1.14]).¹¹ In contrast, Søgaard et al¹⁶ reported there was no meaningful difference between DOACs and warfarin on the risk of dementia, even though there was significant interaction in different age subgroups (60–69, 70–79, and ≥80 years) with DOACs being less effective in those aged ≥80 years. On the contrary, Friberg et al showed more favorable results for DOACs compared to warfarin in reducing dementia, amongst patients at low-risk (CHA₂DS₂-VASc scores of 0 or 1, HR 0.47 [95% CI, 0.18–1.22]) or relatively young age subgroup (aged 65–74 years, HR 0.66 [95% CI, 0.24–1.79]).^8,15 In a previous observational study based on the US health care claims database, in a well-matched cohort using PS matching method, rivaroxaban and apixaban were significantly associated with a lower risk of dementia compared with warfarin.¹⁷

In our study, the pooled DOAC group did not show a significant difference in the risk of dementia compared with warfarin, consistent with previous results.^8,11,16 In subgroup analyses, among age subgroups, DOAC showed a lower risk of dementia than warfarin only in patients aged 65 to <75 years, consistent with previous studies.^8,11,16 The role of AF on the increased relative risk of dementia was higher in the younger age group (<70 years) and attenuated in those aged ≥80 years.³ Age is the strongest risk factor for dementia, and the contribution of OAC treatment for preventing dementia might be significant from the patients aged ≥65 years. In a previous study by Søgaard et al,¹⁶ DOAC showed nonsignificantly lower rates of dementia than warfarin in patients aged 60 to <80 years; however, DOAC use was associated with a higher risk of dementia compared with warfarin in those aged ≥80 years. The possible cause of this phenomenon is that the study population >80 years of age who was OAC naive and without prior neurological diagnoses might be a highly selected population. As mentioned by the authors, potential residual confounding related to selected prescription (DOAC versus warfarin) would be considered to interpret these findings.

In our study, when compared with warfarin, DOACs might be more beneficial in patients with prior stroke, not in those without prior stroke. The impact of AF on the increased risk of dementia was accentuated in poststroke patients (relative risk, 2.7 [95% CI, 1.82–4.00]) than in the general population (relative risk, 1.40 [95% CI, 1.19–1.64]).⁵ The protective effect of OAC treatment on the risk of dementia was similar in patients with prior stroke.^8,10 There have been no previous data about the different effects of preventing dementia between DOACs and warfarin according to the presence of prior stroke. Considering the low quality of anticoagulation control among warfarin users in Korea (with median time in therapeutic range of 30.4%),^30,31 the superior effect on the stroke prevention of DOAC over warfarin in patients with a history of stroke could provide a plausible explanation for this result.³²

When balancing all available covariates among 4 DOAC groups, warfarin, rivaroxaban, dabigatran, and apixaban showed a comparable risk of dementia with warfarin, while edoxaban was associated with a lower risk of dementia than warfarin. These results should be interpreted cautiously because there could be potential residual confounding factors in the observational study. Although we considered these limitations, different DOACs have different profiles in their effectiveness and safety. In a previous study from our group, we found that edoxaban was associated with the largest risk reduction in ischemic stroke compared with warfarin among 4 DOACs.²⁰ Several studies have also showed that different OACs have variable effect sizes on vascular protection.^33–36 Further investigation is needed whether different OACs have different risk of silent or subclinical stroke (eg, on the brain imaging), and these findings would be associated with subjects’ cognitive function.

Limitations

First, the study design was a retrospective observational study, so we could only show associations between the treatment and the risk of dementia. To confirm causality, we need further investigations, such as well-designed large prospective randomized clinical trials. Second, we balanced the groups using the IPTW method to compare the risk of dementia among OAC treatments. Even then, there might be residual confounding factors that could not be fully adjusted in our database. Third, there might be misclassifications between the actual diagnosis and the diagnosis using operational definitions based on the claims database. Although we used the definition and the classification of dementia that was widely used in previous studies,^8,9,15,24,37 there remains a possibility of inaccuracy. We tried to overcome this limitation by combining diagnostic codes and prescription records. Fourth, the actual drug adherence rate was not measured owing to the inherent limitations of claims data. Fifth, among DOAC users, the proportion of off-label underdosed DOAC prescriptions in the Asian population was substantial.³⁸ However, we could not evaluate the impact of off-label underdosed DOAC on dementia risk because of the lack of information, including body weight and serum creatinine of the total study population. In our study population, body weight and body mass index information were available in 70% of the total study population. Although this population was a subgroup of our study, we found that almost 40% of patients were low body weight, defined as weighing ≤60 kg (Table X in the Data Supplement). Thus, a substantial proportion of patients might have been prescribed appropriately reduced low-dose DOACs. Although the clinical impact of off-label underdosed DOAC is still controversial, off-label underdosed DOAC might be associated with the increased risk of ischemic stroke.³⁹ Further investigation is needed to find the clinical impact of off-label underdosed DOACs in Asian patients with AF, especially for the risk of dementia. Lastly, this study had a relatively short follow-up duration (mean, 1.3±1.1 years) to observe the effect of OAC treatments on the development of dementia.

Conclusions

In this large Asian population with AF, DOAC showed a comparable risk of dementia with warfarin overall. DOACs appeared more beneficial than warfarin, in those aged 65 to 74 years or with a history of stroke. For specific DOACs, only edoxaban was associated with a lower risk of dementia than warfarin.

Supplemental Materials

Online Tables I–X

Online Figure I

Footnotes