Population Dementia Incidence and Direct Oral Anticoagulant Use in a Representative Population With Atrial Fibrillation

Abstract

Background and Objectives

Incidence and prevalence of atrial fibrillation (AF), a risk factor of dementia, have been increasing over time. Oral anticoagulation reduces risk of stroke and other negative outcomes of AF and may reduce dementia health inequities. The objective of this study was to estimate dementia incidence in patients with newly-diagnosed AF and taking an anticoagulant as use of direct oral anticoagulants (DOACs) increased.

Methods

We used a retrospective cohort design with annual incident AF cohorts of community-dwelling Medicare Fee-for-Service beneficiaries, enrolled in Parts A, B, and D from 2007 to 2017. The sample was limited to beneficiaries aged 67 years and older with incident AF; no prior dementia; and use of anticoagulants warfarin, dabigatran, rivaroxaban, apixaban, or edoxaban in year t.

Results

A total of 1,083,338 beneficiaries were included in the study, 58.5% female, with mean (SD) age 77.2 (6.75) years. Among anticoagulated, incident AF cohorts, use of DOACs increased from 10.6% in their first year of availability (2011) to 41.4% in 2017. Among incident AF cohorts taking any oral anticoagulant, 3-year dementia incidence did not change significantly over the cohorts after adjusting for confounders. For example, incidence was 9.1% (95% CI 8.9–9.4) among White persons diagnosed with AF in 2007 and 2008 and 8.9% (95% CI 8.7–9.1) in 2017. Across cohorts, dementia incidence was consistently highest for Black persons, followed by American Indian/Alaska Native and White persons, and lowest for Asian persons. In 2017, 10.9% (95% CI 10.4–11.3) of Black persons in the cohort developed dementia within 3 years, 9.4% (95% CI 8.0–10.9) of American Indian/Alaska Native, 8.9% (95% CI 8.7–9.1) of White, 8.7% (95% CI 8.2–9.1) of Hispanic, and 6.9% (95% CI 6.4–7.4) of Asian persons. Across race/ethnicity, 3-year stroke risk decreased consistently over time; however, the increasing availability of DOACs did not alter the trend.

Discussion

Increased use of DOACs among incident AF cohorts from 2007 to 2017 was not associated with significant declines in dementia or stroke risk. Consideration of similar stroke and dementia risk, as well as differences in cost, is warranted when weighing the risks and benefits of available oral anticoagulants.

Introduction

Incidence and prevalence of atrial fibrillation (AF), the most common arrhythmia, have been increasing over time in the US population and worldwide.^1,2 Current estimates indicate 3–6 million people are living with AF in the United States alone, which is projected to increase to as many as 16 million by 2050.³ AF is associated with a significantly increased risk of stroke, which can lead to cognitive impairment and dementia, but AF has also been identified as risk factor of dementia independent of stroke. Several mechanisms connecting AF and dementia have been proposed including, but not limited to, occurrence of: cerebral infarction, microbleeds, cerebral hypoperfusion, and inflammation.^1,4-7 Preventing these adverse events through timely and effective anticoagulation may reduce subsequent risk of dementia in newly diagnosed persons with AF and, likewise, dementia incidence in the population.^1,8

There are currently 2 types of oral anticoagulants used to treat AF: warfarin and direct oral anticoagulants (DOACs)—dabigatran, rivaroxaban, apixaban, and edoxaban. The newer DOACs began entering the market in late 2010 and have become an increasingly common form of treatment in patients with AF, with use among adults with AF in 1 study increasing from 4.7% in 2011 to over 47.9% in 2020.⁹ Prior studies have demonstrated reduced risk of stroke, mortality, and dementia in patients with AF with any oral anticoagulation compared with no anticoagulation.^10-12 DOACs are easier for patients to adhere to than warfarin. Whereas warfarin requires dietary stability and regular monitoring of patients' international normalized ratio by blood tests, DOACs may be taken without such monitoring to maintain more consistent anticoagulation.¹ Furthermore, cerebral microbleeds, hypothesized as 1 mechanism connecting cognitive impairment and AF, predict future intracerebral hemorrhage and cognitive decline.^5,7 Use of warfarin has been associated with increased incidence of cerebral microbleeds in patients with AF,¹³ whereas no relationship with microbleeds have been found with DOACs.¹⁴ As a result, DOACs are hypothesized to have fewer adverse effects than warfarin in patients with AF, including stroke, bleeding, and subsequent dementia.¹⁵ Indeed, some studies have demonstrated reduced risk of ischemic stroke and hemorrhage, both of which are associated with dementia.^16,17

