See related article, p 3459

Risk factors for nonvalvular atrial fibrillation (AF) include age, male sex, hypertension, diabetes, obesity, and ischemic heart disease, and AF is often accompanied by frailty and multimorbidity.¹ AF increases the likelihood of stroke and also dementia, the latter occurring largely as a result of stroke and also shared risk factors.^1–3 The relative risks of dementia appear the highest in younger people with long exposure to AF.^2,3 Oral anticoagulants (OACs) substantially reduce AF-associated stroke and death in high-risk individuals.¹ Available data suggest that OACs also impact dementia: rates of AF-associated dementia have fallen over the last 3 decades⁴ with increasing OAC use, and increased time in therapeutic range is protective, consistent with prevention of clinical and subclinical (silent) stroke.^5–7 In the acute stroke population, AF has inconsistently been related to new poststroke dementia possibly because of variation in adjustment for baseline stroke severity and other confounders, and high rates of poststroke anticoagulation.^8,9

Until recently, vitamin K antagonists including warfarin were the mainstays of OAC treatment for nonvalvular AF, but these are rapidly being supplanted by direct oral anticoagulants (DOAC)s.¹⁰ DOACs do not require regular blood testing and dose adjustment making them much simpler to administer and more attractive especially to older patients for whom regular blood testing may be difficult. DOACs also have a more predictable anticoagulant effect than warfarin being less often affected by other medications, diet, and acute illness with a lower risk of serious bleeding including intracranial bleeding.¹¹ It should be noted, however, that the short half life of DOACs mean that there is a rapid loss of anticoagulation even after omission of a single dose whereas warfarin effects disappear more gradually over a few days. DOACs are also not straightforward to reverse and are contraindicated in severe liver or kidney failure.

Given the increasing use of DOACs, it is important to understand their effect on risk of future dementia in comparison to vitamin K antagonists. Existing data suggest that DOACs may prevent more strokes than warfarin, and this may be particularly the case in real-world use of warfarin outside clinical trials where time in therapeutic range may be reduced. It might therefore be expected that DOACs would provide better dementia prevention than warfarin. However, previous studies have not consistently shown advantage for DOACs over warfarin although one study showed a benefit for dabigatran, rivaroxaban, and apixaban but with no individual DOAC being clearly superior over another.¹²

In the current issue of Stroke, Lee et al¹³ present the findings of a large retrospective observational study comparing the incidence of new dementia in patients on warfarin versus DOACs (rivaroxaban, dabigatran, apixaban, or edoxaban). The study used a large real-world sample from the nationwide South Korean medical claims database with careful adjustment for baseline differences between DOAC and warfarin groups. Patients prescribed DOAC were older and had higher CHADS-VASc scores than those prescribed warfarin suggesting that the practical considerations described earlier did indeed influence DOAC prescription. The use of OACs was identified using pharmacy prescription records and only initially OAC naive patients were included avoiding the inclusion of patients started on warfarin but switched to DOAC because of increasing frailty or cognitive impairment. The primary analysis was the effect of DOACs (all 4 combined) versus warfarin on incident dementia, which was further subclassified as Alzheimer’s or vascular dementia. Overall, they found that incidence of dementia was similar in patients on DOACs and warfarin, for all dementia and for the different dementia subtypes. Subgroup analyses showed a greater protective effect of the pooled DOACs versus warfarin in those with prior stroke, and younger old people aged 65 to 74 years. Considering individual DOACs separately versus warfarin, dementia prevention was superior for edoxaban, and for edoxaban and rivaroxaban after censoring for incident clinical stroke on follow-up. Both rivaroxaban and edoxban versus warfarin reduced the rates of vascular dementia.

It is important to note that in this study and many such Big Data studies, the definitions of both covariates and dementia outcomes were based on diagnostic (International Classification of Diseases-Tenth Revision [ICD-10]) codes.¹⁴ Use of such secondary data in research has the advantage of obtaining large representative samples at relatively low cost and at no additional burden to patients. However, there are some important issues to consider in the use of ICD-10 diagnostic coding, particularly in the ascertainment of dementia. To code patient episodes, relevant data are extracted by the coder from real-world clinical records, and ICD-10 codes are then allocated using strict criteria. The responsibility for coding varies according to health care setting and country and ranges from the managing senior medical professional (eg, United States) to trained administrative staff working independent of the clinical team (eg, United Kingdom). The accuracy of coding in general is therefore dependent on the condition being coded, the case-mix (coding is more challenging in older multimorbid patients), quality of the clinical record and the time, resources and training of staff.

