Frailty and subsequent adverse outcomes in older patients with atrial fibrillation treated with oral anticoagulants: The Shizuoka study

Background In older patients with atrial fibrillation (AF), frailty is frequently prevalent. However, the prognostic value of frailty for adverse events after initiation of oral anticoagulants (OACs) is unclear. Objectives We assessed whether frailty at the time of OAC initiation is associated with subsequent bleeding or embolic events. Methods We extracted patients aged ≥65 years with nonvalvular AF in whom OACs were initiated from a universal administrative claims database incorporating primary and hospital care records in Shizuoka, Japan, between 2012 and 2018. Frailty was assessed using the electronic frailty index (eFI). The association of frailty with bleeding events and ischemic stroke/transient ischemic attack were evaluated using the Fine-Gray model and restricted cubic spline model. Results Among 12,585 patients with AF, 7.8% were categorized as fit, 31.5% as mildly frail, 34.8% as moderately frail, and 25.9% as severely frail. The risk of bleeding was associated with a higher eFI (adjusted subdistribution hazard ratio [95% CI] for fit or mild frailty: 1.15 [1.02-1.30]; moderate frailty: 1.42 [1.24-1.61]; and severe frailty: 1.86 [1.61-2.15]), whereas the association was weaker for ischemic stroke/transient ischemic attack. The spline models demonstrated that the relative hazard for bleeding increased steeply with increasing eFI. Conclusion Patients with frailty in whom OAC therapy is initiated have higher risk of bleeding, highlighting the importance of discussing this increased risk with patients with AF who have frailty and assessing frailty at the time of OAC initiation.


| I N T R O D U C T I O N
Atrial fibrillation (AF) is associated with a substantial risk of systemic embolic events [1,2], and the risk is particularly high in older patients with AF [2,3]. Oral anticoagulants (OACs) have been shown to reduce the rate of embolic complications, and their use is strongly supported in international clinical practice guidelines [4,5]. However, they remain underused in clinical practice, particularly for frail patients [6][7][8], despite recommendations to use OACs regardless of the frailty status [9,10].
The efficacy of OACs is assumed to be achieved at the expense of an increased risk of bleeding. Frailty is an age-related cumulative decline in physiologic systems and is known to be associated with higher risk of falls, fractures, gastrointestinal bleeding, and cerebral trauma, frequently precluding the initiation of OACs [11,12].
However, the precise association between frailty and subsequent embolic or bleeding events among patients with AF treated with OACs in the community population is unknown. Previous studies that evaluated the association of frailty with outcomes in patients with AF treated with OACs were predominantly small-scale studies or their assessments were not made in community settings [13][14][15].
Furthermore, there are gaps between knowledge from randomized controlled trials (RCTs) and that from real-world clinical practice [16] because RCTs often include patients with fewer chronic medical conditions, less frailty, and lower risk of adverse outcomes than the real-world population.
We, therefore, investigated the association between frailty and clinical outcomes among patients with AF treated with OACs in a large community-based cohort derived from administrative claims data, incorporating primary and hospital care records, in Japan.

| Study design and data source
We conducted a cohort study using the Shizuoka Kokuho Database (SKDB) between April 2012 and September 2018 [17]. SKDB is an administrative claims database of beneficiaries in the municipal government insurance program (national health insurance and late-stage medical care system for the elderly) in Shizuoka Prefecture, Japan. All data in SKDB were anonymized. Of all residents aged <75 years, 22.3% are enrolled in the national health insurance system (eg, selfemployed and unemployed), and all residents aged ≥75 years are enrolled in the late-stage medical care system for the elderly. We utilized basic information (ie, age, sex, and date of death) and health insurance claims (eg, monthly claims for patients' diagnoses, procedures, laboratory tests ordered, drugs dispensed, and dates of hospital admissions) from SKDB. This study was approved by the Ethnic Committee of Shizuoka Graduate University of Public Health (Shizuoka, Japan) (#SGUPH_2021_001_006).

