Differential effect of anticoagulation according to cognitive function and frailty in older patients with atrial fibrillation

Abstract

Background:

In older patients with atrial fibrillation (AF), cognitive impairment and frailty are prevalent. It is unknown whether the risk and benefit of anticoagulation differ by cognitive function and frailty.

Methods:

A total of 1,244 individuals with AF with age ≥ 65 years and a CHADSVASC score ≥ 2 were recruited from clinics in Massachusetts and Georgia between 2016–18 and followed until 2020. At baseline, frailty status and cognitive function were assessed. Hazard ratios of anticoagulation on physician adjudicated outcomes were adjusted by the propensity for receiving anticoagulation and stratified by cognitive function and frailty status.

Results:

The average age was 75.5 (± 7.1) years, 49% were women, and 86% were prescribed oral anticoagulants. At baseline, 528 (42.4%) participants were cognitively impaired and 172 (13.8%) were frail. The adjusted hazard ratios of anticoagulation for the composite of major bleeding or death were 2.23 (95% confidence interval: 1.08 – 4.61) among cognitively impaired individuals and 0.94 (95% confidence interval: 0.49 – 1.79) among cognitively intact individuals (P for interaction = 0.08). Adjusted hazard ratios for anticoagulation were 1.84 (95% confidence interval: 0.66–5.13) among frail individuals and 1.39 (95% confidence interval: 0.84 – 2.40) among not frail individuals (P for interaction = 0.67)

Conclusion:

Compared with no anticoagulation, anticoagulation is associated with more major bleeding episodes and death in older patients with AF who are cognitively impaired.

Atrial fibrillation (AF) is the most common heart rhythm disorder in older individuals and is a major risk factor for stroke. With the aging of the American population, the burden of AF is projected to further increase[1].

Anticoagulation is the cornerstone of stroke prevention in AF. Stroke risk in AF increases with age. Based on CHADSVASC score, age > 65 is adequate in itself for consideration of anticoagulation[2]. Therefore, the challenge in anticoagulation prescription in older individuals with AF lies mostly in weighing the risk of a major or clinically relevant bleeding over a potentially catastrophic stroke. However, the clinical utility of existing bleeding risk prediction tools is limited because their performance is modest[3] and because many variables in these prediction tools overlap with the those in the CHADSVASC score[4].

Cognitive impairment and frailty are also common in older individuals with AF[5]. We have reported that in older individuals with AF who are prescribed anticoagulants, frailty and cognitive impairment are associated with increased risk of bleeding and death[6]. In this report, we additionally include individuals not prescribed anticoagulation and examined whether the therapeutic harm associated with anticoagulation differs by cognitive and functional status in older individuals with AF.

Methods

Study sample

The details of the SAGE (Systematic Assessment of Geriatric Elements) - AF study have been previously described[5]. The inclusion criteria were: 1) age 65 years or older, 2) ambulatory visit at one of four Central Massachusetts practices (University of Massachusetts Memorial Health Care internal medicine, cardiology, or electrophysiology, Heart Rhythm Associates of Central Massachusetts), one practice in Eastern Massachusetts (Boston University cardiology), or two practices in Central Georgia (Family Health Center and Georgia Arrhythmia Consultants), 3) AF present on an electrocardiogram or Holter monitor or noted in any clinic note or hospital record, and 4) a CHA2DS2VASC[2] risk score ≥ 2. Participants were not eligible if they had an absolute contraindication to oral anticoagulation, had an indication for oral anticoagulation other than AF (i.e., mechanical heart valve), could not provide informed consent, did not speak English, had a planned invasive high bleeding risk procedure, were incarcerated, or were unwilling or unable to participate in planned follow-up visits. The enrollment flow chart is shown in Supplementary Figure S1. The length of follow up was 2.0 (±0.4) years.

All participants provided informed written consent. Study protocols were approved by the Institutional Review Boards at the University of Massachusetts Medical School, Boston University, and Mercer University.