In addition to reducing dementia risk by treating AF, DOACs are hypothesized to have a pleiotropic effect on Alzheimer disease (AD) pathology; however, this is still currently under clinical investigation. DOACs directly inhibit clotting factors thrombin (dabigatran) or Factor Xa (apixaban, rivaroxaban, edoxaban). Pathologic thrombin has been implicated in cerebrovascular changes that precede amyloid β and Tau accumulation—the 2 hallmarks of AD—and subsequent neurodegeneration.^18-20 By acting directly on cerebrovascular factors that precede amyloid β and Tau, DOACs may reduce risk of dementia more than warfarin, but the pleiotropic effect may be small. However, comparisons of the relationship between DOACs and warfarin and dementia risk have been inconsistent. Some observational studies reported decreased risk of dementia in DOAC users compared with warfarin users,^12,21-23 whereas others found no difference^24,25 or only found differences among some groups of patients, and the impact of DOACs on dementia remains an open question. A recent randomized controlled trial comparing dementia risk in persons taking dabigatran or warfarin found no difference in 2-year dementia risk.²⁶

Differences in AF prevalence, treatment and subsequent outcomes across racial/ethnic background suggest the impact of DOACs on dementia may vary. Black and Hispanic persons have lower risk of AF than White beneficiaries, but those who develop AF are at higher risk of stroke and death.^27,28 Recent studies examining prescriptions among persons with AF in the outpatient setting,²⁹ and those hospitalized for ischemic stroke,³⁰ found Black persons were more likely to receive warfarin and less likely to receive DOACs than White persons. Among those receiving DOACs, Black and Hispanic persons were at higher risk than White persons of ischemic stroke and bleeding 1 year after hospitalization.³⁰ However, the relationship between DOAC use and dementia risk across diverse population subgroups has not been quantified.

DOACs may present an opportunity to improve dementia health equity, yet the relationship between DOAC use and dementia in representative populations has not been systematically explored. Our study quantified population-level rates of dementia risk associated with DOAC use among persons with incident AF. We used retrospective study of cohorts of persons with incident AF and differential access to DOACs driven by timing of their market entry and subsequent diffusion. We hypothesized that if DOACs are superior to warfarin in protecting against dementia, we would expect a decrease in dementia incidence as their use in the AF population increases over time. Results from this analysis will have important implications in the treatment of AF, particularly among those at higher risk of stroke and/or dementia.

Methods

Data and Sample

The data for this study came from a 100% sample of Medicare Fee-for-Service Parts A, B, and Part D claims from 2007 to 2017. The sample included persons 67 and older with incident AF and no dementia in year t. We further included beneficiaries who were enrolled from time t − 2 to t + 3 or death (relative to an incident AF diagnosis in year t). Persons that died after an incident AF diagnosis were included in our analysis. If an individual died without a dementia diagnosis, they remained part of the at-risk population (i.e., without dementia). This ensured that we only captured new diagnoses of AF and subsequent dementia. Data from 2005 to 2006 was used to exclude persons with prevalent AF or dementia for the 2007 and 2008 cohorts, while data through 2021 was used to verify a second claim for dementia for the 2017 cohort (see measure description below). See Figure 1 for sample selection for the 2017 cohort (see eTable 1 for sample selection for all cohorts). Race and ethnicity were ascertained using Research Triangle Institute (RTI) race code variable in the claims data, which is based on an algorithm developed by the RTI. The algorithm uses first and last names, along with Social Security Administration data, to identify race and ethnicity and has been shown to have higher sensitivity in classifying Hispanic and Asian beneficiaries than prior methods.³¹ The final sample of persons with incident AF from 2007 to 2017 consisted of 1,083,338 persons, including 982,647 White, 36,310 Black, 32,914 Hispanic, 18,941 Asian, 2,938 American Indian/Alaska Native beneficiaries, and 9,588 persons with missing or other race/ethnicity.