Studies on the accuracy of diagnostic dementia coding against the clinical Gold standard are few and are largely confined to United States and UK health systems, but available data suggest that dementia is under-coded: sensitivity is around 70%, although specificity is high, and coding tends to be more often applied to advanced versus milder dementia.^15–17 However, coding can only be applied to cases of diagnosed dementia and since the condition is substantially under-diagnosed both in community and hospital settings,^18,19 the sensitivity of dementia diagnostic coding for the true prevalence of dementia may be much lower at around 30% to 40%. Lee et al used dementia medication prescription as an adjunct to identification of dementia using diagnostic coding, but this does not avoid the problem of under-diagnosis. Further, dementia subtyping using diagnostic coding will be subject to substantial uncertainty since clinical subtyping is challenging and documentation will be inconsistent.

Therefore, in the Lee study,¹³ use of diagnostic coding to ascertain dementia will likely have resulted in inclusion of some individuals with dementia at baseline and under-detection of new dementia on follow-up as well as inaccuracy in subtyping as acknowledged by the authors. It should be noted that previous studies of DOAC and dementia used similar methodology. These factors will have reduced the overall power of studies to detect differences in dementia incidence between groups. Substantial bias is unlikely although some patients with undocumented cognitive impairment may have been more likely to be prescribed a DOAC leading to residual confounding despite adjustment for baseline differences. To try and address this issue and to allow for the fact that dementia is a condition that develops insidiously over time, Lee et al undertook sensitivity analyses excluding dementia occurring within the first year which did not change the results.

Notwithstanding the uncertainties around coding for dementia subtype, Lee et al nevertheless found that in the comparisons of individual DOACs with warfarin, rivaroxaban and edoxaban reduced the incidence of vascular dementia.¹³ These DOACs were also associated with lower incidence of all dementia in the absence of incident stroke on follow-up. Taken together with the observed lower dementia incidence in those with prior stroke, the findings suggest that DOACs may be superior to warfarin in real-world prevention of vascular dementia through reduction in clinical stroke but also of subclinical stroke and microembolism. If so, the superiority of DOACs over warfarin in prevention of dementia in nonvalvular AF may become more apparent in the longer term: follow-up in the Lee study was relatively short at a median of just over 1 year. Further studies are required, including studies with prospective clinical ascertainment of dementia, to determine the effects of DOACs on long-term dementia risk and whether specific DOACs are superior. In the context of a global aging population, optimal treatment of AF constitutes a key intervention in reducing the burden of dementia.

Disclosures None.

The opinions expressed in this article are not necessarily those of the editors or of the American Heart Association.

This manuscript was sent to Elizabeth Warburton, Guest Editor, for review by expert referees, editorial decision, and final disposition.

For Disclosures, see page 3470.

参见相关文章，第 3459 页

非瓣膜性心房颤动 (AF) 的危险因素包括年龄、男性、高血压、糖尿病、肥胖和缺血性心脏病，而 AF 常伴有虚弱和多发病。¹ AF 增加了中风和痴呆的可能性，后者的发生主要是中风的结果，也是共同的危险因素。^1-3长期暴露于 AF 的年轻人患痴呆症的相对风险似乎最高。^2,3口服抗凝剂 (OAC) 可显着减少高危人群的 AF 相关中风和死亡。¹现有数据表明，OAC 也会影响痴呆症：过去 3 年中，房颤相关痴呆症的发病率有所下降⁴随着 OAC 使用的增加和治疗范围内时间的增加，具有保护作用，与临床和亚临床（无症状）中风的预防一致。^5-7在急性卒中人群中，AF 与新发卒中后痴呆的相关性不一致，这可能是因为对基线卒中严重程度和其他混杂因素的调整存在差异，以及卒中后抗凝治疗的比率较高。^8,9