Data Abstraction and assessment of geriatric conditions

Socio-demographic information, clinical data, and relevant laboratory results were abstracted from the medical record by trained staff with appropriate quality control measures.

At baseline, frailty was assessed by the Fried frailty scale[7]. Its components include weight loss/shrinking, exhaustion, low physical activity, slow gait speed, and weakness. Each component receives one point and the scale ranges from 0–5 (0: not frail, 1–2: pre-frail, 3 or more: frail).

The Montreal Cognitive Assessment Battery (MoCA)[8], a 30-item screening tool designed to detect mild cognitive impairment, was used to assess patients’ cognitive status. Lower scores indicate poorer cognitive function, with a score < 23 indicating cognitive impairment[9]. Instead of the traditional cutoff of < 26 in MoCA, we chose the cutoff at <23 because it was previously shown to correspond to the Mini-Mental State Examination (MMSE) score[9] cutoff for mild cognitive impairment.

Data abstraction and assessment of geriatric conditions was performed from 2016 to 2018.

Outcome assessment

The primary outcome was the composite endpoint of major bleeding or death. Secondary outcomes included: 1) death; 2) major bleeding; 3) clinically relevant bleeding; and 4) composite of death, major bleeding, and stroke.

Clinical outcomes during the 2-year follow-up were adjudicated by a committee of physicians from the medical records and death certificates. Bleeding events were graded according to the International Society on Thrombosis and Hemostasis scale[10].

Major bleeding included fatal bleeding, symptomatic bleeding in a critical area or organ (e.g., intracranial, spinal, ocular, pericardial, articular, retroperitoneal, or intramuscular with compartment syndrome), or bleeding that resulted in a fall in hemoglobin of 2 g/dL, or transfusion of ≥2 units of whole blood.

Clinically relevant bleeding included major bleeding and cases of overt bleeding not meeting our major criteria, but requiring medical intervention, unscheduled contact with a physician, temporary interruption of anticoagulation, pain, or impairment of daily function. For individuals with recurrent clinically relevant bleeding or major bleeding, the first event was used in analyses.

Medical records were reviewed on a scheduled basis for 2 years post enrollment for events and deaths for all enrolled subjects allowing complete follow-up of our cohort for these outcomes.

Statistical Analysis

Baseline characteristics were compared according to whether the participants were taking anticoagulants using χ2, analysis of variance, and t tests for discrete and continuous variables, respectively. The distributions of the variables were deemed appropriate for the above tests.

To address the potential confounding by treatment indication (anticoagulation versus no), a propensity score for receiving anticoagulation was calculated for each participant. Variables in the propensity score included age, race, CHADS-VASC score, HAS-BLED score, Charlson Co-Morbidity Index[11], type of AF (paroxysmal, persistent, or permanent), fall in the past 6 months, history of myocardial infarction, history of peripheral vascular disease, history of stroke, history of liver disease, estimated glomerular filtration rate, dual antiplatelet therapy, total medication count, provider type (internist, cardiologist, or electrophysiologist), and site. The variables were selected based upon clinical relevance and statistical significance in Table 1.

Table 1.