Figure 1. Sample Selection, 2017 Cohort.

FFS = Fee-for-Service.

Measures

Outcomes

The primary outcome was incident dementia diagnosis. Our method for identifying persons with dementia is described elsewhere.³² Briefly, we required 1 diagnosis of AD and related dementias (ADRDs) defined by the Chronic Conditions Warehouse (using ICD-9 and ICD-10 codes for ADRDs), which was confirmed by a second diagnosis or a diagnosis of dementia symptoms within 2 years or death within 1 year. A full list of diagnosis codes is included in eTable 2. In a secondary analysis, we estimated stroke incidence among users of DOACs and warfarin—one of the mechanisms connecting AF and dementia. Incident stroke was identified using ICD-9 and ICD-10 diagnosis codes for stroke and transient ischemic attack as defined in the Chronic Conditions Warehouse 2019 update.

Anticoagulant and Other Prescriptions

Use of anticoagulants was ascertained using Part D claims for warfarin, dabigatran, rivaroxaban, apixaban, and edoxaban. Beneficiaries were considered users of the drug if they had claims for 2 fills and at least 30 days' supply total within the calendar year. We also reported percent of each cohort taking other drugs commonly prescribed to treat patients with AF—including antiplatelets (aspirin, cilostazol, clopidogrel, dipyridamole, prasugrel, ticagrelor, and vorapaxar) and subcutaneously injected anticoagulants (dalteparin, enoxaparin, fondaparinux, and heparin)—and those taking no prescriptions to treat AF. Users of injected anticoagulants and antiplatelets, and nonusers, were excluded from the main analysis.

Confounders

Models were adjusted for indicators of sex, race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian, American/Alaska Native, missing/other), age group (<75, 75–84, 85+), and comorbidities common to AF and dementia (stroke, acute myocardial infarction, diabetes, hyperlipidemia, and hypertension). Stroke indicators were excluded in the secondary stroke analysis. Indicators of Medicare/Medicaid dual eligibility and low-income subsidy were included as proxies for beneficiaries' socioeconomic status.

Statistical Analysis

We identified cohorts of persons with incident AF and no prior history of dementia (or stroke in the stroke analyses). First, we described the characteristics of each cohort, including their use of DOACs, warfarin, or other drugs to treat AF. For years before the introduction of DOACs to the market (2007–2010), we pooled 2 years together to report cohort sample characteristics. Thereafter, we reported annually (2011–2017). Next, we restricted the sample to only incident AF users of DOACs and warfarin in order to compare incidence over time just among persons under anticoagulation treatment. We estimated multivariable logistic regressions adjusting for confounders listed above and predicted 3-year dementia (or stroke) incidence for each cohort separately. Our design exploited the timing of approval for each of the 4 DOACs: dabigatran (October 2010), rivaroxaban (July 2011), apixaban (December 2012), and edoxaban (January 2015). We reported annual rates of dementia adjusted for the full set of covariates in persons anticoagulated with either warfarin or DOACs, comparing cohorts with no access with DOACs (2007–2010), less access (2011–2014), and full access (2015–2017). We tested the sensitivity of our results to different observation windows for dementia and stroke risk and to the inclusion of the CHA₂DS₂-VASc (congestive heart failure, hypertension, age, diabetes mellitus, prior stroke or TIA or thromboembolism, vascular disease, age, sex category) score—a risk score for newly diagnosed patients with AF.

Standard Protocol Approvals, Registrations, and Patient Consents

The study was reviewed by the University of Southern California's Institutional Review Board and approved under expedited review.

Data Availability

Data for this study comes from Centers for Medicare and Medicaid Services (CMS) claims data. Our Data Use Agreement (DUA) stipulates that we may not share data with other researchers. A researcher can request access to the same data and obtain a DUA through the CMS Data Request Center (resdac.org/research-identifiable-files-rif-requests). In addition, we share annotated code used to create our analytical file and run analyses from this study on GitHub upon publication (github.com/USCSchaefferCenterDataCore).