直到最近，包括华法林在内的维生素 K 拮抗剂还是非瓣膜性 AF 的 OAC 治疗的主要药物，但这些药物正迅速被直接口服抗凝剂 (DOAC) 所取代。¹⁰ DOAC 不需要定期血液检测和剂量调整，这使得它们更易于给药且更具吸引力，尤其是对难以定期进行血液检测的老年患者而言。与华法林相比，DOAC 还具有更可预测的抗凝作用，受其他药物、饮食和急性疾病的影响较小，严重出血（包括颅内出血）的风险较低。¹¹然而，应该指出的是，DOAC 的半衰期短意味着即使在省略单次剂量后，抗凝作用也会迅速丧失，而华法林的作用会在几天内逐渐消失。DOACs 也不能直接逆转，严重肝或肾功能衰竭时禁用。

鉴于 DOAC 的使用越来越多，与维生素 K 拮抗剂相比，了解它们对未来痴呆风险的影响非常重要。现有数据表明，与华法林相比，DOAC 可以预防更多的中风，在临床试验之外实际使用华法林的情况下尤其如此，在这种情况下，治疗范围内的时间可能会缩短。因此，可以预期 DOAC 将提供比华法林更好的痴呆预防。然而，之前的研究并未一致显示 DOAC 优于华法林，尽管一项研究显示达比加群、利伐沙班和阿哌沙班有益处，但没有单个 DOAC 明显优于另一项。¹²

在当前一期的 Stroke 中，Lee 等人¹³介绍了一项大型回顾性观察研究的结果，该研究比较了华法林与 DOAC（利伐沙班、达比加群、阿哌沙班或依度沙班）患者新发痴呆的发生率。该研究使用了来自韩国全国医疗索赔数据库的大型真实世界样本，并仔细调整了 DOAC 和华法林组之间的基线差异。开具 DOAC 处方的患者年龄较大且 CHADS-VASc 评分高于开具华法林处方的患者，这表明之前描述的实际考虑确实影响了 DOAC 处方。使用药房处方记录确定 OAC 的使用，并且仅包括最初未使用 OAC 的患者，避免包括开始使用华法林但由于虚弱或认知障碍增加而改用 DOAC 的患者。主要分析是 DOAC（所有 4 种组合）与华法林对痴呆发生的影响，后者进一步细分为阿尔茨海默病或血管性痴呆。总体而言，他们发现，对于所有痴呆症和不同的痴呆症亚型，使用 DOAC 和华法林的患者的痴呆症发生率相似。亚组分析显示，合并后的 DOAC 与华法林相比，在既往卒中患者和 65 至 74 岁的年轻老年人中具有更大的保护作用。单独考虑单独的 DOAC 与华法林，在对随访中的临床卒中事件进行审查后，依度沙班、依度沙班和利伐沙班预防痴呆的效果更好。利伐沙班和依度班与华法林相比均降低了血管性痴呆的发生率。进一步细分为阿尔茨海默病或血管性痴呆。总体而言，他们发现，对于所有痴呆症和不同的痴呆症亚型，使用 DOAC 和华法林的患者的痴呆症发生率相似。亚组分析显示，合并后的 DOAC 与华法林相比，在既往卒中患者和 65 至 74 岁的年轻老年人中具有更大的保护作用。单独考虑单独的 DOAC 与华法林，在对随访中的临床卒中事件进行审查后，依度沙班、依度沙班和利伐沙班预防痴呆的效果更好。利伐沙班和依度班与华法林相比均降低了血管性痴呆的发生率。进一步细分为阿尔茨海默病或血管性痴呆。总体而言，他们发现，对于所有痴呆症和不同的痴呆症亚型，使用 DOAC 和华法林的患者的痴呆症发生率相似。亚组分析显示，合并后的 DOAC 与华法林相比，在既往卒中患者和 65 至 74 岁的年轻老年人中具有更大的保护作用。单独考虑单独的 DOAC 与华法林，在对随访中的临床卒中事件进行审查后，依度沙班、依度沙班和利伐沙班预防痴呆的效果更好。利伐沙班和依度班与华法林相比均降低了血管性痴呆的发生率。亚组分析显示，合并后的 DOAC 与华法林相比，在既往卒中患者和 65 至 74 岁的年轻老年人中具有更大的保护作用。单独考虑单独的 DOAC 与华法林，在对随访中的临床卒中事件进行审查后，依度沙班、依度沙班和利伐沙班预防痴呆的效果更好。利伐沙班和依度班与华法林相比均降低了血管性痴呆的发生率。亚组分析显示，合并后的 DOAC 与华法林相比，在既往卒中患者和 65 至 74 岁的年轻老年人中具有更大的保护作用。单独考虑单独的 DOAC 与华法林，在对随访中的临床卒中事件进行审查后，依度沙班、依度沙班和利伐沙班预防痴呆的效果更好。利伐沙班和依度班与华法林相比均降低了血管性痴呆的发生率。