Participant characteristics according to the receipt of oral anticoagulants

Characteristic		Anticoagulant Use		P-value
		No	Yes
		(n=180)	(n=1064)
Age (M, SD) *		74.4 (7.1)	75.7 (7.1)	0.02
65–74 years		107 (55%)	519 (49%)	0.03
75–84 years		53 (29%)	401 (38%)
85 years or older		20 (11%)	144 (14%)
Female		82 (46%)	525 (49%)	0.35
Non-Hispanic White*		154 (86%)	902 (85%)	0.66
College graduate or above		89 (52%)	438 (43%)	0.02
CHADSVASC score (M, SD) *		4.0 (1.6)	4.5 (1.6)	<0.0001
HAS-BLED score (M, SD) *		3.2 (1.2)	3.3 (1.1)	0.37
Charlson Co-Morbidity Index (M, SD) *		5.6 (2.8)	6.1 (2.5)	0.02
Atrial fibrillation type*	Paroxysmal	141 (78%)	600 (56%)	<0.0001
	Persistent/Long Standing	12 (7%)	297 (28%)
	Permanent	4 (2%)	69 (7%)
Fall in Past 6 Months*		36 (20%)	234 (22%)	0.55
Medical history
Major bleeding		35 (19%)	209 (20%)	0.95
Intracranial hemorrhage		2 (6%)	12 (6%)	0.99
Gastrointestinal bleed		24 (69%)	123 (59%)	0.27
Bleeding requiring transfusion		7 (20%)	36 (17%)	0.71
Heart failure		53 (29%)	410 (39%)	0.02
Myocardial infarction*		30 (17%)	212 (20%)	0.30
Coronary artery disease		47 (26%)	301 (28%)	0.54
Peripheral vascular disease*		29 (16%)	150 (14%)	0.48
Hypertension		155 (86%)	967 (91%)	0.06
Diabetes		40 (22%)	306 (29%)	0.07
Hyperlipidemia		147 (82%)	849 (80%)	0.56
Stroke*		14 (8%)	108 (10%)	0.31
Alcohol use		51 (28%)	333 (31%)	0.42
Anemia		57 (32%)	334 (31%)	0.94
Chronic lung disease		48 (27%)	268 (25%)	0.67
Liver disease*		7 (4%)	24 (3%)	0.22
Renal disease		46 (26%)	310 (29%)	0.32
Implantable cardiac device		55 (31%)	366 (34%)	0.31
Creatinine (milligram per deciliter)		1.07 (0.54)	1.10 (0.55)	0.43
Hemoglobin (gram per deciliter)		13.0 (1.8)	13.1 (1.9)	0.54
Estimated GFR*		73.2 (22.0)	70.1 (22.3)	0.13
Platelet (x 109 per liter)		217.4 (76.1)	209.0 (69.3)	0.22
Systolic blood pressure (mmHg)		133.0 (17.4)	131.1 (19.8)	0.19
Aspirin		134 (74%)	308 (29%)	<0.0001
Dual Antiplatelet Therapy*		13 (7%)	15 (1%)	<0.0001
Total Medication Count (M, SD) *		10.7 (4.4)	11.6 (4.7)	0.03
Provider Type*	Internist	6 (3%)	24 (2%)	0.04
	Cardiologist	99 (55%)	488 (46%)
	Electrophysiologist	75 (42%)	552 (52%)
Site*	Massachusetts	148 (82%)	818 (85%)	0.10
	Georgia	32 (18%)	246 (23%)
Frailty category	Robust	65 (36%)	348 (33%)	0.66
	Pre-frail	92 (51%)	567 (53%)
	Frail	23 (13%)	149 (14%)
Cognitive impairment		78 (43%)	447 (42%)	0.74
MoCA score		23.5 (4.1)	23.6 (4.1)	0.83

M: mean; SD: standard deviation; GFR: glomerular filtration rate; mmHg: millimeter of mercury

^*denotes factors included in propensity score calculation.

MoCA: Montreal Cognitive Assessment

Survival function curves adjusted for the propensity to receiving anticoagulation were plotted according to cognitive impairment and frailty.

Cox regression analysis was used to calculate the hazard ratios for our primary and secondary study outcomes. The hazard ratios were adjusted by the propensity score and presented in forest plots. Interaction terms of anticoagulation and frailty and of anticoagulation and cognitive impairment were tested. For the major and clinically relevant bleeding and mortality outcomes, the numbers needed to harm of anticoagulation were calculated, stratified by status of cognitive impairment and frailty. For strokes, the number needed to treat of anticoagulation was calculated. To account for the differences in those receiving anticoagulation in numbers needed to harm and treat, Cox regression models adjusted for propensity score of receiving anticoagulation were used to determine each subjects’ predicted probability of bleeding and mortality outcomes by cognitive impairment and frailty. The predicted probability was used to calculate the relative risk and risk difference included in the calculation of number needed to harm and treat[12].