| Study population and follow-up
We selected patients with nonvalvular AF in whom OAC therapy (warfarin or DOACs: apixaban, dabigatran, edoxaban, or rivaroxaban) was initiated between April 2013 and March 2018. We designated the month of dispensation of the first OAC as the "index month." We included patients aged ≥65 years in whom OAC monotherapy had been initiated in an outpatient setting and who had at least 1 record of AF diagnosis (International Classification of Diseases, 10th Revision, code I48x but not the disease code for valvular AF or AF after surgery) in the preceding 12 months. We excluded patients who had been continuously enrolled in an insurance plan for <12 months (baseline period) before OAC initiation or with alternate potential indications for OAC therapy besides nonvalvular AF based on previous diagnoses of the following conditions: venous thromboembolism, rheumatic mitral valve disease, intracardiac thrombosis, mechanical or bioprosthetic heart valve, mitral valve repair, or valvular AF (Supplementary Table S1). Follow-up of outcomes started in the index month (first OAC prescription). Each patient was censored at the occurrence of the outcome of interest, death, end of enrollment in the plan, or end of the study period (September 2018), whichever came first. Embolic or bleeding events were not competing risks for each outcome of interest. For example, if bleeding occurred before an embolic event, the patient was not censored at the time of bleeding but at the time of the embolic event.

| Patient characteristics
Information on the demographics (ie, sex and age) of each patient was extracted from the insurance subscriber list. We assessed the patients' baseline comorbidities and previous medication use using recorded diagnoses and dispensation claims during the 12 months preceding the index month and in the index month ( Figure 1; Supplementary   Table 2A, B). We also collected data on the daily dose of DOACs in

Essentials
• The prognostic value of frailty for adverse events after initiation of oral anticoagulants (OACs) is unclear.
• The association of frailty with bleeding and embolic events was evaluated.
• Patients with frailty in whom OAC therapy is initiated have a higher risk of bleeding.
• The increased risk of bleeding in frail patients with AF should be shared with these patients. the index month among patients initiating DOAC therapy. Drug dosing was classified as reduced dose (apixaban, 2.5 mg twice a day; dabigatran, 110 mg twice a day; edoxaban, 30 mg every day; and rivaroxaban, 10 mg every day) or standard dose (apixaban, 5 mg twice a day; dabigatran, 150 mg twice a day; edoxaban, 60 mg every day; and rivaroxaban, 15 mg every day). The approved dose of rivaroxaban in Japan was 10 mg once daily for patients with a creatinine clearance rate of 15 to 49 mL/min or 15 mg once daily for patients with a creatinine clearance rate of ≥50 mL/min, based on pharmacokinetic modeling data and the results of the J-ROCKET AF trial (ie, a trial compared the safety of a Japan-specific rivaroxaban dose with warfarin) [18]. Dose data of patients in whom neither the standard nor reduced dose was initiated were treated as missing.
To determine the risk scores for bleeding and thromboembolism, we utilized the HAS-BLED and CHA 2 DS 2 -VASc scores, respectively.
The scores were calculated based on diagnosis claims recorded in the index month and the previous 12 months, using the following factors: age over 65 years, hypertension, abnormal renal and liver function, prior stroke, bleeding history or predisposition, labile international normalized ratio, and drugs/alcohol concomitantly for HAS-BLED; and congestive heart failure, hypertension, age of 75 years or older, diabetes mellitus, stroke, valvular disease, age 65 to 74 years, and sex category for CHA 2 DS 2 -VASc (Supplementary Table 3A, B). The labile international normalized ratio, a component of the HAS-BLED score, was excluded from the calculation because this datum was unavailable in SKDB.

| Frailty assessment
We evaluated the patients' frailty using the electronic frailty index (eFI) [19]. eFI is a coding-based algorithm based on the cumulative deficit model as the theoretical framework. This algorithm was developed from electronic health records in the United Kingdom. The cumulative deficit model is an accumulation of age-related deficits: signs, symptoms, diseases, disabilities, and polypharmacy. eFI was validated for relevant outcomes (eg, mortality, nursing home admission, and frailty phenotype). Because 1 of the 36 variables included in the index (ie, falls) is not available in the Japanese claims coding system, we calculated the score based on 35 variables using the records of diagnoses, according to the International Classification of Diseases, 10th Revision, codes, and the dispensation records for the previous 12 months and the index month (Supplementary Table 4) [20]. eFI is used to categorize patients into 4 groups: fit (eFI score of 0-0.12), mildly frail (>0.12-0.24), moderately frail (>0.24-0.36), and severely frail (>0.36) [19].