值得注意的是，在本研究和许多此类大数据研究中，协变量和痴呆结果的定义均基于诊断（国际疾病分类 - 第十次修订 [ICD-10]）代码。¹⁴在研究中使用此类二手数据的优势在于，可以以相对较低的成本获得大量具有代表性的样本，而且不会给患者带来额外的负担。然而，在使用ICD-10诊断编码时需要考虑一些重要问题，特别是在确定痴呆症方面。为了对患者发作进行编码，编码员从真实世界的临床记录中提取相关数据，并使用ICD-10然后使用严格的标准分配代码。编码的责任因卫生保健环境和国家而异，范围从管理高级医疗专业人员（例如，美国）到独立于临床团队工作的训练有素的行政人员（例如，英国）。因此，编码的准确性通常取决于所编码的病症、病例组合（编码在老年多病患者中更具挑战性）、临床记录的质量以及工作人员的时间、资源和培训。

针对临床黄金标准诊断痴呆编码准确性的研究很少，主要限于美国和英国的卫生系统，但现有数据表明痴呆编码不足：敏感性约为 70%，尽管特异性很高，并且编码往往更常应用于晚期而非轻度痴呆。^15-17然而，编码只能应用于诊断出的痴呆症病例，并且由于该病症在社区和医院环境中^均未得到充分诊断，^18,19痴呆症诊断编码对痴呆症真实患病率的敏感性可能低得多，约为 30% 至 40%。Lee 等人使用痴呆症药物处方作为使用诊断编码识别痴呆症的辅助手段，但这并不能避免诊断不足的问题。此外，使用诊断编码进行痴呆亚型分型将受到很大的不确定性，因为临床亚型分型具有挑战性，并且文档不一致。

因此，在 Lee 研究中，¹³使用诊断编码来确定痴呆症可能会导致纳入一些基线时患有痴呆症的个体，并且在随访中未能检测到新的痴呆症，以及作者承认的亚型分类不准确。应该指出的是，以前对 DOAC 和痴呆症的研究使用了类似的方法。这些因素将降低研究检测组间痴呆发病率差异的整体能力。尽管对基线差异进行了调整，但一些未记录的认知障碍患者可能更有可能被开具 DOAC 处方，导致残留混杂，但不太可能出现实质性偏倚。为了尝试解决这个问题并考虑到痴呆症是一种随着时间的推移而悄悄发展的状况，

尽管痴呆亚型编码存在不确定性，但 Lee 等人发现，在将个体 DOAC 与华法林进行比较时，利伐沙班和依度沙班降低了血管性痴呆的发生率。¹³这些 DOAC 还与在随访中没有发生卒中的情况下所有痴呆的发病率较低有关。结合观察到的既往卒中患者的痴呆发病率较低，研究结果表明，在通过减少临床卒中以及亚临床卒中和微栓塞来预防血管性痴呆的现实世界中，DOAC 可能优于华法林。如果是这样，从长远来看，DOACs 在预防非瓣膜性 AF 痴呆方面优于华法林可能会变得更加明显：Lee 研究的随访时间相对较短，中位时间仅为 1 年多。需要进一步的研究，包括对痴呆症进行前瞻性临床确定的研究，以确定 DOACs 对长期痴呆症风险的影响以及特定的 DOACs 是否更好。

披露无。

本文中表达的观点不一定是编辑或美国心脏协会的观点。

这份手稿已发送给客座编辑伊丽莎白·沃伯顿 (Elizabeth Warburton)，以供专家审阅、编辑决定和最终处置。

有关披露，请参阅第 3470 页。