Statistical analyses were performed using SAS 9.4 (SAS Institute, Inc., Cary, NC). A two-sided p value < 0.05 was considered statistically significant.

Results

Study population

Between 2016 and 2018, a total of 1,244 participants were enrolled and included in the analyses. The average age was 75.5 (±7.1) years old and 49% were women. The mean CHA2DS2VASC score was 4.4 (±1.6) and HAS-BLED score was 3.2 (±1.1).

Anticoagulants were prescribed to 1064 (86%) participants prior to study entry, of whom 598 (56%) were treated with a vitamin K antagonist, and the remaining 466 patients were treated with direct oral anticoagulants (DOACs). Compared to participants not receiving anticoagulants (n=180), participants who were treated with anticoagulants were older, less educated, and less likely to have paroxysmal atrial fibrillation. They also had higher CHA2DS2VASC score. No significant difference was observed in HAS-BLED score or history of bleeding. Their Charlson Co-morbidity score was higher, and they were more likely to have heart failure. In addition, participants receiving anticoagulants were less likely to receive dual antiplatelet therapy (Table 1).

The median MoCA score was 24 (interquartile interval: 21–26, range 11–30) and the mean was 23.6 (standard deviation 4.1). The median Fried frailty scale was 1.0 (interquartile interval: 0–2), and the mean was 1.2 (standard deviation 1.1).

During an average 2-year follow up (standard deviation 0.4 years), there were 108 (8.7%) deaths, 105 (8.4%) major bleeding events, and 319 (25.6%) clinically relevant bleeding events. The primary outcome was the composite endpoint of major bleeding or death which occurred in 191 (15.4%) participants while a total of 19 (1.5%) developed a stroke. Bleeding events by dosing of direct oral anticoagulant and warfarin time in therapeutic range were presented in Supplementary Table S1. Number of events of clinical outcomes by cognition and frailty were in Supplementary Table S2.

Anticoagulation and cognitive impairment

At baseline, 528 (42.4%) participants were cognitively impaired. After adjustment for the propensity score to receiving anticoagulation, anticoagulation was prospectively associated with a higher risk of developing the composite endpoint of major bleeding or death among individuals with cognitive impairment (p =0.03) but not among individuals without cognitive impairment (p=0.80) (Figure 1).

Figure 1.

Absence of major bleeding or death, by anticoagulation status, stratified by cognitive function. The survival function was adjusted by the propensity score to receiving oral anticoagulation (OAC).

The adjusted hazard ratios of anticoagulation for the composite endpoint of major bleeding or death were 2.23 (95% confidence interval: 1.08 – 4.61) among individuals with cognitive impairment and 0.94 (95% confidence interval: 0.49 – 1.79) among those without cognitive impairment. The p value of the interaction term between coagulation and cognitive impairment on outcome was 0.08.

The associations with respect to our pre-specified secondary outcomes are shown in Figure 2. Anticoagulation was associated with significantly more clinically relevant bleeding and the composite of clinically relevant bleeding or death among cognitively impaired individuals but not among cognitively intact individuals. The interaction term of anticoagulation and cognitive function was statistically significant (p= 0.03) for clinically relevant bleeding.

Figure 2.

Hazard ratios of anticoagulation for bleeding, death, and stroke among older patients with atrial fibrillation by cognitive function. Hazard ratios were adjusted for the propensity score to receive oral anticoagulation. . The reference group was patients who did not receive anticoagulation.

Among patients without cognitive impairment, the adjusted hazard ratio of DOAC compared with warfarin for major bleeding and death was 0.60, 95% confidence interval (0.35–1.02). For patients with cognitive impairment, the adjusted hazard ratio was 1.17, 95% confidence interval (0.78–1.76), p for interaction = 0.05. (Supplementary Tables S3 and S4).