Results

Sample Characteristics of Incident AF Cohorts

Across cohorts, a total of 1,083,338 persons with incident AF were included in our analysis. The 2007 and 2008 cohort included 171,652 persons with incident AF, with a mean (SD) age 77.7 (6.7). The 2-year cohort was 62.2% female with high levels of comorbid hypertension, 86.6%, and hyperlipidemia, 64.0% (Table 1). Over time, the cohort size and percent with hyperlipidemia grew steadily and included fewer female patients, but rates of hypertension and other comorbid conditions remained similar. The 2011 1-year cohort (n = 89,044) was 60.9% female, 71.8% had comorbid hyperlipidemia, and 88.4% had hypertension. The 2014 cohort was about 10,000 persons larger (n = 99,328), including 56.6% female, 73.8% with comorbid hyperlipidemia, and 88.2% with comorbid hypertension. The final cohort in 2017 was the largest (n = 131,013) and had the least number of female patients, 53.7%. Rates of comorbid hypertension were similar to other cohorts (87.8%), but comorbid hyperlipidemia grew to 75.5% of the cohort. The racial/ethnic composition of the of the incident AF cohorts was similar over time. Among persons in the first cohort, 2007 and 2008, 90.4% were non-Hispanic White, 3.8% non-Hispanic Black, 3.3% Hispanic, 1.7% Asian and 0.3% American Indian/Alaska Native. By 2017, persons in our sample were 90.4% non-Hispanic White, 3.1% non-Hispanic Black, 2.8% Hispanic, 1.8% Asian and 0.2% American Indian/Alaska Native. Other demographic characteristics changed over time; cohorts were younger over the period—the earliest cohort in 2007 and 2008 included 35.4% of persons under 75, whereas 43.9% of the 2017 cohort were under 75.

Table.

Incident AF Cohort Sample Characteristics

Outcomes	Year of incident AF
	2007–2008	2009–2010	2011	2012	2013	2014	2015	2016	2017
Dementia incidence within 3 y of AF diagnosis, age-adjusted, %	13.00	12.80	12.39	12.15	12.49	12.35	12.57	11.98	11.39
Drug use, %
DOACs	0.00	0.09	10.57	13.34	19.58	26.68	31.74	38.01	41.44
Warfarin	47.30	46.54	35.38	34.11	28.74	23.25	19.85	15.49	12.83
Other	10.45	10.65	10.39	9.57	9.17	8.78	8.40	7.85	7.27
Nonuser	42.25	42.72	43.66	42.98	42.51	41.30	40.01	38.65	38.47
Female, %	62.17	60.93	59.89	58.56	57.33	56.57	54.98	54.11	53.69
Race/ethnicity, %
White	90.55	90.68	90.68	90.76	90.94	90.96	91.02	90.59	90.40
Black	3.78	3.50	3.37	3.33	3.27	3.24	3.19	3.17	3.05
Hispanic	3.28	3.18	3.25	3.18	3.02	2.97	2.72	2.86	2.81
Asian	1.65	1.82	1.80	1.75	1.77	1.73	1.72	1.72	1.79
AIAN	0.31	0.26	0.29	0.29	0.30	0.28	0.26	0.24	0.23
Missing/other	0.44	0.56	0.61	0.69	0.71	0.84	1.08	1.41	1.72
Age group, %
<75	35.38	37.37	38.80	39.83	40.17	41.07	42.12	43.15	43.93
75–84	47.43	45.73	44.47	43.70	43.31	42.88	42.16	41.94	41.88
85+	17.19	16.90	16.73	16.47	16.52	16.05	15.71	14.91	14.19
Comorbidities, %
Acute myocardial infarction	4.86	4.55	4.43	4.43	4.39	4.21	4.47	4.72	4.82
Diabetes	33.45	34.34	35.27	35.04	34.83	34.81	34.60	34.37	33.89
Hyperlipidemia	64.01	68.30	71.76	72.81	73.59	73.84	73.94	73.94	75.46
Hypertension	86.59	87.94	88.38	88.57	88.72	88.17	88.12	88.11	87.80
Dual eligible, %	19.35	17.92	17.11	16.25	14.70	13.71	11.82	10.80	10.05
Low-income subsidy, %	5.86	4.99	4.37	3.95	3.47	3.13	2.69	2.44	2.17
N	171,652	168,361	89,044	91,292	95,201	99,328	115,564	121,883	131,013

Abbreviations: AF = atrial fibrillation; AIAN = American Indian or Alaska Native; DOAC = direct oral anticoagulant.