| Outcomes
The outcomes of interest were bleeding events and embolic events defined using diagnostic codes. The primary bleeding outcome was a composite outcome of major and minor bleeding. Major bleeding was defined as events including intracranial bleeding, gastrointestinal bleeding, or bleeding with shock in an inpatient setting. Minor bleeding was defined as other bleeding events that are not classified as major bleeding, recorded in an inpatient or outpatient setting (Supplementary Table 5). The embolic outcome was a secondary outcome, including ischemic stroke and transient ischemic attack (TIA), diagnosed in an inpatient setting. Major bleeding was also assessed as a secondary outcome.

| Statistical analysis
To describe patients' characteristics and the prevalence of the 35 deficits evaluated to calculate eFI, continuous variables are presented as medians and IQRs and categorical variables are presented as numbers and percentages. There were no missing data, except for data on dose category. To assess the associations of bleeding and thromboembolism risk scores with frailty, we tabulated and displayed the HAS-BLED and CHA 2 DS 2 -VASc scores according to the eFI groups. Spearman correlation coefficients of 2 risk scores with eFI were also estimated. We compared the frequencies of the deficits used to calculate eFI across the categories of both these scores (HAS-BLED and CHA 2 DS 2 -VASc). We assessed the percentage of patients on DOACs in whom treatment with a reduced-dose regimen was initiated according to the eFI groups.
We evaluated the outcomes using cumulative incidence functions for up to 64 months and compared the curves across eFI categories using the Gray test. Death was considered a competing risk. We evaluated the association between frailty and outcomes using univariable and multivariable Fine-Gray subdistribution hazard models after adjustment for the following covariates: sex, baseline comorbidities (cancer; chronic kidney disease; chronic obstructive pulmonary disease; depression; diabetes mellitus; heart failure; hypertension; liver disease; peptic ulcer; peripheral arterial disease; previous admission for myocardial infarction, bleeding, and stroke; rheumatoid arthritis; and sleep apnea syndrome), and the use of medications (antihypertensive drugs, antidiabetic drugs, nitrates, F I G U R E 1 Study design timeline. CHA 2 DS 2 -VASc, congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, stroke, valvular disease, age 65 to 74; HAS-BLED, age>65 years, hypertension, abnormal renal and liver function, prior stroke, bleeding history or predisposition, labile international normalized ratio, and drugs/alcohol concomitantly. NISHIMURA ET AL.   with the exception of eFI components that were not related to stroke. To depict the associations between eFI as a continuous variable and the outcomes, we constructed restricted cubic spline models with 4 knots at the quintile points, adjusted for sex, baseline comorbidities, and the use of medications. We used an eFI score of 0.12, which is the threshold between fit and mild frailty, as the reference point. All statistical analyses were performed using SAS, version 9.4 (SAS Institute).  The frequency of each deficit used to calculate eFI was higher among those in the severely frail category (Supplementary Table 6). The prevalence of each of the 35 deficits used to calculate eFI, not only those included in the bleeding and stroke risk scores (eg, hypertension, diabetes mellitus, and cerebrovascular disease) but also frailty-specific deficits (eg, activity limitation, fragility fracture, weight loss, and anorexia), was higher in patients with higher risk scores ( Supplementary   Figures 3 and 4). Patients in the higher eFI categories were more likely to receive reduced-dose DOACs than the fit patients ( Figure 2).
The annual event rates among patients who were fit and those who were severely frail were 11.2 and 21.0 per 100 patient-years for bleeding, respectively, 1.0 and 1.7 per 100 patient-years for ischemic stroke/TIA, respectively, and 0.7 and 1.5 per 100 patient-years for major bleeding, respectively ( Table 2). Figure 3 shows the cumulative incidence functions for the primary and secondary bleeding outcomes compared across the frailty categories. The cumulative incidence curves for the frailty categories were substantially separated for the bleeding end points but minimally different for ischemic stroke/TIA.
For major bleeding, the cumulative incidence curves for the moderately and severely frail patients differed only slightly but differed considerably from the curve for the fit patients.  Table 8