Frailty and anticoagulation

At baseline, 172 (13.8%) participants were frail. The association between the receipt of anticoagulation and the adjusted survival from the composite endpoint of major bleeding or death did not differ by anticoagulation status in either frail or non-frail participants. (Supplementary Figure S2). The adjusted hazard ratios of anticoagulation for the primary and secondary outcomes stratified by frailty status are presented in Figure 3. None were statistically significant. There was not a statistically significant interaction of anticoagulation and frailty on outcome.

Figure 3.

Hazard ratios of anticoagulation for bleeding, death, and stroke among older patients with atrial fibrillation by frailty status. Hazard ratios were adjusted for the propensity score to receive oral anticoagulation. The reference group was patients who did not receive anticoagulation.

Among patients without frailty, the adjusted hazard ratio of DOAC compared with warfarin for major bleeding and death was 0.76, 95% confidence interval (0.51–1.12); for patients with frailty, the adjusted hazard ratio was 1.26, 95% confidence interval (0.69–2.30), p for interaction = 0.16 (Supplementary Tables S3 and S4).

Numbers needed to treat and harm

To facilitate the risk benefit discussion of bleeding and stroke prevention in anticoagulation prescription, we calculated the numbers needed to treat and harm for our primary and secondary study outcomes as well as stroke, when possible (Table 2). The numbers were adjusted by the propensity to receiving anticoagulation and stratified by cognitive and frailty status. Overall, the numbers needed to treat for stroke prevention ranged from 30.9 to 42.4.

Table 2.

Risk and benefit of anticoagulation among older patients with atrial fibrillation according to cognition and frailty status

		Cognition		Frailty
		Impaired	Intact	frail	not frail
Number needed to harm	Clinically relevant bleeding or death	5.0	22.0	10.0	16.3
	Major bleeding episode or death	8.4	103.0	10.3	25.3
	Clinically relevant bleeding	7.2	22.8	13.6	27.4
	major bleeding episode	12.8	842.6	15.3	33.0
	death	17.8	93.4	40.4	59.0
	Major bleeding, death, or stroke	10.0	103.5	9.0	36.1
Number needed to treat	stroke	30.9	n/a*	34.9	42.4

Numbers needed to harm and treat were calculated with adjustment for the propensity score to receive the oral anticoagulant therapy.

^*There were too few strokes in the cognitively intact group to calculate the number needed to treat.

For cognitively impaired individuals, the number needed to harm for the composite outcome of a major bleeding episode or death was 8.4, for major bleeding was 12.8, and for death was 17.8. Among cognitively intact individuals, the number needed to harm for the composite of major bleeding or death was 103.0, for major bleeding was 842.6, and for death was 93.4.

The numbers needed to harm for adverse outcomes were lower in frail individuals than in not frail individuals, ranging from 10.0 to 59.0.

Discussion

Among older ambulatory patients with AF, we found that 42% of individuals were cognitively impaired and 14% were frail. After 2 years of follow up, anticoagulation was associated with a greater risk of dying or developing major bleeding in cognitively impaired individuals but not in cognitively intact individuals. Among cognitively impaired individuals, the risk of bleeding from anticoagulation appeared to outweigh its benefit of stroke prevention. Frail individuals also had more bleeding but the interaction between frailty status and anticoagulation did not reach statistical significance.

Anticoagulation is a class I guideline indication for stroke prevention in atrial fibrillation[3]. In the historical trials demonstrating the efficacy of warfarin for stroke prevention in AF, the annual stroke rates were consistently above 2% per year in the warfarin groups[13]. In the pivotal trials establishing DOAC as the first-line anticoagulant therapy for stroke prevention in AF, the annual stroke rates in the DOAC group were 1 to 2%[14]. In contrast, SAGE participants had an annual stroke risk was less than 1% per year despite of higher CHADSVASC score (average 4.4). Meanwhile, the risk of major bleeding (8.7% over 2 years) and all-cause mortality (8.4% over 2 years) were significant. By only referring to the data from clinical trials, clinicians may overestimate the stroke risk and underestimate risk of bleeding and comorbidities, which influences the risk benefit consideration in anticoagulation prescription.