OAC Use Over Time

Use of any oral anticoagulant (warfarin or DOACs) increased slightly over the cohorts from 57.7% in 2007–2008 to 61.5% in 2017 (Table 1). In 2011, the first full year of DOAC availability, the share of persons with newly diagnosed AF taking a DOAC was 10.6%, which increased over the period to 41.4% in 2017. At the same time, use of warfarin decreased significantly, from 35.4% in 2011 to 12.8% in 2017. Before DOAC availability, 2007–2008, use warfarin was 47.3%. Use of other drugs to treat AF (i.e., antiplatelets and subcutaneously injected anticoagulants) decreased over time from 10.5% in the 2007–2008 cohort to 7.3% in 2017. Dementia incidence across incident AF cohorts was relatively similar, but decreased slightly over the period from 13.0% in 2007–2008 to 11.4% in 2017.

Model Estimates of Incident Dementia Diagnosis

Figure 2 displays the results from multivariable analysis of 3-year diagnosed dementia risk in a sample of persons with incident AF and taking either DOACs or warfarin (left y-axis), and the share of persons using DOACs relative to warfarin (right y-axis) each year from 2007 to 2017. The share of the sample taking DOACs was 0.0% in 2007 (i.e., 100% of the sample was using warfarin), rose to 23.0% in 2011, and was 76.4% in 2017. The time trend in dementia incidence (adjusted for covariates) was unchanged over the period for all racial/ethnic populations. Three-year dementia incidence was 9.1% (95% CI 8.9–9.4) among White persons diagnosed with AF in 2007 and 2008 and 8.9% (95% CI 8.7–9.1) in 2017. Throughout the period, dementia incidence was consistently highest for Black persons, followed by American Indian/Alaska Native and White, and lowest for Asian persons. In 2017, 10.9% (95% CI 10.4–11.3) of the Black persons in the cohort developed dementia within 3 years, 9.4% (95% CI 8.0–10.9) of American Indian/Alaska Native, 8.9% (95% CI 8.7–9.1) of White, 8.7% (95% CI 8.2–9.1) of Hispanic, and 6.9% (95% CI 6.4–7.4) of Asian persons. Only incidence among Black and Asian persons was statistically different than that of White persons. Note that persons with incident AF are at higher risk of dementia, thus these incident rates are much higher than observed in the full population. eTable 3 compares dementia incidence among all persons enrolled in traditional Medicare and with those with AF.

Figure 2. Incident Dementia in 3 Years (Left Y-Axis) and Use of DOACs, by Cohort Year and Race/Ethnicity, Fully Adjusted, Among DOAC and Warfarin Users Only.

Postestimation prediction for logistic regressions run separately for each incident AF cohort year. Models adjusted for sex, race/ethnicity, age group, acute myocardial infarction, diabetes, hyperlipidemia, and hypertension, dual eligibility, and low-income subsidy. AF = atrial fibrillation; AIAN = American Indian or Alaska Native; DOAC = direct oral anticoagulant.

Model Estimates of Stroke Risk

Figure 3 shows estimates of the percentage of persons newly diagnosed with AF who experienced an incident stroke in the following 3 years. Unlike dementia, stroke incidence decreased over time for all groups. For example, incidence among White persons with AF decreased from 11.0% (95% CI 10.7–11.4) in 2007 to 9.6% (95% CI 9.3–9.9) and 8.4% (95% CI 8.1–8.6) in 2011 and 2017, respectively. However, the slope of the trend did not change once DOACs were available from 2011 on. That is, among persons with AF on anticoagulants, stroke was consistently decreasing over the period and the increasing use of DOACs did not alter the trend. Rates of stroke were consistently highest among Black persons with AF and lowest among American Indian/Alaska Native and White persons (these 2 lines overlap in the figure). In 2017, 11.1% (95% CI 10.6–11.7) of Black persons with incident AF had a stroke in the subsequent 3 years, followed by 10.4% (95% CI 9.9–11.0) of Hispanic, 9.7% (95% CI 9.0–10.4) of Asian and 8.4% of American Indian/Alaska Native (95% CI 6.9–10.1) and White (95% CI 8.1–8.6) persons.