| D I S C U S S I O N
In this contemporary population-based study of older patients with AF aged ≥65 years who had initiated OAC therapy, the prevalence of frailty was common in the community setting, and 26% of the subjects were severely frail. We found a strong association between frailty and increased risk of bleeding. Our results suggest that the increased risk of bleeding in frail patients with AF should be shared with these patients.
Overall, 26% of patients with AF aged ≥65 years on OACs were severely frail, whereas 7.8% were fit. A systematic review reported that the prevalence of frailty among patients with AF was 50.4% to 75.4%, depending on the age distribution of the target and how frailty was defined [18]. A self-reported frailty scale and other frailty indices were included in these definitions, but no frailty measure based on health care data was reported. However, the distribution of eFI and the median age in our study were similar to those in a previous study that included patients with AF in a primary care setting (median age, 80 years; severe frailty, 23%; and fit, 10.5%) [12].
F I G U R E 2 Associations between frailty and direct oral anticoagulant dose. DOAC, direct oral anticoagulant; eFI, electronic frailty index.
We identified a considerable association between frailty and increased risk of bleeding. A higher risk of bleeding among frail patients with AF has been a concern when initiating OAC [21,22], and previous studies have reported a positive association between frailty and bleeding among patients with AF. This association has been most evident in observational, real-world studies [12,13,22,23]. Singlecenter studies from Italy [13] and Japan [22] as well as a post hoc analysis from an RCT [15] found that frailty was associated with major bleeding, defined based on International Society on Thrombosis and Hemostasis criteria. Although large-scale and community-setting studies were limited, a study from the United Kingdom using primary care records reported the association between frailty and gastrointestinal bleeding [12]. Although the definitions of frailty and bleeding have variations, the result of the association between frailty and bleeding in the present study is consistent with those in previous studies. Several observational studies have reported that age is an independent predictor of stroke but not of bleeding in older patients with AF [24,25]. Our observation of an association between frailty and bleeding indicates that frailty should be considered a risk factor for bleeding, as great as or greater than chronologic age. Although "falls," which is also included as a predictor in eFI (but was not available in the present study), is a component of frailty, a prospective cohort study found that a high risk of falls was not associated with major bleeding [26]. Frailty, a cumulative decline in physiologic systems, may contribute to bleeding rather than the high risk of fall itself.
Importantly, our study does not support the withholding or with-   [24,25,29]. Moreover, frailty should be incorporated in risk assessment as part of shared decision making for OAC initiation. The high bleeding risk in frail patients must be recognized.
In the present study, we found that patients with moderate-tosevere frailty had a limited increase in the incidence of ischemic stroke/TIA compared with those who were not frail after OAC initiation. The association between frailty and stroke is controversial [23,29]. An unclear association between frailty and increased risk of stroke in patients with AF has predominantly been reported in studies conducted in primary care settings [12,30]. Other studies that have reported an association between frailty and stroke in patients with AF were prospective cohort studies or post hoc analyses from RCTs [12,15,24,25,31,32]. In our study of patients with AF aged ≥65 years (median age, 80 years), frailty was not strongly associated with ischemic stroke/TIA after adjustment for sex, comorbidities, and medication use. However, when indicators of eFI were not included in the adjustments, frailty was found to be associated with ischemic stroke/TIA. This suggests that the association between frailty and stroke events was weakened by adjusting for comorbidities and concomitant medication use, which are also included in eFI.
In this study, which was based on a contemporary claims database, DOACs were more frequently prescribed (n = 10,340; 82%) than warfarin and more frequently prescribed than in previous studies [12,15,16,22,31,32]. A recent meta-analysis of RCTs and an observational study showed that DOACs were preferable over warfarin in terms of the risk of bleeding [4,16,33]. In the United States clinical practice guidelines [4], DOACs were strongly recommended as firstline therapy in preference to warfarin among eligible patients [4,5].
The proportion of DOAC prescriptions has increased rapidly since the entry of DOACs into the market in Japan, the United Kingdom, and the United States [34][35][36][37]. The current prescription trends must be taken into consideration when interpreting our findings.

| Limitations
This study should be interpreted in the context of several limitations.
First, the validity of disease coding in administrative claims databases has been reported to have suboptimal accuracy. However, in a previous study, we found that the eFI calculated from Japanese claims accurately predicted mortality and the use of long-term care services.

| C O N C L U S I O N
Increased severity of frailty is associated with a higher risk of bleeding. The increased risk of bleeding in frail patients with AF should be shared with these patients, and personalized management strategies should be implemented to minimize the bleeding risk.

DATA AVAILABILITY
Based on the data use agreement with the regional insurers in the prefecture of Shizuoka, we are unable to make the analytical data accessible to readers.