AF is associated with two-to-three-fold increased risk of cognitive impairment and dementia in multiple cross-sectional and prospective studies[15]^,[16]. In our study, the high prevalence of cognitive impairment (42%) was comparable to that in prior studies[17]. Despite the increasing awareness about the link between atrial fibrillation and cognitive impairment, assessment of cognitive function is not included in the AF clinical guidelines. This is partly because it is unclear how the results of assessment of cognitive function should affect clinical management. We recently reported that cognitive impairment was prospectively associated with bleeding and death but not stroke in older individuals with atrial fibrillation already on anticoagulation[6]. The present study additionally includes individuals not taking anticoagulation and elucidates the role of cognitive function in anticoagulation decision making: to prevent 1 stroke, 3.7 patients would suffer major bleeding and 1.7 would die. These findings are useful in informed decision making about anticoagulation therapy for older patients with AF and should encourage the assessment of cognitive function in patients with AF.

The interaction between frailty status and the receipt of anticoagulation did not reach statistical significance. However, the numbers needed to harm were consistently higher with respect to bleeding and death among individuals who were not frail than in their frail counterparts. This is consistent with our previous finding that frailty was prognostic for bleeding and death older patients with AF[6] and suggests a better safety profile of anticoagulation for not frail individuals. Our findings in this propensity- score adjusted analysis do not rule out that frailty may predispose to the above adverse events, and the lack of statistical significance does not preclude this. Our findings suggest that, despite of the wider acknowledgement of frailty than cognitive function as a risk factor for bleeding[3], the differential effect of anticoagulation by frailty was not as strong as by cognitive function. Of note, only 14% of participants were frail in our cohort. Therefore, lack of power could be contributing to lack of statistical significance, reflected by the wide confidence intervals of the hazard ratio estimates.

Limitations

Important limitations deserve attention when interpretating these results. Despite extensive efforts of propensity score matching, confounding by indication for the underlying difference between individuals who did and did not take anticoagulation was likely not eliminated. In addition, given the observational nature of our study, the findings should be considered hypothesis-generating and need to be confirmed in randomized controlled trials focusing on older patients with cognitive impairment. The outcome adjudication committee was not blinded to whether the patient was receiving anticoagulation, which may have introduced bias in adjudicating the clinically relevant non-major bleeding outcome. The numbers needed to harm presented in Table 2 were adjusted by the propensity to receiving anticoagulation. Therefore, they should not be compared to those reported in randomized clinical trials. Instead of the absolute numbers needed to harm in each group, our paper focused on the striking difference in numbers needed to harm between cognitive impaired and intact individuals. Also, less than 20% participants were frail, so the power of frailty analyses was limited. Lastly, the study was underpowered for the outcome of stroke. Additional studies are needed, preferably randomized controlled trials, to better define the risk-benefit ratio of anticoagulation in older patients with cognitive impairment and/or frailty.

Conclusions

Among older patients with AF at high risk for stroke, anticoagulation is associated with a greater risk for developing episodes of major bleeding or death in cognitively impaired individuals but not in cognitively intact individuals. Assessing cognitive function has a role in the benefit risk calculation for anticoagulation prescription.

IMPACT STATEMENT

We certify that this work is novel of recent novel clinical research.

This research adds to the literature by demonstrating that in older patients with atrial fibrillation, anticoagulation was associated with more major bleeding episodes or death among cognitively impaired individuals, but not among cognitively intact individuals. The differential effect of anticoagulation was not seen by frailty status. This provides strong evidence that clinicians should consider cognitive function when prescribing anticoagulants to older patients with atrial fibrillation.