Figure 3. Incident Stroke in 3 Years (Left Y-Axis) and Use of DOACs, by Cohort Year and Race/Ethnicity, Fully Adjusted, Among DOAC and Warfarin Users Only.

Postestimation prediction for logistic regressions run separately for each incident AF cohort year. Models adjusted for sex, race/ethnicity, age group, acute myocardial infarction, diabetes, hyperlipidemia, and hypertension, dual eligibility, and low-income subsidy. AF = atrial fibrillation; AIAN = American Indian or Alaska Native; DOAC = direct oral anticoagulant.

Sensitivity

We tested the sensitivity of our results to the inclusion of the CHA₂DS₂-VASc. Adding this measure of stroke risk did not change our main results and the estimates were not statistically different than our main analysis for any of the cohorts. For example, 3-year dementia risk in our 2017 cohort was 8.9% (95% CI 8.7–9.1) in our main model and 8.8% (95% CI 8.6–9.0) when adding the CHA₂DS₂-VASc score as a covariate. In addition, we examined whether excluding persons with prior history of stroke changed our main results on dementia, given that those patients with AF may be a lower risk of dementia. Excluding persons with prior stroke did not change the time trends of our main results, but the prevalence level was lower across cohorts. Three-year dementia risk in our 2017 cohort was 8.9% (95% CI 8.7–9.1) in our main model and 7.7% (95% CI 7.5–7.9) when excluding persons with prior history of stroke. eTable 4 includes a comparison of our main results with these analyses. Finally, we tested the sensitivity of the main results to different observation windows. Specifically, we reran our analysis of dementia and stroke risk over 4 years and 5 years and the time trends remained the same—unchanged for dementia and decreasing stroke incidence. Results are available on request.

Discussion

This study used an annual cohort study design to analyze trends in dementia risk before and after the introduction of DOACs to the market for the treatment of AF. We hypothesized that if DOACs are superior to warfarin in protecting against dementia, dementia incidence would decrease over time as DOAC use, relative to warfarin use, increased across cohorts. We found that as use of DOACs increased, 3-year dementia incidence among newly diagnosed patients with AF did not change. The finding was consistent across all racial/ethnic populations. Similarly, 2 studies recently reported no significant differences in dementia between users of DOACs and warfarin in the United Kingdom and Sweden. These results are not consistent with the hypothesis that DOACs are more protective than warfarin because of better overall anticoagulation and studies that found decreased risk of dementia.^12,21,33 For example, one study reported lower risk of DOAC users³³; however, this study used a shorter follow up period (0.7–2.2 years), found less of an effect in older persons, and did not account for individual unobserved differences. Furthermore, there were no changes in stroke risk trends among patients with AF in our sample as a result of greater access to DOACs—suggesting this mechanism linking AF and dementia was also unaffected. Our results suggest that any reduction in dementia risk is a result of oral anticoagulation in general, with no added benefit of DOACs.

Our study opens new lines of inquiry by estimating dementia risk among racially and ethnically representative older persons in the United States. Although we found no difference across racial and ethnic populations in the relationship between increased DOAC use, both dementia and stroke incidences were consistently higher among Black, Hispanic, American Indian, and Alaska Native persons with AF than similar White persons. These findings are consistent with the literature on the so-called AF Paradox, which documents higher prevalence of risk factors and adverse outcomes among under-represented racial and ethnic groups.²⁸ Differences in use of anticoagulants have also been documented, including lower likelihood of prescription receipt and fills of warfarin and lower time in therapeutic range (TTR) among Black patients with AF. Similarly, Black, Hispanic, and Asian persons have been shown to be less likely to receive DOACs than similar White persons.^29,30 While representative of the incident AF population in Medicare, our sample is majority White. If the observed increase in DOAC use was primarily driven by White beneficiaries, we may not see benefits of DOAC use to Black, Hispanic, American Indian, and Alaska Native persons and future study may shed additional light on the relationship between DOACs and dementia in these populations and their role in promoting health equity. Our study design allowed us to abstract from individual differences and report only trends in the full incident AF population. The study also had some limitations. First, we measured drug use as only 2 fills and at least 30 days' supply. We did not require subsequent use. Thus, it is possible that some people in the sample discontinued use. We examined the distribution of days' supply for both DOACs and warfarin in our cohorts and found days supplied was similar, for example, median days supplied was 332 days for DOACs and 316 days for warfarin in our 2015 cohort. In addition, greater TTR for warfarin has been associated with reduced dementia risk. Our study uses fills to identify use and thus could not measure TTR. Our empirical strategy assumes that any differences in days' supply or TTR did not vary significantly across cohorts; future individual-level analyses may consider whether variation in use intensity/adherence affects dementia risk in persons with AF. Furthermore, we included indicators of common comorbidities that are risk factors of AF and dementia (e.g., hyperlipidemia) in our models, but not medications to treat them. Excluding these medications or any other common medications (e.g., those with anticholinergic properties) only affects our results if they affect dementia risk and are taken differentially across warfarin and DOAC users. Given the high comorbid rates and use of these medications in the population, this is not likely to be a significant issue. Similarly, there are AF and dementia risk factors we do not observe in Medicare claims data (e.g., smoking, alcohol abuse). Our analysis implicitly assumes these risk factors are similar across warfarin and DOAC users. Finally, our data come from Traditional Medicare, so they may not be generalizable to the growing population of Medicare Advantage beneficiaries.

As evidence grows linking AF to dementia, early identification of those patients with AF at highest risk of cognitive impairment is important to prevent or slow future cognitive decline. The state of the evidence at present does not allow us to prevent dementia in patients with AF, but it would be important to screen patients with AF for cognitive function and to incorporate risk scoring to guide appropriate use of oral anticoagulation in patients with AF. Dementia risk was indeed higher in persons with incident AF in our sample than the full population; however, we did not find evidence of that risk being mediated by use of DOACs as opposed to warfarin. Oral anticoagulation addresses many of the mechanisms hypothesized to cause dementia in patients with AF, reducing hypercoagulable state seen in AF, ischemia-caused infarcts due to microbleeds or clots, and inflammation-caused blood clots. Because DOACs are associated with less bleeding than warfarin, one would expect there would be less cerebral microhemorrhages; however, we do not find evidence of this. It could be that mechanism does not greatly contribute to dementia in older persons. Future research is needed to test and better understand these hypothesized mechanisms.

Finally, our results have implications for clinical practice. DOACs are indeed easier for patients to adhere to than warfarin, but they are also much costlier. When weighing the risks and benefits of different oral anticoagulants, consideration should be given to similar stroke risk and dementia, particularly among older persons.

Glossary

AD

Alzheimer disease

ADRD

AD and related dementia

AF

atrial fibrillation

CMS

Centers for Medicare and Medicaid Services

DOAC

direct oral anticoagulant

DUA

Data Use Agreement

ICD-9

International Classification of Diseases, Ninth Revision

ICD-10

International Classification of Diseases, Tenth Revision

RTI

Research Triangle Institute

TTR

time in therapeutic range

Appendix. Authors

Name	Location	Contribution
Johanna Thunell, PhD	Schaeffer Center for Health Policy and Economics, and Sol Price School of Public Policy, University of Southern California, Los Angeles	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Kathryn Wood, PhD, RN	Woodruff School of Nursing, Emory University, Atlanta, GA	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Whitney Wharton, PhD	Department of Neurology, School of Medicine, Emory University, Atlanta, GA	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design
Geoffrey Joyce, PhD	Schaeffer Center for Health Policy and Economics, and Mann School of Pharmacy, University of Southern California, Los Angeles	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Patricia Ferido, MS	Schaeffer Center for Health Policy and Economics, University of Southern California; Sol Price School of Public Policy, Los Angeles, CA	Analysis or interpretation of data
Julie Zissimopoulos, PhD	Schaeffer Center for Health Policy and Economics, and Sol Price School of Public Policy, University of Southern California, Los Angeles	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data

Study Funding

This work is supported by NIA grant P30AG066589.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.